INTRADAY

LIQUDITIY RISK:
INSIGHTS AND
MODELLING
Compiled by:
Gaby Frangieh

Risk Management, Finance and Banking – Senior Advisor

Intraday Liquidity Risk: Insights and
Modelling
Compilation Contents
1. Liquidity-adjusted Intraday Value at Risk modeling and risk
management: An application to data from Deutsche Börse
2. Liquidity Risk Management: Recent Trends
3. Intraday Liquidity Around the World
4. Monitoring Intraday Liquidity Risks in a Real Time Gross Settlement
System
5. Intraday liquidity Presentation: implementation challenges, AI usage
and RTGS evolution
6. Payment throttling: a balancing act for managing intraday liquidity risk
7. ALMA Intraday Liquidity Masterclass Presentation: Learn the dark art
of intraday, how it fits into your broader liquidity management
framework and the impact to your technology strategy
8. Intraday liquidity modelling using statistical methods
9. Intraday liquidity: risk and regulation
10. Estimating the intraday liquidity risk of financial institutions: a Monte
Carlo simulation approach
 Liquidity-adjusted Intraday Value at Risk modeling and risk
management: An application to data from Deutsche Börse☆
Georges Dionnea*, Maria Pacurarb, Xiaozhou Zhouc
aCanada Research Chair in Risk Management, HEC Montréal, Montréal, Canada H3T 2A7
bRowe School of Business, Dalhousie University, Halifax, Canada B3H 4R2
cDépartement de Finance, HEC Montréal, Montréal, Canada H3T 2A7

First version: March 05, 2013

This version: February 28, 2014
Abstract

This paper develops a high-frequency risk measure, the Liquidity-adjusted Intraday Value at Risk
(LIVaR). Our objective is to explicitly consider the endogenous liquidity dimension associated
with order size. Taking liquidity into consideration when using intraday data is important because
significant position changes over very short horizons may have large impacts on stock returns.
By reconstructing the open Limit Order Book (LOB) of Deutsche Börse, the changes of tick-by-
tick ex-ante frictionless return and actual return are modeled jointly using a Log-ACD-VARMA-
MGARCH structure. This modeling helps to identify the dynamics of frictionless and actual
returns, and to quantify the risk related to the liquidity premium. From a practical perspective,
our model can be used not only to identify the impact of ex-ante liquidity risk on total risk, but
also to provide an estimation of VaR for the actual return at a point in time. In particular, there
will be considerable time saved in constructing the risk measure for the waiting cost because
once the models have been identified and estimated, the risk measure over any time horizon can
be obtained by simulation without re-sampling the data and re-estimating the model.

Keywords: Liquidity-adjusted Intraday Value at Risk, Tick-by-tick data, Log-ACD-VARMA-

MGARCH, Ex-ante Liquidity premium, Limit Order Book.
JEL classification: C22, C41, C53, G11

☆
We thank Yann Bilodeau for his help in constructing the dataset and comments. We also thank Diego Amaya,
Tolga Cenesizoglu, Philippe Gregoire, Benoit Perron, Gabriel Yergeau and participants at the 53e congrès annuel de
la Société canadienne de science économique and the 2013 Canadian Economics Association Conference for their
remarks. Georges Dionne and Maria Pacurar acknowledge financial support from the Social Sciences and
Humanities Research Council (SSHRC), Xiaozhou Zhou thanks Fond de Recherche sur la société et la culture de
Québec (FRQSC) and Centre interuniversitaire sur le Risque, les Politiques, Economiques et l’Emploi (CIRPEE) for
financial support.
* Corresponding author. Canada Research Chair in Risk Management, HEC Montréal, Montreal, Canada H3T 2A7.

Tel.: +1 514 340 6596; Email addresses: [email protected] (G. Dionne), [email protected] (M. Pacurar),
[email protected] (X. Zhou)
1. Introduction
With the help of computerization, many main exchanges around the world such as
Euronext, the Tokyo Stock Exchange, Toronto Stock Exchange and the Australian
Stock Exchange organize trading activities under a pure automatic order-driven
structure: there are no designated market-makers during the continuous trading
and the liquidity is fully guaranteed by market participants via an open Limit Order
Book (LOB hereafter). In other main exchange markets including NYSE,
NASDAQ and Frankfurt Stock Exchange, the trading activities are carried out
under the automatic order-driven structure and the traditional floor-based quote-
driven structure. Nevertheless, most trades are executed under the automatic
order-driven structure due to its advantage of transparency, efficiency and
immediacy. Consequently, the frequency of trading is becoming shorter and
trading activity has become easier than ever before.

One type of trading behavior is active trading or day-trading by which traders trade
during the day and liquidate all open positions before market closing. Besides high-
frequency traders, financial institutions also need intraday risk analysis (Gouriéroux
and Jasiak (2010)) for internal control of their trading desks. As a result, an active
trading or day-trading culture requires institutional investors, active individual
traders and even regulators of financial markets to pay more and more attention to
intraday risk management. 1 However, traditional risk management has been
challenged by this trend towards high-frequency trading because the low-frequency
measures of risk such as Value at Risk (VaR), which is usually based on daily data,
struggle to capture the potential liquidity risk hidden in very short horizons.
Typically, the well recognized description of a liquid security is that there is the
ability to convert the desired quantity of the financial asset into cash quickly and with
little impact on the market price (Demsetz (1968); Black (1971); Kyle (1985); Glosten
and Harris (1988)). Four dimensions are implicitly included in this definition: volume

1
High-frequency trading was estimated to make up 51% of equity trades in the U.S. in 2012 and
39% of traded value in the European cash markets (Tabb Group).

1
(significant quantity), price impact (deviation from the best price provided in the
market), time (speed to complete the transaction) and resilience (speed to backfilling).
In an automated order-driven trading system, because the liquidity is fully provided
by the open LOB, we can investigate liquidity risk by monitoring the evolution of
LOB and exploring the corresponding embedded information.

The traditional VaR can be interpreted as estimation of the potential loss on a

predetermined portfolio over a relatively long fixed period, that is, a measure of
price risk. This is not a liquidation value since it does not take into account the
volume dimension but solely a ‘paper value’ for a frozen portfolio. However,
liquidity risk is always present when the transaction is not yet realized.
Furthermore, from a high-frequency market microstructure perspective, the
transaction price is an outcome of information shock, trading environment, market
imperfections and the state of the LOB. For very short horizons, all these
microstructure effects could cause the transaction price to deviate from the
efficient price. Therefore, if we concentrate on the actual liquidation value resulting
from an active trading operation, the traditional VaR may be characterized by a
serious omission of liquidity, especially when the liquidation quantity is large. To
make the VaR measure more accurate in evaluating the liquidation value, one
should include an additional dimension of risk, the ex-ante liquidity risk. From a
practical perspective, this introduction of a liquidity dimension can offer a more
accurate risk measure for intraday or active traders and market regulators who aim
to closely monitor the total risk of the market.

Few studies have focused on high-frequency risk measures. Dionne, Duchesne and
Pacurar (2009) are the first to consider a ultra-high-frequency market risk measure,
Intraday Value at Risk (IVaR) based on all transactions. In their study, the
informative content of trading frequency is taken into account by modeling the
durations between two consecutive transactions. One important practical
contribution is that, instead of being restricted to traditional one- or five-minute
horizons, their model allows for computing the IVaR measure for any horizon.

2
The authors found that ignoring the effect of durations can underestimate risk.
However, as they noted, similar to other VaR measures, the IVaR ignores the
ex-ante liquidity dimension by only taking into account information about
transaction prices.

On the other hand, in spite of the rapid development of electronic trading

platforms, high-frequency risk measures hardly incorporate the endogenous
relationship between open LOB dynamics and frictionless prices, although some
recent theoretical and empirical works find that the LOB is informative about the
volatility of efficient price (Pascual and Veredas (2010)), and that the
volume-volatility relationship is related negatively to the LOB slope (Næs and
Skjeltorp (2006)). One exception is Giot and Grammig (2006) who consider the
ex-ante liquidity provided by the LOB using data from Xetra, an automated
auction system. They focus on the ex-ante liquidity risk faced by an impatient
trader acting as a liquidity demander by submitting a market order. The quantity to
liquidate is fictive and the choice of quantity is motivated by descriptive statistics
and trading statistics of the underlying stock. The use of open LOB data allows
them to construct an actual return that is a potential implicit return for a
predetermined volume to trade over a fixed time horizon of 10 or 30 minutes. The
ex-ante liquidity risk is then quantified by comparing the standard VaR based on
frictionless return, i.e. mid-quote return, and the liquidity-adjusted VaR inferred
from the actual return. Since the proposed VaR is computed on data from
regularly spaced intervals of 10 or 30 minutes and the snapshot of the LOB is
taken at the beginning and end of the interval, all the information within the
interval is omitted.

The development of liquidity-adjusted risk measures goes back to Bangia et al.

(1999) who first consider the liquidity risk in VaR computation. Their liquidity risk
is measured by the half spread, which is furthermore assumed to be uncorrelated
with market risk. Actually, the total risk they attempt to identify is the sum of
market risk and trading cost associated to only one share since their liquidity risk

3
measure does not consider the volume dimension. In line with Bangia et al. (1999),
Angelidis and Benos (2006) estimate the liquidity-adjusted VaR by using data from
the Athens Stock Exchange. They find that liquidity risk measured by the bid-ask
spread accounts for 3.4% for high-capitalization stocks and 11% for low-
capitalization stocks. Using the same framework as Bangia et al. (1999), Weiβ and
Supper (2013) address liquidity risk of a five-NASDAQ stock portfolio by
estimating the multivariate distribution of both log-return and spread using the
vine copulas to account for the dependence between the two across firms over a
regular interval of 5 minutes. They evidence strong extreme comovements in
liquidity and tail dependence between bid-ask spreads and log-returns across the
selected stocks.

Our paper is related to the market microstructure literature that investigates the
role of high-frequency liquidity risk. However, our study differs from previous
papers in several ways. First, we examine the ex-ante liquidity risk by focusing on
the tick-by-tick frictionless return and actual returns derived from open LOB. By
ex-ante we mean that the transaction does not really occur but we can still obtain a
potential transaction price issued from a predetermined volume to trade if we have
information from the reconstructed LOB. Most papers in the literature are
interested in ex-post liquidity, which is already consumed by the market or
marketable orders when the trades complete. Compared to ex-post liquidity,
ex-ante liquidity is more informative and relevant as it measures the unconsumed
liquidity in LOB. Besides, the existing literature that analyzes ex-ante measures of
liquidity (e.g., Giot and Grammig, 2006) is based on a regular time-interval and
thus ignores the information during the intervals. In our study, we evidence that
the durations between two consecutive observations do have a positive relation
with volatilities of actual returns and frictionless returns.

Second, our study addresses the questions of what is the relationship between
liquidity risk and market risk and how does the ex-ante liquidity evolve during the
trading day. There are several empirical papers that attempt to identify the

4
proportion of risk associated with liquidity as discussed above. However, to the
best of our knowledge, their identified liquidity is ex-post liquidity without studying
the dynamics of ex-ante liquidity derived from LOB and its relation with market
risk. Nevertheless, one challenge of directly modeling frictionless returns and
actual returns (issued from the best bid/ask price and potential liquidation price,
respectively) is that they are not time-additive. Therefore, in this paper, we model
the frictionless return changes and actual return changes using an econometric
system characterized by the Logarithmic Autoregressive Conditional Duration,
Vector Autoregressive Moving Average and Multivariate GARCH processes
(denoted by Log-ACD, VARMA and M-GARCH hereafter). The structure will not
only capture the joint dynamics of both frictionless return changes and actual
return changes, but also quantify the impact of ex-ante liquidity risk on total risk by
further defining IVaRc and LIVaRc as the VaRs on frictionless return changes and
actual return changes, respectively. In order to make the model more flexible, we
allow for the time-varying correlation of volatility of the frictionless return and
actual return.

Third, from a practical perspective, our proposed risk measure aims at providing a
global view of ex-ante liquidity that can help high-frequency traders develop their
timing strategies during a particular trading day. Our model is first estimated on
deseasonalized data and then validated on both simulated deseasonalized and
re-seasonalized data. The time series of re-seasonalized data is constructed by
re-introducing the deterministic seasonality factors. One advantage of simulated
re-seasonalized data is that risk management can be conducted under calendar
time. In addition, as the model is estimated using tick-by-tick observations and
takes into account the durations between two consecutive transactions,
practitioners can construct the risk measure for any desired time horizon.

The rest of the paper is organized as follows: Section 2 describes the Xetra trading
system and the dataset we utilize. Section 3 briefly presents the procedure used to
test the model and compute Impact coefficient of ex-ante liquidity risk and Ex-

5
ante liquidity premium. Section 4 defines the actual return, frictionless return, IVaR
and LIVaR. In Section 5, we present the econometric model used to capture the
dynamics of duration, frictionless return changes, actual return changes and their
correlation. Section 6 applies the econometric model proposed in Section 5 to data
for stocks RWE AG, Merck and SAP and reports the estimation results. The
model performance is assessed by the Unconditional Coverage test of Kupiec
(1995) and the Independence test proposed by Christoffersen (1998). In Section 7,
we discuss the ex-ante liquidity risk in LIVaRc and compare the proposed LIVaR
with other high-frequency risk measures. Section 8 concludes and provides
possible new research directions.

2. Xetra trading system and dataset

As mentioned above, electronic trading systems have gained popularity in many

stock exchanges during the two last decades. The present study uses data from the
automated order-driven trading system Xetra, which is operated by Deutsche
Börse at Frankfurt Stock Exchange (FSE) and has a similar structure to Integrated
Single Book of NASDAQ and Super Dot of NYSE. It is the main trading platform
in Germany and realizes more than 90% of total transactions at German
exchanges. The trading and order processing (entry, revision, execution and
cancellation) of the Xetra system are highly computerized and maintained by the
German Stock Exchange. Since September 20, 1999, trading hours have been from
9h00 to 17h30 CET (Central European Time). However, during the pre- and post-
trading hours, operations such as entry, revision and cancellation are still
permitted.

To ensure trading efficiency, Xetra operates with different market models that
define order matching, price determination, transparency, etc. One of the
important parameters of a market model is the trading model that determines
whether the trading is organized in a continuous or discrete way or both. During
normal trading hours, there are two types of trading mechanisms: call auction and

6
continuous auction. Call auction could occur one or several times during the
trading day in which the clearance price is determined by the state of LOB and
remains as the open price for the following continuous auction. Furthermore, at
each call auction, market participants can submit both round-lot and odd-lot
orders, and both start and end time for a call auction are randomly chosen by a
computer to avoid scheduled trading. For the stocks in DAX30 index, there are
three auctions during a trading day—the open, mid-day and closing auctions. The
mid-day auction starts at 13h00 and lasts around 2 minutes. Between the call
auctions, the market is organized as a continuous auction where traders can only
submit round-lot-sized limit orders or market orders.

For blue chip and other highly liquid stocks, during the continuous trading there
are no dedicated market makers like the traditional NYSE specialists. Therefore,
the liquidity comes from all market participants who submit limit orders in LOB.
In the Xetra trading system, most of the market models impose the
Price-Time-Priority condition where the electronic trading system places the
incoming order after checking the price and timestamps of all available limit orders
in LOB. Our database includes up to 20 levels of LOB information except the
hidden part of an iceberg order, which means that by observing the LOB, any
trader and registered member can monitor the dynamic of liquidity supply and
potential price impact caused by a market or marketable limit order. However, all
the trading and order submission are anonymous, that is, the state and the updates
of LOB can be observed but there is no information on the identities of market
participants.

The raw dataset that we have access to contains all the events that are tracked and
sent through the so-called data streams. There are two main types of streams: delta
and snapshot. The former tracks all the possible updates in LOB such as entry,
revision, cancellation and expiration. Traders can be connected to the delta stream
during the trading hours to receive the latest information, whereas the second aims
at giving an overview of the state of LOB and is sent after a constant time interval

7
for a given stock. Xetra original data with delta and snapshot messages are first
processed using the software XetraParser developed by Bilodeau (2013) in order to
make Deutsche Börse Xetra raw data usable for academic and professional
purposes. XetraParser reconstructs the real-time order book sequence including all
the information for both auctions and continuous trading by implementing the
Xetra trading protocol and Enhanced Broadcast.2 We further convert the raw LOB
information into a readable LOB for each update time and then retrieve useful and
accurate information about the state of LOB and the precise timestamp for order
modifications and transactions during the continuous trading. Inter-trade durations
as well as LOB update durations are irregular. The stocks SAP (SAP), RWE AG
(RWE) and Merck (MRK) that we choose for this study are blue chip stocks from
the DAX30 index. SAP is a leading multinational software corporation with a
market capitalization of 33.84 billion Euros in 2010. RWE generates and
distributes electricity to various customers including municipal, industrial,
commercial and residential customers. The company produces natural gas and oil,
mines coal and delivers and distributes gas. In 2010, its market capitalization was
around 15 billion Euros. Merck is the world’s oldest operating chemical and
pharmaceutical company with a market capitalization of 4 billion Euros in 2010.

3. Procedure used for computing the risk measures

To compute the proposed risk measures, the model will be first estimated using
deseasonalized data and then the tests will be carried out on both deseasonalized
and seasonalized data. We present the flowchart where we illustrate the steps to
follow. More precisely, our study proceeds as follows:

[Insert Flowchart here]

2
See SolutionXetra Release 11.0 – Enhanced Broadcast solution and Interface specification for a detailed
description.

8
(a) We first compute the raw tick-by-tick durations defined as the time interval
between two consecutive trades and tick-by-tick frictionless returns and actual
returns based on the data of open LOB and trades.
(b) We further compute the frictionless return changes and actual return changes
by taking the first difference of frictionless returns and actual returns,
respectively. This step is required because the frictionless return and actual
return are not time-additive and cannot be modeled directly.
(c) We remove seasonality from durations, frictionless return changes and actual
return changes to obtain the corresponding deseasonalized data.
(d) The deseasonalized data are modeled by the LogACD-VARMA-MGARCH
model.

(e) Once we estimate the model, we simulate the deseasonalized data based on
estimated coefficients and construct VaR measures at different confidence
levels for backtesting on out-of-sample deseasonalized data.
(f) The seasonal factors are re-introduced into deseasonalized data to generate the
re-seasonalized data.
(g) As done in (e), we construct different quantiles for backtesting on out-of-
sample seasonalized data.
(h) We construct Impact coefficients of ex-ante liquidity risk based on the
simulated re-seasonalized data.
(i) We further compute the IVaR and LIVaR, which are defined as the VaR for
frictionless return and actual return, respectively.

(j) Based on IVaR and LIVaR, we can finally compute the ex-ante liquidity
premium by taking the ratio of the difference between IVaR and LIVaR over
LIVaR.

The following sections will explain each step in detail.

4. Frictionless return, actual return and the corresponding high-frequency VaRs

We take into account all information available in the continuous auction by

modeling tick-by-tick data. The first characteristic in tick-by-tick data modeling is
that the durations between two consecutive transactions are irregularly spaced.

9
Consider two consecutive trades that arrive at ti −1 and ti , and define duri as the
duration from ti −1 to ti . Based on this point process, we can further construct two
return processes. One is frictionless return and the other is actual return. More
specifically, the frictionless return is defined as the log ratio of best bid price, bi (1)
at moment i and previous best ask price, ai −1 (1) . The frictionless return is an ex-
ante return indicating the tick-by-tick return for selling only one unit of stock.

bi (1)
RiF = ln( ) (1)
ai −1 (1)

The actual return is defined as the log ratio of selling price for a volume v and
previous best ask price.3

bi (v)
RiB = ln( )
ai −1 (1)

K −1

∑b v + bK ,i vK ,i
k ,i k ,i K −1
where bi (v) = k =1
and vK ,i = v - ∑ vk ,i (2)
v k =1

bk ,i and vk ,i are the kth level bid price and volume available, respectively. vK ,i is the

quantity left after K-1 levels are completely consumed by v. The consideration of
quantity available in LOB is in line with other ex-ante liquidity measures in market
microstructure literature (Irvine, Benston and Kandel (2000); Domowitz, Hansch
and Wang (2005); Coppejans, Domowitz and Madhavan (2004), among others).
The choice of volume v is motivated by transaction volume and volume available
in LOB. Explicitly, for each stock we first compute the cumulative volume
available over the 20 levels at each transaction moment for the bid side of LOB,
and then we choose the minimum cumulative volume as the maximum volume to
construct the actual returns. By doing so, we avoid the situation where the actual
price does not exist for a given volume. One concern that may arise involves the
iceberg orders, which keep a portion of quantity invisible to the market
3 The frictionless return and actual return from the point of view of buyers can be defined similarly.

10
participants. In this study, we assume that the liquidity risk is faced by an impatient
trader and the possibility of trading against an iceberg order will not influence his
trading behavior. Furthermore, as noted by Beltran-Lopez, Giot and Grammig
(2009), the hidden part of the book does not carry economically significant
informational content. The difference between frictionless return and actual return
is that actual return takes into account the desired transaction volume, which is
essential for the liquidity measure. Intuitively, the actual return measures the ex-
ante return when liquidating v units of shares.

One characteristic of our defined frictionless returns and actual returns is that they
do not possess the time-additivity property as traditional log-returns. To
circumvent this difficulty, we model the frictionless return changes and actual
return changes instead of modeling the actual return and frictionless return
directly. More specifically, let r=
i
f
RiF − RiF−1 and r=
i
b
RiB − RiB−1 be the tick-by-tick

frictionless return changes and actual return changes. Following this setup, the L-
step forward frictionless return and actual return can be expressed as follows:

L
bi + L (1) b (1)
RiF+ L = ln( ) = ln( i ) + ri+f1 + ri+f2 + ...ri+fL = RiF + ∑ ri+fm (3)
ai + L −1 (1) ai −1 (1) m =1

L
bi + L (v) bi (v)
R B
i+ L = ln( ) = ln( ) + ri+1 + ri+2 + ...ri+L = Ri + ∑ rib+m
b b b B
(4)
ai + L −1 (1) ai −1 (1) m =1

L L
The terms ∑r
m =1
f
i+m
and ∑r
m =1
b
i+m
are the sum of all tick-by-tick changes in return over a

predetermined interval and can be considered as the waiting cost related to

4
frictionless returns and actual return. More specifically, they measure the
costs/gains associated with the latter instead of immediate liquidation of 1 share
and v shares of stock, respectively. Moreover, if we define the IVaRc and LIVaRc as
L L
the VaR for ∑ ri+fm
m =1
and ∑r
m =1
b
i+m
, respectively and compute them from a

4 The waiting cost could be positive or negative, which indicates loss and gain, respectively.

11
predetermined time interval, then the IVaRc and LIVaRc will provide the possible
loss over a given interval for an investor that trades at frictionless or actual return.
In other words, the IVaRc and LIVaRc will estimate the possible loss in terms of
frictionless return and actual return, which are related to market risk and total risk
(market risk and ex-ante liquidity risk), respectively. Mathematically, consider a
realization of a sequence of intervals with length int and let yint,t
f b 5
and yint,t be the
sum of tick-by-tick changes of returns ri f and rib over t-th interval,

τ ( t ) −1 τ ( t ) −1
y f
int,t ∑ r ; y
=
τ ( t −1)
j=
j
bf
int ,t ∑
τ
j=
( t −1)
rjb (5)

where τ (t ) is the index for which the cumulative duration exceeds t-th interval with
length int for the first time. By definition

τ ( t ) −1 τ (t )
int ≥ ∑
τ ( t −1)
j=
j dur and int ≤ ∑
τ ( t −1)
j=
jdur . (6)

The process of duration allows aggregating the tick-by-tick data to construct the
dynamic of frictionless return and actual return for a predetermined interval that
will allow for consideration of risk in calendar time. Accordingly, the IVaRc and
LIVaRc 6 for frictionless return changes and actual return changes with confidence
level 1 − α for a predetermined interval int are defined as

f
Pr ( yint,t < IVaRint,t
c
(α ) I t ) =
α

,t < LIVaRint,v,t (α ) I t ) =
α
b c
Pr ( yint

5
yintf and yint
b
are defined on both seasonalized return changes and deseasonalized return changes.
6 IVaRc and LIVaRc are also defined on both seasonalized return changes and deseasonalized return changes.
Moreover, our IVaRc and LIVaRc can also be used in the strategy of short selling where IVaRc and LIVaRc will be
the 1-α quantiles of the distributions.

12
I t is the information set until moment τ (t − 1) . Similar to the traditional definition

of VaR, IVaRint,t
c
(α ) and LIVaRint,v,t
c
(α ) are the conditional α-quantiles for yint,t
f
and

,t .
b
yint

We can further define the IVaR and LIVaR as the VaR for the frictionless return
and actual return as following:

=
IVaRint,t RτF(t −1) + IVaRint,t
c

=
LIVaRint,v,t RτB(t −1) + LIVaRint,v,t
c

where RτF(t −1) and RτB(t −1) are the frictionless return and actual return at the
beginning of the t-th interval. Consequently, IVaR and LIVaR estimate the α-
quantiles for frictionless return and actual return at the end of the t-th interval.

5. Methodology

In our tick-by-tick modeling, there are three random processes: the duration, the
changes of frictionless return, and the changes of actual return. The present study
assumes that the duration evolution is strongly exogenous but has an impact on
the volatility of frictionless and actual return changes. The joint distribution of
duration, frictionless return change and actual return change can be decomposed
into the marginal distribution of duration and joint distribution of frictionless and
actual return change conditional on duration. More specifically, the joint
distribution of the three variables is:

f d , f ,b (duri , ri f , rib duri −1 , ri −f1 , rib−1 ; Θ f )

= f d (duri duri −1 , ri −f1 , rib−1 ;θ d ) f f ,b
(ri f , rib duri , duri −1 , ri −f1 , rib−1 ;θ d , θ f ,b ) (7)

where f d , f ,b is the joint distribution for duration, frictionless return change and
actual return change. f d (⋅) is the marginal density for duration and f f ,b
(⋅) is the

13
joint density for actual and frictionless return changes. Consequently, the
corresponding log-likelihood function for each joint distribution can be written as:

n
L(θ d , θ f ,b ) ∑ log f
i =1
d
(duri duri −1 , ri −f1 , rib−1 ;θ d ) + log f f ,b
(ri f , rib duri −1 , ri −f1 , rib−1 , duri ;θ f ,b ) (8)

In the next subsections, we specify marginal density for dynamics of duration and
joint density for frictionless return and actual return changes. We present the
model for deseasonalized duration, frictionless and actual return changes. The
deseasonalization procedure is described in detail in Section 6.

5.1 Model for duration

The ACD model used to model the duration between two consecutive transactions
was introduced by Engle and Russell (1998). The GARCH-style structure is
introduced to capture the duration clustering observed in high-frequency financial
data. The basic assumption is that the realized duration is driven by its conditional
duration and a positive random variable as error term. Let ψ i = E (duri I i −1 ) be the
expected duration given all the information up to i-1, and ε i be the positive
random variable. The duration can be expressed as: dur=i ψ i ⋅ ε i . There are several
possible specifications for the expected duration and the independent and
identically distributed (i.i.d.) positive random error (see Hautsch (2004) and
Pacurar (2008) for surveys). In order to guarantee the positivity of duration, we
adopt the log-ACD model proposed by Bauwens and Giot (2000). The
specification for expected duration is

 p q

ψi =exp  ω + ∑ α j ε i − j + ∑ β j lnψ i − j  . (9)
=  j 1 =j 1 

For positive random errors, we use the generalized gamma distribution, which
allows a non-monotonic hazard function and nests the Weibull distribution
(Grammig and Maurer (2000); Zhang, Russell and Tsay (2001)):

14
 γ1
 γ ε γ1γ 2 −1  εi 
 1 i
exp[−   ], ε i > 0,
p (ε i γ 1 , γ 2 ) =  γ 3γ1γ 2 Γ(γ 2 )  γ3  (10)

0 , otherwise.

1
where γ 1 , γ 2 > 0 , Γ(.) is the gamma function, and γ 3 =
Γ(γ 2 ) / Γ(γ 2 + ) .
γ1

5.2 Model for frictionless return and actual return changes

The high-frequency frictionless return and actual return changes display a high
serial correlation. To capture this microstructure effect, we follow Ghysels and
Jasiak (1998) and adopt a VARMA (p,q) structure:

(=
r , r ) , E (e , e )
' '
=Ri i
b
i
f
i
b
i i
f

p q

=
Ri =
m 1=n 1
∑ Φ m Ri −m + Ei − ∑ ∆ n Ei −n (11)

where Φ m and ∆ n are matrix of coefficients for Ri −m and Ei −n , respectively. As

mentioned in Dufour and Pelletier (2011), we cannot directly work with the
representation in (11) because of an identification problem. Consequently, we
impose the restrictions on ∆ n by supposing that the VARMA representation is in
 δ11( n ) 0 
diagonal MA form. More specifically, ∆ n = (n) 
where δ11( n ) and δ 22( n ) are the
 0 δ 22 

coefficients for nth-lag error terms of the actual return change and frictionless
return change respectively.

Furthermore, we assume the volatility part follows a multivariate GARCH process.

 eib 
 f  = H i zi
1/ 2

 ei 

where zi is the bivariate normal distribution that has the following two moments:
= Var ( zi ) I 2 . The normality assumption for the error term is supported by
E ( zi ) 0,=

backtesting. The other distributions we tried, such as Normal Inverse Gaussian

15
 (NIG), Student-t or Johnson, overestimated the error distribution in our
simulation tests.

H i is the conditional variance matrix for Ri that should be positive definite. To

model the dynamic of H i , we use the DCC structure proposed by Engle (2002) in
which H i is decomposed as follows:

H i = Di Ci Di
where :
γf
=  Di diag ( h11i , =h22i ), h11i σ= 2
duriγ b , h22i σ 222 i duri
 11i


C = (diagQ ) −1/ 2 Q (diagQ ) −1/ 2
 i i i i

ek ,i −1
Qi = (1 − θ1 − θ 2 )Q + θ1ε i −1ε i'−1 + θ 2Qi −1 , ε k ,i −1 = , k = 1, 2 (12)
hkk ,i −1

Qi is the unconditional correlation matrix of {ε i }i =1 , h11i ( h22i ) is the conditional

N

variance for actual (frictionless) return change, and γ b (γ f ) measures the impact of
duration on the volatilities of actual and frictionless return changes, respectively.

A useful feature of the DCC model is that it can be estimated by a two-step

approach. Engle and Sheppard (2001) show that the likelihood of the DCC model
can be written as the sum of two parts: a mean and volatility part, and a correlation
part. Even though the estimators from the two-step estimation are not fully
efficient, the one iteration of a Newton-Raphson algorithm applied to total
likelihood provides asymptotically efficient estimators.

In the DCC framework each series has its own conditional variance. For both
actual and frictionless return changes, we adopt a NGARCH(m,n) process as
proposed by Engle and Ng (1993) to capture the cluster as well as the asymmetry
in volatility. The process can be written as:

16
 γb
h=
11,i σ 11,
2
i ⋅ duri
m n
σ 11, i = ω + ∑ α j σ 11,i − m + ∑ β j σ 11,i − j (ε i −1 − π )
2 b b 2 b 2 b b 2
(13)
=j 1 =j 1

and

γf
h=
22,i σ 22,
2
i ⋅ duri
m n
σ 22,
2
i= ω
f
+ ∑ α jf σ 22,
2
i − m + ∑ β j σ 22,i − j (ε i − j − π )
f 2 f f 2
(14)
=j 1 =j 1

where π and π are used to capture the asymmetry in the conditional volatilities.
b f

When π f or π b = 0 , the model will become a standard GARCH model, whereas a

negative π f or π b indicates that a negative shock will cause higher conditional
volatilities for the next moment.

Our structure also explicitly introduces the duration dimension in conditional

volatilities. The first study of the impact of duration on volatility by Engle (2000)
assumes that the impact is linear, that is h=
11,i σ 11,
2
i ⋅ duri . As suggested in Dionne,

Duchesne and Pacurar (2009), this modeling for the unit of time might be
restrictive for some empirical data for which conditional volatility can depend on
duration in a more complicated way. To make the model more general, we follow
Dionne, Duchesne and Pacurar (2009) by assuming the exponential form
h=
11,i σ 11,2 i ⋅ duriγ b and h=
22,i σ 22,
2 f γ
i ⋅ duri . When
γ f or γ b = 0 , the volatility will become

a standard NGARCH process whereas when γ f or γ b = 1 , it transforms to the

similar model studied in Engle (2000).

As mentioned above, our model has three uncertainties: variation in duration, price
uncertainty and LOB uncertainty. Deriving a closed form of LIVaRc would be
complicated for multi-period forecasting in the presence of three risks, especially
for non-regular time duration. Therefore, once the models are estimated, we follow

17
Christoffersen (2003) and use Monte Carlo simulations to make multi-step
forecasting and to test the model’s performance.

6. Empirical Results

6.1 Seasonality adjustment

It is well known that high-frequency data behave very differently from

low-frequency data. Table 1 presents the descriptive statistics of raw and
deseasonalized duration, frictionless return changes and actual return changes for
which various volumes are chosen for the three studied stocks. From Panel A, we
can observe that for the entire sample period (July 2010), SAP is the most liquid
stock as the average duration is the shortest and the number of observations is the
largest. MRK is the least liquid one. Moreover, as the variables are constructed on
tick-by-tick frequency, all three stocks have an average of zero and a very small
standard deviation for frictionless return changes and actual return changes. All
three stocks present high kurtosis due to the fact that most of the observations are
concentrated on their average and co-exist with some extreme values. In addition,
the raw data are also characterized by extremely high autocorrelation for both first
and second moments for all of the variables.

High-frequency data are characterized by seasonality, which should be cleaned out

before estimating any model. To do so, several approaches have been proposed in
the literature: Andersen and Bollerslev (1997) use Fourier Flexible Functional
(FFF) form to take off seasonality, Dufour and Engle (2000) remove seasonality by
applying a simple linear regression with a dummy, and Bauwens and Giot (2000)
take off seasonality by averaging over a moving window and linear interpolation.

[Insert Table 1 here]

However, as found in Anatolyev and Shakin (2007) and Dionne, Duchesne and
Pacurar (2009), the high-frequency data could behave differently throughout the
day as well as between different trading days. Therefore, in order to fully account

18
 for the deterministic part exhibited in data, we apply a two-step deseasonalisation
procedure, interday and intraday. Besides, it should also be noted that there exists
an open auction effect in our continuous trading dataset similar to the one
mentioned by Engle and Russell (1998). More precisely, for each trading day, the
continuous trading follows the open auction in which specialists set a price in
order to maximize the volume. Once the open auction is finished, the transactions
are recorded. Consequently, the beginning of continuous trading is contaminated
by extremely short durations. In addition, these short durations could produce
negative seasonality factors of duration that are based on previous observations
and cubic splines. To address this problem, the data for the first half hour are only
used to compute the seasonality factor and then discarded.

The interday trend is extracted under a multiplicative form:

duri , s ri ,fs rib, s

duri,inter = ,r f
i,inter = ,r b
i,inter = (15)
durs (rs f ) 2 (rsb ) 2

where duri , s , ri ,fs , rib,s are the ith duration, frictionless and actual return change for

day s , respectively and durs , (rs f )2 , (rsb ) 2 are the daily average for day s for duration,
squared frictionless, and actual return changes, respectively.

Based on inter-day deseasonalized data, the intra-day seasonality is removed by

following Engle and Russell (1998):

duri,inter ri ,finter rib,inter

=duri ,intra = f
, ri,intra = , rib,intra (16)
f
E (duri,inter ) I i −1 E ((ri,inter ) 2 ) I i −1 b
E ((ri,inter ) 2 ) I i −1

where E (duri,inter ) , E ((ri,inter

f 2
) and E ((ri,inter
b
) 2 ) are the corresponding deseasonality

factors constructed by averaging the variables over 30-minute intervals for each
day of the week and then applying cubic splines to smooth these 30-minute
averages. The same day of week shares the same intra-deseasonality curve.

19
However, it takes different deseasonality factors according to the moment of
transaction. Figure 1 illustrates the evolution of the seasonality factors of RWE for
duration, frictionless, and actual return changes when v = 4000.7 It is not surprising
to see that the frictionless and actual return changes have similar dynamics for the
reason that the actual return changes contain the frictionless return changes.
However, the magnitude is different for frictionless and actual return changes.
Panel B of Tables 1.1, 1.2 and 1.3 report descriptive statistics of deseasonalized
durations, frictionless and actual return changes. The raw frictionless and actual
return changes have been normalized to have the mean equal to zero and standard
deviation equal to one. However, other statistics such as skewness, kurtosis and
auto-correlation are not affected by this normalization process. The high kurtosis
and auto-correlation will be captured by the proposed models.

[Insert Figure 1 here]

6.2 Estimation results

We use the model presented in Section 5 to fit SAP, RWE and MRK
deseasonalized data. The data cover the first week of July 2010. The data from the
second week are used as out-of-sample data to test the model’s performance. As
previously mentioned, the estimation is realized jointly for frictionless and actual
return changes. The likelihood function is maximized using Matlab v7.6.0 with
Optimization toolbox.

Tables 2.1, 2.2 and 2.3 report the estimation results for actual return changes for
SAP, RWE and MRK for v = 4000, 4000 and 1800 shares, respectively. It should
be noted that for each stock, the frictionless return changes and actual return
changes are governed by the same duration process, which is assumed to be strictly
exogenous. The high clustering phenomenon is indicated in deseasonalized data by
the Ljung-Box statistic (see Table 1, Panel B). Furthermore, the clustering in
duration is confirmed by the Log-ACD model. To better fit the data, we retain a
7
Results on other volumes for RWE, SAP and MRK are available upon request.

20
log-ACD (2,1) specification for SAP durations, a Log-ACD(3,1) model for RWE
durations and a log-ACD(1,1) model for MRK durations. The Ljung-Box statistic
on standardized residuals of duration provides the evidence that the Log-ACD
model is capable of removing the high autocorrelation identified in deseasonalized
duration data. The Ljung-Box statistic with 15 lags is dramatically reduced to 25.08
for SAP, to 39.7 for RWE and to 21.79 for MRK.

Frictionless and actual return changes of the three stocks are also characterized by
a high autocorrelation in level and volatility. Moreover, the Ljung-Box statistics
with 15 lags on deseasonalized return change and its volatility reject the
independence at any significance level for the three stocks. Taking the model
efficiency and parsimony into consideration, a VARMA(4,2)-
8
MGARCH((1,3),(1,3)))) model is retained for SAP, a VARMA(5,1)-
MGARCH((1,3),(1,3)) model for RWE, and the specification of VARMA(2,2)-
MGARCH((1,3),(1,3)) for MRK. The model adequacy is assessed based on
standardized residuals and squared standardized residuals. Taking MRK as an
example, the Ljung-Box statistics for standardized residuals and squared
standardized residuals of actual return changes computed with 5, 15, 15, 20 lags,
respectively, are not significant at the 5% level. The Ljung-Box statistic with 15
lags has been significantly reduced after modeling to 7.72. Similar results are
obtained for stocks RWE and SAP.

Regarding the estimated parameters, the sum of coefficients in each individual

GARCH model is close to one, indicating a high persistence in volatility.
Furthermore, non-zero of π f and π b in both structures provides the evidence of
asymmetry, that is, a negative shock generates a higher conditional volatility for the
next moment. It also should be noted that γ b and γ f are both positive for the
three stocks. This means that a longer duration will generate a higher volatility for
both actual return and frictionless return changes. In addition, due to the fractional

8 NGARCH (1,3) for actual return changes and NGARCH(1,3) for frictionless return changes.

21
exponent, the volatility increases with a decreasing speed when duration becomes
longer. It should be noted that in our model the volatility is the product of
no-duration scaled variance and duration factor, and the γ b of actual return
changes are higher than γ f of frictionless return changes for the three stocks. This
implies that the duration factor has a larger impact on actual returns than
frictionless returns.

The use of dynamic conditional correlation is justified by the fact that θ1 and θ 2 in
equation (12) are both significantly different from zero for the three stocks. As
expected, the conditional correlation of actual return and frictionless return
changes is time-varying. The sum of two parameters around 0.8 confirms the high
persistence of conditional correlation.

[Insert Table 2 here]

6.3 Model performance and backtesting

In this section, we present the simulation procedure and backtesting results on

simulated deseasonalized and re-seasonalized frictionless and actual return changes.
Once the model is estimated on tick-by-tick frequency, we can test the model
performance and compute frictionless IVaRc and LIVaRc by Monte Carlo
simulation. One of the advantages of our method is that once the model is
estimated, we can compute the simulated deseasonalized IVaRc and LIVaRc for any
horizon without re-estimating the model. In addition, we can compute the
simulation-based re-seasonalized IVaRc and LIVaRc in traditional calendar time
using the available seasonal factors.

We choose different time intervals to test the model performance. The interval
lengths are: 40, 50, 60, 80, 100, 120 and 140 units of time for more liquid stocks
SAP and RWE and 20, 30, 40, 50, 60, 80 and 100 for less liquid stock MRK. As the
model is applied to deseasonalized data, the simulated duration is not in calendar
units. However, they are related in a proportional way. It should also be noticed

22
that, according to the trading intensity, the simulated duration does not correspond
to the same calendar time interval. For a more liquid stock, the same simulated
interval relates to a shorter calendar time interval. For instance, in the case of
MRK, the interval-length 50 re-samples the one-week data for 190 intervals and
corresponds to 13.42 minutes and the 100 interval-length relates to 95 intervals
and corresponds to 26.84 minutes. However, for a more liquid stock such as SAP,
50 interval-length corresponds to 5.45 minutes and 140-interval-length to 15.27
minutes.

The simulations for frictionless and actual return changes are realized as follows:

1) We generate the duration between two consecutive transactions since we assume

that the duration process is strongly exogenous.

2) With the simulated duration and estimated coefficients of the VARMA-

NGARCH model, we obtain the corresponding return changes.

3) We repeat steps 1 and 2 for 10,000 paths and re-sample the data at each path
according to the predetermined interval.

4) For each interval, we compute the corresponding IVaRc and LIVaRc at the
desired level of confidence. To conduct the backtesting for each given interval, we
also need to construct the return changes for original out-of-sample data.

To validate the model, we conduct the Unconditional Coverage and Independence

tests by applying Kupiec test (1995) and Christoffersen test (1998). The Kupiec
test checks whether the empirical failure rate is statistically different from the
failure rate we are testing, whereas the Christoffersen test aims at evaluating the
independence aspect of the violations. More specifically, it rejects VaR models that
generate clustered violations by estimating a first order Markov chain model on the
sequence. Table 3 reports the p-values for the Kupiec and the Christoffersen tests
upon simulated data for confidence levels of 95%, 97.5%, 99% and 99.5%. The
time interval varies from 5 minutes to 15 minutes for SAP, from 5 minutes to

23
16.67 minutes for RWE and from 5 minutes to 27 minutes for MRK. Most of p-
values are higher than 5% indicating that, in general, the model captures well the
distribution of frictionless and actual return changes.

[Insert Table 3 here]

Since most of trading and risk management decisions are based on the calendar
time and raw data, it might be difficult for practitioners to use simulated
deseasonalized data to conduct risk management. To this end, we conduct another
Monte Carlo simulation that takes into account the time-varying deterministic
seasonality factors. The process is similar to that used for simulating
deseasonalized data. However, the difference is that we re-introduce the seasonality
factors for duration, actual return changes and frictionless return changes. As
seasonality factors vary from one day to another, the simulation should take the
day of week into account. More precisely, for the first day, simulated durations are
converted to a calendar time of that day and then the corresponding timestamp
will identify the seasonality factors for actual and frictionless return changes. The
simulation process will continue until the corresponding timestamp surpasses the
closing time for the underlying day. In the case of a multiple-day simulation, the
processes will continue for another day. Based on the simulated re-seasonalized
data, we also compute our IVaRc and LIVaRc by repeating the same algorithm.

Table 4 presents the backtesting results on the re-seasonalized simulated data. The
time interval varies from 5 minutes to 10 minutes for the three stocks and the
confidence levels to test are 95%, 97.5%, 99% and 99.5%. Similar to the test results
for simulated deseasonalized data, both the Unconditional Coverage test and
Independence test suggest that the simulated re-seasonalized data can also provide
reliable high-frequency risk measures for all chosen confidence levels over intervals
from 5 to 10 minutes.

[Insert Table 4 here]

24
7 Risks for Waiting Cost, Ex-ante Liquidity Risk and Various IVaRs

7.1 Risks for waiting cost

As shown in Section 4, the sums of tick-by-tick frictionless return changes and

actual return changes over a given interval can be viewed as the waiting costs
related to market risk and total risk, which contains market risk and ex-ante
liquidity risk. Consequently, the corresponding IVaRc and LIVaRc estimate the risk
of losses on these waiting costs. Based on simulated re-seasonalized data from
previous section that contain the determinist (seasonal factor) and random (error
term) elements, we can further investigate the effect of the ex-ante liquidity risk
embedded in the open LOB on the total risk. To this end, we define an impact
coefficient of ex-ante liquidity risk 9:

�𝑐 − 𝐼𝑉𝑎𝑅
𝐿𝐼𝑉𝑎𝑅 �𝑐
𝚤𝑛𝑡,𝑣 𝚤𝑛𝑡
𝛤𝑖𝑛𝑡,𝑣 = (17)
�
𝐿𝐼𝑉𝑎𝑅 𝑐
𝚤𝑛𝑡,𝑣

As mentioned above, IVaRint,t

c
and LIVaRint,v,t
c
are the VaRs for frictionless return
changes and actual return changes of volume v for the t-th interval. As we simulate
the data in the tick-by-tick framework, we can compute the IVaRint,t
c
and LIVaRint,v,t
c

�𝚤𝑛𝑡
for any desired interval. Accordingly, 𝐼𝑉𝑎𝑅 𝑐 �𝚤𝑛𝑡,𝑣
and 𝐿𝐼𝑉𝑎𝑅 𝑐
are the averages of
c
IVaRint,t and LIVaRint,v,t
c
. As a result, Γint,v assesses, on average, the impact of the
ex-ante liquidity risk of volume v on total risk for a given interval. Figure 2 shows
how the impact coefficients of ex-ante liquidity perform for intervals from 3
minutes to 10 minutes for the three stocks.

[Insert Figure 2 here]

There are two interesting points to stress after observing the plots. First, the curve
is globally increasing, that is, in most of the times, the impact coefficient of the ex-
ante liquidity increases when interval increases and will finally converge to its long-
9 The risk measures in equation (17) are computed for confidence level 95%; in practice, other confidence levels can
also be used.

25
run level. It should be noted that the relation of frictionless return changes and
actual return changes can be explicitly expressed by r=
i
b
ri f + ri LOB , where ri LOB is the

volume-dependent LOB return change for i-th transaction. Econometrically, as the

sum of two ARMA structures is also an ARMA structure (Engel (1984)), the
difference of the frictionless return changes and the actual return changes follow
implicitly another ARMA structure. Accordingly, the relationship of IVaRint,t
c
and
c
LIVaRint,v,t can be generally written as:

c
LIVaRint,v,t = c
IVaRint, t + LOBIVaRint,v,t + Depint,v,t
c F,LOB
(18)

c
LOBIVaRint,v,t measures the risk associated with the open LOB and Depint,v,t
F,LOB
presents
the dependence between the frictionless return changes and the actual return
changes, which can stand for various dependence measures. However, in our
specific modeling, Depint,v,t
F,LOB
is the covariance between ri f and ri LOB . Therefore, the
numerator of equation (17) is the sum of LOBIVaRint,v,t
c
and Depint,v,t
F,LOB
. The fact that the
curve is globally increasing in time is due to the higher autocorrelation in LOB
return changes over time, which are the changes of magnitude in LOB caused by a
given ex-ante volume.

The convergence means that, for the long run, the sum of LOBIVaRint,v,t
c
and Depint,v,t
F,LOB

is proportional to LIVaRint,v,t
c
. Recall that in a general GARCH framework, the
forward multi-step volatility will converge to its unconditional level. In the present
study, once the intervals include sufficient ticks for which the volatilities of both
LOB return changes and frictionless return changes reach their unconditional
levels, the volatilities of the sum of both return changes will increase with the same
speed and therefore the impact coefficient of the ex-ante liquidity will converge to
its asymptotic level.

Second, it is interesting to observe that the impact coefficients of ex-ante liquidity

of RWE stock are negative for a volume of 1000 shares and become positive when

26
volumes are 2000, 3000 and 4000 shares. A negative impact coefficient of ex-ante
liquidity indicates that volatility for the actual return changes is less than that of the
frictionless return changes. In other words, the ex-ante liquidity risk embedded in
LOB offsets the market risk. This again results from the fact that off-best levels of
LOB are more stable than first level. Based on equation (18), when the volume is
small, the negative correlation between frictionless return change and LOB return
change plays a more important role in determining the sign of the impact
coefficients of ex-ante liquidity. However, for a higher ex-ante volume, the risk of
LOB return change also increases but faster than its interaction with frictionless
return change. Consequently, the impact coefficients of ex-ante liquidity become
positive.

7.2 High-frequency ex-ante liquidity premium

Based on the tick-by-tick simulation, we can also compute the high-frequency

IVaR and LIVaR for frictionless return and actual return. Using IVaR and LIVaR
of the same stock, we can further define a relative interval-dependent liquidity
premium as follows:

LIVaR − IVaRint,t
Λ int,v,t = int,v,t (19)
LIVaRint,v,t

Similar to the liquidity ratio proposed in Giot and Grammig (2006), IVaRint,t and
LIVaRint,v,t are the VaR measures for frictionless return and actual return at the end

of t-th interval and int is the predetermined interval such as 5-min, 10-min, etc. It
should be noticed that our defined actual return and frictionless return do not have
the time-additivity property but the frictionless return changes and actual return
changes do. Even though the VaR based on frictionless return changes and actual
return changes can be used directly in practice, in some situations, practitioners
might want to predict their potential loss on frictionless return or actual return
instead of frictionless and actual return changes for a precise calendar time point.

27
To illustrate how our model can be used to provide the ex-ante risk measure for
frictionless return and actual return, we first compute the return changes for
frictionless returns and actual returns, then calculate the instantaneous frictionless
return and actual return at the beginning of the given interval using equations (3)
and (4). Once we know the frictionless return and actual return at the beginning of
a given interval, we can obtain the frictionless return and actual return for the end
of the interval. Figure 3 illustrates how the frictionless return and actual return at
the end of an interval are computed. Figure 4 then illustrates the evolutions of
IVaR and LIVaR associated with a large liquidation volume for SAP, RWE and
MRK during one out-of-sample day, July 12th, 2010. For the three stocks, the IVaR
and LIVaR both present an inverse U shape during the trading day. However, for
the more liquid stock SAP, the IVaR and LIVaR are less volatile than those of the
less liquid stocks RWE and MRK. It also seems that the total risk is smaller during
the middle of day. Nonetheless, it should be noted that the smaller VaR in absolute
terms does not mean we should necessarily trade at that moment. The IVaR and
LIVaR only provide the estimates of potential loss for a given probability at a
precise point in time.

[Insert Figure 3 here]

[Insert Figure 4 here]

In addition, the difference between curves on each graph, which measures the risk
associated with ex-ante liquidity, varies with time. This is due to the fact that LOB
interacts with trades and also changes during the trading days. Smaller (bigger)
difference indicates a deeper (shallower) LOB. More specifically, for the least liquid
stock, MRK, the ex-ante liquidity risk is more pronounced even for a relatively
smaller quantity of 1800 shares. Regarding the more liquid stocks such as SAP and
RWE, the ex-ante liquidity risk premiums are much smaller even for the relatively
larger quantities of 4000 shares. This again suggests that the ex-ante liquidity risk

28
becomes more severe when the liquidation quantity is large and the stock is less
liquid.

7.3 Comparison of LIVaR and other intraday VaRs

Our proposed IVaR and LIVaR, which are validated by backtesting, allow us to
conduct more analysis on ex-ante liquidity and compare them with other
high-frequency risk measures in existing literature.

The standard IVaR proposed by Dionne, Duchesne and Pacurar (2009) is based on
a transaction price that is similar to the closing price in daily VaR computation.
However, the resulting IVaR serves as a measure of potential loss of ‘paper value’
for a frozen portfolio and it omits the ex-ante liquidity dimension. To some extent,
the IVaR accounts for an ex-post liquidity dimension; more specifically, it
measures the liquidity already consumed by the market. However, active traders are
more concerned with ex-ante liquidity because it is related to their liquidation
value. For any trader, the risk related to liquidity is always present and the omission
of this liquidity dimension can cause a serious distortion from the observed
transaction price especially when the liquidation volume is large.

Another major difference is that before obtaining LIVaR, we should compute

LIVaRc for actual return changes, which gives the potential loss in terms of waiting
costs over a predetermined interval. Accordingly, the resulting LIVaR provides a
risk measure for actual return at a given point in time while the standard IVaR is
based on tick-by-tick log-returns, which have time-additive property. It thus
directly gives a risk measure in terms of price for a given interval.

The high-frequency VaR proposed by Giot and Grammig (2006) is constructed on

mid-quote price and ex-ante liquidation price over an interval of 10 or 30 minutes.
The mid-quote price is usually considered as the efficient market price in market
microstructure theory. However, from a practical perspective, the traders can rarely
obtain the mid-quote price during their transactions. Therefore, the use of mid-

29
quote price will underestimate the risk faced by high-frequency traders. Instead of
taking the mid-quote price as the frictionless benchmark, we take a more realistic
price, best bid price, as the frictionless price for traders who aim at liquidating their
stock. Consequently, for active day-traders, our LIVaR can be considered an upper
bond of risk measure that provides the maximum p-th quantile in absolute value
when liquidating a given volume v. Figure 5 illustrates the difference in
constructing the frictionless returns and actual returns.

[Insert Figure 5 here]

8. Conclusion

In this paper, we introduce the ex-ante liquidity dimension in an intraday VaR

measure using tick-by-tick data. In order to take the ex-ante liquidity into account,
we first reconstruct the LOB for three blue chip stocks actively traded in Deutsche
Börse (SAP, RWE and MRK) and then define the tick-by-tick actual return that is
the log ratio of ex-ante liquidation price computed from a predetermined volume
over the previous best ask price. Correspondingly, the proposed IVaRc and LIVaRc
are based on the frictionless return changes and actual return changes and relate to
the ex-ante loss in terms of actual return. In other words, both risk measures can
be considered as the waiting costs associated with market risk and liquidity risk.

In order to model the dynamic of actual return, we use a logACD-VARMA-

MGARCH structure that allows for both the irregularly spaced durations between
two consecutive transactions and stylized facts in changes of actual return. In this
setup, the time dimension is supposed to be strongly exogenous. Once the model
is estimated, Monte Carlo simulations are used to make multiple-steps forecasts.
More specifically, the log-ACD process generates first the tick-by-tick duration
while the VARMA-MGARCH simulates the corresponding conditional tick-by-tick
frictionless return changes and actual return changes. The model performance is
assessed by using the tests of Kupiec (1995) and Christoffersen (1998) on both
simulated deseasonalized and re-seasonalized data. Both tests indicate that our

30
model can correctly capture the dynamics of frictionless returns and actual returns
over various time intervals for confidence levels of 95%, 97.5%, 99% and 99.5%.

Our LIVaR provides a reliable measure of total risk for short horizons. In addition,
the simulated data from our model can be easily converted to data in the calendar
time. Practically, the potential users of our measure could be the high-frequency
traders that need to specify and update their trading strategies within a trading day
or the market regulators who aim to track the evolution of market liquidity, and the
brokers and clearing houses that need to update their clients’ intraday margins.

Future research can continue in several directions. Our study is focused on a single
stock ex-ante liquidity risk. A possible alternative is to investigate how the IVaR
and LIVaR evolve in the case of a portfolio. In particular, the no-synchronization
of the durations between two consecutive transactions for each stock is a
challenge. Another direction is to test the role of ex-ante liquidity in different
regimes. Our study focuses on the liquidity risk premium in a relatively stable
period. It could also be interesting to investigate how the liquidity risk behaves
during a crisis period. This study will require a more complicated econometric
model to take into account different regimes.

31
 References

Anatolyev, S., Shakin, D., 2007. Trade intensity in the Russian stock market: dynamics,
distribution and determinants. Applied Financial Economics 17, 87-104
Andersen, T.G., Bollerslev, T., 1997. Intraday periodicity and volatility persistence in financial
markets. Journal of Empirical Finance 4, 115-158
Angelidis, T., Benos, A., 2006. Liquidity adjusted value-at-risk based on the components of the
bid-ask spread. Applied Financial Economics 16, 835
Bangia, A., Diebold, F.X., Schuermann, T., Stroughair, J.D., 1999. Modeling liquidity risk with
implications for traditional market risk measurement and management. The Wharton
Financial Institutions Center WP 90-06
Bauwens, L., Giot, P., 2000. The logarithmic ACD model: an application to the bid-ask quote
process of three NYSE stocks. Annals of Economics and Statistics / Annales
d'Économie et de Statistique, 117-149
Beltran-Lopez, H., Giot, P., Grammig, J., 2009. Commonalities in the order book. Financial
Markets and Portfolio Management 23, 209-242
Bilodeau, Y., 2013. XetraParser [computer software]. HEC Montréal
Black, F., 1971. Toward a fully automated stock exchange. Financial Analysts Journal 27, 28-44
Christoffersen, P., 1998. Evaluating interval forecasts. International Economic Review 39, 841-
862
Christoffersen, P., 2003. Elements of financial risk management. Academic Press, San Diego,
214
Coppejans, M., Domowitz, I., Madhavan, A., 2004. Resiliency in an automated auction. Working
Paper ITG Group
Demsetz, H., 1968. The cost of transacting. Quarterly Journal of Economics 82, 33-53
Dionne, G., Duchesne, P., Pacurar, M., 2009. Intraday Value at Risk (IVaR) using tick-by-tick
data with application to the Toronto stock exchange. Journal of Empirical Finance 16,
777-792
Domowitz, I., Hansch, O., Wang, X., 2005. Liquidity commonality and return co-movement.
Journal of Financial Markets 8, 351-376
Dufour, A., Engle, R.F., 2000. Time and the price impact of a trade. The Journal of Finance 55,
2467-2498
Dufour, J.-M., Pelletier, D., 2011. Practical methods for modelling weak VARMA processes:
Identification, estimation and specification with a macroeconomic application.
Discussion Paper McGill University (Department of Economics), CIREQ and CIRANO

32
Engel, E.M.R.A., 1984. A unified approach to the study of sums, products, time-aggregation and
other functions of ARMA processes. Journal of Time Series Analysis 5, 159-171
Engle, R.F., 2000. The econometrics of ultra-high-frequency data. Econometrica 68, 1-22
Engle, R.F., 2002. Dynamic conditional correlation: A simple class of multivariate generalized
autoregressive conditional heteroskedasticity models. Journal of Business & Economic
Statistics 20, 339-350
Engle, R.F., Ng, V.K., 1993. Measuring and testing the impact of news on volatility. The Journal
of Finance 48, 1749-1778
Engle, R.F., Russell, J.R., 1998. Autoregressive conditional duration: A new model for irregularly
spaced transaction data. Econometrica 66, 1127-1162
Engle, R.F., Sheppard, K., 2001. Theoretical and empirical properties of dynamic conditional
correlation multivariate GARCH. NBER Working Papers
Ghysels, E., Jasiak, J., 1998. GARCH for irregularly spaced financial data: The ACD-GARCH
model. Studies in Nonlinear Dynamics & Econometrics 2, 133
Giot, P., Grammig, J., 2006. How large is liquidity risk in an automated auction market?
Empirical Economics 30, 867-887
Glosten, L.R., Harris, L.E., 1988. Estimating the components of the bid/ask spread. Journal of
Financial Economics 21, 123
Gourieroux, C., Jasiak, J., 2010. Local likelihood density estimation and Value-at-Risk. Journal of
Probability and Statistics 2010, 26Pages
Grammig, J., Maurer, K.-O., 2000. Non-monotonic hazard functions and the autoregressive
conditional duration model. Econometrics Journal 3, 16-38
Hautsch, N., 2004. Modelling irregularly spaced financial data: theory and practice of dynamic
duration models. Springer-Verlag
Irvine, P.J., Benston, G.J., Kandel, E., 2000. Liquidity beyond the inside spread: Measuring and
using information in the Limit Order Book. SSRN eLibrary
Kupiec, P., 1995. Techniques for verifying the accuracy of risk measurement models. Journal of
Derivatives 2, 73-84
Kyle, A.S., 1985. Continuous auctions and insider trading. Econometrica 53, 1315-1335
Næs, R., Skjeltorp, J.A., 2006. Order book characteristics and the volume-volatility relation:
Empirical evidence from a limit order market. Journal of Financial Markets 9, 408-432
Pacurar, M., 2008. Autoregressive conditional duration models in finance: A survey of the
theoretical and empirical literature. Journal of Economic Surveys 22, 711-751

33
Pascual, R., Veredas, D., 2010. Does the open limit order book matter in explaining
informational volatility? Journal of Financial Econometrics 8, 57-87
Weiß, G.N.F., Supper, H., 2013. Forecasting liquidity-adjusted intraday Value-at-Risk with vine
copulas. Journal of Banking & Finance 37, 3334-3350
Zhang, M.Y., Russell, J.R., Tsay, R.S., 2001. A nonlinear autoregressive conditional duration
model with applications to financial transaction data. Journal of Econometrics 104, 179-
207

34
 Flowchart for Computing Impact Coefficient of Ex-ante Liquidity Risk
and Ex-ante Liquidity Premium

(a): Compute durations, frictionless returns and actual returns based on raw data.

(b): Construct the frictionless return changes and actual return changes.

(c): Remove seasonality from durations, frictionless return changes and actual return changes.

(d): Estimate the model.

(e): Backtest deseasonalized IVaRc and LIVaRc.

(f): Compute the re-seasonalized data by re-introducing the seasonal factors.

(g): Backtest re-seasonalized IVaRc and LIVaRc.

(h): Compute the Impact coefficient of ex-ante liquidity risk.

(i): Compute the IVaR and LIVaR for frictionless return and actual return.

(j): Derive the ex-ante liquidity risk premium.

35
 Table 1.1: Descriptive Statistics for SAP Raw and Deseasonalized Data
The table shows the descriptive statistics for raw durations, actual return changes when Q=2000, 4000, 6000, 8000
and frictionless return changes. The sample period is the first 2 weeks of July 2010 with 44,467 observations.

Panel A : SAP Raw data

Mean Std.Dev Skew Kurt Min Max LB(15) LB2(15)
Duration 6.43 13.21 4.58 37.28 1.00E-03 292.75 6909.93 1426.06
FR Change 6.39E-09 1.93E-04 0.10 6.01 -1.98E-03 1.47E-03 6408.48 7082.94
AR(Q=8000) -7.77E-09 1.66E-04 0.13 8.13 -1.92E-03 2.14E-03 5096.10 6512.58
AR(Q=6000) -6.25E-09 1.66E-04 0.10 7.43 -1.93E-03 1.72E-03 5252.56 6021.21
AR(Q=4000) -4.73E-09 1.68E-04 0.09 7.79 -1.96E-03 2.11E-03 5374.51 5185.26
AR(Q=2000) -2.76E-09 1.72E-04 0.11 7.15 -1.98E-03 1.90E-03 5514.05 5156.26
Panel B : SAP Deseasonalized Data
Duration 1.00 1.88 4.21 37.82 0.00 52.68 4345.54 1528.45
FR Change -1.33E-04 1.01E+00 0.07 5.99 -10.96 7.21 6411.40 5923.61
AR(Q=8000) -4.32E-05 1.04E+00 0.08 9.99 -15.37 11.25 5043.46 6599.80
AR(Q=6000) -1.05E-05 1.04E+00 0.08 9.45 -13.07 12.60 5184.79 7189.66
AR(Q=4000) -6.65E-05 1.03E+00 0.07 8.33 -12.78 10.76 5324.84 5661.21
AR(Q=2000) -1.37E-04 1.02E+00 0.09 7.26 -12.78 9.69 5501.73 4684.47

Table 1.2: Descriptive Statistics for RWE Raw and Deseasonalized Data
The table shows the descriptive statistics for raw durations, actual return changes when Q=1000, 2000, 3000, 4000
and frictionless return changes. The sample period is the first 2 weeks of July 2010 with 37,394 observations.

Panel A : RWE Raw data

Mean Std.Dev Skew Kurt Min Max LB(15) LB2(15)
Duration 7.64 15.88 4.75 37.53 1.00E-03 296.58 10061.17 2479.01
FR Change -3.10E-08 2.51E-04 0.12 7.51 -3.07E-03 2.49E-03 5091.89 7971.04
AR(Q=4000) -4.33E-08 2.23E-04 0.22 10.01 -2.96E-03 2.38E-03 3940.66 10253.80
AR(Q=3000) -4.05E-08 2.25E-04 0.18 10.00 -3.29E-03 2.37E-03 4022.16 10239.86
AR(Q=2000) -3.70E-08 2.27E-04 0.18 9.19 -3.22E-03 2.33E-03 4131.99 10306.19
AR(Q=1000) -3.12E-08 2.32E-04 0.16 8.75 -3.19E-03 2.42E-03 4364.74 11046.85
Panel B : RWE Deseasonalized Data
Duration 0.97 1.73 3.71 24.51 2.45E-05 28.63 3101.10 577.62
FR Change -7.80E-05 0.98 0.12 6.00 -8.55 7.80 5123.94 5093.30
AR(Q=4000) -1.23E-04 0.99 0.20 7.32 -8.61 10.11 3930.82 4781.76
AR(Q=3000) -1.13E-04 0.98 0.18 7.08 -9.52 9.60 4002.22 4796.94
AR(Q=2000) -1.01E-04 0.98 0.17 6.66 -9.40 9.14 4101.41 4888.15
AR(Q=1000) -8.61E-05 0.98 0.17 6.38 -9.27 8.52 4315.98 5519.57

36
 Table 1.3: Descriptive Statistics for MRK Raw and Deseasonalized Data
The table shows the descriptive statistics for raw durations, actual return changes when Q=900, 1800, 2700 and
frictionless return changes. The sample period is the first 2 weeks of July 2010 with 17,472 observations.

Panel A : MRK Raw data

Mean Std.Dev Skew Kurt Min Max LB(15) LB2(15)
Duration 16.31 36.40 5.16 47.06 0.001 718.35 2157.42 905.39
FR Change -3.83E-09 3.11E-04 0.00 15.52 -5.03E-03 4.25E-03 2551.10 4125.30
AR(Q=2700) -1.33E-08 2.74E-04 -0.02 24.71 -4.71E-03 3.75E-03 1672.50 2303.16
AR(Q=1800) -9.58E-09 2.79E-04 0.17 24.70 -4.77E-03 4.74E-03 1754.34 2308.44
AR(Q=900) -6.82E-09 2.84E-04 0.24 21.74 -4.82E-03 4.50E-03 1890.60 3030.64
Panel B : MRK Deseasonalized Data
Duration 0.97 1.95 4.13 29.42 1.581E-05 26.65 736.41 150.75
FR Change -2.55E-04 1.01 0.06 13.44 -13.87 11.61 2537.61 4884.19
AR(Q=2700) -2.78E-04 1.03 0.14 15.67 -14.65 11.65 1670.95 2537.68
AR(Q=1800) -2.81E-04 1.03 0.20 17.60 -14.15 15.63 1730.01 2201.63
AR(Q=900) -2.68E-04 1.02 0.33 16.44 -13.79 16.86 1862.13 2289.72

37
 Table 2.1: Estimation Results SAP (v = 4000)
The column Statistics reports the Ljung-Box statistic on standardized residuals of duration, actual return changes
and squared standardized residuals for different lags. The bold entries are the estimation coefficients that are not
significant from zero and the Ljung-Box statistics that reject the no-correlation in the residuals.

Estimation log-ACD(2,1)-VARMA(4.2)- NGARCH((1.3).(1.3)) (Obs =21089)

Parameters Estimation Std Error Statistics

𝛼1 0.127 0.009 LB test on Residuals
-0.049 0.009 Lags Statistic C_Value
ACD(2.1)

𝛼2
𝛽1 0.963 0.004 5 5.364 11.070
𝛾1 0.812 0.022 10 13.045 18.307
𝛾2 0.419 0.016 15 25.077 24.996
𝜔 -0.081 0.004 20 30.561 31.410
(1) 1.043 0.088
𝜑11
(1)
LB test on Residuals
𝜑12 -0.322 0.010
(2)
𝜑11 -0.154 0.061 Lags Statistic C_Value
(2)
𝜑12 0.089 0.031 5 5.291 11.070
VARMA(4.2)-NGARCH((1.3).(1,3))

(3)
𝜑11 -0.031 0.016 10 13.424 18.307
Actual Return Changes

(3)
𝜑12 0.047 0.014 15 16.106 24.996
(4) 0.032 0.011 20 18.769 31.410
𝜑11
(4)
𝜑12 0.008 0.012
(1)
LB test on Squared Residuals
𝛿11 -1.338 0.087
(2)
𝛿11 0.353 0.085 Lags Statistic C_Value
𝜔𝑏 0.088 0.004 5 3.395 11.070
𝛼1𝑏 0.412 0.022 10 15.405 18.307
𝛽1𝑏 0.159 0.006 15 18.925 24.996
𝜋𝑏 0.404 0.022 20 21.266 31.410
𝛼3𝑏 0.308 0.018
𝛾𝑏 0.073 0.002
(1)
𝜑22 0.6055 0.0693
(1)
LB test on Residuals
𝜑21 0.1754 0.0093
(2)
𝜑22 -0.0401 0.0175 Lags Statistic C_Value
ARMA(4,2)-GARCH((1,3),(1,3))

(2)
𝜑21 -0.0593 0.0187 5 15.811 11.070
Frictionless Return Changes

(3)
𝜑22 0.0488 0.0133 10 21.046 18.307
(3)
𝜑21 -0.0199 0.0103 15 23.203 24.996
(1)
𝛿22 -1.4122 0.0690 20 32.595 31.410
(2)
𝛿22 0.4265 0.0669
LB test on Squared Residuals
𝜔 𝑓 0.0716 0.0037
𝑓
𝛼1 0.4799 0.0305 Lags Statistic C_Value
𝑓
𝛽1 0.1612 0.0074 5 7.794 11.070
𝜋𝑓 0.3466 0.0254 10 17.916 18.307
𝑓
𝛼3 0.2472 0.0237 15 21.902 24.996
𝛾𝑓 0.0377 0.0025 20 24.743 31.410
DCC 𝜃1 0.0791 0.0032
parameter
𝜃2 0.7537 0.0113

38
 Table 2.2: Estimation Results RWE (v = 4000)
The column Statistics reports the Ljung-Box statistic on standardized residuals of duration,
actual return changes and squared standardized residuals for different lags. The bold entries are
the estimation coefficients that are not significant from zero and the Ljung-Box statistics that
reject the no-correlation in the residuals.

Estimation log-ACD(3,1) VARMA(5.1)- NGARCH((1.3).(1.3)) (Obs =15320)

Parameter Estimation StdError Statistics

𝛼1 0.108 0.010
LB test on Residuals
𝛼2 -0.020 0.013
ACD(3.1)

𝛼3 -0.024 0.009 Lags Statistic C_Value

𝛽1 0.970 0.004 5 9.209 11.070
𝛾1 1.145 0.028 10 20.280 18.307
𝛾2 0.280 0.010 15 39.712 24.996
𝜔 -0.068 0.004 20 44.304 31.410
(1)
𝜑11 0.710 0.012
(1)
𝜑12 -0.333 0.011 LB test on Residuals
(2)
𝜑11 0.047 0.014
(2)
𝜑12 0.005 0.013 Lags Statistic C_Value
VARMA(5.1)-NGARCH((1.3).(1,3))

(3)
𝜑11 0.006 0.014 5 10.161 11.070
(3)
Actual Return Changes

𝜑12 0.024 0.013 10 14.653 18.307

(4)
𝜑11 0.006 0.014 15 27.042 24.996
(4)
𝜑12 0.009 0.013 20 30.095 31.410
(5)
𝜑11 0.041 0.012
(5)
𝜑12 0.007 0.012 LB test on Squared Residuals
(1)
𝛿11 -0.980 0.002
𝜔 𝑏 0.099 0.006 Lags Statistic C_Value
𝛼1𝑏 0.437 0.030 5 4.356 11.070
𝛽1𝑏 0.156 0.008 10 8.474 18.307
𝜋𝑏 0.461 0.029 15 11.071 24.996
𝛼3𝑏 0.260 0.023 20 14.102 31.410
𝛾𝑏 0.067 0.003
(1)
𝜑22 0.2324 0.0123
(1)
𝜑21 0.1447 0.0109
(2)
LB test on Residuals
𝜑22 0.0374 0.0132
(2)
𝜑21 0.0048 0.0131 Lags Statistic C_Value
VARMA(5.1)-NGARCH((1.3).(1,3))

(3)
𝜑22 0.0746 0.0135 5 6.378 11.070
Frictionless Return Changes

(3)
𝜑21 -0.0428 0.0133 10 9.420 18.307
(4)
𝜑22 0.0305 0.0138 15 13.824 24.996
(4)
𝜑21 -0.0168 0.0132 20 22.560 31.410
(5)
𝜑22 0.0205 0.0128
(5)
𝜑21 0.0179 0.0114
(1)
𝛿22 -0.9805 0.0018
LB test on Squared Residuals
𝜔 𝑓 0.0711 0.0049
𝑓
𝛼1 0.4586 0.0343 Lags Statistic C_Value
𝑓
𝛽1 0.1436 0.0080 5 7.403 11.070
𝜋𝑓 0.4531 0.0359 10 16.284 18.307
𝑓
𝛼3 0.2731 0.0276 15 19.750 24.996
𝛾 𝑓 0.0299 0.0029 20 22.811 31.410
DCC 𝜃1 0.0813 0.0043
parameter 𝜃2 0.7791 0.0135

39
 Table 2.3: Estimation Results MRK (v = 1800)
The column Statistics reports the Ljung-Box statistic on standardized residuals of duration, actual return changes
and squared standardized residuals for different lags. The bold entries are the estimation coefficients that are not
significant from zero and the Ljung-Box statistics that reject the no-correlation in the residuals.

Estimation log-ACD(1,1)-VARMA(1.1)-NGARCH((1.3).(1.3)) (Obs =7653)

Statistics
Parameter Estimation StdError LB test on Residuals
𝛼1 0.048 0.005 Lags Statistic C_Value
ACD(1.1)

𝛽1 0.984 0.004 5 8.240 11.070

𝛾1 0.784 0.034 10 17.239 18.307
𝛾2 0.382 0.025 15 21.796 24.996
𝜔 -0.049 0.005 20 25.006 31.410
(1)
𝜑11 0.9401 0.0599 LB test on Residuals
(1)
𝜑12 -0.3679 0.0149 Lags Statistic C_Value
VARMA(2.2)-NGARCH((1.3).1)

(2)
𝜑11 -0.0316 0.0515 5 11.971 11.070
Actual Return Changes

(2)
𝜑12 0.0945 0.0264 10 25.105 18.307
(1)
𝛿11 -1.1788 0.0584 15 27.396 24.996
(2)
𝛿11 0.2028 0.0565 20 29.201 31.410
𝑏 0.1543 0.0109 LB test on Squared Residuals
𝜔
𝛼1𝑏 0.3863 0.0341 Lags Statistic C_Value
𝛽1𝑏 0.1877 0.0115 5 1.818 11.070
𝜋𝑏 0.4853 0.0328 10 3.084 18.307
𝛼3𝑏 0.2238 0.0289 15 7.719 24.996
𝛾𝑏 0.0861 0.0032 20 10.320 31.410
(1)
𝜑22 0.4737 0.0725 LB test on Residuals
(1)
𝜑21 0.1822 0.0129 Lags Statistic C_Value
Frictionless Return Changes
ARMA(1,3)-GARCH((1,3),1)

(2)
𝜑22 0.0422 0.0228 5 20.248 11.070
(2)
𝜑21 -0.0652 0.0179 10 28.844 18.307
(1)
𝛿22 -1.2728 0.0712 15 30.197 24.996
(2)
𝛿22 0.3020 0.0677 20 34.977 31.410
𝑓
𝜔 0.1551 0.0119 LB test on Squared Residuals
𝑓
𝛼1 0.4034 0.0378 Lags Statistic C_Value
𝑓
𝛽1 0.2373 0.0163 5 4.348 11.070
𝑓
𝛼3 0.1670 0.0272 10 11.419 18.307
𝑓 0.0039 22.855
𝛾 0.0525 15 24.996
DCC 𝜃1 0.1009 0.0027 20 35.233 31.410
parameter 𝜃2 0.6775 0.0159

40
 Table 3: Backtesting on Simulated Deseasonalized Data
Panel A: SAP Out-of-sample Backtesting on Deseasonalized Actual Return Change (v = 4000)

Interval Kupiec Test Christoffersen Test Nb of Time interval

(in
units) 5% 2.50% 1% 0.50% 5% 2.50% 1% 0.50% Intervals in minutes
40 0.001 0.010 0.717 0.191 0.407 0.724 0.769 0.953 585 4.36
50 0.097 0.133 0.160 0.321 0.302 0.669 0.896 0.948 468 5.45
60 0.907 0.090 0.286 0.971 0.174 0.747 0.919 0.919 390 6.54
80 0.313 0.796 0.548 0.264 0.376 0.530 0.773 0.803 292 8.73
100 0.078 0.062 0.320 0.872 0.709 0.852 0.926 0.926 234 10.90
120 0.342 0.679 0.971 0.980 0.503 0.723 0.838 0.919 195 13.08
140 0.200 0.231 0.573 0.196 0.619 0.825 0.912 # 167 15.27

Panel B: RWE Out-of-sample Backtesting on Deseasonalized Actual Return Change (v = 4000)

Interval Kupiec Test Christoffersen Test Nb of Time interval

(in
5% 2.50% 1% 0.50% 5% 2.50% 1% 0.50% Intervals in minutes
units)
40 0.001 0.002 0.095 0.239 0.477 0.832 0.902 0.951 535 4.77
50 0.134 0.589 0.891 0.923 0.313 0.557 0.783 0.891 428 5.96
60 0.217 0.292 0.362 0.874 0.341 0.679 0.881 0.881 357 7.14
80 0.106 0.258 0.239 0.761 0.510 0.762 0.931 0.931 267 9.55
100 0.109 0.877 0.577 0.126 0.591 0.661 0.811 0.811 214 11.92
120 0.145 0.460 0.522 0.909 0.590 0.748 0.915 0.915 178 14.33
140 0.626 0.299 0.646 0.797 0.538 0.817 0.908 0.908 153 16.67

Panel C: MRK Out-of-sample Backtesting on Deseasonalized Actual Return Change (v = 1800)

Interval Kupiec Test Christoffersen Test Nb of Time interval

(in
5% 2.50% 1% 0.50% 5% 2.50% 1% 0.50% Intervals in minutes
units)
20 0.006 0.120 0.384 0.699 0.430 0.647 0.845 0.845 476 5.36
30 0.026 0.119 0.152 0.617 0.547 0.782 0.936 0.936 317 8.04
40 0.021 0.058 0.309 0.857 0.642 0.854 0.927 0.927 238 10.71
50 0.213 0.384 0.942 0.959 0.530 0.756 0.836 0.918 190 13.42
60 0.470 0.613 0.747 0.820 0.490 0.732 0.820 0.910 158 16.14
80 0.055 0.179 0.122 0.275 0.793 0.896 # # 119 21.43
100 0.378 0.800 0.959 0.506 0.656 0.768 0.883 0.883 95 26.84

The table contains the p-values for Kupiec and Christoffersen tests for the stocks SAP, RWE and MRK.
Interval is the interval length used for computing the LIVaR. Nb of intervals is the number of intervals
for out-of sample analysis and Time interval in minutes are the corresponding calendar times. Bold entries
indicate the rejections of the model at 95% confidence level. When the numbers of hits are less than two,
the p-values are denoted by #.

41
 Table 4: Backtesting on Simulated Re-seasonalized Data
Panel A: SAP Out-of-sample Backtesting on Raw Actual Return Change (v = 4000)

Interval Kupiec Test Christoffersen Test Nb of

(in
5% 2.50% 1% 0.50% 5% 2.50% 1% 0.50% Intervals
mins)
5 0.445 0.420 0.091 0.017 0.612 0.444 0.130 0.675 485
6 0.132 0.483 0.256 0.986 0.334 0.570 0.888 0.888 405
7 0.072 0.160 0.803 0.838 0.053 0.332 0.818 0.878 345
8 0.588 0.566 1.000 0.663 0.297 0.651 0.841 # 300
9 0.670 0.923 0.209 0.220 0.500 0.132 0.697 0.832 270
10 0.066 0.726 0.362 0.520 0.081 0.552 0.752 0.898 245

Panel B: RWE Out-of-sample Backtesting on Raw Actual Return Change (v = 4000)

Interval Kupiec Test Christoffersen Test Nb of

(in mins) 5% 2.50% 1% 0.50% 5% 2.50% 1% 0.50% Intervals
5 0.173 0.341 0.364 0.778 0.266 0.604 0.897 0.927 485
6 0.132 0.071 0.583 0.986 0.352 0.777 0.888 0.921 405
7 0.570 0.563 0.772 0.379 0.259 0.618 0.791 0.851 345
8 0.070 0.156 0.537 0.697 0.651 0.776 0.908 0.908 300
9 0.307 0.247 0.857 0.601 0.430 0.795 0.863 0.931 270
10 0.043 0.156 0.733 0.175 0.647 0.856 0.856 0.856 245

Panel C: MRK Out-of-sample Backtesting on Raw Actual Return (v = 1800)

Interval Kupiec Test Christoffersen Test Nb of

(in mins) 5% 2.50% 1% 0.50% 5% 2.50% 1% 0.50% Intervals
5 0.187 0.201 0.140 0.298 0.344 0.628 0.897 0.949 485
6 0.010 0.071 0.980 0.220 0.476 0.723 0.777 0.777 405
7 0.273 0.175 0.394 0.548 0.312 0.701 0.878 0.939 345
8 0.267 0.566 0.537 0.083 0.359 0.620 0.870 # 300
9 0.098 0.247 0.654 0.752 0.513 0.728 0.863 0.931 270
10 0.016 0.050 0.765 0.520 0.647 0.856 0.856 0.856 245

The table contains the p-values for Kupiec and Christoffersen tests. Intervals are regularly time-spaced
from 5 minutes to 10 minutes. Bold entries indicate the rejections of the model at 95% confidence level.
When the numbers of hits are less than two, the p-values are denoted by #.

42
 Figure 1: Seasonality Factor For RWE
Panel A: Seasonality Factor for Duration

Panel B: Seasonality Factor for Actual Return Changes (v = 4000)

43
Panel C: Seasonality Factor for Frictionless Return Changes

44
Figure 2: Impact Coefficients of Ex-ante Liquidity for Different Time Intervals
Panel A: SAP

Panel B: RWE

45
 Panel C: MRK

Panels A, B and C illustrate how the impact coefficients of ex-ante liquidity evolve for intervals from 5 minutes to
10 minutes for stocks SAP, RWE and MRK. The selected volumes for the actual return changes are 2000, 4000,
6000, and 8000 shares for SAP, 1000, 2000, 3000, and 4000 shares for RWE, and 900, 1800 and 2700 shares for
MRK.

46
 Figure 3: Computation of the Frictionless Return and the Actual Return for the
End of an Interval

Figure 3 illustrates the computation of the frictionless return and the actual return for the end of an interval. I
indicates the transaction. At the beginning of the interval, we compute the frictionless return and actual return using
the real data from market. Each I corresponds to a frictionless (actual) change which comes from the simulations.
Consequently, the frictionless return (actual return) at the end of the interval is the sum of the initial frictionless
return (actual return) and all the corresponding changes in the interval.

47
Figure 4: IVaR and LIVaR of 5-minute for July 12, 2010

Panel A: SAP with LIVaR (v =4000)

Panel B: RWE with LIVaR (v =4000)

48
 Panel C: MRK with LIVaR (v =1800)

Panels A, B and C present the VaRs for frictionless returns and actual returns at the end of each 5-minute interval
on July 12th, 2010, for the three stocks of SAP, RWE and MRK, respectively. The selected volumes for the actual
returns are 4000 shares for SAP, 4000 shares for RWE and 1800 shares for MRK.

49
 Figure 5: Frictionless Returns and Actual Returns

This figure presents the difference between our frictionless (actual) return and the frictionless (actual) return proposed by Giot and Grammig (2006). Arrow a
presents the starting price and end price in constructing our frictionless return, while arrow b shows the starting price and end price for the actual return given a
liquidation quantity v. Both frictionless returns and actual returns take the previous best ask price as starting price. Arrows c and d give the starting price and end
price for computing the frictionless return and the actual return (for quantity v) proposed by Giot and Grammig (2006). Their frictionless return takes the previous
mid-quote as the starting price and the following mid-quote as the end price. Their actual return takes the previous mid-quote as the starting price and the actual
price as the end price.

50
Liquidity Risk Management

Recent Trends

Sidharth Reddy, Genpact

Anirban Naskar, Genpact

2017
Table of contents

1. INTRODUCTION ...................................................................................................... 6

2. LIQUIDITY RISK MANAGEMENT CYCLE .............................................................. 7

3. LIQUIDITY RISK MANAGEMENT STRATEGY, FRAMEWORK, METRICS AND

POLICY LIMITS ............................................................................................................ 10

4. LIQUIDITY RISK MEASURING MODELS AND REPORTS .................................. 12

5. INTRADAY LIQUIDITY RISK MANAGEMENT ...................................................... 21

6. BASEL III FRAMEWORK: LCR AND NSFR ......................................................... 25

7. MARKET LIQUIDITY RISK AND ITS RELATIONSHIP WITH FUNDING LIQUIDITY

RISK.............................................................................................................................. 26

8. CONCLUSION: ...................................................................................................... 27

ABOUT THE AUTHORS............................................................................................... 28

Page 2 of 28
Abstract

With the evolution of Basel reforms, risk management is an increasingly key part of the
decision-making process for financial institutions. The crisis of 2008 has shown that having
sufficient capital is not adequate to remain solvent under market stress—indeed, inadequate
management of liquidity has been found to be the sole reason behind the crisis. The Basel
Committee of Banking Supervision (BCBS), along with its member countries and their banking
regulators, has been more vigilant than before in institutionalizing sound liquidity risk
management practices.

In this paper, we have provided the details of these guidelines and relevant liquidity risk
management activities which are generally observed in banks around the world. This paper also
contains certain aspects of liquidity risk which do not stem from regulatory guidelines, but rather
fall under prudent risk management techniques.

Acknowledgements

We would like to thank our colleagues, both from Genpact and the industry at large, with whom
we have discussed this subject on various occasions.

Keywords: Liquidity risk management, liquidity risk, liquidity risk management cycle, liquidity risk
management strategy, asset and liability management, liquidity risk framework, liquidity risk
policy, liquidity risk models, liquidity risk reporting, behavioral modelling, liquidity risk modelling,
liquidity stress testing, liquidity risk ICAAP, liquidity risk scenario analysis, new volume
projections, funding execution plan, early warning indicators, EWI, intraday liquidity
management, Basel III, Basel 3, liquidity coverage ratio, LCR, net stable funding ratio, NSFR,
Genpact

Page 3 of 28
Page 4 of 28
EXHIBIT 1: LIQUIDITY RISK MANAGEMENT CYCLE ................................................................................................ 7
EXHIBIT 2: FUNCTIONAL RESPONSIBILITIES ........................................................................................................ 8
EXHIBIT 3: LRM POLICY AND PRACTICES START AT THE TOP ............................................................................ 10
EXHIBIT 4: LIQUIDITY RISK MANAGEMENT MODELS ........................................................................................... 12
EXHIBIT 5: REGULATORY COMPLIANCE OF BEHAVIORAL MODELS ...................................................................... 14
EXHIBIT 6: “BEST PRACTICE” STEPS IN BEHAVIORAL MODELS ............................................................................ 15
EXHIBIT 7: VARIANTS OF BEHAVIORAL MODELS ................................................................................................ 15
EXHIBIT 8: SOURCES OF LIQUIDITY RISK FROM BALANCE SHEET DYNAMICS........................................................ 16
EXHIBIT 9: THREE-STAGE APPROACH TO SCENARIO IDENTIFICATION AND DEVELOPMENT ................................... 17
EXHIBIT 10: STEP-BY-STEP ACTIVITIES IN NEW VOLUME PROJECTIONS .............................................................. 19
EXHIBIT 11: SCHEMATIC OF AN INTRADAY LIQUIDITY MODEL ............................................................................. 22
EXHIBIT 12: MARK-TO-FUTURE (MTF) DISTRIBUTION ....................................................................................... 23
EXHIBIT 13: MARK-TO-FUTURE (MTF) DISTRIBUTION AND CASH REQUIREMENT ................................................. 23

Page 5 of 28
1. INTRODUCTION
The core activities of a bank are to raise funds through deposits and market borrowings, and
deploy the same through loans and advances and various investments. Banks tend to take
advantage of upward-sloping yield curve by sourcing funds in the short term and deploying
those in the long term. In this way, banks end up collating liquidity risk in their books.

Regulators across the globe have designed various standards to help banks mitigate such risk.
Invariably, all of those standards are linked to the Basel guidelines.1 The root cause of the 2008
global financial crisis was liquidity risk mismanagement. Subsequently, the Basel Committee
developed the Basel III guidelines to address the need for sound liquidity risk management
under stressed conditions.2

Another area of focus for regulators is managing intraday liquidity risks.3

In addition to these, funding liquidity risk has a deep-rooted connection with the funds transfer
pricing (FTP) process, as the treasury is mandated to remove business groups from the burden
of managing their funding risk. In this regard, Bank for International Settlement (BIS) had
published a paper to describe best practice of incorporating liquidity risk charges, called,
liquidity transfer pricing (LTP)4.

This paper analyzes the practices that banks follow to comply with these regulatory
requirements. This paper also describes current industry practices for liquidity risk management.

1 Basel Committee for Banking Supervision (BCBS), Principles for Sound Liquidity Risk Management and
Supervision, September 2008
2 BCBS, Basel III: A global regulatory framework for more resilient banks and banking systems, June

2011
3 BCBS, Monitoring tools for intraday liquidity management, April 2013
4 Joel Grant, Australian Prudential Regulation Authority, Liquidity transfer pricing: a guide to better

practice, Financial Stability Institute, BIS, December 2011

Page 6 of 28
2. LIQUIDITY RISK MANAGEMENT CYCLE
Risk management at banks involves having a sound policy and framework, as well as
measurement metrics, calculation techniques, and monitoring mechanisms. Such processes are
closely monitored by senior management. Risk management, as a tool, is used to manage
liquidity risk by managing specific levers. Exhibit 1 explains the processes involved in liquidity
risk management.

Exhibit 1: Liquidity risk management cycle

Liquidity risk management

strategy and framework
 Stress testing
KRI, risk tolerances, risk and scenario
limit thresholds analysis
 Behavioral
Liquidity risk measuring modelling
models, setup, reporting

Funding plans, FTP/LTP

Intraday liquidity risk, EWI

These activities are performed in close coordination across various functions at a bank. The
detailed functionalities are given in Exhibit 2.

Page 7 of 28
Exhibit 2: Functional responsibilities
# Functions Activities Design and execution
responsibility

1 Strategy and Risk appetite statement CFO and asset and liability
framework management (ALM),5 CRO
and risk management

2 KRI, risk Risk indicators (e.g., loan to ALM and risk management
tolerances, deposits, reliance on wholesale
risk limit funding/macro environment,
thresholds bucket-wise gaps, projected stock
of liquidity, etc.)

Liquidity limits per maturity

buckets – forecast balance
sheets, forecast risk positions,
apply constraints (e.g., ability and
reputation to borrow in the
market, etc.)

3 Liquidity risk Behavioral models – non-maturity Risk management and

measuring deposits (NMD), prepayment, etc. business groups
models

Stress testing models (e.g.,

ICAAP models, scenario analysis
models, etc.), short-term and
protracted stress, bank- and
market-specific scenarios

4 Model setup Setting up ALM software and Middle-office operations

underlying database and technology

5 ALM includes funding execution desk

Page 8 of 28
# Functions Activities Design and execution
responsibility

Connecting the system with

various banking systems and
accounting systems

5 Reporting Regulatory reports - LCR, NSFR, Middle-office operations

static/dynamic gap, liquidity
projections

Internal MIS, trend analysis, root

cause analysis, commentary,
ALCO reporting

6 Funding plans, Methodology to price liquidity risk ALM, finance, risk

FTP/LTP collateral, calculate liquidity management
utilizations

Liquidity buffer management

Funding contingency plan

7 Intraday Design of databases for internal ALM, operations

liquidity risk and external data sources

Reconciliation systems for tallying

data between internal records
and counterparty records

Development of models, BAU and

stress

8 EWI Design of EWI and operational ALM, risk management,

protocols operations

Monitoring of EWI

Page 9 of 28
3. LIQUIDITY RISK MANAGEMENT STRATEGY,
FRAMEWORK, METRICS AND POLICY LIMITS
It is a regulatory requirement and of strategic importance to the risk management function, the
CRO, the CFO, and the board, that a bank has and reviews its entire liquidity risk management
(LRM) framework and processes periodically. Accordingly, as a prudent practice, all banks
review their LRM policy (or some other equivalent policy like asset and liability management
(ALM) Policy), liquidity risk stress testing Policy, LRM limits and controls in accordance with the
risk appetite statement, LRM measurement metrics, models, monitoring, and reporting
processes.

This paper aims to provide basic guidance for such policies and processes. At the same time, it
is important to emphasize that each bank is unique, and that objectives, products, and risk
management practices will therefore be different from bank to bank.

The board and management have to set up the right structure and governance mechanism,
which will be responsible for the entire pyramid of liquidity risk management. The “must take”
steps are shown in Exhibit 3: LRM policy and practices start at the top.

Exhibit 3: LRM policy and practices start at the top

Key requirements for LRM Steps to be followed at the enterprise level

 LRM activities to be closely controlled by  Right "organization structure"

board-level committees
 Balance sheet management to be − Specific subsidiary requirements
performed under the supervision and for foreign banks (e.g., in US);
boundaries laid down by the committees governance procedure for foreign
 Adequate focus on risk measurement entities and banking/non-banking
and reporting by implementing required subsidiaries
IT infrastructure − Board-level committees like risk
 Compliance with regulatory guidelines management committee of board
 For global and diversified banks/ bank (RCB) and asset and liability
holding companies, to ensure that management committee (ALCO)
governance structure comply with norms
at respective geographies − The quorums, permanent
attendees, and roles and
responsibilities of such
committees
 Group-level governance framework, such
as group liquidity management
framework (GLMF)

Page 10 of 28
  Relationship between the group entities
and the branches; policies around the
credit lines through entities

Diligent monitoring and reporting are crucial to effectively managing liquidity risks. The following
section contains a few measures which are observed in many banks.

 Flow approach

− Liquidity gap, bucket wise, and cumulative

− Currency-wise, country-wise, and consolidated reports

 Stock approach

− Liquidity projections
− Various ratios capturing: the extent to which volatile money
supports bank’s basic earning assets; the extent to which assets
are funded through stable deposit base; the extent to which illiquid
assets are financed out of core deposits, etc.

 Stress testing, Pillar 2 capitals, ICAAP

− LCR and NSFR

− Design and monitoring of ICAAP (in some countries ILAAP and
ILAA) scenarios
− Pillar 2 capital provisioning

Page 11 of 28
4. LIQUIDITY RISK MEASURING MODELS AND REPORTS

The core liquidity risk management models, such as static gaps, ratios, and dynamic liquidity
projections, are automated models. In general, these models are implemented either in
sophisticated liquidity risk management systems, such as OFSAA, QRM, and Bancware, or in
simple spreadsheets, depending upon the complexity of the balance sheet and the business
structure. However, there are various models which are implemented outside of such systems.
Examples of such models include new volume/yield projections, behavioral projections, capital
calculation models, etc. Exhibit 4, below, explains the role of various models in the liquidity risk
reporting cycle.

Exhibit 4: Liquidity risk management models

Transactional Behavioral Models for regulatory
and master models and strategic risk
data sources management

On B/S - Ledger − Non-maturity - Static and dynamic

data deposits risk metrics
− Optionality in - Stock and flow
term deposits measures (e.g., gap,
− Prepayment liquidity projections,
On B/S -
of term loans key ratios, etc.)
Transactional
− Off-balance-
data
sheet to on-
balance-sheet
conversion
Off-balance- Capital models – Pillar 2
sheet exposure capital (ICAAP)

New volume Regulatory calculation of

projections LCR, NSFR, and stress
testing

Market data Strategic purposes:

models (IR and funding execution plan
FX curves), for future weeks/months,
volatility smile early warning indicators
(EWI) and liquidity
transfer pricing (LTP)

Page 12 of 28
 Models built outside liquidity risk management systems

4.1 Liquidity risk arising due to balance sheet dynamics

The risk that cash flows deviate from contractual cash flows drives a significant portion of
liquidity risk.

 Cash flow reporting: The liquidity gap reports cash inflows and outflows and is used to
determine the net cash position at any point in time. Cumulative cash position and
maximum cash outflows highlight the maximum inflow and outflow in any forecasting
period.
 Term loans: Liquidity risk in the term loans portfolios entails modelling the probabilities of
partial prepayment, loan extension, and full prepayment. All or any one of the above
changes contractual cash flows and gives rise to liquidity risk.
 Residential mortgage portfolios entail modelling the probabilities of curtailment (partial
prepayment), loan conversion/reduction, and full repayment.
 Maturing assets and liabilities: The liquidity risk in term deposit portfolios is related to
early withdrawal (termination before contractual end date) and rollover risk (investing in
the same product after maturity date, renewal). Events such as stress tests, scenario
analysis, and changes in macro-economic variables impact contractual maturities and cash
flow projections.
 Non-contractual assets and liabilities: Estimating liquidity for accounts without a maturity
date is challenging, yet critical. The risk is that the total outstanding relative to the limit (i.e.,
utilization) strongly impacts liquidity. Liquidity risk in non-contractual liability is related to the
run-off risk and to occasionally erratic fluctuations in the outstanding balance of the
accounts.
 Trading assets – securities: Liquidity risk in trading assets is associated with unexpected
cash flows resulting from changes in haircuts and initial margins.
 Collateralized derivatives: Liquidity risk in collateralized derivative exposure is related to
initial/variation margins.

4.2 Behavioral modelling

Behavioral modelling is essential to capturing liquidity risks. In order to effectively assess the
liquidity risk of various products, banks need well-specified econometric models combined with
a consistent approach across all on- and off-balance-sheet items. Typical items for which
behavioral models are required are:

 Maturing assets: residential mortgages, term loans - prepayment risk, conversion,

maturity extension/reduction

Page 13 of 28
  Non-maturing assets: current accounts debit - utilization risk
 Capital markets: securities and CSA derivatives - haircuts, initial margins, variation
margins
 Non-maturing liabilities: Checking/current accounts, savings accounts - attrition risk,
balance fluctuation
 Maturing liabilities: term deposits - roll-over risk, early withdrawal risk
 Modelling approaches:

− Map financial accounts to generic modelling categories

− Identify key behavioral liquidity risks for the “in-scope” portfolios
− Define business segmentation per portfolio (e.g., corporate, retail,
international clients, etc.)
− Estimate behavioral liquidity risks via well-specified econometric
models (emphasis on diagnostics)
− Report liquidity cash flow mismatches via the behavioral calendar

General expectations from Basel and other regulators are given in Exhibit 5.

Exhibit 5: Regulatory compliance of behavioral models

Regulatory requirements Road to compliance

In measuring liquidity risk, key  Developing behavioral model (after analyzing various
behavioral and modelling alternative model methodologies, to ensure tailor-made
assumptions should be fully models for the portfolio)
understood, conceptually
sound, and clearly
documented. Such  Validation of existing behavioral models
assumptions should be
rigorously tested and aligned
with the bank’s business  Periodic enhancements of these models as per
strategies.
management/audit/regulatory findings

 Model documentation and recommendations on potential

areas of enhancements

Exhibit 6 contains the high-level steps which are generally followed at big banks as best
practices.

Page 14 of 28
Exhibit 6: “Best practice” steps in behavioral models
Analyze  Determine importance of the product at the portfolio/account level
product  Understand the factors affecting movements in product composition
features

Portfolio  Classify the product portfolio (e.g., deposit) into homogenous yet
segmentation materially distinct segments
 Geographical/customer segmentation (e.g., retail/wholesale
segmentation)

Statistical  Establish quantitative relationship between products with

modelling determinants across scenarios
 Use different techniques, such as time series analysis, portfolio
optimization, etc.

Integration with  Business-as-usual and stress scenarios

ALM system  Static and dynamic gap reports
 Acts as input to all ALM model

For different parts of the balance sheet and various product characteristics, multiple behavioral
studies are carried out. Some of those are given in Exhibit 7.

Exhibit 7: Variants of behavioral models

Non- Time Borrowings Loans and Non- Off-balance-
maturity deposits advances maturity sheet items
deposits assets
Core/volatile Premature Financial Prepayments Utilizations LC, SBLC,
withdrawal covenants undrawn WC

Transactional Rollover Embedded Delinquency Repayments Derivatives

deposits options

Liquidity risk management focuses primarily on balance sheet dynamics and behavioral
modelling in order to predict cash flows and risks associated with each type of inflow and
outflow. A summary of classification, product type/behaviour, and exposure to risk type is
summarized in the table below.

Page 15 of 28
Exhibit 8: Sources of liquidity risk from balance sheet dynamics
Classification Product type/behavior Risk type

Liquidity Liquidity risk and Balance sheet Contractual cash flows Maturities
risk balance sheet cash flow
management
cycle Amortization

Behavioral cash flows Loan prepayments

Deposit attrition

Contingent cash flows Run offs

Stress payments

Margin calls

Behavioral Maturing Term loans and Prepayment risk

liquidity - assets residential mortgages
modeling
Conversion risk

Maturity
extension/reduction
risk

Non- Current account debit Utilization risk

contractual
assets

Maturing Term deposits Early withdrawal

liabilities risk

Roll-over risk

Page 16 of 28
 Classification Product type/behavior Risk type

Non- Current Balance volatility

contractual accounts/savings
liabilities accounts
Attrition risk

Capital Securities and Initial margin risk

markets derivatives

Variation margin
risk

Haircuts

4.3 Capital models/stress testing/scenario analysis

These models aim to generate “extreme-but-plausible” scenarios that a bank could experience,
both for short and protracted periods of time. These models are built primarily using time-series
analysis, studying historical stress periods, monitoring ongoing economic situations, and
applying sound business judgments. The stress testing approach involves calculating liquidity
gaps under such scenarios. A three-stage Analyze-Identify-Develop approach is outlined in
Exhibit 9 below.

Exhibit 9: Three-stage approach to scenario identification and development

1 Analyze assets, liabilities, and off-balance-sheet exposures

 Express all items as a percentage of total assets, paying attention to high-

percentage items
 Find out key liquidity ratios based on underlying business rationale (e.g., if the
loan book is strategically funded by the deposit book, loans to deposits are a key
risk metric)

Page 17 of 28
 2 Identify sources of liquidity risk

 Of all the percentage compositions and key metrics, identify those which pose
significant risk to the balance sheet (e.g., high percentage of deposits, high
percentage of loans being funded by wholesale funds, etc.)
 Identify balance sheet growth and strategic areas where there may be sources
of liquidity risk
 Identify macroeconomic events which could pose a threat to those items on the
balance sheet which may not currently be at risk

3 Develop scenarios addressing the sources of risk

 Scenarios should address the risk areas identified in step two

 Scenarios should be extreme but plausible

Stressed economic events may arise from macroeconomic sources as well as from the specific
financial/business situation within a particular bank. Accordingly, banks are supposed to
consider market-specific, bank-specific, and combined scenarios to assess their financial
preparedness for handling such situations. Some indicative list stress scenarios, which are
observed in many banks, are given below.

 Market-specific stress scenarios -

− Illiquidity in interbank money market and wholesale funding market

− Drying up of loan securitization market
− Central bank imposing restrictions on reserve maintenances
− Central bank imposing restrictions on liquidity facilities
− Deterioration of credit risk of the market creating higher utilizations
of working capital limits
− Worsening credit worthiness of selected sectors of economy

 Bank-specific scenarios -

− Higher margin calls from outstanding derivative contracts

Page 18 of 28
 − Run-on-the-bank
− Defaulting on key covenants, such as rating downgrade and
reduced solvency ratio, forcing repayment of borrowings
− Top “n” depositors redeeming earlier than maturity
− Top “n” borrowers becoming delinquent or defaulting

These scenarios need to be converted to low-level scenarios, which can help quantify the
impacts of such scenarios. These models can also be created using some time series statistical
methods, say, multiple equation vector-auto-regression (VAR) models and their impulse
responses to different shocks. The following are a few methods practiced by many banks:

 Using historical percentile from available time series, for example, 99th percentile of daily
deposit outflows
 Develop a model of required variables with important macro variables (say house prices
and unemployment rate). Subsequently, feed the worst “n”-year path available of the macro
variables into the model. Alternatively, feed into the model a bad scenario from the
regulatory stress testing. The resulting “stressed output” is the desired scenario.

Liquidity risk can result from a variety of shocks, and these need to be monitored, controlled,
and mitigated at all times.

4.4 Models for projecting new volume

New volume projection models do not follow a specific pattern. For instance, depending upon
business complexity, small banks can benefit from simulations based on the models of larger
and more complex banks. These steps are generally followed using the following kinds of
models (Exhibit 10):

Exhibit 10: Step-by-step activities in new volume projections

Activities Solutions

Balance sheet growth  This is largely strategic decision taken by the board
projections  For a bottom-up approach to new volume forecasts, this acts
as a constraint

Business growth  Balance sheet growth if further allocated to various

projections businesses by CFO’s office
 Models are used for segment-wise growth projections. Such
models consider macro-economic projections as inputs.
Such macro-economic projections are sourced from internal

Page 19 of 28
 research departments, market research, industry reports,
economic surveys, etc.
 Scenario analysis and Monte Carlo simulations are also run
for various estimates of projections under conservative,
realistic, and optimistic scenarios

Constraint of various  Before finalizing the growth projections, it is ensured that

risk and other limits such growth will not exceed any policy limits
 Such risk positions, derived from hypothetical growth
scenarios, are presented to key decision-making bodies,
such as ALCO, before being finalized and communicated to
individual businesses

4.5 Models for strategic purposes

4.5.1 Funding execution plan

Treasury needs to forecast the funding needs for future months and quarters in order to
efficiently run borrowing, lending, or securitization programs at optimized cost. These kinds of
models take input information from various asset and liability businesses, such as deposits,
loans, and treasury. The model adds the information over the contractual maturities. Thus, it
produces the funding gaps at future time periods. There are certain constraints apart from the
inputs gathered from the business units, such as maintenance of regulatory or management
mandated liquidity buffer, minimum LCR and NSFR requirements, etc. Models of this kind are
recursive in nature, as the funding execution plan for period t will be treated as input for the
period t+1. These models are also process-wise complicated in nature, as inputs are
judgmentally generated by various business units located across geographies.

4.5.2 Early warning indicators

Banks need to have identified indicators that serve as guides to the likelihood of impending
liquidity events. Such indicators are of two types: either they are instigated from the bank’s own
financial positions, or they are indicated by market dynamics. Examples of market indicators are
a bank’s share price, banking sector index, bank’s CDS spread, exchange rate volatility,
systemic credit deterioration/country rating, wholesale funding market turnover/rates, etc.,
whereas the same from bank’s balance sheet dynamics can be outflows from non-maturity
deposits, early redemption of term deposits, reputation of bank, adverse news or industry
reports, etc.

Page 20 of 28
5. INTRADAY LIQUIDITY RISK MANAGEMENT
Managing intraday liquidity (IDL) presents a real challenge for banks in terms of data, KPIs,
measurements, analytics, and reporting. BCBS mandated that regulators establish certain
minimum standards to monitor the intraday liquidity management of banks.6 These were
developed in consultation with the Committee on Payment and Settlement Systems (CPMI).

 Tools applicable for all reporting banks:

− Daily maximum liquidity usage

− Available intraday liquidity at the start of the business day
− Total payments
− Time-specific obligations

 Tools applicable to banks providing corresponding banking services:

− Value of payments made on behalf of corresponding banking

customers
− Intraday credit lines extended to customers

 Tools applicable for direct participants:

− Intraday throughput

Intraday liquidity models involve intense interactions between various systems on a near-real-
time basis. A high-level schematic of an intraday liquidity model is given below.

6 BCBS, Monitoring tools for intraday liquidity / BCBS248, April 2013

Page 21 of 28
Exhibit 11: Schematic of an intraday liquidity model
Data Flows Systems and Models Reporting

Internal / External Data Centralized Intraday Liquidity Monitoring

Business-as-Usual
• Identify Scope of
Monitoring
• Consolidate all
Internal External transactional flows (A)
• Stock of liquidity (B) Seven Intraday
- Central • Relationship ( A , B )
- Cash & coins at Parameters (As
Banks • Limits, Utilizations, Triggers
currency chest applicable) and
- Net / Gross
- Vostro Stress Testing
Settlements Stress Period
- Customers A/c
- Nostro Underlying models • Static / Dynamic Scenario
- Temporary
- Counterparty • Impact of Scenarios
Overdraft
- Disbursement
- Liquid Assets
JUDGMENTAL OVERLAYS

Intraday Predictive Analytics to forecast * Diagram’s simplicity abstracts

away the true complexity of the
Flows small ticket high volume flows process

Derivatives are an integral part of banking businesses. Therefore, it is important to capture

potential liquidity risk comprehensively by projecting cash flows from margin requirements
arising out of derivative contracts. Margin requirements need to cover two parts:

 Variation margin - or current exposure - defined as (unrealized) profit or loss in the portfolio
 Initial margin - or potential future exposure (PFE) - defined as the PFE of the portfolio, that
is, the potential maximum loss in the portfolio over the time till close out of the portfolio.
PFE at certain confidence levels requires the simulation system to calculate mark-to-future
(MtF) distribution, which needs a scenario generating and pricing method.

Although the variation margin takes care of mark-to-market movements, initial margin factors
the potential maximum loss over the time till close out of the deal.

Page 22 of 28
Exhibit 12: Mark-to-future (MtF) distribution
Mark to Future

Future Scenarios

To forecast future cash flow need/release, banks must generate probable future scenarios, and
subsequently, calculating mark-to-future distribution generation. From this distribution,
depending on choice of confidence level which is derived from risk appetite, a risk-based MtF
amount is calculated. The broad steps are as follows:

Exhibit 13: Mark-to-future (MtF) distribution and cash requirement

Scenario generation Mark-to-Future Distribution
Generation

 Use stochastic models of each risk factor  Calculating derivative price on each
 Estimate parameters from historical data scenario and generate MtF distribution
 Get PFE at certain confidence level from
MtF distribution

Page 23 of 28
Below are additional calculations to consider:

 Changes in valuation of underlying derivatives using mark-to-future calculation (PFE)

 Changes in valuation/credit rating of existing collateral, creating the need to post additional
or higher-rated collateral
 Excess collateral placed by counterparty included in HQLA which can be taken back by
counterparty
 Change of collateral by counterparty with non HQLA (if allowed to have other instruments)
 Difference between master-netting and without-master-netting agreements
 Extra collateral required due to change in credit rating/financial condition of bank

Page 24 of 28
6. BASEL III FRAMEWORK: LCR AND NSFR
Guidelines on the Basel III framework have been issued since 2010. This is a comprehensive
set of reform measures primarily strengthening the capital and liquidity risk management of
banks. Major reform in liquidity risk has been introduced through two ratios, liquidity coverage
ratio (LCR)7 and net stable funding ratio (NSFR).8 LCR is already being calculated and reported
in a phased manner; NSFR reporting will begin in 2018.

6.1 LCR
LCR measures a bank’s resilience to survive for a 30-day period under severe liquidity stress.
This stress event/scenario is largely defined by the regulators. Banks are required to hold high-
liquid assets amounts equal to or greater than their net cash outflow over a 30-day period. As
an additional requirement, banks are required to hold sufficient intraday cash and collateral to
survive net cash outflows caused by crisis events. Since the scenarios are largely defined by
regulators, banks have little role in developing models for scenarios. However, banks can
calculate the outstanding balances of various balance sheet components. Models are used for
calculating these components - for example, stock of stable deposits.

6.2 NSFR
NSFR measures the structural liquidity risk of a bank’s businesses and balance sheet positions.
It measures whether a bank has enough stable funding (such as equity capital) to ensure
uninterrupted activities on its on- and off-balance-sheet components. At a higher level, it
considers the relationship between a bank’s settlement obligations (longer term) and available
funding. NSFR is used to examine bank’s resilience over a protracted stress period, which is, of
more than one year Like LCR, this is largely driven by regulator-created stress scenarios, with
banks playing very little role in the development of such scenarios.

7 BIS, Basel III: The Liquidity Coverage Ratio and liquidity risk monitoring tools / BCBS238, January 2013
8 NIS, Basel III: the net stable funding ratio / BCBSd295, October 2014

Page 25 of 28
7. MARKET LIQUIDITY RISK AND ITS RELATIONSHIP WITH
FUNDING LIQUIDITY RISK
Market liquidity risk is an important risk for capital market participants, such as proprietary
trading of commercial banks, investments banks, and corporate treasury functions. Market
liquidity risk has an important impact on funding liquidity risk. Such funding requirements
are driven by the following factors:

Funding requirement =
 Current position + (haircuts + initial margin) + contingency funding needs
 Collateral securities annexure (CSA) + collateral requirement + variation margin
 CSA VaR

 Haircuts and initial margins: When physical hedges like stocks and bonds are traded and
refinanced through SBL/Repo, a haircut is required depending on the underlying asset.
When futures are traded to hedge the exposure, an initial margin is to be posted to the
exchange.
 Contingency funding needs: It is expected that initial margins and haircuts might change
in times of stress.
 Collateral securities annexure (CSA), variation margin, and cash pool security-based
lending (SBL) collateral: These requirements are mainly due to market movements and
are to be “pledged” on a daily basis.

− CSA is a result of the movement in OTC derivatives

− Variation Margin is daily payable/receivable by the exchange
through clearing
− Cash pool SBL collateral is the daily posting of collateral when
stocks or bonds are used to hedge security exposure out of the
sold products to the clients

Page 26 of 28
8. CONCLUSION:
Liquidity risk can result from a variety of shocks, and it can materialize at any time. A sound
liquidity risk management policy along with constant monitoring of balance sheet dynamics
is an absolute must to navigate through the peaks & troughs of liquidity risk management.
Neglecting even a simple of critical item can lead to significant exposure to liquidity risk and
make a financial institution insolvent. Banks need to strike a balance between the numerous
business benefits of enhanced visibility of their liquidity risks and any so-called unintended
consequences. Banks with strong liquidity management processes in place and superior
visibility of available liquidity and exposures will march ahead of the curve but those banks
that continue to operate blind without visibility of their exposures, will be left behind the
curve and won’t understand what risks they are exposed to. Hence there is a constant need
to understand the liquidity risk management cycle & closely monitor balance dynamics.

Diligent monitoring and reporting are key to effectively managing liquidity risks and there is
always a need to identify and constantly evaluate the optionality of balance sheet items &
regularly review all behavioural risks inherent within the balance sheet. This requires a
strong risk management policy that is constantly monitored in addition to a strategic &
proactive balance sheet management.

But before banks can realise the strategic and operational benefits of monitoring their
liquidity risk, they need to overcome various challenges, the most difficult one is to capture
timely and quality data from their own internal IT and operational silos, as well as from
correspondent banks. Real-time liquidity & intra day liquidity reporting is already a reality &
overcoming the data & system challenges will be a key first step to start a robust liquidity
risk management strategy at any financial institution.

Page 27 of 28
ABOUT THE AUTHORS
Sidharth Reddy is Vice President and global head of Model Risk Management & Market Risk
practice in Genpact. He has 18 years of industry experience across the globe. In his career, he
has held front- and middle-office roles in various banks in USA and India. He has also consulted
with numerous commercial and investment banks across the globe in the field of ALM, FTP,
Liquidity Risk, and Market Risk Management. He has an MBA from Illinois State University, as
well as a Bachelor of Arts (Economics) and Master of Arts (Economics) from Loyola College,
Chennai, India.

Anirban Naskar is a Senior Manager in Genpact. He has eight years of industry experience. In
his career, he has worked in asset and liability management (ALM) and market risk
management for banks and consulting organizations. He holds a Bachelor of Technology (with
honors) as well as a Master of Technology from the Indian Institute of Technology, Kharagpur,
India, and received his Master of Management (MBA equivalent) from the Indian Institute of
Technology, Bombay, India. He is also a National Talent Scholar, awarded by the government
of India.

Sidharth Reddy Anirban Naskar

E: [email protected] E: [email protected]

Page 28 of 28
 Intraday Liquidity Around the World∗
Biliana Alexandrova Kabadjovaa Anton Badevb
Saulo Benchimol Bastos o Evangelos Benosc,d,†
Freddy Cepeda-Lópeze James Chapmanf,† Martin Diehlg
Ioana Duca-Raduh Rodney Garratti Ronald Heijmansj
Anneke Kossef Antoine Martink Thomas Nellenl
Thomas Nilssonm,n Jan Paulickg Andrei Pustelnikovc
Francisco Rivadeneyraf Mario Rubem do Coutto Bastoso
Sara Testih
September 13, 2021

Abstract
We study intraday liquidity usage and its determinants using a unique cross-country
data set on large-value payments. We document that the amount of intraday liquidity
that depository institutions around the world use each day equals, on average, 15%
of their total daily payment values or 2.8% of their countries’ GDP. We then define
and calculate system-level measures of liquidity efficiency and inequality in liquidity
provision. We show that these measures vary systematically with participants’ degree
of payment coordination, the quantity and opportunity cost of central bank reserves
and institutional characteristics such as incentives for early payment submissions and
Liquidity Saving Mechanism (LSM) design features. Our results are consistent with
payment system participants actively managing intraday liquidity and acting strategically
∗
Among the authors of this paper are members/alternates of the user groups with access to TARGET2
data in accordance with Article 1(2) of Decision ECB/2017/2080 of 22 September 2017 on access to and use of
certain TARGET2 data. The European Central Bank, the Deutsche Bundesbank, the Market Infrastructure
Board (MIB) and the Market Infrastructure and Payments Committee (MIPC) have checked the paper
against the rules for guaranteeing the confidentiality of transaction-level data imposed by the MIB pursuant to
Article 1(4) of the above mentioned issue. The views expressed in the paper are solely those of the authors and
do not necessarily represent the views of the authors’ affiliations, listed below, or any other person associated
with these institutions. The authors wish to thank Derek Brito, Adam Epp, Othón Moreno González, Aldo
Juárez Acevedo, Carlos León-Rincón, Clara Machado-Franco, Carlos Navarro Ramirez, Kristian Nørgaard
Bentsen, Fabio Ortega-Castro, Patrick Papsdorf, Michael Pywell, Matthieu Rüttimann, Teresa Samuel,
Takeshi Shirakami and Flurina Strasser for their contributions to the project.
a
Banco de México; b Federal Reserve Board; c Bank of England; d University of Nottingham; e Banco
f g h
de la República, Colombia; Bank of Canada; Deutsche Bundesbank; European Central Bank;
i j k
University of California Santa Barbara; De Nederlandsche Bank; Federal Reserve Bank of New York;
l
Swiss National Bank; m Danmarks Nationalbank; n Bank for International Settlements; o Bank of Brazil
†
Corresponding authors: [email protected], [email protected].

1
 in order to do so. System participants also appear to condition their payment behavior
on specific LSM characteristics, which may weaken some of the LSMs’ intended effects.

1 Introduction
Financial institutions manage their intraday liquidity to meet obligations that arise
during the day. These obligations are typically associated with (large) payments that
financial institutions make to one another, in central bank money, using dedicated electronic
networks collectively referred to as large-value payment systems (LVPSs). LVPSs in most
jurisdictions settle their participants’ payments on a gross basis, meaning that any payments
need to be pre-funded.1 While settlement of transactions on a gross basis helps to reduce
credit risk [Bech and Hobjin (2007), Kahn and Roberds (1998)] it is also liquidity-intensive
with intraday liquidity needs arising whenever there are timing mismatches between same-day
incoming and outgoing payments. These intraday liquidity needs can be sizeable and usually
much larger than banks’ net daily obligations, especially during times of stress. This became
evident during the financial crisis of 2008-09 by the collapse of Lehman Brothers.2 In
response, regulatory reforms since the financial crisis of 2008-09 have sought to address
risks arising from intraday liquidity shortfalls.3
This paper is the first to measure intraday liquidity usage across multiple jurisdictions
over a long period of time and study its determinants. For this purpose, we use proprietary
data on individual LVPS transactions, between financial institutions and other payment
system participants, in each jurisdiction. This data is used to measure, for each jurisdiction
and on a daily frequency, the aggregate amount of liquidity that institutions use to cover
their intraday obligations and also to describe overall payment activity. Our data spans
several major economies, which account collectively for more than 45% of the world’s GDP
and as such is highly representative of global payment activity and intraday liquidity usage.
The jurisdictions in our sample vary with respect to such institutional arrangements as the
terms under which the central bank provides intraday credit, the presence of incentives for
banks to settle their payments earlier in the day, as well as the presence and the particular
design features of liquidity saving mechanisms (LSMs).4 This allows us to compare and
1
For this reason such payment systems are also referred to as Real-Time Gross Settlement (RTGS)
systems.
2
According to the examiner’s post-mortem for Lehman, the company had a total liquid asset pool of $25
billion as of September 12, 2008. Of this, $16 billion was pledged as collateral with clearing and custodian
banks with the exclusive purpose of covering intraday liquidity needs. This meant that a substantial fraction
of Lehman’s liquid assets were encumbered and unavailable to meet other obligations, which on that day
exceeded Lehman’s available liquidity and led to its demise. See Valukas (2010).
3
Intraday liquidity risk is directly addressed by the “Principles for Sound Liquidity Risk Management and
Supervision” issued by the Basel Committee on Banking Supervision [BIS (2008)]. For instance, Principle
8 states that “A bank should actively manage its intraday liquidity positions and risks to meet payment
and settlement obligations on a timely basis under both normal and stressed conditions and thus contribute
to the smooth functioning of the payment and settlement systems”. For an overview of intraday liquidity
regulation see Ball et al. (2011).
4
LSMs are mechanisms that offset or net payments on a frequent basis during the day. As such, they
allow payment system participants to economize on intraday liquidity. The exact payment offsetting/netting
criteria may vary across LSMs. We provide more details on LSM design characteristics in Section 4.4.

2
contrast their impact on intraday liquidity usage. Furthermore, our data spans an eventful
17-year period (2003-20) that includes the financial crisis of 2008-09 and the subsequent
central bank interventions that took place in several of the jurisdictions of our sample.
We start our analysis by measuring, for each jurisdiction, aggregate amounts of
liquidity used by banks and other participants to cover their payment obligations during the
day. Participants can meet their obligations by either using their own reserves balances, or
by obtaining intraday credit from the central bank, or by recycling incoming payments from
other system participants.5 We find that the amounts of intraday liquidity that participants
use by tapping their reserves or by borrowing from the central bank, is economically significant.
For example, in the US, participants (mostly banks) with Fed reserve accounts collectively
use, during our sample period, on average, $630 billion each day, with a maximum value
of about $1 trillion. For the Eurosystem, the daily average and maximum values are $443
billion and $800 billion, respectively. On average, across jurisdictions and over our entire
sample period, participants use, each day, the equivalent of 15% of aggregate daily payment
values or about 2.8% of their countries’/jurisdictions’ GDP in order to cover their intraday
liquidity needs. These numbers are large and highlight the financial stability relevance of
intraday liquidity.
Given that participants often choose when to settle their payments on a given day,
and given that incoming payments can be used to fund outgoing ones, LVPSs may give rise
to strategic effects since recycling incoming payments allows participants to economize on
their own liquidity usage [Bech and Garratt (2003)]. For this reason, we calculate, for each
jurisdiction, a liquidity efficiency measure as in Benos et al. (2014), defined as the ratio of
aggregate payment values to aggregate intraday liquidity used. This measure captures the
degree to which intraday liquidity needs are met by payment recycling in a given LVPS.
We show that this ratio varies substantially across systems and over time. For example, for
every unit of intraday liquidity used, UK participants make on average 13 units worth of
payments, whereas participants in Denmark make 2.5 units worth of payments.
Similarly, given that payments are, to some extent, recycled in an LVPS, we examine
whether liquidity is provided by system participants in a manner proportional to their own
payment obligations, or whether most participants rely on just few other ones to make
payments first and supply liquidity to the LVPS for the rest to (re)use. This matters for
financial stability because a high degree of inequality in liquidity provision could mean
that the LVPS is reliant on just a few participants to function smoothly. Should these
liquidity-supplying participants choose to hoard on liquidity or otherwise not be able to
supply it, during times of stress, this could potentially affect the ability of the other participants
to meet their obligations. As such, we calculate a Gini coefficient of relative liquidity
provision as in Denbee et al. (2015) and show that this coefficient also varies over time
and across jurisdictions. Why do such differences in efficiency and inequality in liquidity
provision arise? This motivates the rest of our analysis.
Since intraday liquidity usage is a direct function of participant activity in an LVPS,
we construct two additional variables to capture key elements of this activity, namely the
timing and degree of coordination of payments. As such, we calculate, for each jurisdiction,
the value-weighted average settlement time as well as a measure of payment dispersion
5
This includes cases where participants obtain credit from one another.

3
across system participants. In panel regressions, estimated across days and jurisdictions, we
condition these activity variables on a number of regressors that include both time-varying
factors, such as the total amount of reserves available each day and the opportunity cost of
holding them, as well as time-invariant institutional characteristics, such as the presence of
incentives to settle payments early, the central bank intraday credit regime and the various
LSM design features, which pertain to the criteria and algorithms used to offset payments.6
Our cross-system data includes different combinations of LSM design characteristics which
allows us to compare them and assess their impact on intraday liquidity usage.
We find that participant activity correlates significantly with most of these variables
in a manner mostly consistent with theoretical predictions. For example, higher levels of
reserves balances are associated with earlier settlement times and reduced payment coordination
both of which are consistent with reduced participant incentives to economize on liquidity.7
On the other hand, increases in the opportunity cost of reserves are associated with later
settlement times as well as reduced payment coordination, which is consistent with liquidity
hoarding, particularly at times of stress.
Regarding institutional characteristics, although incentives to settle payments early
are associated with earlier settlement times, as one might expect, the rather unexpected
and thus most interesting finding here is that there is a stronger statistical correlation with
payment coordination. A potential explanation for this is that the incentives to pay early
also increase the incentives for payment coordination as participants may not wish to deviate
too much from average behavior, as this might stigmatize them. This might be especially
true if the incentives for early payment submission take the form of penalties to which
deviating participants might be liable to. The possibility to obtain intraday credit from
the central bank at a lower collateral cost is associated with reduced payment coordination,
also consistent with fewer incentives to economize on liquidity but paradoxically this is also
associated with later settlement times. This is surprising, as one might expect that a lower
collateral cost for central bank liquidity would decrease participants’ incentives to delay their
payments.
We also explore how our activity (i.e. payment timing and coordination) variables are
related to LSMs. Overall, the presence of an LSM queue correlates only weakly with these
variables; and while the effects are in a direction consistent with theoretical predictions they
are not statistically significant. However, specific LSM design features are strongly correlated
with our activity variables. For example, liquidity saving features of LSMs (such as the ability
of LSMs to bypass the priority of payments in the queue in order to maximize offsetting
benefits and also the ability to offset payments on a multilateral basis) are both associated
with earlier settlement times. This is consistent with theoretical predictions (Martin and
McAndrews (2008)) suggesting that LSMs reduce banks’ incentives to delay their outgoing
payments in anticipation of incoming ones. Some of our results also suggest that system
participants could potentially be conditioning their behavior on these LSM features in a way
that reduces (or negates) their intended effect. For instance, we find that the FIFO bypass
functionality, which allows LSMs to more flexibly offset payments, is associated with reduced
6
For more details, see BIS (1997) and BIS (2005).
7
This is also consistent with the findings of Bech et al. (2012) who look at the impact of reserve balances
on settlement times in the US Fedwire system.

4
payment coordination. One potential explanation of this is that, in its presence, participants
have less of an incentive to coordinate their payments.
In the last part of our analysis, we use our activity variables as regressors in specifications
where the dependent variables are our our measures of liquidity efficiency and inequality
in liquidity provision. Given that our activity variables may not capture all aspects of
LVPS participant behaviour, we also include in these specifications as regressors the same
time-varying and institutional variables described above. Consistent with theory, our results
show that efficiency is strongly and positively correlated with the degree of payment coordination.
This provides empirical support to the literature that casts LVPS interactions in a game-theoretic
setting (e.g. Bech and Garratt (2003)). Our results also suggest that the amount of available
liquidity as well as its opportunity cost affect intraday liquidity efficiency primarily via their
effect on payment coordination.
Institutional characteristics also matter for intraday liquidity efficiency. Incentives for
early settlement help banks economize on liquidity, largely by inducing them to coordinate
the timing of their payments. Thus, if payment incentives have the effect of inducing
coordination, as discussed above, this results in a clear improvement in liquidity efficiency.
Our results also show that intraday liquidity usage is correlated with the collateral costs
of central bank intraday credit. In regimes where intraday credit can be obtained on an
uncollateralized basis, or where the cost of collateral is lower, intraday liquidity efficiency is
also lower and this is almost entirely driven by the reduced coordination among participating
institutions that was also discussed earlier. Finally, the presence of an LSM in an LVPS is
statistically uncorrelated with our liquidity efficiency measure. This is consistent with our
earlier finding that LSMs are overall also uncorrelated with our activity variables. There
are a couple of potential non-exclusive explanations for this. First, a large segment of our
data overlaps with periods of ample reserves balances in many jurisdictions, as a result
of central bank QE programs. This means that incentives to economize on liquidity have
likely been lower in these jurisdictions with LVPS participants using LSMs less intensively
as a result.8 Alternatively, different LSM features are influencing participant behavior and
liquidity usage in different ways, thus potentially minimizing the overall intended effect of
LSMs. For example, in addition to our finding that participant payment coordination is
negatively correlated with the FIFO bypass functionality, we also find that it is positively
correlated with the multilateral offsetting one. Given that payment coordination strongly
correlates with liquidity efficiency, this suggests that the two LSM functionalities could be
related with coordination and liquidity efficiency in opposite ways, thus weakening the overall
relationship between the LSM and liquidity efficiency.
Regarding our measure of inequality in liquidity provision, we find that the Gini
coefficient is lower when the opportunity cost of liquidity increases, when reserves balances
are higher and when there are incentives for early payments in place. These results are
consistent with the idea that some participants may choose to hoard on liquidity when
liquidity is expensive and that an abundance of reserves alongside requirements for early
payments mitigate this problem.
Overall, our paper is the first to study, on a cross-LVPS basis, the relation between
intraday liquidity and institutional arrangements such as the central bank’s intraday credit
8
Unfortunately, we do not observe LSM usage and as such, we cannot empirically verify this.

5
regime, the presence of incentives for early payment submissions and the design of LSMs.
Such an empirical test requires data from multiple jurisdictions and our cross-country data
allows us to do precisely that.
The rest of the paper proceeds as follows: In Section 2 we review the literature; in
Section 3 we describe the data. In Section 4 we define and measure intraday liquidity usage
as well as payment system activity variables. This is followed in Section 5 by our empirical
analysis and results. We conclude and discuss future work in Section 6.

2 Literature review
The literature on intraday liquidity is relatively scarce not least because there are no
explicit intraday money markets and banks typically obtain intraday credit from the central
bank at a low cost. This is in contrast to overnight liquidity where in most jurisdictions,
central banks primarily rely on the overnight market to supply and allocate reserves. This
discrepancy, however, has motivated a number of theoretical studies on whether it is optimal
to supply liquidity via a market mechanism or via the central bank at a pre-determined,
fixed price. Freeman (1999) sets up a model where the asynchronous presence of borrowers
and lenders in the money market creates a need for liquidity which can be met by the central
bank either via a standing facility (at a pre-determined price) or by open market operations.
Abstracting from moral hazard, Freeman (1999) shows that open market operations enable
better risk sharing. Chapman and Martin (2013) extend this model to account for moral
hazard and show that open market operations continue to yield more efficient outcomes as
long as the central bank interacts with a select subset of banks that are themselves unaffected
by moral hazard. That way, the price of liquidity reflects information available to market
participants. While these arguments suggest that an explicit market for intraday liquidity
might be desirable, there are also arguments in favor of the current regime: A lower cost of
intraday liquidity, as typically provided by most central banks, reduces banks’ incentives for
delaying their payments and also protects them from costly intraday overdrafts that banks
could have faced in an explicit intraday money market. Martin and McAndrews (2010)
provide a comprehensive overview of these arguments. The unique data across jurisdictions
that our paper utilizes, allow us to directly test some of these theoretical predictions.
Although no explicit intraday money markets exist, a number of papers have looked at
whether such markets nevertheless implicitly exist through differentiated prices of overnight
loans with different settlement times. Studies that use data prior to the 2007-08 financial
crisis generally find small but positive implied intraday rates. For example, Furfine (2001)
estimates the hourly intraday unsecured rate in Fedwire to be around 0.9 bps whereas
Kraenzlin and Nellen (2010) estimate the Swiss hourly repo rate to be around 0.43 bps.
Similarly, Baglioni and Monticini (2008) detect economically small intraday rates in the
unsecured Italian e-MID market. However, these implied rates significantly increased during
the financial crisis. Baglioni and Monticini (2010) and Jurgilas and Zikes (2014) both
report more than ten-fold increases in implied unsecured intraday rates in the Italian e-MID
and Sterling markets respectively during the crisis. They argue that these rises reflect the
increased opportunity costs of pledging collateral with the central bank. While our paper
does not estimate implied intraday rates, it complements this literature by calculating the

6
aggregate quantities of intraday liquidity deployed in each jurisdiction and by examining
their determinants.
Our paper is also closely linked to the literature of payment system design, of strategic
behavior of participants within payment systems and of central bank policies regarding the
provision of intraday credit. On the theoretical side, Kahn and Roberds (2009) compare
and contrast the properties of pure RTGS systems with those that settle payments on a
net basis (also known as deferred net settlement or DNS systems). Their paper formalizes
the key tradeoff between RTGS and DNS systems, namely that while the latter are more
liquidity-efficient, they also give rise to intraday credit exposures among system participants
which may create moral hazard.
Given the prevalence of RTGS systems around the world, Bech and Garratt (2003)
describe the incentives and equilibrium strategies of their participants. Their key intuition is
that since incoming payments can be recycled in order to fund outgoing ones, this creates a
rich set of possible interactions between system participants, the nature of which depends on
the conditions under which the central bank provides intraday liquidity. Bech and Garratt
(2003) show that in a collateralized (and free of fees) credit regime, the strategies and
payoffs faced by RTGS system participants are those of a “prisoner’s dilemma” whereas in
a priced (and uncollateralized) credit regime, they are those of a “stag hunt”. The authors
make the key assumption that in a collateralized credit regime participants always bear a
collateral opportunity cost, whereas in a priced credit regime they only do so when their
payment requests are not coordinated and therefore not offset. Under this assumption,
delaying payments is always socially inefficient when credit is collateralized because the cost
of collateral is sunk. On the contrary, when credit is uncollateralized and priced, delays can
be efficient if they lead to liquidity savings. In all cases, Bech and Garratt (2003) consider
single-stage games, meaning that in the repeated versions of the “prisoner’s dilemma”, one
might expect efficient (cooperative) equilibria to arise. Building on their approach, Mills and
Nesmith (2008) study payments equilibria when participants strategically interact both in
payment and security settlement systems. Their analysis suggests that settlement risk may
lead to late-day concentration of payments and, moreover, in the presence of settlement risk,
an overdraft fee can have a greater impact on the concentration of transactions. Nellen (2019)
also studies the intraday liquidity management game in the presence of credit regimes with
fixed and variable cost focusing on different designs of intraday liquidity facilities provided
by a central bank and associated incentives (or disincentives) for early settlement. In his
model, a variable cost credit regime leads to late settlement as it is assumed to have a
positive marginal cost, in contrast to a regime with fixed credit cost which has zero marginal
cost. A fixed cost credit regime then eliminates incentives to coordinate payments provided
that intraday liquidity borrowed is available until the end of the day. In that scenario, a
strictly positive transaction fee for late settlement will incentivize early settlement. Finally,
Bech and Garratt (2012) theoretically show that a wide-scale disruption (caused either by
operational outages or credit events) can lead to a breakdown in coordination among RTGS
system participants and thus lead to an increase in the amount of intraday liquidity used.
The empirical evidence from a number of different jurisdictions suggests that banks
coordinate their payments to some extent in order to economize on intraday liquidity. For
instance, McAndrews and Rajan (2000) and Becher et al. (2008) show that in the US and UK
RTGS systems respectively, banks use incoming payments to fund outgoing ones. Becher

7
et al. (2008) attribute the high level of payment recycling in the UK system to its small
membership and the throughput rules that are in place which require banks to make a
certain amount of payments by various points of time during the day. However, payment
coordination between RTGS system participants is not always a given and can break down
due to some external event. Several papers have examined actual coordination failures
and their impact. McAndrews and Potter (2002) document that in the days following the
September 11, 2001 terrorist attacks, payment coordination between Fedwire participants
dropped substantially, resulting in increased liquidity usage and triggering a short-term
liquidity injection by the Fed.9 Bech and Garratt (2012) and Benos et al. (2014) document
coordination failures in the wake of Lehman’s default, in the US and UK RTGS systems
respectively. In both cases, payments were delayed with the evidence from the UK RTGS
system further suggesting that these delays were targeted at banks with perceived higher
credit risk.
Finally, our paper is also related to the literature on liquidity saving mechanisms
(LSMs). This literature generally examines theoretically how the presence of an LSM
affects the tradeoff between the cost of delaying payments and the cost of obtaining intraday
liquidity in order to settle payments early. This typically involves a game-theoretic setup
which captures the impact of an LSM on RTGS system participants’ incentives. In this
respect, Martin and McAndrews (2008) show that an LSM reduces banks’ incentives to delay
payments and this leads to payments being made earlier on average which, in turn, improves
liquidity efficiency and welfare. However, there can also be instances where welfare is reduced
in the presence of an LSM. This happens because LSMs reduce the degree of strategic
complementarity relative to a pure RTGS system, since its offsetting functionality reduces
participants’ incentives to coordinate their payments. When a high degree of payment
coordination is desirable (e.g. when a few participants have to make large payments that are
urgent and cannot be queued) welfare is reduced because banks are forced to obtain intraday
liquidity from the central bank at a cost. Jurgilas and Martin (2013) argue that such cases
do not arise in collateral-based credit regimes where it is assumed that the opportunity cost
of pledging collateral with the central bank is low. Evidence of strategic complementarity
is also provided by Nellen et al. (2018) who show that as LSM queuing times in the Swiss
RTGS were reduced, as a result of higher settlement balances, payment submissions into
the queue were delayed thus offsetting the reduced queuing time. Our paper complements
this literature by studying the impact of particular LSM design features on the incentives
and behavior of RTGS system participants and how these features ultimately affect intraday
liquidity usage.

3 Data
The primary source of data in our paper are payment messages from the large-value
payment systems (LVPSs) of nine different jurisdictions: Brazil, Canada, Colombia, Denmark,
the Eurosystem, Mexico, the United Kingdom, the United States and Switzerland. The
payments in our data are typically large in value and are made among financial and other
9
For instance, McAndrews and Potter (2002) report that the ratio of daily payment values to reserve
balances dropped from more than 100 before September 11 to only 18 on September 14, 2001.

8
institutions with access to central bank reserve accounts. As such, they are made using
central bank reserve balances in the local currency and, once processed, are final and
irrevocable. Typical payments that are settled via LVPSs include unsecured wholesale
money market loans and foreign exchange transactions. Many LVPSs also settle the cash
legs for wholesale repo and securities transactions. LVPS systems are also regularly used to
settle margin payments to clearing houses and support other payment systems to settle net
obligations. These payments can be made or received either on the system participants’ own
account or on behalf of their customers. Furthermore, some LVPSs are also used to settle
retail transactions.10
These data are available to central banks operating and/or overseeing their respective
payment systems and contain information on the identities of payers and payees as well as
the value, date and settlement time of the payments being made.11 These granular data
are confidential and for this reason are aggregated, in this study, across LVPS participants
and used to construct, on a daily frequency, the variables described in Section 4. Since the
aggregated data do not contain participant-specific information, they can be shared and put
together to form a panel data set. Our data cover the period from 2006 to 2018 but data from
some systems cover only sub-periods within this time range, due to availability constraints.
Table 1 shows the data time range available for each jurisdiction, along with the number of
daily observations and the local currency of payment denomination. Our final panel consists
of 21,544 observations.

Table 1: Payment data sources. This table shows the large-value payments systems included in
our study, their jurisdiction, their number of daily observations, their data range as well as the
currency in which payments are denominated.

System name Jurisdiction N First date Last date Currency

CHAPS United Kingdom 3148 2006-01-03 2018-06-18 GBP
CUD Colombia 2444 2008-07-01 2018-06-29 COP
Fedwire United States 2523 2008-06-02 2018-06-26 USD
Kronos Denmark 3120 2006-01-03 2018-06-29 DKK
LVTS Canada 3170 2006-01-03 2018-07-31 CAD
SIC Switzerland 2568 2008-06-03 2018-07-30 CHF
SPEI Mexico 1967 2008-06-02 2016-06-29 MXN
STR Brazil 4650 2003-01-02 2020-12-31 BRL
TARGET2 Eurosystem 2604 2008-06-02 2018-07-31 EUR
10
One such system is the Swiss SIC. Nellen et al. (2018) show that retail payments in SIC help recycle
settlement balances thus contributing to increased liquidity efficiency.
11
In the presence of an LSM, payment settlement times are different from payment submission times. The
latter refers to the point in time that a payment is submitted to the LSM queue. Banks receiving instructions
by clients to make payments on their behalf, may also choose to settle these payments at a later time during
the day. In both cases, we only use in our analysis settlement times as only these are consistently observed
across systems.

9
 To construct our variables of interest, we apply a set of filters on the raw payments
data in a consistent manner across systems. Given our focus on intraday liquidity, we only
use, as a general rule, transactions that affect a participant’s liquidity position, that is, the
funds available across a participant’s accounts to make RTGS payments at a given time.
This includes all interbank payments between settlement banks, payments that banks settle
on behalf of their customers, as well as liquidity transfers to and from accounts reserved
for ancillary systems.12 Similarly, central bank transactions with system participants are
included if they alter the participants’ liquidity position, but are excluded if they are purely
administrative or technical in nature.13 Finally, central banks and ancillary systems such
as CLS and securities settlement systems are out of scope of our analysis, as stand-alone
entities, since they do not behave strategically as credit institutions may do. For this reason,
we do not count them as system participants and do not examine their activity, although
their interactions with the other system participants are within scope and included in our
data.
Finally, we complement our payments data with information on the number of participants
active in each RTGS system, the aggregate value of central bank reserve balances and the
overnight unsecured interbank borrowing rate (or alternatively the central bank policy rate)
as a proxy for the opportunity cost of liquidity.

4 Variables and summary statistics

4.1 Payment system activity and intraday liquidity
We start our analysis by calculating, for each system, the aggregate value of payments
made and the amount of intraday liquidity used. Let t ∈ {1, 2, ..., T } be a time partition of
the daily business hours of each payment system. Let also xi,j s (t) be the value of payments
sent by participant i to participant j on day s and in time interval t. Then, the total value
of payments made in that system on day s, is:
X
Ps ≡ xi,j
s (t) (1)
i,j,t

Given a participant’s incoming and outgoing payments, the amount of intraday

liquidity used by that participant is equal to the amount of liquidity that the participant
needs to have in place, in the form of reserves or intraday credit from the central bank, in
order to meet its payment obligations for the day. This is the maximum cumulative net
debit position that this participant attains during the day. Using the above notation, the
net debit position of participant i, on day s and at time t is:
t X
X
Nsi (t) ≡ xi,j j,i
s (k) − xs (k)
k=1 i6=j

12
Additional information on the ancillary systems linked to the LVPS in each jurisdiction, is provided in
the Appendix.
13
For example, a repayment of an overnight central bank loan is included whereas a liquidity transfer
between two central bank reserve accounts held by the same participant is excluded.

10
 and therefore the amount of intraday liquidity used by this participant for the day is:

Lis ≡ max{Nsi (t), 0}

t

The aggregate amount of intraday liquidity used in the payment system is then the
sum, across system participants, of the individual amounts of own liquidity used:
X
Ls = Lis (2)
i

Figure 1 plots the aggregate values of payments made and liquidity used for each
system in our sample. To facilitate comparison across systems, all values are reported in
USD. Daily aggregate payment values are large and measured in the trillions of USD for
the larger systems (Fedwire and TARGET2) and in the tens or hundreds of billions for the
other systems. It is notable that payment values are elevated at times and in jurisdictions
experiencing financial stress. This is evident, for example, in Fedwire (US), CHAPS (UK)
and Kronos (Denmark) during the financial crisis of 2007-09 and in TARGET2 (Eurozone)
during the European sovereign debt crisis of 2011-13. This likely reflects increased activity in
financial markets as a result of higher market volatility. For instance, some LVPSs settle the
cash leg of security transactions which tend to increase in volatile conditions, while LVPSs
may also facilitate margin payments, either between market participants directly or via a
clearing house, which also increase with market volatility.
The amount of intraday liquidity used in each system is also economically significant.
For instance, in the US, the amount of intraday liquidity used in Fedwire fluctuates around
$630 billion daily with values reaching as high as $1 trillion. The corresponding values for
Eurozone’s TARGET2 are $443 and $800 billion respectively. Across systems and over our
sample time, daily intraday liquidity usage accounts for about 2.8% of GDP.
It is also evident that the amount of liquidity used is invariably lower than the total
value of payments made as system participants may use incoming payments to fund outgoing
ones, thus forgoing the need to use their own liquidity for this purpose. Although in all
systems intraday liquidity used is highly correlated with the value of payments, as one
would expect, the ratio of the two varies substantially across systems and over time. This
motivates our measure of intraday liquidity efficiency.

4.2 Efficiency and inequality in intraday liquidity usage

Following Benos et al. (2014), we define intraday liquidity efficiency, at the system
level, to be the ratio of aggregate payment values over aggregate intraday liquidity used. In
terms of our notation this is expressed as:
Ps
Qs ≡ (3)
Ls
This ratio captures the value of payments that are made for each unit of intraday
liquidity used. If system participants can meet their daily payment obligations with minimal
liquidity usage, the ratio takes on higher values and the system is considered to be more
liquidity-efficient. As mentioned earlier, liquidity efficiency will depend on the degree to

11
Figure 1: Daily aggregate values (in USD billions) of payments made (P ) and liquidity used (L)
by payment system. The two variables are defined in equations (1) and (2) respectively.

Fedwire TARGET2

5000
Payments and ID Liquidity Used (USD bn)
4000
3000
2000
1000
0

06 08 09 11 13 14 16 18 06 08 09 11 13 14 16 18

Payments ID Liquidity Used

(a) “Large” systems

CHAPS CUD Kronos

500 1000
Payments and ID Liquidity Used (USD bn)
0

LVTS SIC SPEI

500 1000
0

06 09 11 13 16 18 06 09 11 13 16 18

STR
500 1000
0

06 09 11 13 16 18

Payments ID Liquidity Used

(b) “Smaller” systems

12
which payments are recycled as well as on the extent to which payment obligations are
matched and offset via an LSM.14 Figure 2 plots daily values of this measure for each
system. For most systems (CUD, Fedwire, SIC, TARGET2) liquidity efficiency takes on
values between four and six whereas CHAPS and Kronos appear to have higher and lower
values, than the other systems, respectively. In all cases, there is variation over time with
some systems (e.g. CHAPS, SIC) exhibiting higher variability than others. A key goal of
our study is to understand why this variability arises and what makes systems more or less
liquidity efficient.
Given that payments settle on a gross basis in all systems, liquidity efficiency is
ultimately determined by the extent to which payments are coordinated and thus recycled
in each system and over time. This assumes that system participants have the option to time
and coordinate their payments and thus economize on liquidity. Whether they choose to do
so should, in turn, depend on the cost of liquidity usage as well as the easiness of coordinating
their payments. The former will likely depend on the unit opportunity cost (and available
quantity) of reserves balances whereas the latter could depend on the presence of an LSM
which is intended to incentivize the early submission of payments.
In our empirical analysis we explore these potential determinants of cross-sectional
and time variability in liquidity efficiency. Our prior is that liquidity efficiency will increase
with payment coordination and that payment coordination will itself depend on participants’
incentives to coordinate. For instance, we expect that an increase in reserves balances will
tend to reduce participants’ incentives to coordinate their payments and thus negatively
affect liquidity efficiency. Given that several jurisdictions in our sample engaged in quantitative
easing programs that drastically increased the amount of aggregate reserves, it could be for
this reason that efficiency is trending downward in the UK, the US and the Eurozone.
The opportunity cost of reserves is another potential determinant of efficiency though
it is not a-priori clear what its effect could be. On the one hand, a higher opportunity cost
could incentivize system participants to coordinate their payments, leading to higher liquidity
efficiency. On the other hand, it could incentivize them to hoard on liquidity which could
lead to a breakdown of coordination and a drop in efficiency.
Finally, we expect that payment coordination and efficiency will likely also depend on
a number of institutional features such as incentives to submit payments early, the collateral
cost of obtaining intraday credit from the central bank, the degree of tiering and the presence
of an LSM with its various design features. As such, we expect that a lower collateral cost
of intraday credit will dis-incentivize payment coordination (and reduce efficiency) whereas
incentives for early payments will likely have the opposite effect. We are otherwise agnostic
as to what is driving the sizeable time and cross-sectional variation in liquidity efficiency
shown in Figure 2.
Finally, we measure how intraday liquidity usage is distributed across payment system
participants and in particular whether it is proportional to the value of payments made by
each participant. This is our measure of inequality in intraday liquidity usage. For this, we
follow Denbee et al. (2015) and calculate the Gini coefficient over relative liquidity usage
14
A system with a higher value for liquidity efficiency, as defined in this paper, does not imply that this
system is overall “superior” to one with a lower value. This is because there are additional LVPS features
that matter (e.g. settlement times) that are not captured by the liquidity efficiency measure.

13
Figure 2: Liquidity efficiency (Q) across systems. Liquidity efficiency is defined in equation (3).

CHAPS CUD Fedwire

30
20
10
0
Liquidity efficiency (Q)

Kronos LVTS SIC

30
20
10
0

SPEI STR TARGET2

30
20
10
0

06 09 11 13 16 18 06 09 11 13 16 18 06 09 11 13 16 18

(or liquidity cost) across participants in each jurisdiction. Let Psi ≡ j,t xi,j
P
s (t) be the total
i
value of payments made by participant i on day s and let Ls be the participant’s amount of
intraday liquidity used (as defined above). Then, we define the relative liquidity usage (or
liquidity cost) of that participant to be:
Lis
lsi ≡
Psi
and the Gini coefficient of relative liquidity usage across participants is then the
volume-weighted average of the pairwise differences in relative liquidity usage:
!
1 XX
Gs = 2
mi mj |lsi − lsj | (4)
2M µ i j

where mj is the number of payments made by participant j, M is the total number of

payments made by all participants, µ is the average relative liquidity cost of all payments
and lj is the relative liquidity usage of participant j. The Gini coefficient takes a minimum
value of zero when each participant uses the same amount of intraday liquidity relative to
their payments. On the other extreme, it takes a value of one when all of the intraday
liquidity cost is incurred by a single participant. In Figure 3 we plot daily values of the
Gini coefficients, over our sample period, for each jurisdiction. The Gini coefficient varies
substantially over time and also across systems. In our empirical analysis we explore the
determinants of this variability.

14
Figure 3: Inequality in intraday liquidity usage (G) across systems. Inequality is captured by the
Gini coefficient of relative liquidity usage defined in equation (4).

CHAPS CUD Fedwire

1.00
0.50
Inequality in Liquidity porvision (G)
0.00

Kronos LVTS SIC

1.00
0.50
0.00

SPEI STR TARGET2

1.00
0.50
0.00

06 09 11 13 16 18 06 09 11 13 16 18 06 09 11 13 16 18

4.3 Payment timing and dispersion

Given that intraday liquidity usage depends on the degree of payment recycling which,
in turn, depends on payment timing, we construct two additional variables to capture the
value-weighted average settlement time and the degree of payment coordination or, inversely,
dispersion. Let Ps (t) be the aggregate value of payments made on day s and in time interval
t. Then the value-weighted average settlement time on that day is:
PT
tPs (t)
Ts ≡ Pt=1T
(5)
t=1 Ps (t)
where the denominator is the total value sent through the system on day s. To quantify
payment coordination, we first calculate, for a given system, the ten deciles of daily payment
timing. Payment decile Ds (d) on day s is defined as:
Pk
Ps (t)
Ds (d) ≡ arg min Pt=1
T
−d≥0
k
t=1 Ps (t)

15
which is the earliest point in time during the day by which a fraction d of total daily payment
value has been made.15 Our measure of payment dispersion is then defined as:
1
Tdiffs ≡ [Ds (0.7) + Ds (0.8) − Ds (0.2) − Ds (0.3)] (6)
2
This variable can be interpreted as a proxy for system participants’ degree of payment
coordination with higher values implying a smaller degree of coordination and vice versa.
Since it is based on payment value deciles, our dispersion measure does not depend on the
time of the day that coordination in payments might take place. However, since we only
observe payment settlement times and not payment submission times, this variable is a noisy
proxy of the degree to which payment submission is coordinated among participants.
Both the value-weighted average settlement time and payment dispersion are plotted,
for each system, in Figures 4 and 5. Average settlement time is relatively stable for most
systems over the longer run and centered around the middle of each system’s business hours.16
However, there appears to be a downward trend for some systems (LVTS, Fedwire, SIC).
Our prior is that payment timing will mainly be influenced by the quantity and opportunity
cost of reserves. The more reserves are available and/or the lower their opportunity cost,
the earlier payments will be settled as participants will have less of an incentive to delay
their payments to economize on liquidity. The increase in reserves balances as a result of
quantitative easing programs in Canada, the US and Switzerland may be the reason for the
downward trends in timing that we observe in their systems.
Regarding our dispersion measure, this too varies substantially across systems and
over time. We hypothesize that dispersion will generally increase with the quantity of reserves
balances as in those cases there will be less of a need for system participants to economize
on liquidity by coordinating their payment submission times. On the other hand, it is
less clear what the relationship between dispersion and the opportunity cost of reserves
should be. An increase in the opportunity cost could incentivize participants to coordinate
their payments more or it could incentivize them to hoard on liquidity thereby increasing
dispersion. Finally, while payment dispersion will depend to some degree on participants’
incentives to coordinate their payments, it will also be influenced by exogenous factors beyond
the control of participants. For example, payments to and from ancillary systems (e.g.
securities settlement systems) often take place at pre-determined times.

15
These deciles are similar to Table 1 and Chart 6 in Armantier et. al. (2008). This would be the system
equivalent to Intraday throughput [C(i) of table 1 in BCBS (2013)]
16
The only exception to that is SIC where average settlement time appears to be later in the day. This
happens because SIC has much longer business hours with the system opening in the afternoon of the
previous business day.

16
Figure 4: Value-weighted average settlement time (T ) across systems. Value-weighted average
settlement time is defined in equation (5).

CHAPS CUD Fedwire

1.00
0.50
0.00
Average Settlement Time (T)

Kronos LVTS SIC

1.00
0.50
0.00

SPEI STR TARGET2

1.00
0.50
0.00

06 09 11 13 16 18 06 09 11 13 16 18 06 09 11 13 16 18

Figure 5: Payment dispersion (Tdiff ) across systems. Payment dispersion is defined in equation
(6).

CHAPS CUD Fedwire

0 .2 .4 .6 .8
Payment Dispersion (Tdiff)

Kronos LVTS SIC

0 .2 .4 .6 .8

SPEI STR TARGET2

0 .2 .4 .6 .8

06 09 11 13 16 18 06 09 11 13 16 18 06 09 11 13 16 18

17
Table 2: Summary statistics. This table shows summary statistics, by payment system, for the key
variables of interest. The sample properties for each system are summarised in Table 1. Variables
P (in USD bn), L (in USD bn), Q, G, T and Tdiff are defined in equations (1)-(6). IBOR (in %) is
either the unsecured overnight interbank rate or the central bank policy rate. Reserves (in $ bn) is
the total size of reserve balances held with the central bank by payment system participants. The
table continues on the next page.

CHAPS (UK) Fedwire (US)

mean sd min max mean sd min max
P (USD bn) 388.33 80.45 186.45 867.14 2734.88 460.16 624.01 4440.73
L (USD bn) 31.81 6.98 11.96 77.34 629.54 172.47 173.26 1087.55
Q 12.55 2.78 5.30 26.90 4.55 0.98 2.84 9.10
G 0.22 0.04 0.09 0.47 0.48 0.09 0.34 0.85
T 0.50 0.03 0.40 0.90 0.60 0.06 0.45 0.78
Tdiff 0.38 0.03 0.07 0.53 0.51 0.07 0.26 0.60
Members 17.42 3.64 13 28 5628.18 340.23 4800 6184
IBOR(%) 1.53 1.97 0.16 6.49 0.39 0.50 0.04 2.97
Reserves (USD bn) 241.10 100.13 62.40 450.42 1718.59 690.37 1.88 2699.97

LVTS (Canada) Kronos (Denmark)

mean sd min max mean sd min max
P (USD bn) 139.21 28.10 4.44 267.66 43.76 21.64 14.48 192.16
L (USD bn) 16.73 3.56 1.67 47.79 18.71 13.21 6.01 97.95
Q 8.46 1.77 2.09 15.84 2.52 0.40 1.35 3.90
G 0.33 0.07 0.12 0.61 0.31 0.09 0.1 0.94
T 0.55 0.03 0.41 0.70 0.43 0.08 0.23 0.69
Tdiff 0.45 0.03 0.28 0.73 0.42 0.10 0.20 0.86
Members 15.91 0.80 15 17 109.01 15.04 86 130
IBOR(%) 1.51 1.35 0.23 4.55 1.09 1.71 -1.98 6.97
Reserves (USD bn) 240.67 49.58 119.07 570.97 10.02 10.86 -20.46 61.45

CUD (Colombia) TARGET2 (Eurosystem)

mean sd min max mean sd min max
P (USD bn) 16.36 3.43 0.22 36.63 1835.55 444.71 774.12 4931.80
L (USD bn) 3.14 0.85 0.13 12.37 442.63 74.03 261.98 799.14
Q 5.36 1.04 1.64 10.65 4.13 0.56 2.17 6.71
G 0.52 0.06 0.28 0.82 0.49 0.04 0.37 0.67
T 0.51 0.03 0.32 0.60 0.52 0.01 0.47 0.60
Tdiff 0.25 0.05 0.00 0.47 0.53 0.04 0.43 0.63
Members 132.55 8.81 12 161 969.13 99.04 747 1076
IBOR(%) 5.02 1.82 2.95 10.05 0.36 0.97 -0.37 4.60
Reserves (USD bn) 8.99 2.04 4.14 14.37 545.11 417.41 88.34 1718.05

4.4 Institutional characteristics and LSM design features

Payment systems around the world have in place different institutional arrangements
that can affect intraday liquidity usage. In this paper we study two key arrangements: the
presence of incentives for LVPS participants to settle their payments early and the conditions
under which central banks provide intraday liquidity. The details of these are as follows:
• Early payment incentives: Incentives to pay early may affect payment timing and
intraday liquidity usage. Incentives to settle payments earlier rather than later are

18
 Table 2 continued

SIC (Switzerland) SPEI (Mexico)

mean sd min max mean sd min max
P (USD bn) 122.65 31.05 41.48 266.26 84.79 16.99 16.14 145.18
L (USD bn) 6.60 6.88 1.12 53.74 9.51 2.65 3.12 55.36
Q 4.31 1.97 2.40 16.51 9.22 2.05 1.75 21.00
G 0.47 0.08 0.20 0.76 0.48 0.11 0.00 0.79
T 0.69 0.08 0.51 0.92 0.43 0.06 0.16 0.80
Tdiff 0.19 0.12 0.02 0.62 0.30 0.10 0.00 0.68
Members 363.92 14.44 331 387 49.84 2.26 48 53
IBOR(%) -0.17 0.61 -1.58 3.00 4.46 1.37 3.00 8.25
Reserves (USD bn) 274.63 174.04 3.07 530.01 19.09 1.37 14.83 23.04

STR (Brazil) Pooled Statistics

mean sd min max mean sd min max
P (USD bn) 26.26 12.30 3.03 92.78 649.92 968.71 0.22 4931.80
L (USD bn) 8.14 3.89 0.78 31.23 139.06 234.95 0.13 1087.55
Q 3.28 0.76 1.30 8.77 6.44 3.67 1.35 26.90
G 0.38 0.11 0.11 0.80 0.4 0.13 0 0.94
T 0.52 0.04 0.35 0.71 0.53 0.10 0.16 0.92
Tdiff 0.40 0.06 0.10 0.66 0.39 0.12 0.00 0.86
Members 172.69 32.07 138.00 273.00 859.89 1765.92 12 6184
IBOR(%) 11.68 4.91 1.90 26.35 3.39 4.75 -1.98 26.35
Reserves (USD bn) 15.13 6.56 4.71 41.00 374.83 591.46 -20.46 2699.97

N = 26065

typically motivated by the fact that LVPS participants have an inherent incentive
to delay their payments so as to economize on their liquidity by recycling incoming
payments. They may also be motivated by operational risk considerations in that the
earlier payments are released and settled during the day, the less likely they are to be
affected and potentially delayed by an operational incident later on in the day. These
incentives can take different forms ranging from throughput rules, which require banks
to make a certain amount of payments by various points of time during the day, to
late settlement fees.

• Central bank credit regime: The conditions under which central banks provide intraday
liquidity to payment system participants vary across jurisdictions. The two ways
for doing so is either on a collateralized or an uncollateralized basis. In the first
case, system participants need to pledge collateral with the central bank to cover
the full amount of intraday liquidity that they may obtain. In the latter case, there
are arrangements in place that allow system participants to obtain intraday liquidity
from the central bank either on an uncollateralized basis or at a substantially lower
collateral cost. This may include collateral pooling or more generally the ability to
use collateral at low (or zero) marginal cost. For example, system participants may
use unencumbered collateral pledged with the central bank for term funding in order
to obtain intraday credit. The opportunity cost of pledging collateral will also depend

19
 on whether “double duty” is permitted or not.17 If it is permitted and banks can
borrow funds by pledging collateral that they have to hold, in any case, for regulatory
purposes, then the marginal cost of obtaining intraday liquidity, is zero.

An additional important determinant of intraday liquidity usage is the presence of

Liquidity Saving Mechanisms (LSMs) and their specific design features. LSMs are queuing
mechanisms that store, prioritize and offset payments. To offset payments, LSMs typically
settle participants’ outgoing payments if and when these are fully or partially offset (or
matched) by incoming ones. This implies that participants’ liquidity is fully or partially
replenished after each payment is made, which ensures that participants do not have to use
large amounts of their own liquidity, or borrow them from the central bank, in order to make
their daily payments. This, in turn, should incentivize participants to submit their payments
earlier on as the LSM removes the need for participants to wait for incoming payments before
submitting their own. In addition, LSMs may facilitate netting of any partially or fully
offsetting payments, further reducing the liquidity burden of these payments.18
LSMs typically run multiple matching cycles during the day but they can differ
significantly across payment systems in terms of their matching frequencies, the algorithms
and criteria being used to facilitate matching, whether or not these criteria can be bypassed
by system participants, etc. Furthermore, the extent to which LSMs are used by participants
to settle payments may also vary substantially across systems. Overall, the institutional
characteristics and specific LSM design features that we include in our analysis are:

• FIFO bypass: This characterizes the priority rules by which payments are processed in
an LSM queue. Typically, payments submitted first in the queue will also be processed
first on a first-in-first-out (FIFO) basis. However, some LSM matching algorithms may
apply exceptions to this priority rule in order to maximize liquidity efficiency. If, for
example, the first payment in the queue of one LVPS member is similar in size to the
third payment in the queue of one of its counterparties, then an LSM that can bypass
the FIFO protocol would be able to match these two payments despite the fact that
for the counterparty this is the third payment in the queue.

• Multilateral offsetting: The simplest way to offset payments in an LSM queue, between
system participants, is do so on a bilateral basis. That is, payments in the queue
from participant A to participant B can only be offset against payments submitted by
participant B and intended to credit participant A. Some LSMs however, may also offset
payments on a multilateral basis which means that as long as the gross outstanding
payment value between a subset of participants is larger than the net outstanding
value, then these payments can be offset. Multilateral offsetting improves liquidity
efficiency as it enables netting for a wider set of payment queue configurations.
17
“Double duty” is a bank practice of using regulatory liquid asset buffers (typically measured as at the
end of the day) to support intraday payment system activity. In the run-up to the financial crisis of 2008-09,
double duty was permitted in several jurisdictions around the world. For more details, see Ball et al. (2011).
18
When netting is possible and payments are only partially offset, the LSM ensures that payments are
only released when there is sufficient liquidity to cover the net outstanding obligation.

20
 • Priority setting: Payments submitted to an LSM may have to wait in the queue to
be settled. LVPS participants however may wish to expedite specific time-sensitive
payments whose delay would otherwise be costly. For this reason, some LSMs make it
possible to alter the priority of payments already submitted in the queue.19

• Liquidity reservations: In some LVPSs with an LSM, participants can also reserve
liquidity to make payments outside the LSM. Typically this is liquidity earmarked for
the most urgent payments.
Panel A of Table 3 lists these institutional characteristics and LSM features for which
we construct dummy variables to use in our empirical analysis. Panel B of the same table
shows how these characteristics vary across jurisdictions. All systems included in our sample,
except Fedwire, feature LSM queuing - central queues that allow for a varying set of system
features listed above, whereas CHAPS introduced one during our sample period, in April
2013. However, there is more variation in institutional characteristics and LSM features
across systems, which allows us to empirically compare and contrast their effects on intraday
liquidity usage.

5 Empirical analysis
Our empirical analysis proceeds in two steps. First, we examine the determinants
of payment timing and coordination as measured by the T and Tdiff variables defined in
equations (5) and (6) respectively. These two variables are intended to capture the most
important decisions that payment system participants can make: when to submit their
payments and whether to coordinate with other participants in doing so. Thus, we are
interested in examining whether these decisions are correlated with market-wide variables
such as the opportunity cost of reserves and the aggregate amount of available liquidity but
also with the institutional characteristics of each jurisdiction as summarized in Table 3. For
this reason, we first estimate a number of models where timing and dispersion are treated
as dependent variables with the market-wide ones and the institutional characteristics as
independent.
In the second step, we examine the determinants of liquidity efficiency (defined in
equation 3) and inequality of liquidity provision across participants (defined in equation 4).
In principle, both of these variables are entirely determined by the payment patterns of
system participants and for this reason, our timing and dispersion metrics are now included
as explanatory variables in these specifications.
A couple of caveats are however in order. First, our timing and dispersion variables
are imperfect proxies of participants’ payment patterns. For instance, if payments were
highly coordinated at a few points of time that are relatively far apart (e.g. early and
late on the business day), our dispersion metric would fail to capture that and instead
would take on a relatively higher value. Similarly, it could be the case that for liquidity
efficiency, specific timing percentiles matter as much (if not more) as average settlement
19
System participants also have the option to immediately settle their payments outside the LSM queue.
However this is a common feature across all systems and as such we cannot use it to draw cross-system
comparisons.

21
Table 3: Institutional characteristics and LSM design features. This table describes the various
institutional characteristics and LSM design features that we study, shows the relevant dummy
variables that we construct and also the values that these variables take for each jurisdiction in our
sample. The date range for each system is shown in Table X. In the UK an LSM was introduced
on April 22, 2013. In Switzerland, the ability to reserve liquidity was introduced on June 18, 2016.

Panel A: Dummy variable definitions

Dummy variable Definition
Institutional characteristics
Incentives Equals 1 if there are in place incentives or requirements for settling payments early
Credit Equals 1 if the central bank can provide intraday credit on an uncollateralized basis
or at a lower collateral cost (e.g. via collateral pooling)

LSM design features

LSM Equals 1 if there is an LSM in place
FIFO bp Equals 1 if the LSM allows for the FIFO protocol to be bypassed
Offsetting Equals 1 if the LSM enables multilateral offsetting
Priority Equals 1 if it is possible to change the priority of payments in the LSM queue
Reservations Equals 1 if it is possible to reserve liquidity for payments outside the LSM

Panel B: Dummy variable values by jurisdiction

System name Jurisdiction Incentives Credit LSM FIFO bp Offsetting Priority Reservations
CHAPS United Kingdom 1 0 0/1 0/1 0/1 0/1 0/1
CUD Colombia 1 0 1 1 1 1 0
Fedwire United States 0 1 0 0 0 0 0
Kronos Denmark 1 0 1 1 1 1 0
LVTS Canada 1 1 1 0 1 0 0
SIC Switzerland 1 0 1 0 0 1 0/1
SPEI Mexico 0 0 1 0 1 0 0
STR Brazil 0 0 0/1 0/1 0/1 0/1 0
TARGET2 Eurosystem 0 0 1 1 1 1 1

time. Second, as mentioned earlier, we only observe settlement times rather than submission
times. The former are contaminated by the random settlement processing times and as
such capture less accurately participants’ behavior. For these two reasons, our liquidity
efficiency and inequality specifications also include as controls the same set of market-wide
and jurisdiction-specific variables used in the first set of regressions. The goal is to allow
these variables to capture any effects that our payment timing and coordination proxies
might fail to properly account for.

5.1 Payment timing and coordination

RTGS system participants generally have the flexibility to choose the timing of their
outgoing payments with the exception of: (i) any payments that are deemed urgent either
by themselves or by institutions for which they may act as correspondents and (ii) payments
submitted to ancillary systems such as securities settlement systems and clearing houses.
Given that incoming payments may be used to fund outgoing ones, this flexibility can give
rise to strategic behaviors as it implies that system participants may time their payments in

22
a way that minimizes their own liquidity commitment (e.g. Bech and Garratt (2003)).
Therefore, in this section we examine the determinants of average settlement times
(as captured by T ) and the degree of payment dispersion across participants (as captured
by Tdiff ). For this purpose, we estimate the following panel specifications across systems
and over time:

Tit =a1 + d1 Xit0 + uit (7)

Tdiff it =a2 + b2 Tit + d2 Xit0 + eit (8)

where i denotes systems and t denotes days. The vector X 0 includes regressors that are
motivated by the theoretical literature on payment timing as well as economic intuition.
For instance, explanatory variables include the aggregate value of payments made (P ), the
total number of system participants (Members), innovations in the overnight interbank rate
(∆IBOR), the total amount of reserves (Reserves) and dummies for the various institutional
characteristics and LSM design features as described in Table 3. These regressors are
natural candidates for our timing specifications. A higher value of outgoing payments could
mean that they have to be spread more evenly during the day, affecting both their average
settlement time as well as their dispersion. Similarly, the number of system participants
may affect their individual incentives to settle earlier and their ability to coordinate their
payments. Given that payments in RTGS payment systems are liquidity-intensive and that
participants have the ability to recycle incoming payments, the timing and dispersion of
payments should also in theory depend on the amount of available liquidity (Reserves) and
its opportunity cost (IBOR). The same is true of institutional features such as the central
bank credit regime and the various LSM characteristics: the presence of a mechanism that is
designed to help participants economize on liquidity could reduce participants’ incentives to
delay or concentrate payments. We estimate both models (5) and (6) using random effects
and inference is done by clustering at the system level.20
The results of these specifications are shown in Tables 4 and 5. With respect to
average settlement time (Table 4), wider system membership is associated with payments
being made later on in the day in most specifications. This could be because incentives to
delay payments might increase in the number of direct participants.21 Alternatively, it could
be because the marginal cost of liquidity and the associated incentives for delay are higher for
smaller participants who tend to be present in systems with wider participation. However,
the effect of wider membership disappears in the full specification (column 12).
On the other hand, increases in the opportunity cost of reserves (as captured by
changes in IBOR) are associated with later settlement times (columns 4, 6, 12). If the
opportunity cost of reserves increases, then system participants would have an incentive to
delay their outgoing payments in anticipation of incoming ones.
20
We use random effects because our models feature time-invariant characteristics. Owing to the large
time and small cross-sectional dimensions of our sample, in practice the random effects estimators are mostly
determined by the within (time) variation of our sample and are therefore unlikely to be biased because of
group unobserved, time-invariant characteristics.
21
Given the public good nature of intraday liquidity, the incentives of LVPS participants are similar to those
faced by players in a volunteering game. In such setups, the probability that any given player volunteers,
decreases with the number of players. See Diekmann (1985).

23
 The negative relationship between available liquidity (as captured by aggregate reserves
balances) and payment settlement times (columns 5, 6, 12) is also consistent with this idea
since aggregate reserves balances are a key instrument that central banks use to target short
term interest rates. However, the fact that reserves balances are significant after controlling
for ∆IBOR suggests that there is an incremental effect, not necessarily associated with
the opportunity cost of liquidity. The negative relationship between reserves balances and
payment timing confirms similar findings from Fedwire reported in Bech et al. (2012) and
Bech and Garratt (2012). This finding is important because several central banks engaged
in quantitative easing programs, during our sample period, after having reduced interest
rates to historically low levels. These programs increased significantly the reserves balances
in their respective systems and our results suggest that this likely had a positive effect on
payment systems as it induced earlier payment submissions.22
The presence of incentives to settle payments early is weakly associated with earlier
settlement times and only the full specification (column 12), whereas the possibility to obtain
intraday credit from the central bank, at a lower collateral cost, is associated with a later
average settlement time (columns 8, 10, 12). If collateral has opportunity cost, this is
contrary to what one might expect as the costlier the central bank liquidity, the more likely
LVPS participants would be to delay their payments in anticipation of incoming ones.
We also find a negative, albeit statistically insignificant, association between the
presence of an LSM and payment timing (columns 9, 10). This appears to be driven by
the opposite association with timing of specific LSM characteristics. For instance, average
settlement time is strongly negatively correlated with FIFO bypass and multilateral offsetting
(columns 11, 12). Since these are the features that increase the potential for liquidity savings
in the LSM queue, they could reduce participants’ incentives to delay submitting their
outgoing payments in anticipation of incoming ones. On the other hand, priority setting
in an LSM is associated with later average settlement times (columns 11, 12). This could
be because the ability to prioritize specific payments in the LSM queue allows for the more
urgent payments to be settled faster so that there is less of an incentive to expedite the
submission of all payments in the queue.23 Alternatively, it could be that prioritizing
payments within the LSM queue results in poorer offsetting matches or liquidity being
blocked for high-value high-priority payments which, on average, delays settlement.24
Table 5 shows the results on payment dispersion. Dispersion tends to decrease with
average settlement times although the effect is not statistically significant. This might arise
because payments that are made later have to settle within a narrower time frame before
the end of the business day. Dispersion is also negatively associated with LVPS membership
(columns 11, 14). It is not immediately clear what drives this effect but it could be purely
22
Earlier payment submission and settlement in the day is desirable not the least because it lowers the
impact of a potential operational outage in the LVPS that might occur during the day and which might
prevent participants from sending or receiving payments. If a larger volume of payments has been processed
before such an outage occurs, its impact will be smaller.
23
This explanation is consistent with the findings of Nellen et al. (2018) who study the relationship between
submission and settlement times in the Swiss payment system.
24
A reduction in matching efficiency might result if urgent payments are larger than non-urgent ones and
if only a few system participants have to make them on any given day so that there is less of a chance
that they are netted in the LSM queue. Unfortunately, we cannot test this specific hypothesis as we cannot
distinguish between urgent and non-urgent payments in our data.

24
mechanical: as the number of participants increases, the number of payments that coincide
may also increase, which would tend to decrease dispersion.
Dispersion increases with the level of reserves balances (columns 5, 6, 14) and with
changes in their opportunity cost (∆IBOR; columns 4, 6, 11, 14). The first effect is
consistent with the idea that an abundance of liquidity reduces participants’ incentives to
coordinate their payments whereas the second effect is consistent with liquidity hoarding
at times of stress. That is, when market conditions deteriorate (i.e., liquidity becomes
more expensive) then system participants may withhold or delay outgoing payments with
the result being that payment coordination weakens and dispersion increases. The positive
relation between ∆IBOR and payment timing discussed earlier, seems to corroborate this.
Dispersion is also higher when it is possible to obtain credit from the central bank
at a lower collateral cost (columns 8, 10, 11, 13, 14), suggesting that a lower collateral
cost potentially reduces participants’ incentives to coordinate their payment submissions to
economize on liquidity. Additionally, incentives for early payment are associated with lower
dispersion in the full specification (column 14). Although our results showed no substantial
correlation between such incentives and average settlement timing, it could be the case
that these incentives induce more uniform payment patterns especially if delaying payments
consistently carries a penalty. One could hypothesize that in the presence of such rules,
system participants might have an incentive to avoid being an outlier in terms of their
payment patterns which would result in lower average dispersion.
Finally, several LSM features are also related to payment dispersion (columns 12-14).
FIFO bypass is associated with higher payment dispersion. This effect is unlikely to be
mechanical because, if anything, FIFO bypass increases the opportunities for settlement
which should decrease dispersion. As such, we suspect that system participants might be
endogenously modifying their behavior in the presence of this functionality. In particular,
FIFO bypass could be reducing participants’ incentives to coordinate their payment submissions
because even if payments are submitted in the LSM queue, at different points in time, this
functionality will reschedule them so as to maximize any offsetting opportunities.
The same effect, in reverse, could explain the negative correlation between priority
setting and dispersion: if priority setting overrules the offsetting algorithm and, as a result,
decreases offsetting efficiency, this might increase participants’ incentives to coordinate their
payments so as to economize on liquidity.25
Multilateral offsetting is also associated with reduced dispersion. This could be either
a mechanical effect or a behavioral one (or both). Multilateral offsetting, by definition,
enables a wider set of payments submitted by multiple participants to settle simultaneously
which would reduce our dispersion measure. Alternatively, given that this functionality is
more effective when more participants’ payments are in the queue at any given point in time,
it could ex-ante incentivize participants to coordinate their payment submission. Since we
do not observe payment submission times, unfortunately we cannot disentangle the effect of
multilateral offsetting on the dispersion of submission times versus that of settlement times
and thus see which of the two explanations drives our result.26
25
This could be the case whether LVPS participants actually use the LSM bypass functionality or not. The
fact that the functionality is available could create ex-ante incentives for participants to coordinate more.
26
See Nellen et al. (2018) for a study of the Swiss SIC payment system utilizing data on both submission
and settlement times.

25
 Overall, our results suggest that an LSM (or particular LSM design features) can
potentially attenuate the degree of strategic complementarity that is found in pure RTGS
environments. If, for example, a FIFO bypass functionality reduces LVPS participants’
incentives to better coordinate their payments, this could explain the higher dispersion in
settlement times. Interestingly however, our findings suggest that particular LSM features
(e.g. multilateral offsetting) could also be increasing the degree of strategic complementarity
between LVPS members by increasing the benefits of coordination in payment submissions.

5.2 Intraday liquidity efficiency

In this section we study the determinants of intraday liquidity efficiency. Given
that RTGS systems with larger flows of payments can naturally be expected to be more
liquidity intensive, we examine how liquidity efficiency, defined in equation (3), is influenced
by payment system behavior (i.e. payment timing and dispersion), the opportunity cost and
quantity of available liquidity as well as a number of specific LSM features.
As discussed earlier, our measure of liquidity efficiency would in principle only depend
on the value, the timing and degree of coordination of payments in an LVPS. Payment
system activity, in turn, would depend on (and might also influence) the price and quantity
of available liquidity. However, our two variables of LVPS activity (average settlement time
and dispersion), while informative, likely do not capture all aspects of LVPS activity and,
as such, additional variables are included as regressors in our specification for Q to account
for any ultimate effects on efficiency from market variables that our timing and dispersion
variables fail capture.
With this in mind, we estimate the following panel specification:

Qit = a3 + b3 Tit + c3 Tdiff it + d3 Xit0 + vit (9)

where i denotes systems, t denotes days and X 0 contains the same regressors as in model
specifications (7) and (8). In addition, Tdiff is included as a regressor because payment
coordination allows RTGS system participants to recycle payments and thus economize on
liquidity usage.
Table 6 presents the results of this specification. Indeed, a higher degree of payment
dispersion is empirically associated with lower levels of liquidity efficiency (columns 2, 12, 13)
confirming the idea that in liquidity-intensive RTGS systems, payment recycling is a way for
system participants to economize on liquidity. Otherwise, the timing of payments does not
seem to be significantly associated with liquidity efficiency whereas the number of system
members seems to only be indirectly associated with efficiency through other variables, as
its effect disappears once additional controls are included.
Changes in the opportunity cost of reserves (as captured by ∆IBOR) are positively
associated with efficiency only after we control for timing effects (average settlement time
and dispersion). We saw earlier that increases in IBOR are strongly positively correlated
with dispersion which negatively predicts efficiency. Thus, it is not surprising that the
effect of IBOR becomes more positive in the presence of these controls. It is not clear
however how the opportunity cost of reserves would affect liquidity efficiency other than via
payment coordination. On the other hand, while reserves balances are negatively correlated

26
with efficiency in some specifications, their effect disappears once additional controls are
included (columns 11-13) suggesting that reserves balances might be related with efficiency
only indirectly by influencing (negatively) payment coordination.
The presence of incentives for early settlement is associated with higher liquidity
efficiency (columns 12-14) which is likely partially driven by the negative correlation between
those incentives and dispersion discussed earlier. On the other hand, the ability to obtain
uncollateralized (or partially collateralized) credit, seems to correlate with efficiency only via
dispersion as the inclusion of our Tdiff variable eliminates its significance (columns 11-13).
Finally, the presence of an LSM is uncorrelated with liquidity efficiency in our sample
(columns 8, 9). This is somewhat surprising as LSMs’ intended purpose is precisely, to help
economize on liquidity. Looking at specific LSM design features, FIFO bypass is uncorrelated
with efficiency despite its strong positive correlation with our dispersion measure. On
the other hand, multilateral offsetting is positively correlated with efficiency (columns 10,
11) with this effect being likely driven by the negative effect of multilateral offsetting on
dispersion since the inclusion of T dif f as a control eliminates the significance of the former.
The negative relation between priority setting and liquidity efficiency, after controlling for
dispersion, could be because changing the priority of payments within the LSM queue is
driven by payment urgency rather than liquidity saving considerations, resulting in poorer
payment offsets. In other words, to the extent that the LSM matching algorithm releases
payments in a way that minimizes liquidity usage, altering payment priority may result in
less liquidity-efficient matches. Finally, the ability to reserve liquidity for urgent payments
could result in some liquidity being released in the payment system early in the day and
subsequently being recycled to fund additional payments. This effect arises after controlling
for payment dispersion and other variables, since payment reservations are themselves also
positively associated with dispersion.
Overall, these empirical regularities imply that the various LSM features may not all
have the desired effect of improving liquidity efficiency with some features potentially even
being detrimental to it.

5.3 Inequality in intraday liquidity usage

In the final part of our empirical analysis we look at the determinants of inequality in
the usage (and provision) of intraday liquidity. We estimate a number of panel specifications
where the dependent variable is the Gini coefficient of the relative usage of intraday liquidity,
defined in equation (4). This variable captures the extent to which some participants tap into
their own available liquidity more (or less) relative to their daily payment obligations. To the
extent that some participants use more of their own liquidity in relation to the value of their
payments, these participants contribute to system-wide liquidity that can be subsequently
recycled by other participants in order to meet their payment obligations.
Our empirical specification is thus:

Git = a4 + b4 Tit + c4 Tdiff it + d4 Xit0 + uit (10)

where i denotes systems, t denotes days and X 0 contains the same regressors as in the
previous models. T and Tdiff are again included as a regressors to gauge if payment timing

27
and the degree of payment coordination are associated with overall equality (or lack thereof)
in liquidity provision.
Table 7 presents the results of these specifications. The first thing to notice is that
the Gini coefficient is unrelated to our timing and dispersion variables. This suggests
that the Gini index is not associated with the overall timing and degree of coordination
in payments. However, it is negatively correlated with ∆IBOR (columns 4, 9, 11, 13). This
result suggests that when the opportunity cost of reserves increases, there is less reliance
on fewer participants for liquidity and instead more participants commit their own liquidity.
This is consistent with the earlier results that ∆IBOR is positively correlated with average
settlement times and dispersion. If a higher opportunity cost of liquidity results in payment
delays and hoarding, then one would indeed expect more participants to be forced to commit
their own liquidity in order to meet their payment obligations.
The aggregate amount of reserves and the presence of incentives for early settlement
are both associated with a lower Gini coefficient across models. This is likely because a higher
amount of reserves reduces participants’ incentives to rely on recycled liquidity provided
by other participants. Additionally, the presence of incentives for early settlement limits
the degree to which certain participants can recycle liquidity provided by only a few other
participants, since every participant has an incentive (or obligation) to make some early
payments and thereby inject liquidity in the payment system. Finally, the presence of an
LSM is positively correlated with the Gini index with the effect being driven almost entirely
by the LSM priority setting functionality (columns 10-13). This is a rather unexpected
finding since LSMs are intended to level the playing field of liquidity provision by ensuring
that all submitted payments are offset by incoming ones and that no participants bear the
brunt of consistently supplying liquidity to the system.

6 Summary and conclusions

This is the first paper to systematically study intraday liquidity usage, by financial
institutions, across several jurisdictions and over a long period of time. Using a unique
cross-country data set, we measure intraday liquidity usage on a daily frequency and at
system level and assess its drivers. We find that intraday liquidity usage is highly economically
significant accounting, on average, for 15% of daily aggregate payment values or about 2.3%
of local GDP.
Consistent with the theoretical literature, we also find that intraday liquidity usage
depends on the way participants interact with one another in an LVPS. For instance, a higher
degree of payment coordination is associated with higher liquidity efficiency, i.e. a higher
value of payments made for every unit of intraday liquidity used. Participant interaction
in turn, depends on both policy-related variables such as the overall supply of central bank
reserves balances as well as system-specific institutional characteristics.
Regarding the former, we find, for example, that higher aggregate reserves balances
(largely as a result of quantitative easing programs in several of the jurisdictions in our
sample) are associated with reduced incentives among participants to coordinate their payments
and thus economize on intraday liquidity usage. On the flip side, higher reserves balances
appear to induce earlier payment submissions and also reduce the reliance on just a few

28
system participants to provide liquidity to the rest. Both of these effects are desirable as
they help reduce the impact of a potential outage in the payment system. In general, the
amounts of excess liquidity that have been injected by central banks in many jurisdictions
appear to have reduced the benefit of liquidity saving and the need to manage liquidity. This
will likely change in the future when central banks start reducing the size of their balance
sheets.
The most novel contribution of our paper however is to study the impact of institutional
and system-specific characteristics on intraday liquidity usage. Given that these characteristics
are generally time-invariant, such an analysis requires cross-country data on large-value
payments, which our paper is the first to assemble. Our analysis yields several new results.
The first is that incentives for early payment submissions seem to have more of an effect on
payment coordination than actual payment submission times. Given that these incentives
often take the form of penalties, it appears that they induce LVPS participants to coordinate
their submission times likely in an attempt to “not stand out from the pack”. Interestingly
however, this increased coordination renders the payment system more liquidity-efficient as
it facilitates payment recycling.
A second novel result is that LVPS participants appear to endogenize some of the LSM
features which in some cases improves liquidity efficiency but in other cases it is detrimental
to it. For example, multilateral offsetting is associated with increased payment coordination.
One explanation of that could be that participants coordinate their payments more in order
to take full advantage of this functionality. Increased participant coordination then further
enhances liquidity efficiency. On the other hand, the presence of a FIFO bypass functionality,
whereby the offsetting algorithm can bypass the time priority of submitted payments, is
associated with reduced payment coordination. This could be because participants are less
incentivized to coordinate their payments in the presence of this functionality. As a result
however, liquidity efficiency is reduced.
Overall, a key insight from our paper is that, in endogenizing the various payment
system design features and institutional arrangements, LVPS participants can influence the
aggregate amount of intraday liquidity they use to fund their payments. We believe that
understanding these endogenous dynamics is important when designing payment systems
and therefore additional research in this area is warranted.

29
 Table 4: Payment timing panel regressions. The dependent variable, T, is the value-weighted average settlement time of payments made
in each system on any given day. It is defined in equation (5). P (in USD bn) is the daily total value of payment made in each system.
∆IBOR (in %) is the first difference in either the unsecured overnight interbank rate or the central bank policy rate. Reserves (in USD
bn) is the total size of reserve balances held with the central bank by payment system participants. The Incentives, Credit and LSM
characteristic dummies are defined in Table 3. The models are estimated using random effects. Robust p-values are reported in the
parentheses. *, ** and *** denote significance at 10%, 5% and 1% levels respectively.

(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12)
T T T T T T T T T T T T
P -0.0000 -0.0000 0.0000
(0.366) (0.859) (0.724)
Members 0.0001*** 0.0001*** 0.0001*** -0.0000
(0.000) (0.000) (0.000) (0.499)
∆IBOR 0.0121*** 0.0123*** 0.0118**
(0.009) (0.009) (0.019)
Reserves -0.0001*** -0.0001*** -0.0001***
(0.001) (0.005) (0.000)

30
Incentives 0.0192 0.0315 -0.0561*
(0.731) (0.472) (0.083)
Credit 0.0636* 0.0683* 0.1933***
(0.090) (0.087) (0.000)
LSM -0.0059 -0.0059
(0.159) (0.160)
FIFO bp -0.1441** -0.0759**
(0.033) (0.030)
Offsetting -0.0269 -0.1451***
(0.617) (0.000)
Priority 0.1817*** 0.1887***
(0.000) (0.000)
Reservations -0.0408* 0.0157
(0.058) (0.620)
cons 0.5425*** 0.4299*** 0.4318*** 0.5282*** 0.5520*** 0.4955*** 0.5159*** 0.5146*** 0.5332*** 0.4970*** 0.5170*** 0.5760***
(0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000)
R2 0.0436 0.0977 0.0980 0.0005 0.0527 0.1004 0.0014 0.0796 0.0171 0.0902 0.3394 0.6457
N 26039 26039 26039 25205 26033 25200 26039 26039 26039 26039 26039 25200
 Table 5: Payment dispersion panel regressions. The dependent variable, Tdiff, is defined in equation (6). T, is the value-weighted
average settlement time of payments made in each system on any given day. It is defined in equation (5). P (in USD bn) is the daily
total value of payment made in each system. ∆IBOR (in %) is the first difference in either the unsecured overnight interbank rate or
the central bank policy rate. Reserves (in USD bn) is the total size of reserve balances held with the central bank by payment system
participants. The Incentives, Credit and LSM characteristic dummies are defined in Table 3. The models are estimated using random
effects. Robust p-values are reported in the parentheses. *, ** and *** denote significance at 10%, 5% and 1% levels respectively.

(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) (13) (14)
Tdiff Tdiff Tdiff Tdiff Tdiff Tdiff Tdiff Tdiff Tdiff Tdiff Tdiff Tdiff Tdiff Tdiff
T -0.3773 -0.2110 -0.5610*** -0.3549
(0.442) (0.685) (0.009) (0.274)
P 0.0000 -0.0000 0.0001*** -0.0000*
(0.585) (0.412) (0.001) (0.054)
Members -0.0001*** -0.0000 -0.0000** -0.0001***
(0.000) (0.813) (0.016) (0.000)
∆IBOR 0.0332*** 0.0356*** 0.0369*** 0.0372***
(0.000) (0.000) (0.001) (0.000)
Reserves 0.0001*** 0.0001*** 0.0000 0.0000**

31
(0.000) (0.005) (0.539) (0.040)
Incentives -0.1004 -0.0799 0.0272 -0.0246 -0.2518***
(0.186) (0.264) (0.684) (0.523) (0.000)
Credit 0.1265** 0.1014** 0.1089*** 0.1108*** 0.4237***
(0.011) (0.034) (0.000) (0.000) (0.000)
LSM -0.0233 -0.0233 -0.0482
(0.360) (0.360) (0.235)
FIFO bp 0.2581*** 0.2725*** 0.4713***
(0.001) (0.000) (0.000)
Offsetting -0.0608 -0.1019*** -0.3855***
(0.326) (0.001) (0.000)
Priority -0.2631*** -0.2364*** -0.1472**
(0.000) (0.000) (0.024)
Reservations 0.1051*** 0.1051*** 0.0604
(0.000) (0.000) (0.133)
cons 0.5813** 0.3719*** 0.4786*** 0.3824*** 0.3490*** 0.4834** 0.4487*** 0.3537*** 0.3997*** 0.4305*** 0.6406*** 0.4347*** 0.4327*** 0.8505***
(0.023) (0.000) (0.000) (0.000) (0.000) (0.043) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000)
R2 0.0602 0.2647 0.1307 0.0016 0.2079 0.1661 0.1308 0.1423 0.0769 0.2462 0.5215 0.4384 0.5491 0.6429
N 26035 26035 26035 25201 26029 25196 26035 26035 26035 26035 25196 26035 26035 25196
 Table 6: Liquidity efficiency panel regressions. The dependent variable, Q, is defined - in equation (3) - as the ratio of the aggregate
value of payments made to the aggregate amount of intraday liquidity used. The independent variables T and Tdiff are defined in
equations (5) and (6) respectively. ∆IBOR (in %) is the first difference in either the unsecured overnight interbank rate or the central
bank policy rate. Reserves (in USD bn) is the total size of reserve balances held with the central bank by payment system participants.
The Incentives, Credit and LSM characteristic dummies are defined in Table 3. The models are estimated using random effects. Robust
p-values are reported in the parentheses. *, ** and *** denote significance at 10%, 5% and 1% levels respectively.

(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) (13)
Q Q Q Q Q Q Q Q Q Q Q Q Q
T 4.8259 2.4497
(0.125) (0.591)
Tdiff -3.5611* -12.5865*** -12.1812***
(0.061) (0.002) (0.007)
Members 0.0013*** -0.0004 0.0013 0.0003 0.0004
(0.000) (0.721) (0.100) (0.678) (0.629)
∆IBOR -0.0383 -0.0287 0.0480 0.4552*** 0.4133*
(0.191) (0.196) (0.612) (0.004) (0.059)
Reserves -0.0016** -0.0017** 0.0003 0.0010 0.0012

32
(0.010) (0.029) (0.825) (0.397) (0.293)
Incentives 0.1142 -1.3430 5.3491** 3.3869* 3.5877**
(0.956) (0.689) (0.042) (0.072) (0.043)
Credit 0.5983 2.6634 -5.7255** -2.2255 -2.8119
(0.764) (0.391) (0.034) (0.266) (0.191)
LSM -0.6718 -0.5579
(0.631) (0.666)
FIFO bp -1.6296 -5.7207 -0.4172 -0.4015
(0.631) (0.149) (0.922) (0.925)
Offsetting 3.0812* 6.5399* 3.3422 3.8002
(0.095) (0.099) (0.393) (0.328)
Priority -0.9878 -4.6873 -7.3624* -7.7391*
(0.581) (0.269) (0.080) (0.066)
Reservations -2.8161*** 4.5631** 5.1767*** 5.1186***
(0.000) (0.040) (0.005) (0.005)
cons 3.4915* 7.4022*** 4.9260*** 6.0355*** 6.6009*** 5.9659*** 5.9090*** 6.5580*** 7.7179** 5.7597*** 2.3485 9.4304** 7.7919
(0.064) (0.000) (0.002) (0.000) (0.000) (0.000) (0.000) (0.001) (0.018) (0.002) (0.560) (0.035) (0.154)
R-sq 0.0036 0.0113 0.0321 0.0000 0.0114 0.0012 0.0155 0.0111 0.1352 0.0637 0.3610 0.4351 0.4363
N 26039 26035 26065 25231 26059 26065 26065 26065 25226 26065 25226 25196 25196
 Table 7: Inequality in liquidity usage panel regressions. The dependent variable, G, is the Gini coefficient in relative intraday liquidity
usage, defined in equation (4). The independent variables T and Tdiff are defined in equations (5) and (6) respectively. ∆IBOR (in
%) is the first difference in either the unsecured overnight interbank rate or the central bank policy rate. Reserves (in USD bn) is the
total size of reserve balances held with the central bank by payment system participants. The Incentives, Credit and LSM characteristic
dummies are defined in Table 3. The models are estimated using random effects. Robust p-values are reported in the parentheses. *, **
and *** denote significance at 10%, 5% and 1% levels respectively.

(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) (13)
G G G G G G G G G G G G G
T 0.0113 0.1361
(0.961) (0.445)
Tdiff -0.1033 -0.1735 -0.1510
(0.487) (0.483) (0.518)
Members 0.0002*** 0.0001 0.0000 0.0000 0.0000
(0.000) (0.193) (0.474) (0.966) (0.888)
∆IBOR -0.0178*** -0.0178*** -0.0192*** -0.0135 -0.0159**
(0.002) (0.001) (0.000) (0.105) (0.036)
Reserves -0.0001** -0.0001** -0.0001*** -0.0001*** -0.0001**

33
(0.015) (0.048) (0.000) (0.005) (0.021)
Incentives -0.1145*** -0.0101 -0.2110*** -0.2380** -0.2269***
(0.009) (0.930) (0.004) (0.011) (0.009)
Credit -0.0054 -0.1589 0.0679 0.1162 0.0837
(0.938) (0.267) (0.365) (0.301) (0.326)
LSM 0.0532** 0.0546***
(0.025) (0.004)
FIFO bp -0.0896 -0.1042 -0.0311 -0.0301
(0.401) (0.103) (0.813) (0.828)
Offsetting 0.0415 -0.0041 -0.0481 -0.0227
(0.567) (0.946) (0.610) (0.747)
Priority 0.1098** 0.2207*** 0.1840** 0.1630*
(0.017) (0.000) (0.010) (0.064)
Reservations -0.0209 -0.0859 -0.0776 -0.0808
(0.527) (0.217) (0.264) (0.257)
cons 0.4047*** 0.4501*** 0.2805*** 0.4109*** 0.4358*** 0.4870*** 0.4118*** 0.3697*** 0.3605*** 0.3660*** 0.4978*** 0.5952*** 0.5042***
(0.000) (0.000) (0.001) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000)
R2 0.0432 0.0241 0.0831 0.0004 0.0170 0.1851 0.0000 0.0331 0.2226 0.0570 0.4369 0.4482 0.4512
N 26021 26017 26047 25215 26041 26047 26047 26047 25210 26047 25210 25180 25180
 Monitoring Intraday Liquidity Risks in a Real Time
Gross Settlement System

Neville Arjani a∗ Fuchun Li b Leonard Sabetti c

a
Canada Deposit Insurance Corporation
b
Payments Canada
c
Bank of Canada

Abstract
The objective of the paper is to propose a tool for predicting intraday liquidity risks
in a real time gross settlement system. To achieve this goal, we construct an intraday
liquidity risk indicator (LRI) to assess intraday liquidity risks of a participant by com-
paring the evolution of the expected liquidity sources of the participant for settling
payments with its expected liquidity requirements in the remainder of the payment
day. If the participant’s expected liquidity requirements are larger than its expected
liquidity sources, the participant is very likely to incur a lack of intraday liquidity for
settlement obligation within the remainder of the day. Otherwise, the available liquid-
ity sources of the participant will be sufficient to cover its expected intraday liquidity
requirements.
Furthermore, based on the LRI, we propose a framework that can predict the likeli-
hood of an intraday liquidity risk event throughout the remainder of the payment day,
where an intraday liquidity risk event is said to occur if the LRI rises above one. Using
data from Canada’s RTGS-equivalent payment system, the Large Value Transfer Sys-
tem, to evaluate the forecasting performance of the LRI, we find that the LRI performs
reasonably well, and we obtain some new empirical findings.

Keywords: Intraday liquidity Risk; Clearing and Settlement Systems; Predicting Payment Transactions.
JEL Classification: G21, G23, C58.

∗
We are grateful to seminar participants at 17th Conference of Bank of Finland Payment and Settlement
System Simulator (2019), Bank of Canada, Donbei University of Economics and Finance, Shandong University,
Segun Bewaji, Shaun Byck, James Chapman, Walter Engert, Ronald Heijmans, Charlie Kahn, Anneke Kosse,
Kimmo Soramaki and our Payments Canada colleagues for many useful comments and suggestions. We thank
Cyrielle Chiron for giving us helpful advice and encouragement for this discussion paper at Payments Canada.
We also thank Pooja Paturi for her excellent research assistance. The views expressed in this paper are those
of the authors. No responsibility for them should be attributed to the Payments Canada, Bank of Canada, and
Canadian Domestic Insurance Company. Neville Arjani: Canada Deposit Insurance Corporation; Fuchun Li
(Corresponding author): Payments Canada; Leonard Sabetti: Bank of Canada.
 1 Introduction

Historically, interbank payments have been settled via deferred netting systems at the end

of the day. As a consequence of the rapid increase in values settled in large-value pay-

ments system over the last few decades, however, policy makers of payments systems be-

come concerned about settlement risk inherent in deferred net settlement systems. In

particular, system participants are concerned about the potential for contagion effects

attributable to the unwinding of net positions that would result if a participant failed to

make its obligation when it is due.1 Consequently, many countries have chosen to mod-

ify the settlement procedure employed by their interbank payment system with a view

of reducing the settlement risk and delivering payments commitment effectively to meet

international security and operating standards for modern payments systems (Bech and

Hobijn, 2007; Bech, Preisig, and Soramaki, 2008).

Since a real time gross settlement (RTGS) system is settled continuously, individually, and

irrevocably on a gross basis throughout the business day, it can reduce the settlement risk.

An RTGS system can also help reduce settlement risk by facilitating payment versus pay-

ment and delivery versus payment in the settlement of forex and securities transactions,

respectively. With this background, it comes as no surprise that RTGS systems have been

used as the favored large-value payment systems in many countries to reduce the settle-

ment risks inherent in clearing procedures. For example, Britain’s large-value payment
1
Unwinding is a procedure followed in certain clearing and settlement systems in which payments trans-
fers are settled on a net basis, at the end of the processing cycle, with all transfers provisional until all partici-
pants have discharged their settlement obligations. If a participant fails to settle, some or all of the provisional
transfers involving that participant are deleted from the system and the settlement obligations from the re-
maining transfers are then recalculated. Such a procedure has the effect of allocating liquidity pressures and
losses attributable to the failure to settle to the counterparties of the participant that fails to settle (Bank for
International Settlements 2003).

1
 system, CHAPS, which previously operated under a deferred netting system, was con-

verted to an RTGS system in April 1996. The European Monetary Union chose an RTGS

system for its large-value funds transfer system, TARGET, in January 1999. Specifically,

Canada is going to replace the current large value transfer system (LVTS) with Lynx, an

RTGS system (Payments Canada, 2017). 2

However, as pointed out by Allsopp et.al (2009) and Bech and Soramaki (2002), in RTGS

systems, the reduction of settlement risk is traded off against an increased need for intra-

day liquidity requirements. As a result, participants in RTGS systems are inevitably con-

fronted with the issue of whether there will be sufficient intraday liquidity to meet pay-

ments and settlement obligations on a timely basis. Supervisory authorities have devoted

much effort to monitoring and managing the intraday liquidity risk in RTGS systems. A key

issue in such supervision is that policy makers need an analytical tool to monitor intraday

liquidity positions and risks (SWIFT, 2014). Under stressed conditions, such a tool can help

them gauge the likely impact of distress of intraday liquidity on payments systems, while

in normal times it is crucial to use such a tool to calibrate prudential instruments, such as

collateral requirements, in accordance to the relative contribution of different participants

to systemic risk.

In this context, Heijmans and Heuver (2014) developed monitoring indicators from large

value payment systems to identify signs of liquidity stress in both individual banks and
2
In developed countries, Canada is the only country in the Group of Ten (G-10) Countries that has decided
not to implement an RTGS system. Instead, Canada opted for a hybrid system, the LVTS, which employs an
advanced settlement algorithm that combines two payment streams: Tranche 1 and Tranche 2. A participant
can either send a payment through the fully collateralized Tranche 1 which involves real-time settlement or
through Tranche 2 in which collateral is pooled, risk is shared, and settlement takes place at the end of the
payment day.

2
 market segments. Using the data from TARGET2, they find their indicators perform well.3

The BCBS (2009) has published “Monitoring Tools for Intraday Liquidity Management”,

where it explicitly includes the management of intraday liquidity risk as a principle and

proposes a set of analytical tools for managing intraday liquidity risk. Leon (2012) esti-

mated the intraday liquidity risk of financial institutions using a Monte Carlo simulation

approach. Li and Perez Saiz (2018) constructed an indicator for monitoring the settlement

risk at the end of day in the LVTS. These analytical tools in above-mentioned papers and

references therein are very powerful for monitoring the current status of intraday liquidity

risk in payments systems, but they cannot be used to predict intraday liquidity risks within

an upcoming period of time in a payment day.

Taking into consideration that an effective prediction for intraday liquidity risk has sub-

stantial value to policy makers by allowing them to detect the future potential weakness

and vulnerabilities in the payment systems, and possibly take pre-emptive policy actions

to avoid a risk event or limit its effects, this paper aims to propose an analytical tool for

predicting intraday liquidity risks in a real time gross settlement system. To achieve this

goal, based on the work of Baek et. al (2014), we construct an intraday liquidity risk indi-

cator (LRI) to assess the upcoming intraday liquidity risks of a participant by comparing

the expected sources of the participant for settling payments with its expected liquidity

requirements in the remainder of the day. If the participant’s expected liquidity require-

ments are larger than its expected sources, the participant is very likely to incur a net debit

position that exceeds its credit limits. Otherwise, the available sources of the participant
3
TARGET2 is a real time gross settlement system and is used by both central banks and commercial banks
to process payments in EURO. TARGET2 replaced the decentralized first-generation TARGET system.

3
 will be sufficient to cover its expected intraday liquidity requirements within the remain-

der of the day. Furthermore, based on the LRI, we propose a framework that can predict

the likelihood of an intraday liquidity risk event within the remainder of the payment day,

where an intraday liquidity risk event is said to occur if the value of the LRI rises above one.

Using the data from the LVTS to evaluate the forecasting performance of the LRI, we find

that the LRI performs reasonably well, suggesting that it is a useful tool for predicting in-

traday liquidity risk in an RTGS system. Additionally, we find that the predicted values of

the LRI are more varied in the late afternoon, which is consistent with our finding that

the probabilities of an intraday liquidity risk event reach peak levels in the late afternoon.

This suggests that participants need to manage their intraday liquidity well to synchronize

their outgoing payments with the incoming funds that they expect to receive in the late

afternoon to avoid the mismatch problem (McAndrews and Rajan, 2000).

The paper is organized as follows. Section 2 outlines the effects of settlement methods

on liquidity risk. In Section 3, we introduce an indicator for each participant to assess the

intraday liquidity risks in the remainder of a payment day. Section 4 proposes a framework

to predict the likelihood of an intraday liquidity risk event during the remainder of the day,

where an intraday liquidity risk event is said to occur if the LRI rises above one. Section 5

concludes.

2 The effects of settlements methods on liquidity needs: real-time

gross settlements versus netting settlements

According to the way settlement takes place, a payment system can be classified into a net

settlement system or a gross settlement system. With a net settlement system, payment

4
 messages are processed continuously in real time, but settlement occurs only at the end

of a clearing cycle, on a net multilateral basis. With a gross settlement system, fund trans-

fers at the settlement stage occur on a bilateral and gross basis. A common form of gross

settlement large-value payment system is the real-time gross settlement system, in which

both information processing and settlement take place continuously in real time.

Under a netting settlement system with end-of-day settlement, the payment is settled only

at the end of the day, at which it would have received all of the day’s incoming funds, as

well as having made all the outgoing transfers. The end-of-day payment implies that send-

ing participants have no incentive to delay sending the payment messages if there are no

other payment-system-imposed constraints. Hence, there should be no costly settlement

delays or gridlock.4 Also, since each participant needs to pay only the net amount at the

end of a day, which usually is much smaller than the value of total outgoing payments the

participant has to make during the day, it needs to hold lower clearing balances as pay-

ment liquidity relative to an RTGS system. However, if the sending participants fail during

the day and cannot make their payments at the end of the day, this may result in a chain of

defaults by other participants in the system. Particularly, the potential spillover of this set-

tlement failure to other payment systems and financial markets could lead to a collapse in

the financial system. Consequently, the main concern associated with a netting settlement

system is the potential settlement risk.

An RTGS system reduces settlement risk in payments systems by settling transactions on

4
Gridlock is a case of payment system illiquidity in which the failure of some transfers to be executed pre-
vents a substantial number of other transfers from other participants from being executed (Soramaki and
Bech, 2001). For example, in the case of two participants, gridlock is a situation where participant A is waiting
for participant B’s payment and participant B is waiting for participant A’s payment, so neither can pay the
other.

5
 bilateral and gross basis, instead of netting payments at the end of the day. A sending

participant, however, is faced with the issue of when to send the payment request. This

decision depends on whether it has sufficient funds in its clearing account to cover the

transfer, when incoming payments will arrive, and whether it needs to save the account

balance for more urgent payment requests. The timing decision of each participant in this

payment system may collectively slow down the speed of funds transfer or may even trig-

ger gridlock of the whole payment system. Therefore, the main concern with an RTGS sys-

tem is whether there will be sufficient liquidity to cover outgoing payments. Participants

must make intraday liquidity available for settlements that could take place throughout

the business day. Otherwise, the cost of a lack of liquidity would be very high, not just for

the participants, but for the payment system as a whole.5

3 Monitoring intraday liquidity risk

3.1 An intraday liquidity risk indicator

Management of intraday liquidity risk is a key element in the overall risk management

framework of a payment system. There are two main questions of managing intraday liq-

uidity risk in a payment system: (i) what is the current status of intraday liquidity risk in a

payment system? and (ii) what is the future status of intraday liquidity risk?

Among the recent contributions to answer (i) are the studies by Heijmans and Heuver

(2014), Li and Perez Saiz (2018), BCBS (2013), and Leon (2012). Particularly, the BCBS
5
Central Banks have mitigated the need for settlement liquidity in real time to an extent by providing intra-
day liquidity. Lending liquidity generates credit risk for the central banks and thus this lending is collateralized
to remove this risk. However, the benefits from reducing the risk associated with netting settlement systems
are considered to exceed the costs of greater liquidity needs (Zhou, 2000). Hence, the number of RTGS systems
has grown. Recent debate has mainly concentrated on the benefits of complementing RTGS with a liquidity-
saving mechanism ( Willison, 2004).

6
 (2013) has developed a set of quantitative tools to monitor participants’ intraday liquid-

ity risk and their ability to meet payment and settlement obligations on a timely basis.

Relative to the literature on developing tools to monitor current status of intraday liquid-

ity risk in payments systems, there has been relatively little effort on prediction analysis for

intraday liquidity risks. However, an effective prediction of the intraday liquidity risk has

substantial value to policy makers as it allows them to detect future potential weaknesses

and vulnerabilities in payment systems, and possibly take pre-emptive policy actions to

avoid occurrence of a risk event or limit its effects. To answer (ii), based on the work of

Baek et. al (2014), we propose an intraday liquidity risk indicator by comparing the par-

ticipant’s expected payment capacity with its expected payment requirements, within the

reminder of the payment day. 6

The setup of our indicator is described as follows. At time t on day j, for a participant i, the

intraday liquidity source consists of its net payment income up to time t on day j, denoted

i , its intraday credit limits at the central banks, denoted by CLi , and its payment
by P It,j t,j

incomes to be received from other participants during the remainder of day j after time

i . Its intraday liquidity requirements during the remainder of day j

t, denoted by RP It,j

i . Its future intraday liquidity risks are managed by the following

are denoted by RP Dt,j

inequality,

i
P It,j + CLit,j + RP It,j
i i
> RP Dt,j . (1)
6
Baek and Soramaki (2014) propose a set of tools to monitor the future status of intraday liquidity risks in
the BoK-Wire system. Unlike the BoK-Wire system, the Bank of Canada does not impose any reserve require-
ments through which it could control interest rates and liquidity. As a result, our indicator is different from the
indicator in Baek and Soramaki (2014). Furthermore, using the data from the LVTS, we find the model that we
use for predicting the expected payment transactions performs better than the model that Baek and Soramaki
(2014) use.

7
 The intraday liquidity risk indicator is defined as,

i
RP Dt,j
i
LRIt,j = i + CLi + RP I i
. (2)
P It,j t,j t,j

In Tranche 1 of the LVTS, to prevent a participant from incurring a situation where its net

debit position is in excess of its net debit cap, the LVTS applies a real-time risk control

test to each payment submitted to the system, which ensures that the submitted payment

value does not exceed the summation of the participant net payment income and credit

limit, i.e.,

i
P It,j + CLit,j > P Dt,j
i
, (3)

i and CLi have the same definitions as in (1) and P D i is the submitted pay-
where P It,j t,j t,j

ment value at time t on day j for the participant i. 7 i is ob-

Compared to (3), the LRTt,j

i with the expected liquidity

tained from extending the risk control test by replacing P Dt,j

i ) and increasing liquidity sources by

demands during the remainder of the day j (RP Dt,j

i ).
the expected payment income during the remainder of the day j (RP It,j

i is greater than one, the predicted liquidity

For a given participant and a time t, if the LRIt,j

needs of the participant are larger than its available sources and thus it will possibly incur

a liquidity risk event where the expected intraday liquidity requirements exceed the ex-

pected intraday liquidity sources. Otherwise, at the given time, the participant’s expected

liquidity sources are sufficient to cover its expected intraday liquidity requirements within
7
In Tranche 2 of the LVTS, the payment submitted will be processed only if it passes two risk control tests:
the bilateral risk control test, and multilateral risk control test. Based on the two risk control tests, we can
build two indicators for predicting whether the bilateral net debit position and multilateral debit position can
be covered by the bilateral credit limits and multilateral intraday line of credit, respectively, in the reminder of
the payment day. This work is beyond the focus of this paper and will be pursued in our future research.

8
 the remainder of the day.

A participant’s intraday credit limits represent the maximum net debit position that the

participant can incur during the remainder of the day. The participants determine the

value of their credit limits and must fully secure this limit with eligible collateral. If a par-

i at time t is greater than one, it should increase its credit limits at any time
ticipant’s LRIt,j

during the payments cycle by apportioning additional collateral on a dollar-for-dollar ba-

i is less than one, it may reduce its credit limits at any time during the pay-
sis. If its LRIt,j

ments cycle. The collateral no longer needed to cover a participant’s appointment be-

comes part of its excess holdings.

In order to implement the intraday liquidity risk indicator, we need to predict the values

i and RP I i . In the following, we outline the method to predict both RP I i

of both RP Dt,j t,j t,j

i . Let T P I i represent the total payment income that participant i has received
and RP Dt,j j

i be the total payment that participant i has received up to time t on day

on day j, and P It,j

j. We have,

i
RP It,j = T P Iji − P It,j
i
. (4)

Similarly, let T P Dji represent the total payment demand that participant i has spent on

i be the total payment that participant i has sent up to time t on day j, we

day j, and P Dt,j

have,

i
RP Dt,j = T P Dji − P Dt,j
i
. (5)

i and P D i in both (3) and (4) are given by the data, to predict the values of
Since P It,j t,j

i and RP D i , we only need to predict both T P I i and T P D i . In next subsection,

RP It,j t,j j j

9
 we will introduce three models for predicting both payments sent and payments received,

and evaluate their forecasting performance.

3.2 Alternative models for predicting payment transactions and their forecast-
ing performance
3.2.1 Alternative models for predicting payment transactions

To calculate both the expected liquidity requirement and the expected liquidity source for

the remainder of the day, we need to predict the total payment income and total payment

demand for each participant. We examine the predicting accuracy of the following three

models commonly used in this literature for predicting payment transactions. The three

models are a linear regression model, an autoregressive integrated moving average model,

and a lognormal diffusion model.

A. A linear regression model

We fit a linear regression model with the payment income to be received as the response

variable, with the days of the week and the holiday being independent variables,

T P Iji = αiR + βiR Dj + γiR Hj + εij , (6)

where αiR is a consistent effect on the total payment incomes on day j for participant i, Dj

is the vector of the week indicators with the exception of Monday (Tuesday, Wednesday,

Thursday, and Friday), and Hj is the indicator of whether day j is a Canadian holiday, and

εij is an error term. Similarly, for T P Dji , we have,

T P Dji = αiD + β D Dj + γ D Hj + εij . (7)

Given the regression equations for participants, we can get a seemingly unrelated regres-

10
 sion (SUR) model which consists of these linear regression equations for different par-

ticipants. We use the feasible general least squared method to estimate the SUR model

(Zellner, 1962).

B. An autoregressive integrated moving average model

An autoregressive integrated moving average model (ARIMA) is a popular and flexible class

of predicting model that utilizes historical information to make predictions.

Let 4T P Iji = T P Iji − T P Ij−1

i , the ARIM A(1, 1, 1) model for predicting payments received

is specified as,

4T P Iji = cIi + αiI 4T P Ij−1

i
+ ij − βiI ij−1 , (8)

where ij is the random shock to bank i occurring at time j.

Similarly, let 4T P Dji = T P Dji − T P Dj−1

i , the ARIM A(1, 1, 1) model for predicting pay-

ments sent is specified as,

4T P Dji = cD D i i D i
i + αi 4T P Dj−1 + j − βi j−1 , (9)

where ij is the random shock to participant i on day j. The maximum likelihood estima-

tion procedure is used to estimate the ARIMA(1, 1, 1) model.

C. A lognormal diffusion process model

The total payment income is modeled by a lognormal diffusion process,

dT P Iji /T P Iji = µIi dt + σiI dwji , (10)

where wji is a standard Brownian motion process, µi is the instantaneous growth rate of

the payment value, and σi is the instantaneous volatility of the growth rate of the payment

11
 value.

Similarly, the total payment sent is modeled as,

dT P Dji /T P Dji = µD D i
i dt + σi dwj . (11)

3.2.2 Performance of alternative models for forecasting payment transactions

Since any net debit position incurred by a participant must be fully collateralized in Tranche

1 of the LVTS, Tranche 1 is very similar to an RTGS system. The focus of our paper is on

the Tranche 1 payment stream. The transaction and credit limit data are obtained from

the Payments Canada over the period from January 4, 2016 to December 29, 2017, with the

exact time of the payments sent, payments received, and collateral (credit limits) pledged

by each participant.

We focus primarily on assessing the forecasting performance of the above three models

for five large participants which account for a significant portion of the Canadian financial

sector. The results for the remaining participants are qualitatively similar to those for the

five participants and available upon request. 8

In order to evaluate both in-sample and out-of-sample forecasting performance of the

three models, we divide our data into two subsamples. The first subsample, from January

4, 2016, to November 30, 2017, is used to estimate the model parameters and to evaluate

the in-sample forecasting performance of the three models. The second subsample from

December 1, 2017, to December 29, 2017, is used to evaluate out-of-sample forecasting

performance of the three models. We use the root mean squared error (RMSE) to evaluate
8
As of 2018, there are 17 financial institutions, including the Bank of Canada participating in the LVTS.

12
 the in-sample and out-of-sample forecasting performance of the three models. 9

For participant 1 (Bank1), participant 2 (Bank2), ..., and participant 5 (Bank5), the esti-

mation results from the SUR model are reported in Tables 1-3. All coefficients of this

model are statistically significant at a significance level of 5%. The estimation results of

the ARIMA(1, 1, 1) model and the lognormal diffusion process model are reported in Table

4 and Table 5, respectively.

For in-sample forecasting, we have,

v
u
uj=2017 November 30
u X
RMSE = u t (yji − ŷji )2 /505, (12)
j=2016 January 4

where yji is for either the total payment sent or the total payment received, and ŷji is either

the in-sample forecasting value for the total payment sent or the in-sample forecasting

value for the total payment received. We have 505 observations over the period from Jan-

uary 4, 2016, to November 30, 2017.

For out-of-sample forecasting, we have,

v
u
uj=2017 December 29
u X
RMSE = u t (yji − ŷji )2 /29 (13)
j=2017 December 1

where yji is for either the total payment sent or the total payment received, and ŷji is ei-

ther the out-of-sample forecasting value of the total payment sent or the out-of-sample

forecasting value of the total payment received. We have 29 observations over the out-of-
9
In-sample analysis is important and can reveal useful information about possible sources of model mis-
specification. In practice, however, what matters most is the evolution of the payment transactions in the
future, not in the past. A model that fits a historical data well may not forecast the future well because of
unforseen structural changes or regime shifts in the data-generating process. Therefore, from both practi-
cal and theoretical standpoints, in-sample analysis alone is not adequate, and it is necessary to examine the
out-of-sample predictive ability of payments transaction models (White, 2000).

13
 sample forecasting period from December 1 , 2017, to December 29, 2017.

The RMSE values of both in-sample and out-of-sample forecasts are reported in both Table

6 and Table 7, respectively. Among all the three models, it is noticeable that the ARIMA(1,

1, 1) model consistently shows the best performance for both in-sample forecasting and

out-of-sample forecasting across all participants. The ARIMA(1, 1, 1) model uses a combi-

nation of past values and past forecasting errors and therefore it offers a potential for fitting

data that could not be adequately fitted by using a linear regression model or a diffusion

process model.

Since the ARIMA(1, 1, 1) model has a better forecasting performance than other two mod-

els, we use the ARIMA(1, 1, 1) model to fit the data and obtain the predicted total payment

transactions for each participant. The predicted total payment transactions are used to

compute the values of the intraday liquidity risk indicator for each participant.

14
 Table 1: Estimation results of regression models for total payment transactions

Payments sent Payments received

Coefficient t-test statistic Coefficient t-test statistic
Bank 1
Intercept 23. 37 1226 23.37 1189

Tue -0.029 -1.636 -0.0239 -3.904

Wed -0.017 -7.235 -0.0203 -2.781

Thu -0.006 -2.970 -0.0046 -1.742

Fri 0.0421 1.827 0.0505 1.885

Holiday 0.0501 1.949 0.0414 1.954

Family-day -2.819 -22.104 -2.8039 -21.723

Aug-civic -3.286 -25.773 -3.0172 -23.375

Bank 2
Intercept 23.98 1271 23.980 1282

Tue -0.0129 -1.505 -0.008 -3.079

Wed 0.04863 1.916 0.046 1.847

Thu -0.0138 -2.653 -0.010 -3.142

Fri 0.04228 1.649 0.038 1.506

Holiday -0.0179 -2.816 -0.016 -3.427

Family-day -3.2845 -26.026 -3.277 -26.672

Aug-civic -4.0521 -32.108 -3.887 -31.634

The data used to estimate the models covers the period from January 4, 2016, to November 30, 2017. The
number of observations is 503. The feasible general least squared method is used to estimate the SUR
model. For Bank1, the model for payments sent: Adjusted R-Squared is 0.70; the model for payments
received: Adjusted R-Squared is 0.68. For Bank2, the model for payments sent: Adjusted R-Squared is 0.78;
the model for payments received: Adjusted R-squared is 0.79.

15
 Table 2: Estimation results of regression models for total payment transactions

Payments sent Payments received

Coefficient t-test statistic Coefficient t-test statistic
Bank 3
Intercept 23.21 719 23.2119 706.10

Tue -0.1460 -3.3552 -0.1336 -3.0182

Wed -0.1089 -2.5106 -0.1039 -2.3498

Thu -0.0719 -1.6553 -0.0672 -1.5178

Fri -0.0037 -0.0769 -0.0025 -0.0549

Holiday 0.04201 0.66827 0.04727 0.73478

Family-day -2.5549 -11.842 -2.5500 -11.896

Aug-civic -2.8739 -13.321 -2.8641 -13.259

Bank 4
Intercept 24.5789 1376 24.5793 1347

Tue 0.00485 0.20121 -007364 0.29968

Wed 0.06343 2.64114 0.06299 2.56661

Thu 0.05074 2.11028 0.05176 2.10645

Fri 0.07829 3.22719 0.07452 2.99338

Holiday 0.06089 1.74943 0.05692 1.59404

Family-day -2.7173 -22.750 -2.7154 -22.650

Aug-civic -3.2647 -27.333 -3.2348 -26.983

The data used to estimate the models covers the period from January 4, 2016, to November 30, 2017.
The number of observations is 503. The feasible general least squared method is used to estimate the
SUR model. For Bank3, for payment sent: Adjusted R-Squared is 0.39, for payment received: Adjusted
R-Squared is 0.49. For Bank 4, Adjusted R-Squared is 0.73, for payment received : Adjusted R-Squared is
0.73.

16
 Table 3: Estimation results of regression models for total payment transactions

Payments sent Payments received

Coefficient t-test statistic Coefficient t-test statistic
Bank 5
Intercept 23.6837 997.790 23.6813 972.343

Tue 0.01669 0.52102 0.02254 0.68707

Wed 0.07031 2.20237 0.06968 2.12706

Thu 0.05118 1.60160 0.05614 1.71173

Fri 0.07775 2.41105 0.07709 2.31986

Holiday 0.03083 0.66636 0.01812 0.38017

Family-day -2.9705 -18.709 -2.8251 -17.653

Aug-civic -3.2237 -20.304 -2.8466 -17.788

The data used to estimate the models covers the period from January 4, 2016, to November 30, 2017. The
number of observations is 503. The feasible general least squared method is used to estimate the SUR
model. For Bank 5, the model for payments sent: Adjusted R-Squared is 0.62; the model for payment
received: Adjusted R-Squared is 0.58.

17
 Table 4: Estimation results for autoregressive integrated moving average models

Payments sent Payments received

Intercept α β Intercept α β
Bank1 23.3504 0.8711 -0.8351 23.3569 0.8791 -0.8410
(0.1096) (0.1211) (0.0182) (0.0183) (0.097) (0.1095)

Bank2 23.9640 -0.7659 0.7412 23.9650 -0.7229 0.6962

(0.0164) (0.4586) (0.4776) (0.016) (0.4844) (0.499)

Bank3 23.1302 -0.0762 0.1644 23.1322 -0.0737 0.1638

(0.0184) (0.3085) (0.3029) (0.0184) (0.3077) (0.2971)

Bank4 24.5983 -0.0083 -0.0086 24.5997 -0.0076 -0.0080

(0.0139) (1.4180) (1.4244) (0.0138) (1.5266) (1.5342)

Bank5 23.7044 -0.2033 0.15310 23.7093 -0.2190 0.1710

(0.0151) (0.5466) (0.5509) (0.0145) (0.5803) (0.5855)
The data used to estimate the models covers the period from January 4, 2016, to November 30,
2017. The number of observations is 503. The model is estimated by using the maximum likelihood
estimation method. The standard errors are reported in the parentheses.

18
 Table 5: Estimation results for diffusion models

Payments sent Payments received

µ σ µ σ
Bank1 0.08 1175.76 1.14 1457.83
(0.04) (557.23) (0.57) (732.17)

Bank2 0.23 1957.71 1.59 2337.55

(0.11) (998.81) (0.78) (1152.12)

Bank3 1.11 3842.34 0.69 3307.82

(0.55) (1760.10) (0.35) (1679.10)

Bank4 0.66 3889.73 2.51 4027.07

(0.39) (1048.90) (1.33) (1983.71)

Bank5 0.42 1968.06 1.54 2198.07

(0.19) (1046.90) (0.76) (1163.22)
The data used to estimate the models covers the period from January 4, 2016, to
November 30, 2017. Number of observations is 503. The model is estimated by using
the maximum likelihood estimation method. The standard errors are reported in the
parentheses.

19
 3.3 Prediction accuracy of the intraday liquidity risk indicator

Typically, for any given time in a payment day, the LRI can predict intraday liquidity risks

for the remainder of the day for each participant. To evaluate the forecasting performance

of the LRI, we split the data set into an in-sample period from 2016 January 4 to 2017

November 30, used for the initial parameter estimations, and an out-of-sample period

from 2017 December 1 to 2017 December 29, used to evaluate forecasting performance

of the LRI. For the five participants (Bank 1, ..., Bank 5), both the actual and forecasted val-

ues of the LRI are reported in Figures 1-3 across different payment times in a payment day.

95% confidence bands are also reported in these figures to evaluate the indicator forecast-

ing performance. Several general conclusions can be drawn from these figures. First, the

actual values of the LRI for each participant is lower than one, indicating that in practice,

the participants’ liquidity sources during the remaining times are sufficient to cover the

liquidity requirements during the period considered. This suggests that these participants

managed their intraday liquidity positions well to meet payment and settlement obliga-

tions on a timely basis, which contributes to the smooth functioning of payment systems.

Second, the predicted values of the LRI across different participants fall into the 95% confi-

dence bands of the actual values of the indicator, which reflects the fact that the predictive

ability of the indicator performs reasonably well. Third, both the actual and predicted val-

ues of the indicator tend to be more varied in the late afternoon. This can be explained

by the fact that payment activity peaking in the late afternoon makes the indicator more

volatile. 10
10
This pattern of payment activity, i.e., payment activity peaking in the late afternoon, can be explained
partially by recognizing two factors that affect participants’ payment activity. First, the timing of participants’

20
 4 Predicting the likelihood of intraday liquidity risk events

The intraday liquidity risk indicator helps us predict whether the intraday liquidity source

is enough to cover the liquidity demand. However, it can only provide the prediction for

the future value of the intraday liquidity risk, which can at most convey some notion of

future intraday liquidity risk, but it cannot provide the possible uncertainty of the future

intraday liquidity risk. For most decision issues, there is a need to provide insights on the

likelihood of occurrence of intraday liquidity risk for a given period of time (Gneiting and

Ranjan, 2011). In this section, we propose an approach that can predict the likelihood of

occurrence of an intraday liquidity risk event within the remainder of the day, which is

defined as an episode where LRI is greater than or equal to a positive constant C.

Different values of C result in different definitions for intraday liquidity risk events. A lower

value of C leads to the identification of more intraday liquidity risk events. Given the in-

formation set It at time t, the probability that an intraday liquidity risk event will occur

during the remaining day can be expressed as,

i
P [LRIt,j ≥ C|It ]. (14)

i , we cannot directly calculate the

Since we do not know the distribution function of LRIt,j

probability in (14). However, we can use a bootstrap method to obtain the empirical distri-

i . The bootstrap procedure used to predict the probability in (14) consists

bution of LRIt,j

of three steps, as described below.

payment activity reflects underlying customer demand. Settlement of financial transactions customarily takes
place in the late afternoon, which tends to cause a demand for payments late in the day. Second, such timing
also reflects participants’ efforts to synchronize their outgoing payments with the large payment inflows they
expect to receive in the late afternoon.

21
 Step 1: Use the original sample to estimate the unknown parameters in the ARIMA(1, 1,

i }29 .
1) models in (7) and (8) and obtain the estimated residuals: {Rt,j j=1

Step 2: Use the nonparametric bootstrapping method by resampling the residuals in (7)

i }29 . Use (7) and (8), and the boot-

and (8) to obtain the bootstrapping residuals {Rt,j ∗ j=1

strapping residuals to calculate the predicted payments sent and payments received, from

i .
which we can build up the bootstrapping intraday liquidity risk indicator: LRIt,j ∗

Step 3: Repeat step 1 and step 2 R times, at time t the predicted probability of an intraday

liquidity risk event within the reminder of a payment day is computed as,

R
1 X i
I(LRIt,j ∗ ≥ C). (15)
R
j=1

Using the sample period from January 4, 2016, to November 30, 2017, to update the model

parameters and taking R = 100, we obtain the out-of-sample predicted probabilities of

an intraday liquidity risk event from 2017 December 1 to December 29 for the five partic-

ipants, which are reported in Tables 4-6, respectively. As we can see from the figures the

probabilities of an intraday liquidity risk event reach a peak in the late afternoon, which

is consistent with the result that the indicator tends to be more variations during that pe-

riod. Thus, participants face more uncertainty of potential settlement failure if they do not

have sufficient funds to cover the transfer, for example, if the income funds that they are

expecting cannot arrive. The potential of whether there will be sufficient liquidity to cover

outgoing payments demand may raise probability of an intraday liquidity risk event in the

late afternoon.

22
 Table 6: In-sample predictive ability for payments from alternative models

Bank1 Bank2 Bank3 Bank4 Bank5

Payments sent from alternative models
Linear regression model
3.0853 1.9665 4.3291 0.7104 1.9920

Autoregressive integrated moving average

0.0002 0.0523 0.0002 0.0252 0.0721

Log normal diffusion process

3.4374 2.4511 4.6129 1.1509 2.6783

Payments received from alternative models

Linear regression model
0.6638 0.5340 0.9494 0.3966 0.6844
Autoregressive integrated moving average
0.0211 0.0167 0.0008 0.0008 0.0143
Log normal diffusion process
0.6867 0.5894 0.9826 0.4510 0.7019
The data used to evaluate the in-sample forecasting performance covers the period from January 4, 2016, to
November 30, 2017.

23
 Table 7: Out-of-sample predictive ability for payments from alternative models

Bank1 Bank2 Bank3 Bank4 Bank5

Payments sent from alternative models
Linear regression model
3.0853 1.9665 4.3291 0.7104 1.9922
Autoregressive integrated moving average
1.5305 0.7014 1.3405 0.3309 0.4819
Log normal diffusion process
1.7137 0.8954 1.4632 0.4202 0.8832

Payments received from alternative models

Linear regression model
0.6532 0.5454 0.6849 0.4011 0.4667
Autoregressive integrated moving average
0.5463 0.3974 0.5680 0.2856 0.8955
Log normal diffusion process
0.6494 0.5409 0.6715 0.4145 0.6589
The data is divided into two subsamples. The first subsample from January 4, 2016, to November 30, 2017, is
used to update the model parameters. The second subsample from December 1, 2017, to December 29, 2017,
is used to evaluate the out-of-sample forecasting performance.

24
 Figure 1: Intraday Liquidity Risk Indicators for Bank 1 and Bank 2

Bank1
0.8

0.6

0.4
Indicator

0.2

0 Upper Band
Forecasted LRI
-0.2 Actual LRI
Lower Band
-0.4
0 2 4 6 8 10 12 14 16 18
Time

Bank2
3

2
Indicator

-1

-2
0 2 4 6 8 10 12 14 16 18
Time

Figure 1 reports the average values of intraday liquidity indicators over the sample period from 2017
December 1 to December 29 for both Bank 1 and Bank 2 across different payment times in a payment day.
The data over the sample period from January 4, 2016, to November 30, 2017, is used to estimate the model
parameters, while the data over the sample period from 2017 December 1 to December 29 is used to estimate
the values of the intraday liquidity risk indicators for Bank 1 and Bank 2. The intraday liquidity risk indicator
is defined in (2).

25
 Figure 2: Intraday Liquidity Risk Indicators for Bank 3 and Bank 4

Bank3
1

0.5
Indicator

0 Upper Band
Forecasted LRI
Actual LRI
Lower Band
-0.5
0 2 4 6 8 10 12 14 16 18
Time

Bank4
3

1
Indicator

-1

-2

-3
0 2 4 6 8 10 12 14 16 18
Time

Figure 2 reports the average values of intraday liquidity indicators over the sample period from 2017
December 1 to December 29 for both Bank 3 and Bank 4 across different payment times in a payment day.
The data over the sample period from January 4, 2016, to November 30, 2017, is used to estimate the model
parameters, while the data over the sample period from 2017 December 1 to December 29 is used to estimate
the values of the intraday liquidity risk indicators for Bank 3 and Bank 3. The intraday liquidity risk indicator
is defined in (2).

26
 Figure 3: Intraday Liquidity Risk Indicator for Bank 5

Bank5
1.5

1
Indicator

0.5

Upper Band
0 Forecasted LRI
Actual LRI
Lower Band
-0.5
0 2 4 6 8 10 12 14 16 18
Time

Figure 3 reports the average values of intraday liquidity indicators over the sample period from 2017
December 1 to December 29 for Bank 5 across different payment times in a payment day. The data over the
sample period from January 4, 2016, to November 30, 2017, is used to estimate the model parameters, while
the data over the period from 2017 December 1 to December 29 is used to estimate the values of the intraday
liquidity risk indicators for Bank 5. The intraday liquidity risk indicator is defined in (2).

27
 Figure 4: Predicted Probability of an Intraday Liquidity Risk Event

The predicted probability of P(LRI 1 )(Bank1)

0.45

0.4

0.35
Probability

0.3

0.25

0.2

0.15
0 2 4 6 8 10 12 14 16 18
Time

The predicted probability of P(LRI 1 )(Bank2)

0.6

0.5
Probability

0.4

0.3

0.2

0.1
0 2 4 6 8 10 12 14 16 18
Time

Figure 4 reports the predicted probability of an intraday liquidity risk event over the sample period from 2017
December 1 to December 30 for both Bank 1 and Bank 2 across different payment times in a payment day.
The data over the sample period from January 4, 2016, to November 30, 2017, is used to estimate the model
parameters, while the data over the sample period from 2017 December 1 to December 29 is used to predict
the probability of an intraday liquidity risk event.

28
 Figure 5: Predicted Probability of an Intraday Liquidity Risk Event

The predicted probability of P(LRI 1 )(Bank3)

0.2

0.15
Probability

0.1

0.05

0
0 2 4 6 8 10 12 14 16 18
Time

The predicted probability of P(LRI 1 )(Bank4)

0.5

0.4
Probability

0.3

0.2

0.1

0
0 2 4 6 8 10 12 14 16 18
Time

Figure 5 reports the predicted probability of an intraday liquidity risk event over the sample period from 2017
December 1 to December 30 for both Bank 3 and Bank 4 across different payment times in a payment day.
The data over the sample period from January 4, 2016, to November 30, 2017, is used to estimate the model
parameters, while the data over the sample period from 2017 December 1 to December 29 is used to predict
the probability of an intraday liquidity risk event.

29
 Figure 6: Predicted Probability of an Intraday Liquidity Risk Event

The predicted probability of P(LRI 1 )(Bank5)

0.4

0.35

0.3
Probability

0.25

0.2

0.15

0.1
0 2 4 6 8 10 12 14 16 18
Time

Figure 6 reports the predicted probability of an intraday liquidity risk event over the sample period from 2017
December 1 to December 30 for both Bank 5 across different payment times in a payment day. The data over
the sample period from January 4, 2016, to November 30, 2017, is used to estimate the model parameters,
while the data over the sample period from 2017 December 1 to December 29 is used to predict the
probability of an intraday liquidity risk event.

30
 5 Conclusion

In this paper, we construct an intraday liquidity risk indicator for monitoring whether a

participant’s expected liquidity sources for settling payments in the remainder of the day

is sufficient to cover its expected liquidity requirements. Using data from the LVTS to eval-

uate the forecasting performance of the intraday liquidity risk indicator, we find that the

intraday liquidity risk indicator performs reasonably well, suggesting that this indicator is

a useful tool for assessing intraday liquidity risk in an RTGS system.

Based on this indicator, we propose a framework to predict the likelihood of occurrence

of an intraday liquidity risk event throughout the remainder of the payment day. Using

data over the period from 2017 January to December 29, we find that an intraday liquidity

risk event is more likely in the late afternoon, suggesting that participants need to manage

their intraday liquidity in order to synchronize their outgoing payments with the payments

inflows they expect to receive in the late afternoon to avoid the occurrence of intraday

liquidity risk events.

In future work, this intraday liquidity indicator can be used as a metric to conduct liq-

uidity stress testing for an RTGS system, like Lynx, to inform a comprehensive assessment

of whether participants’ internal sources are sufficient to withstand adverse shocks in the

payment system. This kind of stress testing could provide insights for monitoring partici-

pants’ intraday liquidity risks under stressed scenarios.

31
Intraday liquidity
EMMEC – 28 Nov 2023

Vincent Caillon
Group Intraday Treasurer
Société Générale

1
Agenda

1 Context

2 Intraday liquidity management - implementation challenges

3 Addressing intraday related to custodian accounts

4 AI usage for intraday liquidity management

5 RTGS evolutions, impacts on intraday

2
Context

3
Managing intraday liquidity is not an option

4
Intraday status

✓Intraday used to be a « specialist » topic with few regulation documents only as sole compass.

▪ It now gets direct scrutiny from regulators during periodic reviews

▪ Efficient IT tools - if not developed internally - are now offered by several vendors
▪ Consistent large settlement volumes in securities markets mobilize large amounts of Intraday liquidity
▪ Sensitivity in payments management is higher after recent banking difficulties – stress approach is required
▪ Through Put ratio put additional constraints on Intraday active management
▪ Evolution in payments practices (extended hours, IP,…) requires agility in liquidity management.

➢Liquidity fluidity is a must and has to be ensured in any circumstances, especially to avoid any negative stigma.

5
Intraday liquidity
implementation
challenges

6
7
Intraday concepts

▪ Concept defined in BCBS 248 can be refined

✓BAU calculation
✓Historical data averages, 95% 99% percentiles
✓Stress shape distortion
✓Prefunding or « close-to close » applied to some flows
✓« Only TSO » intraday liquidity usage
✓Analytical view by Business Unit
✓Statistical approach - surfaces

Source: BIS « Monitoring tools for intraday liquidity management »

April 2013
8
Implementation challenges – must have

▪ Master Intraday liquidity concepts => design an internal policy including all stakeholders

▪ Measure Intraday liquidity usages : High precision fast IT tool necessary to collect all flows, with enriched details

▪ Identify priority payments (Time Specific Obligations) and all priority commitments to clients and counterparts

▪ Reconcile need to use liquidity with frugality and respect ones’ commitments
Review payments management, sequencing, setup cash pooling…

▪ Adapt bank’s payments prioritization to counterparts behavior

▪ Setup a robust modelization of stress and assess related liquidity needs

▪ Ensure availability of enough Intraday resources in any circumstances

9
Implementation challenges

▪ Infuse Intraday « culture » across the organization into each Business Unit including securities activities

▪ Charging Intraday internally – to clients ?

▪ Anticipate expected flows using all available data and modelization => great advantage for active management

▪ Adopt multi-currency approach ?

▪ Design KPIs like « Intraday Liquidity Efficiency » (see BIS document, April 2023…)

10
Intraday @ Custodian
accounts

11
Custodian accounts

➢Why are they so important for intraday liquidity?

▪ Custodian accounts are material, systemic, and have significant stigma in case of cash shortage
▪ Because not all sales and purchases of securities settle simultaneously, Intraday liquidity is used
▪ 2 funding sources = Liquidity injected from main accounts + Secured intraday liquidity provided by the custodian

➢What are the challenges?

▪ Pure cash injections impact main accounts, therefore increase Intraday buffers requirements
▪ Access to Intraday credit from Central Banks requires a very robust setup
▪ Access to Intraday credit from private custodians can be limited and uncertain in stress
▪ « Know how » and cultures are rarely common between payments and securities departments… while cash and
securities are fungible Intraday which requires both knowledge of securities and payments processes
▪ Cutoff times can be different between Central Bank accounts and Custodian accounts
▪ Strict monitoring of balances is required to avoid costly O/N balances on custodian accounts

12
Custodian accounts

▪ Setting KPIs

✓ Daily Funding Efficiency = Max (Daily funding) / Total amount of purchases of the day

✓ Daily Credit Efficiency = Max (Intraday secured credit) / Total amount of purchases of the day

✓Through-Put Ratios = Purchases settled at given time / Total amount of purchases of the day

▪ Benchmarking custodian accounts (iCSDs, Domestic accounts..)

▪ Identifying margins for improvements

13
Concrete actions

▪ Fully involve Business Units in intraday subjects – regular follow-ups

▪ Review practices for securities settlements

▪ Put in place warnings for atypical flows (syndications, large deals,…)

▪ Develop multi-currency active management between cash and securities accounts

▪ Optimize collateral moves to reduce Intraday liquidity usage, focusing on timings and booking efficiency

➢Ultimately gains: reducing usage of Intraday scarce resource and settlements fails, improving client satisfaction
and market reputation

14
AI for Intraday
liquidity management

15
Mobilization of SG stakeholders

MULTIPLICITY OF STAKEHOLDERS

Flow Pilots Short Term Banking

The role of the back- Treasurers Counterparties
office flow pilots is to Manage short term Counterparties
orchestrate funding like money which are facing
and process market deals and end- Societe Generale for
the payments of-day balances debits /
credits flows.

Vincent
Securities Account Intraday Central banks
Managers Treasurers Design rules related
Manage securities
Supervise to liquidity on
accounts for the
intraday liquidity accounts managed
bank - which
management to by banks, and apply
uses liquidity
ensure fluidity and regulation
resource
availability

SOCIETE GENERALE ACTORS EXTERNAL ACTORS

(1) Source : https://www.bis.org/publ/bcbs248.pdf
What was the initial pain point with intraday
liquidity steering ? - reminder

For our clients For internal teams

Stress identification for Limit high costs associated Optimize cash allocation Given the materiality and criticality
dynamic liquidity with Intraday liquidity buffers: between different accounts of flows, absolute need to monitor
management to match actual it is a scarce resource for banks held in each currency, requiring Intraday liquidity, to:
needs (ex: Covid - March 2020) => "frugality" is necessary cash transfers / pooling between 1. Control the usage,
these accounts all along the day 2. Provide fluidity and comfort in
payments management,
3. Respect our commitments to
clients and the market
4. Avoid any stress situation
which could harm confidence in
our institution
Key takeaways

ORGANIZATION: sponsorship of executives,

1
commitment of sponsors, involvement of all stakeholders (R.A.C.I)

AVOID PRODUCING DATA FOR DATA: target concrete

2
actions from the beginning

CONFIDENCE IN AI TOOL: building trust is crucial

3 right from the implementation phase to
ensure involvement of stakeholders and success
RTGS evolutions,
Impacts on intraday

19
RTGS towards 24/7 ?

▪ G20 endorsed in 2020 a roadmap to enhance cross-border payments, developed by the Financial Stability Board
(FSB) in coordination with the Bank for International Settlements Committee on Payments and Market
Infrastructures (CPMI) and other relevant international organisations and standard-setting bodies.

▪ An extension of RTGS operating hours should reduce friction in cross-border payments and is now considered.
Some large Central Banks like FED are already open almost 24/5. The ECB extended already in March 2023 the
opening of Target2 to 2:30 CET, and BoE is considering such a move (recent consultation).

▪ Many initiatives are also seen in the market from private entities looking to offer 24/7 multi currency payments
facilities, that encourage such development efforts.

➢Gradual extension of RTGS opening hours is well on its way.

20
Global settlement window

▪ Source: BIS CPMI consultative

report Nov 2021

21
RTGS towards 24/5 ?

Operational and business reasons are in favor of targeting 24/5

▪ Most participating banks are closed on week-ends (staffing issues,…)

▪ Activity is very tiny on week-ends and demand is limited at this stage
▪ Solving specific difficult situations related to financial stability and resolution policy for troubled institutions
▪ IT upgrades often require time
▪ IP systems cover some of the needs (retail,…)
▪ Screening sanctions filters can be problematic

➢Reg. Intraday liquidity why 24/5 would be preferable to 24/7:

✓Intraday liquidity management could be managed in a much more efficient way…
✓…especially as access to liquidity from other banks is well available in the market if needed.

22
RTGS extension - challenges for Intraday

▪ Leverage on existing experience (Fed 22/24,…)

▪ Adapt to new clients demands
▪ Modify custodian accounts liquidity management
▪ Intraday liquidity active management
✓ Follow the sun process
✓ Automation - priority rules driven
✓ Intraday buffers available at any time
✓ New alternatives to source Intraday liquidity anytime – Central Bank Liquidity Bridges ?

23
Multi-currency intraday approach

24
Multi-currency Intraday approach

✓A consolidated view with stress approach

▪ BCBS248 : « If a bank associated to Central Bank Liquidity Bridges
- can prove to the satisfaction of its supervisor that it can be a solution.
manages liquidity on a cross-currency basis
- and has the ability to transfer funds intraday with ✓Available: Euro Liquidity Bridge between ECB
minimal delay – including in periods of acute stress – and BoE as an example

- then the intraday liquidity positions across currencies

may be aggregated for reporting purposes.” ✓Set up in stress: Reciprocal swap lines
between ECB and several Central Banks like
Fed, BoC, Sveriges,…

Central banks already offer such facilities and these could be developed in the context of longer opening hours.

25
Central Bank Liquidity Bridges

European Central Bank Bank of England

❑ Commercial Bank ❑ Commercial Bank +870m£

1bn€

❑ Bank of England

26
Optimizing Intraday liquidity across currencies

▪ Currency A

▪ Currency B

27
 Payment throttling:
a balancing act for
managing intraday
liquidity risk

Payment prioritization strategically

Introduction
regulates outflows to safeguard a bank’s
intraday liquidity position.
In the financial services industry, the management of
In brief intraday liquidity is the cornerstone of a well-functioning
• Payment throttling is a key strategy payment and settlement system and a key aspect of
financial institutions use to manage overall financial stability. Effective intraday liquidity
intraday liquidity and maintain stability management means that financial institutions can
by prioritizing the timing of transactions. meet their payment obligations at the time expected,
such that there are no systemic delays impacting
• A well‑designed throttling framework
the stability of their customers, counterparties, and
balances the dual objectives of
the broader financial markets. At the same time,
protecting a bank’s own safety and
the institution itself should have sufficient liquidity
soundness and minimizing the impact of
to support its operations throughout the day. The
payment delays on its counterparties.
importance of effective intraday liquidity management
is only going to increase as FedNow and other global
real-time payment mechanisms are quickly changing
the nature of “intraday” liquidity management
and making it more of 24/7 operational liquidity
management requiring more dynamic operational and
analytical capabilities than those that exist today.

Payment throttling: A balancing act for optimal intraday liquidity | 1

The formalization of payment-throttling practices
Key components of an intraday
is driven by the need to harmonize the flow of
transactions and the fluctuations in liquidity. This throttling framework
is particularly important during times of market
volatility and high payment volumes. Intraday liquidity
management practices, specifically the ability to
identify and prioritize time specific obligations and the
Intraday
related operational capabilities continue to be a U.S. throttling
supervisory priority per Regulation YY – Enhanced triggers and
Prudential Standards as well as SR 10-6 Interagency governance
Policy Statement on Funding and Liquidity Risk
Management.
Payment throttling allows institutions to prioritize Intraday Payment
Reporting
payments so that the most critical transactions are throttling prioritization
and
processed first, while others may be delayed pending framework methodology
analytics
sufficient liquidity. The main goal is to avoid systemic
disruption while preserving the firm’s liquidity
position.
This article explores essential elements of intraday Operational
throttling frameworks; regulatory expectations processes,
that govern them; and current industry practices. controls and
testing
Additionally, it examines the strategic approaches
firms should adopt in implementing these frameworks
and the role of artificial intelligence (AI) in optimizing
and future-proofing payment operations.

Payment throttling: A balancing act for optimal intraday liquidity | 2

Intraday throttling triggers Payment prioritization
and governance methodology
A robust governance structure is essential to The prioritization methodology is the systematic
confirming that payment throttling aligns with the approach used to determine the order in which
institution’s risk appetite, regulatory requirements payments are processed. This methodology should
and operational objectives. This structure typically consider various factors such as counterparty and
involves a governance committee, clear roles and client relationships, intraday credit arrangements with
responsibilities, and escalation procedures for customers, negative market signaling, and regulatory
decision‑making during periods of stress. considerations. High‑priority payments, such as those
related to market settlements or critical business
Triggers are predefined conditions or thresholds
operations, are processed first to reduce risk and
that when breached initiate a response such as
maintain business continuity.
the activation of throttling mechanisms. These
triggers should be aligned with the firm’s approved There should be a clear linkage between the
risk appetite statement, with linkage to the prioritization methodology and the throttling triggers.
contingency funding plan (CFP) if stress is severe Such a linkage should consider the stress continuum.
enough to activate safeguards. In calibrating As an example, a minor deterioration in the firm’s
their intraday risk appetite and limits that initiate liquidity position (trigger level) may not warrant a
throttling, firms should consider factors such as throttling action and only serves as an alert level for
intraday buffer and usage amount, contingent heightened monitoring. However, as the bank moves
capacity, overdraft limits, and historically stressed along the stress continuum, reflecting further decline
payment profiles and volume behaviors. in its liquidity position, it can initiate throttling of
certain non‑priority payments to preserve liquidity
Effective governance means that triggers are set
while continuing to process time‑critical payments.
appropriately, monitored continuously, and
adjusted as market conditions or the institution’s
risk profile changes.

Triggers are predefined conditions or

thresholds that when breached initiate a
response such as the activation of throttling
mechanisms. These triggers should be
aligned with the firm’s approved risk
appetite statement, with linkage to the
contingency funding plan (CFP) if stress is
severe enough to activate safeguards.

Payment throttling: A balancing act for optimal intraday liquidity | 3

Operational processes, controls
Banks should look to configure their
and testing systems to classify payments by their
relative priority and execute throttling
Payment‑throttling processes include the
in line with the prioritization framework
mechanisms for monitoring intraday liquidity
approved by the governance committee.
position and payment flows, initiating throttling
and subsequently managing the payment release
Triggers and alerts should be automated
when the liquidity improves. Automation plays a such that a breach in approved threshold
key role in establishing that these processes are results in payments being throttled until
efficient, accurate and scalable. Banks should look reviewed and approved for release per the
to configure their systems to classify payments by escalation protocol.
their relative priority and execute throttling in line
with the prioritization framework approved by the
governance committee. Triggers and alerts should be
automated such that a breach in approved threshold
Reporting and analytics
results in payments being throttled until reviewed
Reporting and analytics are vital to monitoring the
and approved for release per the escalation protocol.
performance of the intraday throttling framework and
Regular testing of the throttling capability is crucial providing insights into payment operations. Reporting
to verify that systematic configuration and controls mechanisms should deliver timely and accurate
are effective. In addition to operational tests, banks information to internal stakeholders and regulators,
should test the governance and coordination of including data on liquidity usage, payment queue
payment throttling through tabletop exercises. status, and breaches of triggers or limits. Firms should
Results should be documented and made available to also incorporate ex‑post reports to inform senior
the governance committees for their review. Lessons management with a summary of the day’s throttled
learned should be applied in improving the capability payments and hold queues.
and informing trigger calibrations.
By leveraging advanced analytics, institutions can
gain a deeper understanding of their payment
flows, identify trends and forecast future liquidity
needs to make more informed decisions regarding
payment throttling.

Payment throttling: A balancing act for optimal intraday liquidity | 4

The role of AI in intraday The way forward:
throttling
Firms looking to enhance their intraday throttling
capability should adopt a strategic approach that
AI can significantly enhance the capabilities of
encompasses the following steps:
intraday throttling frameworks. AI‑driven systems can
analyze vast amounts of data in real time, identifying 1. Assessment of current capabilities: Evaluate
patterns and predicting liquidity needs with a high existing intraday liquidity triggers, their linkage to
degree of accuracy. Machine learning algorithms can payment prioritization as well as the operational
be used to identify and prioritize payments for hold throttling capability to identify gaps and areas for
and release based on near‑real‑time data analytics; improvement.
adaptive modeling and methodology rulesets, 2. Regulatory alignment: Confirm that the framework
including liquidity, regulatory and firm‑specific is designed to meet current and foreseeable
considerations. AI can serve as a copilot for a regulatory expectations.
cash manager who is evaluating several hundred
3. Technology investment: Invest in technologies that
transactions in real-time-payment queues to see
enable real‑time monitoring of intraday liquidity
which ones are to be released within the guardrails of
position and support throttling in line with changes
the firm’s prioritization framework.
in liquidity position.
Moreover, AI can automate complex decision‑making
4. Process optimization: Streamline processes
processes, reducing the need for manual intervention
to reduce latency and improve the speed and
and allowing for more efficient resource allocation.
accuracy of payment processing. Consider
By leveraging AI, firms can achieve a more proactive
generative AI solutions to support throttling
approach to risk management, anticipating and
decisions.
mitigating potential issues before they arise.
5. Operational processes and governance: Develop
comprehensive operational procedures for staff to
execute throttling and establish clear governance
structures to oversee payment operations.

AI can serve as a copilot for a cash

manager who is evaluating several hundred
transactions in real-time-payment queues
to see which ones are to be released within
the guardrails of the firm’s prioritization
framework.

Payment throttling: A balancing act for optimal intraday liquidity | 5

Ernst & Young LLP contacts

Lisa M Maus Gaurav Kohli Abhishek Bhola Ryan Flesher

Principal Managing Director Senior Manager Manager

Financial Services Financial Risk Consulting Financial Services Financial Risk Consulting
Ernst & Young LLP Ernst & Young LLP Ernst & Young LLP Ernst & Young LLP

+12127733099 +1 212 773 7250 +1 212 360 9618 +1 212 773 1916
[email protected] [email protected] [email protected]
[email protected]

EY | Building a better working world

EY exists to build a better working world, helping to create long‑term value for
clients, people and society and build trust in the capital markets.

Enabled by data and technology, diverse EY teams in over 150 countries

provide trust through assurance and help clients grow, transform and operate.

Working across assurance, consulting, law, strategy, tax and transactions, EY

teams ask better questions to find new answers for the complex issues facing
our world today.

EY refers to the global organization, and may refer to one or more, of the member firms of
Ernst & Young Global Limited, each of which is a separate legal entity. Ernst & Young Global
Limited, a UK company limited by guarantee, does not provide services to clients. Information
about how EY collects and uses personal data and a description of the rights individuals have
under data protection legislation are available via ey.com/privacy. EY member firms do not
practice law where prohibited by local laws. For more information about our organization,
please visit ey.com.

Ernst & Young LLP is a client‑serving member firm of Ernst & Young
Global Limited operating in the US.

© 2024 Ernst & Young LLP.

All Rights Reserved.

US SCORE no. 24378-241US_2

2407‑4569081
ED None

This material has been prepared for general informational purposes only and is not intended to be relied upon as accounting,
tax, legal or other professional advice. Please refer to your advisors for specific advice.

ey.com
ALMA Intraday Liquidity Masterclass
Learn the dark art of intraday, how it fits into your broader liquidity
management framework and the impact to your technology strategy

13th May 2021

Copyright © Baringa Partners LLP 2021. All rights reserved. This document is subject to contract and contains confidential and proprietary information.
Today’s session

Objectives
Explore industry liquidity risk management (LRM) best practices
Understand the drivers of intraday risk and how intraday risk should support the overall LRM
Learn how data and visualisation tools can support the process to analyse, explain and understand risk in a wider context and support
scenario building, simulation and planning activities including mitigating actions

Agenda
Section 1: 11:00 – 12:00
Quick Break: 12:00 – 12:05
Section 2: 12:05 – 12:50
Lunch Break: 12:50 – 13:00
Section 3: 13:00 - 14:00

Please ask questions throughout in the chat window with your name and organisation and we will aim to cover these throughout the
session

5
Copyright © Baringa Partners LLP 2021. All rights reserved. This document is subject to contract and contains confidential and proprietary information.
Intraday Liquidity Management (IDLM/ILM) Overview
Pro-active intraday liquidity management allows firms to optimise their resources effectively and reduce funding costs
and liquidity risk
What is Intraday Liquidity Management?
Intraday Liquidity refers to the availability of liquid and quickly monetize assets to ensure that a Firm’s obligations can be met in a timely manner i.e. pay for its transactions when these become
due.
The management of intraday liquidity therefore requires that firms have strong and robust systems, data, processes and controls to ensure that expected obligations are tracked against actual
transactional activity and available liquid resources. This allows firms to better understand:

What liquidity is available e.g. Cash, Unencumbered Collateral, Credit lines

Where liquidity is available and where it is needed e.g. Nostro bank / Custodian Bank / ICSD / entity / country / currency
How it can be sourced / monetised e.g. FX transaction, Cash Sweep, posting collateral, increase credit lines
When it can be sourced and when it is needed for e.g. same day (intraday / end of day), next day, next week etc.

Contributing factors to the costs of intraday liquidity: The costs may be managed or offset through:
Active Management Retrospective Management

Analytics,
1. Maintaining an intraday liquidity buffer Forecasting reporting
& metrics

2. The additional buffer requirement

Intraday Monitoring Stress testing,
and Flow contingency &
3. Lack of transparency over uses and sources of liquidity management buffer
and business behaviours

4. Ineffective processes, systems, controls and data Meet obligations in Funding

timely manner allocation
6
Copyright © Baringa Partners LLP 2021. All rights reserved. This document is subject to contract and contains confidential and proprietary information.
The different elements of intraday liquidity
Intraday Liquidity can be interrupted differently based on the function or process within an organisation - all elements are
key to the overall liquidity framework to ensure governance, efficient risk management and optimisation of cost.

▪ Risk Management: Leveraging the same data points ▪ Cash management: Leveraging the central bank and Nostro
information to forecast cash flow requirements and obligations
from Intraday through to End of Day liquidity provides
across the account population. Activities include monitoring real
the data building blocks and intelligence to run risk Cash time positions and cross entity and account management
management functions; especially important are stress
testing models, assumptions and scenarios. This can be Management
used to feedback into management actions, planning and
policy setting
Risk Intraday
Management Cash Profile
▪ Intraday Cash Profile: Information is used from the
central bank and Nostro bank accounts to provide a cash
profile for a day, this uses time stamps to plot the cash
▪ End of Day Liquidity: Links the intraday view and
Intraday profile over a day, information can be used to provide BCBS
data to the Liquidity Coverage Ratio, and looks at the Liquidity 248 reporting, peaks and troughs as well as more advances
HQLA position that is unencumbered. This should around cash payment control and pricing intraday cash
also feed into risk and wider liquidity management
End of Day Securities
Liquidity Management

▪ Securities Management: Management of the firms

Intraday depot accounts which should also be linked closely to the
Liquidity cash management function. Securities trades generally
▪ Intraday Liquidity Profile: The combined information from the Profile settle Delivery vs Payment (DvP), however can also be
securities management and the cash management elements will realigned as Free of Payment (FoP) creating liquidity
create a view of the true intraday profile. For example, the liquidity mismatches intraday
impact from a cash outflow to purchase HQLA collateral is sized by
the haircut applied to the securities rather than the full cash
outflow

7
Copyright © Baringa Partners LLP 2021. All rights reserved. This document is subject to contract and contains confidential and proprietary information.
Why is ILM a crucial problem to solve?
Firms must understand factors impacting the size of their liquidity buffer and manage these effectively, or a conservative
liquidity buffer will be required
Cash Position
Start of Day

Time through Day

Intraday Calculated using historical data and plotting

Liquidity the average amount of intraday liquidity
Requirement needed to perform day to day activities.

Stressed
Stress testing for specific market and
Intraday
idiosyncratic scenarios ensure that under stress
Liquidity the liquidity requirement is sufficient.
Requirement

Additional This is determined after considering the

accuracy of the information used to calibrate
Buffer the buffer, overlaid with the current
Requirement capabilities to actively monitor and manage.

A comprehensive operating model combined with data and technology strategy

can reduce/mitigate further additional liquidity buffer

8
Copyright © Baringa Partners LLP 2021. All rights reserved. This document is subject to contract and contains confidential and proprietary information.
How have regulations changed over time?
Increasing regulation has been published in response to the financial crisis to define the requirement for tighter
management of a bank’s intraday liquidity risk.

FEDERAL RESERVE’S SR14-1

BCBS 144: Principles for Sound FEDERAL RESERVE’S SR10-6 BCBS 248 “… [firms] should evaluate
Liquidity Risk Management and “…institutions with material “… a set of quantitative tools to the funding requirements
Supervision payment, settlement and clearing enable banking supervisors to necessary to perform ECB’s final guides FEDERAL Bank Recovery and
“… a bank should actively manage activities should actively manage monitor banks’ intraday liquidity critical for banks on their RESERVE’S Resolution
its intraday liquidity positions and their intraday liquidity positions risk and their ability to meet operations, including capital and Living Wills Directive (BRRD)
risks to meet payment and and risks to meet payment and payment and settlement shared liquidity Dodd Frank Harmonise banks’
settlement obligations on a timely settlement obligations on a timely obligations on a timely basis and outsourced services management living wills or resolution and
basis under both normal and basis under both normal and under both normal and stressed and Guidelines resolution recovery
stressed conditions.” stressed conditions." conditions.” access to FMUs.” published. plan for Banks frameworks

2008 2009 2010 2011 2012 2013 2014 2015 2016 2017 2018 2019 2020 2021

BIPRU 12.3 FCA: BIPRU 12.3 Commission Delegated FEDERAL RESERVE’s Reg The PRA’s approach to Bank of England PRA Bank of Resolvability
“A firm must Liquidity risk Regulation (EU) 2015/61 YY supervising liquidity publishes ‘Pillar 2 Liquidity’ Assessment Framework (RAF)
actively manage its management (The European Commission) Enhanced prudential and funding risks The PRA sets out a cash Sets out the organisations’
intra-day liquidity Implementation of The implementation of Basel standards for foreign Definition of the flow mismatch risk ability to create a framework
positions…” “… on a BCBS framework to III in Europe was published, banking organisations Internal Liquidity framework to assess and to manage funding, valuation
timely basis”. the UK. building on the BCBS with total consolidated Adequacy Assessment monitor liquidity and the orderly unwind in a
international framework. assets of $50bn or Process (ILAAP). mismatches during stress resolution scenario
more. scenarios.

Key: Framework

Regional Implementation
9
Copyright © Baringa Partners LLP 2021. All rights reserved. This document is subject to contract and contains confidential and proprietary information.
The importance of intraday
Understanding intraday liquidity flows and impacts is a crucial step to comprehensively manage risk, control liquidity and
support liquidity needs in BAU and any stress periods or events.

Regulatory Reporting
Business as usual –
intraday Management Reporting
management
Strategic Business
Management
Business as usual
– end of day Stress Testing
management
Risk Management
Financial Resource
Times of
Management
Stress
Data Management
Recovery Technology

Resolution
We’ve seen a resolution precedent in Europe with Banco
Popular in 2017 triggered by a deteriorating liquidity situation

10
Copyright © Baringa Partners LLP 2021. All rights reserved. This document is subject to contract and contains confidential and proprietary information.
Recovery and Resolution Planning
Ensuring firms are adequately prepared to deal with potential recovery and resolution scenarios has become a regulatory
priority over the last few years.
Key Elements
 A contingency plan is developed and serves as the document(s) to capture the specific
actions and elements needed to manage the risks during specific events

 It must also become embedded into processes and functions and tested for its ongoing Governance
validity

 The planning and documentation is owned within the Risk function and will incorporate: Operation of RRP in BAU activity
 Contingency Funding Planning

 Business Continuity Planning

Resolution Scenario Definition
 Recovery Resolution Planning

 Many firms will have an independent review of the plan, its components and map to peers
Capabilities:
and industry best practice
Legal Entity Rationalisation
 As well as mitigating against risk, the plan is also a regulatory requirement for organisations
Liquidity
to articulate the clearly defined plan that allows traceability from BAU through to a
Capital
resolution event, the triggers and actions that would be taken and in the worst case how
Customer Activities
clients, customers, counterparts and the wider market is protected. This includes
Payment, Clearing and Settlement
operational, technology and cyber risks.
Shared and outsourced services

11
Copyright © Baringa Partners LLP 2021. All rights reserved. This document is subject to contract and contains confidential and proprietary information.
Liquidity Risk Management (LRM) frameworks

Approach A: Intraday Liquidity as a separate Approach B: Intraday Liquidity embedded in

capability within liquidity management the wider liquidity management framework

Liquidity Risk Strategy Liquidity Risk Strategy

Liquidity Risk Appetite Liquidity Risk Appetite

Organisational structure, policy and procedures Organisational structure, policy and procedures

Governance and Controls

Governance and Controls

Risk Risk
Risk Risk Risk Risk Risk Risk
assessment/ Reporting assessment/ Reporting
identification monitoring management identification monitoring management
methodology methodology
Intraday
Liquidity
Intraday Intraday Intraday Intraday Intraday

Liquidity Optimisation Liquidity Optimisation

Data and Technology Data and Technology

12
Copyright © Baringa Partners LLP 2021. All rights reserved. This document is subject to contract and contains confidential and proprietary information.
Key insights from the 2021 MORS ALM Survey

Clear majority (65%) cited performance and frequency of

data management as a bottleneck:
The majority of banks (58%) stated that they
could not flexibly change assumptions, drivers,
and sensitivities, to facilitate further analysis
once an initial analysis had been completed:
In more than 70% of banks, the Senior Management and the
Board of Directors lack the tools to support them in the limit-
setting process and in understanding the consequences of their
actions in the broadest context:

13
Copyright © Baringa Partners LLP 2021. All rights reserved. This document is subject to contract and contains confidential and proprietary information.
The data challenge…

Earnings and ALM Exposure

Aggregated Monthly Data
Manual export to
Financial Planning
layer Export to
Credit Risk, RWA & Provisioning Reporting
Aggregated Monthly Data and
Inclusion of off- Visualisation
B/S Income Layer
statement items
Liquidity and Funding and FP overlay
Aggregated Daily / Monthly Data

14
Copyright © Baringa Partners LLP 2021. All rights reserved. This document is subject to contract and contains confidential and proprietary information.
The data challenge…

No lineage, slow and labour-

Earnings and ALM Exposure
Aggregated Monthly Data
intensive process. Manual export to
Financial Planning
layer Export to
Reporting
Difficult to change
Credit Risk, RWA & Provisioning
Aggregated Monthly Data
Inclusion of off-
B/S Income
and
Visualisation
Layer
assumptions, scenarios and
Liquidity and Funding
Aggregated Daily / Monthly Data
statement items
and FP overlay

sensitivities.
15
Copyright © Baringa Partners LLP 2021. All rights reserved. This document is subject to contract and contains confidential and proprietary information.
Treasury Data Best Practice

Integrated Income Integrated

Integrated system statement forecast Reporting Engine
utilising the same generation feeding reporting
data for all risk
With strategic and visualisation
surfaces
planning overlay layer online

• Agile, fast and responsive solution for online Financial Planning and
Stress Testing alike
• Data quality and lineage in a BCBS239 consistent way

16
Copyright © Baringa Partners LLP 2021. All rights reserved. This document is subject to contract and contains confidential and proprietary information.
Structure of the rest of today’s session

Liquidity Risk Strategy

Liquidity Risk Appetite

4
Organisational structure, policy and procedures
Governance and Controls

1. How to get the right data

2 3
2. Understanding the risks
Risk
Risk
Risk Risk 3. Managing the risks
assessment/ Reporting
identification monitoring management
methodology 4. Governing the risks
Intraday Intraday Intraday Intraday Intraday 5. Optimisation

5 Liquidity Optimisation

1 Data and Technology

17
Copyright © Baringa Partners LLP 2021. All rights reserved. This document is subject to contract and contains confidential and proprietary information.
How to get the right data

Copyright © Baringa Partners LLP 2021. All rights reserved. This document is subject to contract and contains confidential and proprietary information.
Traditional technology and data challenges
Regulatory requirements and exams continue to add pressure on complex reporting flows from a technological and data
perspective
Legacy Reporting Flows Target State Approach
Multiple internal units use idiosyncratic data
methodology so is no longer Reg. compliant
Internal Modelling of
Data inputted from Front Idiosyncratic Internal Modelling of Data (if needed)
Office Teams is compliant Data from segregated sources
with regulatory and Infrastructure
reporting requirements Treasury

Finance Risk Control for Internal

Front Office Finance Model

If a separate model is
Middle Office Risk
desired, this is kept separate
Back Office from the main process
Treasury
Front Office

Middle Office ETL

Multiple channels of data
flow and internal models Unverifiable output Back Office
leads to unverifiable output
Data inputted from Front
Office Teams is compliant
with regulatory and Reporting Group
Regulatory / Risk reporting requirements
Regulatory Group required
to unwind changes made by Reporting
Finance so that data Accountability of final
Control acts as a check to
complies with regulatory data is low and may
the data output above
standards not be compliant
Regulatory compliant data
Regulatory reported data

19
Copyright © Baringa Partners LLP 2021. All rights reserved. This document is subject to contract and contains confidential and proprietary information.
Triangulation of data
Controlling and managing data for intraday liquidity can be summarised by 3 key data areas; trade capture, settlement
and the real world.

Transaction Capture Transaction Capture

Ownership by the business to understand booking models
Compression and agree transformation policy and procedures to ensure
Affirmation
accurate capture of trading activity at book and entity level.
Netting /
Aggregation Confirmation

Settlement (Internally)
Create the ability to connect the settlement data back to
trade level entries that can provide accurate traceability but
Settlement Real World
also fair representation on business where netting /
Cash and Security aggregation takes place
movements

Data Control
Data across the organisation is transformed and
changed for multiple reasons at various points of the Real World (Point of Truth)
end to end process. This has created challenges for Reconcile real time the external status with the internal
organisations to be able to effectively track and trace representation including accurate timings to allow clear
the true drivers of liquidity and meet the required reporting and transparency back to business around liquidity
regulation to provide transparency as to the business profile / usage and cost
owners of liquidity movements.

20
Copyright © Baringa Partners LLP 2021. All rights reserved. This document is subject to contract and contains confidential and proprietary information.
MORS Demo Bank
To illustrative the importance of the right data, we have created a demo bank

5bn EUR 35,000 10 million

The “MORS Demo- Unique transactions, Cash flows
bank” balance sheet split into…

21
Copyright © Baringa Partners LLP 2021. All rights reserved. This document is subject to contract and contains confidential and proprietary information.
Understanding the risks

Copyright © Baringa Partners LLP 2021. All rights reserved. This document is subject to contract and contains confidential and proprietary information.
Risk identification
Products, transactions, business events and exposures give rise to liquidity and funding risks within the firm. These can
include but are not limited to:

Types of Liquidity and Funding Risks

Reduction in wholesale secured and Intraday liquidity risk A few example data points that can
unsecured funding help identify intraday liquidity risk
Run-off of retail funding Transfer of liquidity across entities, sectors and
countries News
Correlation and concentration of funding Franchise viability risk Early warning indicators from trading
behaviour (e.g. bilateral limit changes)
Early warning indicators from settlement
Contingent off-balance sheet exposures Marketable asset risk behaviour (e.g. fails)
Margin call analysis
Funding tenors Non-marketable asset risk

Deterioration in firm’s credit rating Internalisation risk

FX convertibility and access to FX markets

23
Copyright © Baringa Partners LLP 2021. All rights reserved. This document is subject to contract and contains confidential and proprietary information.
Understanding the drivers for intraday risk
FIs need to develop a view on how the various risk drivers influence their intraday liquidity profiles during BAU and in
times of stress

Identify risk drivers and relative priority Business lines

Understand dynamics of the driver Clients and FMIs (direct

counterparties clearing)

Intraday
Risk Drivers
Identify KPIs to monitor risk

Correspondent
Affiliates banks (indirect
Reflect considerations in stress testing clearing)
framework

24
Copyright © Baringa Partners LLP 2021. All rights reserved. This document is subject to contract and contains confidential and proprietary information.
Managing the risks

Copyright © Baringa Partners LLP 2021. All rights reserved. This document is subject to contract and contains confidential and proprietary information.
Building a Stress Testing capability
Stress testing is a key tool for a firm to predict how its liquidity position will react to differing scenarios assessed against
the controls and limits embedded within their liquidity framework

Problem Statement Planning Future Vision

Planning for times Stress testing to

of stress become BAU
Stress Testing

Governance & Control

Contingency Funding Planning

Recovery & Resolution Planning

Business Continuity planning

Operational
Situation

Contingency
BAU Recovery Resolution
Funding
Timeline
Event

Key Management Action

Real Time Alert

26
Copyright © Baringa Partners LLP 2021. All rights reserved. This document is subject to contract and contains confidential and proprietary information.
Key dimensions of risk reporting
Comprehensive reporting allows firms to act based on data insights as they understand the changing level of risks and
monitor any early warning indicators

Sample areas
for reporting: Credit line Macroeconomic
Liquidity buffer Limit utilisation
utilisation trends

Collateral
FMIs Market trends
requirements

Key metrics should be

defined for each area:

Average Peak Credit Contractual Enhanced

FMI Average Peak Activity Credit Line Amount Secured vs. Unsecured
Line Utilised Requirements
CHAPS
Euroclear
CLS
CREST
Fedwire Securities
DTCC

27
Copyright © Baringa Partners LLP 2021. All rights reserved. This document is subject to contract and contains confidential and proprietary information.
Governing the risks

Copyright © Baringa Partners LLP 2021. All rights reserved. This document is subject to contract and contains confidential and proprietary information.
Operating model requirements
As firms begin to break down operational siloes to support a more holistic approach to liquidity management, there will
be a necessary shift in the underpinning data and technology.
Liquidity management responsibilities are traditionally spread amongst various teams with oversight by Treasury. In order for this to be
effective, senior management oversight is required with clear delineation of roles and responsibilities across the value chain.

Board
Asset and Liability Committee (“ALCO”)
Intraday Liquidity Oversight Committee
Treasury
2nd
Cash Management Operations Liquidity Committees
Funding Desk LOD
Management

Policy Setting

Payment Flow
Management
Policy
Management Forecasting

Short Term Funding

Regulatory Assessments (ILAAP)

Funds Transfer Pricing

Risk Appetite & Buffer Calibration

29
Copyright © Baringa Partners LLP 2021. All rights reserved. This document is subject to contract and contains confidential and proprietary information.
Liquidity risk appetite and strategy
The cost of suboptimal intraday liquidity management can be high due to additional buffer needs to mitigate risk and
ensure sufficient pool available to meet BAU and stressed requirements
Intraday Liquidity Buffer Requirements

1. Measuring active management needs

1 The Intraday Buffer is calculated using historical data plotting the average amount of intraday liquidity needed to perform day to day
activities, this forms the Intraday Liquidity Requirement

2. Calculating retrospective management needs

To ensure that under stress the buffer requirement is sufficient, stress testing for specific market and idiosyncratic scenarios determine
2 what the stressed Intraday liquidity requirement would be.

3. Sizing add-on requirements

3 The accuracy of the information used to calibrate the buffer, overlaid with the current capabilities to actively monitoring and management
will then determine the additional requirement added

Process improvement change Infrastructure Change

Portion of save
used as
2 investment into
infrastructure

3 3

30
Copyright © Baringa Partners LLP 2021. All rights reserved. This document is subject to contract and contains confidential and proprietary information.
Optimisation

Copyright © Baringa Partners LLP 2021. All rights reserved. This document is subject to contract and contains confidential and proprietary information.
The strategic state of liquidity optimisation
Once the other elements are in place, you can drive real benefits using the data to manage, monitor and report the end-
to-end liquidity of the organisation, allocating costs to provide incentives to the business and controlling policy.

Front Office

Products Services
Asset Class Offerings

External Liquidity Data

Silo Optimisation Silo Optimisation

Silo Optimisation Silo Optimisation

Internal Position with external liquidity data

Liquidity Management: Liquidity Risk: enforce policy

Provide holistic Optimal Liquidity: Across all Asset Class and Products using and provide independent
management to business internal Inventory and external data points oversight

Actual Internal position data

External Fails data

Asset Class Offerings

Products Services

Back Office

32
Copyright © Baringa Partners LLP 2021. All rights reserved. This document is subject to contract and contains confidential and proprietary information.
Opportunities
If the challenges around technology and data are resolved, there are multiple use cases for data analytics to generate
insights and opportunities for improvement

Leverage historical data of payments and receipts Based on all intraday liquidity impacts
Predict future behaviours based on previous trends FTP transparency allows:
Allows profile forecast based on expected activity Funds o Incentivisation of decisions based on cost and profile
Behaviour and plotted timings
Transfer o The business to model the cost of intraday
analytics Projected view can then be monitored in real time o Treasury to manage the liquidity profile collaboratively
with management actions against any breach Pricing
with the business

Limit setting in both a granular and flexible way Simulative profiling of scenarios, assumptions and actions
Allows agility with the day to day management, Allows layered scenarios to be overlaid on historic data to
whilst maintaining robust risk factors test assumptions and understand which management
actions mitigate the risks
Allows specific limits to rapidly change as stress or
Smart limits market impacts are identified
Stress Testing Allow users the ability to play out scenarios via simulation
to ensure effective BAU and stress related decision making
Limits will be based on client at the first instance,
rather than at a high-level currency Create real time war gaming with owners participating to
test the risk strategy

33
Copyright © Baringa Partners LLP 2021. All rights reserved. This document is subject to contract and contains confidential and proprietary information.
Key Takeaways

1. How to get the right data – this should be an integrated

Liquidity Risk Strategy view across risk types that is real-time or near-time
Liquidity Risk Appetite 2. Understanding the risks – a holistic view is required
across organisations with consideration to the risk drivers
Organisational structure, policy and procedures that influence the intraday liquidity profiles during BAU
Governance and Controls

and in times of stress

3. Managing the risks – comprehensive stress testing and
Risk
Risk
Risk Risk reporting (particularly of early warning indicators) are
assessment/ Reporting
identification
methodology
monitoring management critical to identify and mitigate risks before they realise
4. Governing the risks – senior management oversight is
Intraday Intraday Intraday Intraday Intraday
required with clear delineation of roles and
Liquidity Optimisation
responsibilities across the liquidity risk management
framework
Data and Technology
5. Optimisation – there are real opportunities to use the
data to manage, monitor and optimise the end-to-end
Please get in touch with any questions or comments using the below details:
liquidity of organisations
Simon Gray, Baringa – [email protected]
Peter Serlachius, MORS – [email protected]

34
Copyright © Baringa Partners LLP 2021. All rights reserved. This document is subject to contract and contains confidential and proprietary information.
Considerations: Business division allocations
Business division allocations must align to the strategic balance sheet management objectives which must be clearly
stated and communicated with the businesses. These objectives should then be reinforced through key design choices in
three key areas – maturity profiles, FTP curves and balances.
Two potential balance sheet management objectives are to (1) minimise
the bank’s funding spread risk and (2) minimise the bank’s structural (i.e.
If the underlying inputs, processes and behavioural) balance sheet gap risk. The allocation framework cannot
associated charges/benefits are made support both simultaneously so deciding and clearly articulating to the
transparent to business divisions and continually businesses which objective will be supported by the framework is crucial.
reviewed by Treasury, this should lead to Balance Sheet
continued optimisation to ensure the businesses Management
are being appropriately incentivised to steer the Objectives
firm-wide balance sheet in the right direction.
Behavioural Defining these profiles of key behavioural parameters
Charges/
Maturity requires deep understanding across products, businesses
Benefits
Profiles and client segments to develop a connection between the
bank’s relationships with its clients and the predictability
Ideally, the balances used in the allocation Business and time horizon of the funding requirement.
process should directly reconcile with the firm’s
official balance sheet. The lack of standardised Division
data flows and processes can lead to Allocations
inconsistencies, harming credibility and leading
to challenges from business divisions. Processes and
FTP Curves
Systems

Scope of
To incentivise the efficient use of funding across the Balances
firm, transfer pricing the funding costs of all material These should be designed in such a way that Treasury has
balance sheet items is key. Treasury policies should available sufficient levers to steer the balance sheet to
clearly define whether each type of asset and liability meet its strategic objectives e.g. distinct pricing between
is in or out of scope for the allocation process. assets and liabilities, specific legal entity treatment.

36
Copyright © Baringa Partners LLP 2021. All rights reserved. This document is subject to contract and contains confidential and proprietary information.
Challenges and opportunities for business division allocations
Solving the complexities of the allocation processes will allow Treasury to steer the firm’s balance sheet through
transparent incentives and conversations with business divisions

Current Allocation Challenges Business

activity
Allocation models and Treasury
trading activity (e.g. XVA)

 Allocation processes and

methodologies are not transparent
and therefore BDs receive unexpected Business Divisions Treasury ✓ Create better incentives By making
allocation charges (indicative View) allocation processes transparent to the
Types of business allocations: business divisions, it creates incentives
 Data inputs and methodologies are to improve business behaviours to
Markets FTP
inconsistent across business divisions achieve balance sheet management
Allocation
and legal entities, leading to Investment Wholesale Calculations objectives
complexities with calculations Bank Bank (Systems/EUAs) Treasury Rates
✓ Economic profit generation The ability
 A lot of manual effort is required for to trace activity through to
OTC Margin
Treasury to understand and explain the charges/benefits can support accurate
drivers for allocation charges/benefits charging, creating a transparent and fair
XVA model for economic profit
when challenged by BDs Corporate and Retail
Allocation ✓ Pre-trade impact Once inputs and
Regulatory Capital
Business divisions are unable to link Corporate
Retail Bank Outputs calculations are standardised, this
Bank
their trading/settlement activity with enables the business to understand the
Liquidity and Funding link between their activity and
the resulting allocation Allocation charges/
charges/benefits. benefits
associated charges/benefits pre-trade

Treasury are unable to steer the firm’s

balance sheet through the allocation Allocation disputes
framework

37
Copyright © Baringa Partners LLP 2021. All rights reserved. This document is subject to contract and contains confidential and proprietary information.
Process improvement: Payment Control
Controlling size and timing cash payment outflows is an important part of Active Intraday Liquidity Management

Current State Post Optimisation

Lack of Payment Control

and management to
Controlled
policy
payment flow in
line with policy

Active Management Optimisation Opportunities

With lack of control and adherence to policy the Intraday Liquidity Ensuring payment outflows adhere to policy that considers impact
Requirement finds its level based on activity and hard stop limits on intraday buffer can result in a reduced Intraday Liquidity
(e.g. credit facilities, STP KPIs) Requirement that will remain within the tolerance profile. This will
also realise a reduction in retrospective management/stressed
scenario needs*

38
Copyright © Baringa Partners LLP 2021. All rights reserved. This document is subject to contract and contains confidential and proprietary information.
Process improvement: Training and Policy
Beyond cash payment management, there are other operational processes that would create optimisation opportunities
once consistent training and policy are applied

Key areas of focus include:

• Limits: Monitoring limits and managing escalations and remediation activities in coordination with impacted stakeholders based on methodology
and guidance set forth by Treasury and Risk within the governance policy

• Daily coordination: Facilitating the pre-positioning and intraday cash and securities movements across various FMUs and correspondent banks
based on limits, reporting and metrics as well as ongoing input into intraday credit line extensions, in line with Treasury’s policy

• Data analytics: Running reporting on historic cash and securities settlement data to identify trends and provide an attribution of any variances or
otherwise unexpected intraday events. The “explain” should be performed by financial market utility (FMU), entity, currency, business line,
customer/counterparty and product to support balance sheet management, intercompany reporting and effective allocation of charges via funds
transfer pricing and new product pricing activities

• Forecasting: Similar to data analytics and explain, forecasts should provide an aggregate view of cash and security settlements by different views
(e.g., FMU, entity, etc.). This view would be based on the settlement forecasts sourced from various operations teams, but would be more than a
simplistic “sum” as it would provide linkages between cash and securities

• Reporting and metrics: Producing planned and ad-hoc reporting in coordination with lines of business, operations, risk and finance

• Reconciliations: Performing reconciliations between cash and security legs of transactions.

39
Copyright © Baringa Partners LLP 2021. All rights reserved. This document is subject to contract and contains confidential and proprietary information.
Considerations: Securities Requirements
Many solutions exclude securities when building the overall liquidity position. Having a view that only considers cash
creates inaccurate reporting and risk management.
Securities Positions Overall Liquidity Position Cash Positions

1 Accurate 1
Delivery vs Payment Negative If both sides are considered the only impact is the
Negative Delivery vs Payment
Cash and securities settle simultaneously Only viewing securities impact Only viewing cash leg Cash and securities settle simultaneously
haircut value of security position

2 2
Free of Payment Negative Accurate Negative Cash Only
Securities realigned will impact Cash and securities moving independently need
Securities settle without cash Only viewing cash Cash payment or receipt
liquidity position to be tracked to build a complete liquidity profile

3 3
Margin substitution Negative Accurate Negative
Securities are placed as collateral for Margin Call
Securities placed can miss the Both cash and securities collateral present a Any securities placed would be
margin cash impact complete liquidity position missed from profile
Cash is posted as collateral

4 4
Securities Position Negative Accurate Negative
Long securities can be unencumbered to Securities only would miss all Securities can be used to allow cash credit lines. Any securities supporting cash
Cash position
support overall liquidity cash long and short This tracking gives a full liquidity profile view would be excluded Long or short cash position

Difference in Liquidity position reporting

By not having the full liquidity profile view, organisations can be in danger of significantly
Most solutions do not include securities activity or positions
under or over stating their intraday liquidity

By combining cash and

Looking at cash only gives a securities it provides a
distorted liquidity view much smoother and
accurate liquidity position

40
Copyright © Baringa Partners LLP 2021. All rights reserved. This document is subject to contract and contains confidential and proprietary information.
 ARGUMENTA OECONOMICA
No 1 (50) 2023
ISSN 1233-5835; e-ISSN 2720-5088

Mária Vojtková*, Patrik Mihalech**

Intraday liquidity modelling using statistical methods

The correct approach to liquidity risk management in banks is essential for securing their financial
stability. The position of liquidity risk among the rest of bank risks is specific because the negative
outcome is not just a loss, but directly the bankruptcy of the institution. Such an occurrence might start
a chain reaction and bring uncertainty into the entire financial system. This paper focused on one source
of liquidity risk, i.e. management of liquidity throughout the day. The management of intraday liquidity
is related to cash inflows and outflows occurring during the business day, their timing and settlement.
In 2013, the BCBS published the document Monitoring tools for intraday liquidity management, often
referred to by the regulatory authorities. It offers basic concepts of intraday liquidity monitoring and
sketchily defines stress scenarios. The author suggests possibilities of how to perform intraday liquidity
stress testing in a bank, which is often required by supervisors, even though no detailed approach or
methodology as to how to proceed was introduced by the regulators. The research was carried out on
anonymised data of cash inflows and outflows recorded on a central bank reserves account of one of the
Slovak commercial banks. Both a base and four stress scenarios were developed and suggested for the
better understanding of expected cashflows in standard conditions and during stress. The author’s aim
was to develop scenarios in a non-traditional way by means of a basic and EWMA historical bootstrap
simulations, respectively. Stress scenarios are supposed to simulate reputation crisis, disruption in
RTGS payment system, increased deposit outflows and bank run. The purpose of the proposed intraday
liquidity monitoring scenarios was to strengthen resilience not only for a concrete bank, but also the
entire financial system. Intraday liquidity monitoring is a key factor in securing stability of the financial
sector.
Keywords: liquidity risk in bank, intraday liquidity, bootstrap simulation, stress testing
JEL Classification: G21, G32, C15, C83
DOI: 10.15611/aoe.2023.1.08
©2023 Mária Vojtková, Patrik Mihalech
This work is licensed under the Creative Commons Attribution-ShareAlike 4.0 International License.
To view a copy of this license, visit http://creativecommons.org/licenses/by-sa/4.0/
Quote as: Vojtková, M., Mihalech, P. (2023). Intraday liquidity modelling using statistical methods.
Argumenta Oeconomica, 1(50), 151-178.

** Department of Statistics, Faculty of Economic Informatics, University of Economics in Bratislava,

Slovakia. ORCID: 0000-0001-6257-901X.
** Department of Statistics, Faculty of Economic Informatics, University of Economics in Bratislava,
Slovakia. ORCID: 0000-0002-1161-5874.
152 M. Vojtková, P. Mihalech

1. Introduction and literature

Due to their business active ties, banks are exposed to a wide range of risk. Given
the fact that banks have a substantial impact on the financial sector and the national
economy as a whole, they are under banking supervision in order to avoid their
bankruptcy. The main goal of banking regulation is to ensure that banks hold
sufficient amounts of capital necessary to cover their risk exposure (Hull, 2018).
Banking regulation cannot eliminate all sources of bankruptcy, since it is not possible,
but its aim is to ensure that risk exposure is reasonable, and the probability of
bankruptcy is sufficiently low (Skoglund and Chen, 2015). By regulating this sector,
governments seek to create a stable economic environment, where households and
enterprises have confidence in the banking sector.
In commercial banks, the need for liquidity results from fact that their cashflow
profile is uncertain. Banks have to make sure that they can cope with increased
outflows and potentially decreased inflows in any given time, for these changes
might be fully unexpected (Smolík, 1995). From the terminology point of view, one
can come across the concepts of liquidity and liquidity risk. Some authors tend to use
these concepts interchangeably, however, it is beneficial to distinguish between the
two. Farahvash (2020), suggested that liquidity can be defined as ability to repay
obligations in time of their maturity and capability to transform any asset to cash by
market price. From this point of view, measuring liquidity stands for the estimation
of the expected development, while the measurement of liquidity risk represents the
estimation of the negative deviation from the expected development with a given
probability. Liquidity risk monitoring was further elaborated by, for example
Cucinelli (2013), Drehman and Nikolaou (2013), Hong et al. (2014), Ippolito (2016)
and Khan (2017).
It is also necessary to distinguish between liquidity and the solvency of a financial
institution (Scannella, 2016). The theoretical concept of both risks is similar but not
the same. Liquidity represents the ability of a bank to manage cash outflows promptly
and economically relevantly, while solvency is related to a bank’s ability to repay its
obligations in the long term, and is linked mostly to a sufficient amount of own funds
of a bank.
The first serious attempt to unify liquidity risk management across different
institutions and countries was A framework for measuring and managing liquidity
created by the Basel Committee for Banking Supervision (1992). However, it did not
succeed in the methodology definition, nor in the motivation of bank institutions to
increase consistency and improve their processes in the field of liquidity risk
monitoring. The BCBS introduced several definitions of liquidity risk and the
development of its management and regulation, but progress in this topic was
remarkably slow and inadequate to the speed of the banking industry’s development.
In 2006, this approach to liquidity risk management was still very notable among
banks, and regulators insisted on the development of different heterogeneous models
for liquidity profile evaluation (Castagna and Fede, 2013).
 Intraday liquidity modelling using statistical methods 153

The crisis that began in 2007 showed that the banking sector was completely
unprepared for the management of strong liquidity shocks, and the models used by
banks for liquidity crisis forecasting turned out to be ineffective. In the same manner,
the models applied by the regulatory institutions were also overly optimistic. The
measurement and management of liquidity risk were not considered a priority among
bankers, and the literature dedicated to this topic also failed to cover this as a whole,
resulting in the non-existence of any integrated management process of liquidity risk
(Giordana and Schumacher, 2013).
Scannella (2016) distinguished between two types of liquidity risk: funding and
market liquidity risk. Market liquidity risk can be caused either by external factors
(such as the condition of the financial markets) or internal factors (such as the size
and structure of the financial institutions’ bond portfolio). Funding liquidity risk is
identified by the fact that the bank is not able to manage the expected or unexpected
cash outflows effectively. In other words, this occurs when a bank cannot satisfy its
obligations in time of maturity. Among the sources of funding liquidity risk one can
include:
• liquidity mismatching risk – a mismatch between the size and maturity of cash
inflows and outflows,
• liquidity contingency risk – future events may cause an increased need for
liquidity,
• intraday liquidity risk – the inability to settle payments throughout the day and
fulfill collateral requirements.
The reviewed studies and publications focus on intraday liquidity monitoring by
means of parametric methods. The aim of this research was to outline the possibility
of using a non-parametric simulation method. Inflows and outflows are not generated
using a known or assumed distribution. This study was based on the use of an
empirical distribution, which allows avoiding erroneous assumptions.

1.1. Intraday liquidity

Apart from strategic liquidity risk management over longer time horizons, banks
must also deal with the availability of liquidity throughout each business day. They
ought to have a sufficient amount of resources to settle all cash operations which are
due on a given day and time of their maturity. The sources of intraday liquidity are
highly liquid assets which are available throughout the banking day for the settlement
of payments. The management of intraday liquidity needs represents a set of metrics
and procedures carried out in order to secure the timely settlement of obligations
(Ball et al., 2011, Farahvash 2020). Banks as intermediary institutions execute
a large number of payments, which can be either cash inflows – the bank is the
recipient of the cashflow/s – the bank sends money to another bank. The volume and
multitude of cash inflows and outflows may significantly vary throughout the day.
154 M. Vojtková, P. Mihalech

An extensive number of cash outflows may lead to problems related to insufficient

resources available for settlement. Managing intraday liquidity must play an
important role in the risk management of a bank (Soprano, 2015).
Settlement of payments entered by clients into the banking system is realised in
minimum reserve requirements (MRR) accounts, which banks must have in the
central bank of a given jurisdiction; for Slovakia, this is the National Bank of
Slovakia (NBS). The minimum reserve requirement is a given amount of money
that banks must hold in the NBS on MRR accounts. The minimum requirements are
determined for 6-week periods, based on the bank balance sheet. The required
pre-set amounts of money have to be available in the central bank, on average over
the maintenance period. Due to the settlement of intraday payments, banks tend to
have much higher resources allocated in MRR accounts as prescribed by the
regulations. Payments between banks are usually realised by means of Real-Time
Gross Settlement payment systems (RTGS). In terms of the eurozone, the most
important is TARGET2. One can simply say that all payments in TARGET2, which
is the bank – either a sender or a recipient – are deducted or added to the MRR
account. For this reason, banks tend to maintain sufficient amounts of cash in MRR
accounts.
Intraday liquidity management came into the domain of the BCBS and in 2013
Monitoring tools for intraday liquidity management was published. This document
defined the basic concepts and approaches to the measurement and management of
liquidity cash flows during the day, according to which banks should perform the
following activities:
• measurement of expected daily gross liquidity inflows and outflows and anticipate
their timing where possible,
• monitoring of intraday liquidity positions in terms of expected activities of
a bank,
• securing of sufficient funding sources to cover intraday liquidity needs,
• management of timing of cash outflows in line with intraday objectives,
• development of plan how to proceed in cases of unexpected intraday cash
outflows.
Sources of intraday liquidity are used throughout the day mostly on payments
realised by payment systems for correspondent banks, dedicated lines offered to
clients for intraday usage and unexpected expenses related to the failure of payment
transactions. Thus the BCBS introduced seven monitoring tools whose aim is to
identify intraday liquidity needs (BCBS, 2013). BCBS here suggested monthly
reporting of these tools to regulators, however, their monitoring by banks is expected
to be carried out on a daily basis. Given that not all of these tools are applicable to
all banks, they were divided into three categories. The monitoring tools and their
respective categories are shown in Table 1, and the briefly described tools are
discussed in this paper.
 Intraday liquidity modelling using statistical methods 155
Table 1
Intraday liquidity monitoring tools

Category A: Tools applicable to all reporting banks

A(i) Daily maximum intraday liquidity usage
A(ii) Available intraday liquidity at the start of the business day
A(iii) Total payments
A(iv) Time-specific obligations
Category B: Tools applicable to reporting banks that provide correspondent services
B(i) Value of payments made on behalf of correspondent banking customers
B(ii) Intraday credit lines expended to customers
Category C: Tool applicable to reporting banks which are direct participants
C(i) Intraday throughput

Source: based on BCBS (2013).

Category A: Monitoring tools applicable to all reporting banks

• A(i) – Daily maximum intraday liquidity usage
This tool allows supervisors to monitor bank’s intraday liquidity usage under
standard operating conditions. The measurement is based on the net balance of all
payments – sent and received during the day in all central bank accounts or with
a correspondent bank. The largest net negative position occurred during the day
stands for maximum intraday liquidity usage. All payments are recorded in order of
settlement and maxima of liquidity usage is calculated at the end of the day. Cash
flow CFt in time t stands for inflow CFt In when cash flow is positive and outflow
CFt Out , when cash flow is negative. Net cumulative liquidity position in time t can be
expressed as the sum of cash flows up to time t:
t t t
CF
=t
Net

=i 1 =i 1
∑=
CFi ∑CFi In - ∑CFi Out ,
=i 1
(1)

CF if CFi > 0 0 if CFi ≥ 0

CF( iIn) =  i , CF(Out
i)
= .
0 if CFi ≤ 0 -CFi if CFi < 0

Maximal usage of intraday liquidity is expressed by the largest negative net

cumulative position and can be written as follows, where n stands for number of
payments occurred on a given day:

Maximal liquidity usage = min {CFt Net } . (2)

0<t ≤ n
156 M. Vojtková, P. Mihalech

• A(iii) – Total payments

This tool is aimed at monitoring of the bank’s payment activity – sent and received
payments during the day through the central bank account, or with correspondent
banks. Total inflows and outflows can be expressed in the following way:
T T
total inflow = ∑CFi In , total outflow = ∑CFi Out. (3)
i =1 i =1

• B(i) – Value of payments made on behalf of correspondent banking customers

This tool is applicable only to banks which provide correspondent banking
services to other financial institutions. It is calculated as the total amount of payments
realised in the name of all correspondent banks’ customers.
• C(i) – Intraday throughput
The last tool is relevant only for banks which are direct participants of payment
systems. The goal is to measure the share of outflows in a given time period (hourly
for example) to total daily outflows. Outflows OuttTot profile up to time t can be
expressed as stated below, where T stands for total number of time intervals into
which daily cash flows were divided (so denominator stands for total 1 day outflow):

∑
t
CFi Out
Out Tot
t = i =1
. (4)
∑
T
i =1
CFi Out

1.2. Intraday liquidity stress testing

The monitoring tools described previously, provide information about intraday

liquidity profile of a bank under standard conditions. However, they say nothing
about how liquidity profiles change in cases of stress situations in the financial
market, respectively, in the case of stress which occurred because of unavailability
of funding sources, or due to reputational risk concerning a specific bank. The BCBS
states that banks and supervisors should also consider intraday liquidity requirements
in stress conditions. Intraday liquidity stress testing was further elaborated for example
by León (2012), Pagratis (2017), Roncalli (2021) and Liermann et al. (2021). The
BCBS proposes general examples of stress scenarios, and banks are encouraged to
choose which of them are most relevant for their business model. The following
scenarios are proposed:
• Own financial stress – the bank encounters stress situations due to counterparties
declining their payments or denying intraday credit lines. The bank faces a stress
situation because it will be forced to use more sources of liquidity to prevent
delaying its own payments.
• Counterparty stress – one of the major counterparties in the intraday banking
payment system faces a stress situation that makes it unable to realise payments.
 Intraday liquidity modelling using statistical methods 157

This results in a situation where the bank will not receive any payments from this
counterparty.
• A customer bank’s stress – which may result in deferring payments to customers,
generating further loss of intraday liquidity.
• Market-wide credit or liquidity stress – in the case of crisis in the financial market,
it might happen that the market value of high liquid assets held for intraday
liquidity purposes will significantly degrade. A severe decrease in the market
value or credit ranking of unencumbered liquid assets may result in the
inaccessibility of intraday liquidity from the central bank because these assets
might not then meet the criteria for intraday loan anymore.
However, the Basel Committee states that these scenarios serve only as an
example, and banks are encouraged to define their own stress scenarios. Stress testing
of intraday liquidity is often required by supervisors during the Supervisory Review
and Evaluation Process (SREP) as well. Despite the fact that stress testing of intraday
liquidity is frequently required, no direct methodology of how to proceed has been
elaborated. In this paper the author introduces the possibilities of using a historical
bootstrap simulation for the estimation of inflows and outflows in the standard and
specific under-stress conditions.

2. Methodology

2.1. Historical bootstrap simulation

Bootstrapping is a computationally demanding technique used for the estimation

of a variety of statistical metrics. Aside from standard statistical approaches where
inference about population is made from a single sample, bootstrap estimate is based
on random sampling with a replacement. Bootstrap belongs among a broader class
of resampling methods and allows to estimate sample distributions of almost any
statistic. The term ‘bootstrap’ was first used by B. Efron in his paper Bootstrap
methods: Another look at the jackknife (1979). The importance of this approach
started to increase with the development of information technology and the creation
of specialised packages devoted to bootstrap techniques in statistical software. Among
the most used packages in programming language R (used to perform computations
in this paper) are the bootstrap package created by Efron and Tibshirani in 1993
(Efron and Tibshirani, 1993), and the package boot, programmed by A. J. Canty. The
popularity of bootstrapping techniques has increased thanks to its high flexibility and
relative simplicity (Hesterberg, 2011).
Suppose one wants to make inference about parameter θ of random variable X
based on sample data ( x1 , x2 ,…, xn ) with distribution function F ( x;θ ). When the
probability distribution of random variable X is not known, one replaces the observed
sample ( x1 , x2 ,…, xn ) with a new sample obtained from the given sample by random
158 M. Vojtková, P. Mihalech

sampling. By following this approach, one obtains one bootstrap sample. To obtain
a bootstrap estimate of parameter θ of random variable X, one proceeds in
a following way (Fox and Weissberg, 2018):
• From observed values ( x1 , x2 ,…, xn ) of random sample ( X 1 , X 2 ,…, X n ) , one
calculates θˆ as an estimate of parameter θ .
• Next, B random bootstrap samples are created by a replacement with sample size
n from observed values ( x1 , x2 ,…, xn ) . The accuracy of the estimate increases
with the increasing number of bootstrap samples. A disadvantage of a large number
of samples is a higher computational complexity.
• For each of the bootstrap samples one can calculate an estimate of parameter θ
denoted θˆi , where= i 1, 2,…, B .
The concept based on repeated random sampling can also be applied to an analysis
of intraday cash flows. In this case, the author does not estimate a single parameter,
but a process – the development of cash flows during the day. The usage of simulatin
methods in the modelling of non-maturing liabilities was described by Kalkbrenner
and Willing (2005) and Castagna and Fede (2013). Next, a way to apply this process
to intraday cash flows is proposed. The main difference is that, while in non-maturing
liabilities modelling only outflows are modelled, for intraday liquidity modelling
purposes one must simulate both outflows and inflows and then create net cash flows.
For modelling the study used following approach:
• First, denote time horizon T and period [0, T]. In the given case, the time horizon
is one business day, and inflows and outflows are divided into hourly intervals in
order to obtain estimates of cashflows on an hourly basis. The start of the business
day in this dataset is at 7.00 and ends at 18.00. Thus, the time horizon is divided
into 11 parts (T = 11).
• Next, simulate B trajectories of inflows and outflows cumulated by hours. Inflows
and outflows are simulated separately, and each trajectory can be understood as
one bootstrap sample.
• Calculate the expected level of cumulated inflows In ( 0, Ti ) and cumulative
outflows Out ( 0, Ti ) for each step in projection i ∈ {0, 1, …, T} by averaging the
B scenarios.
• Next, calculate the stressed level of inflows volumes on confidence level p –
In p ( 0,Ti ) and outflows Out p ( 0,Ti ) for each projection step i ∈ {0, 1, …, T} by
averaging the B scenarios.
• Finally, denote stress scenarios (by means of choosing confidence level p for
inflows and outflows) and calculate net liquidity flows for each scenario as the
difference between cumulated inflows and outflows. For liquidity risk management
purposes, it is relevant to analyse increase net liquidity outflows, i.e. cash inflows
will be lower as in standard conditions and outflows will be higher.
 Intraday liquidity modelling using statistical methods 159
2.2. EWMA historical bootstrap simulation
In the standard bootstrap technique, each element has in every moment the same
probability of being chosen into the bootstrap sample. Therefore, for each element of
the sample, the probability of being selected is 1/n (where n is the total number of
elements in the sample). However, this might not always be the desired state. In this
case, the subject of analysis are the previously recorded inflows and outflows, and it
may happen that newer observations reflect the actual situation more accurately than
older ones, so it could be beneficial to choose newer cash flows into bootstrap
samples more often than older ones. In this case one can assign a vector of probabilities
to the selected sample, assigning each element the probability of being chosen by
using the exponentially weighted moving average – EWMA (Barbe and Bertail,
1995; Hall and Maesono, 2002). Suppose one has time series of outflows and inflows
recorded during a given time period (e.g. several days). One calculates cumulative
inflows and inflows recorded each day divided into hourly intervals, and to each of
the cash flows a weight is assigned, which denotes the probability of being chosen.
In the case of EWMA historical bootstrap simulation, this weight will be exponentially
decreasing as the records are older. The weights assigned to the records can be
expressed as follows:
1- λ
Wλ ( r ) = λ r -1 , (5)
1- λn
where Wλ ( r ) stands for weight assigned to r-th record, r is number of elements
since most actual record to given record, λ is decay factor and n is number of records
in bootstrap sample (sample size). Weights W represent the probability distribution
of a record being chosen into a bootstrap sample in any step of the simulation. The
sum of all weights has to be equal to 1:
n

∑Wλ ( i ) = 1.
i =1
(6)

During each record selection, a stochastic process is implemented and a random

number from 0 to 1 interval is picked. This number (respectively quantile) is approxi-
1- λr
mated by cumulative distribution Q ( r ).=
1- λn
1- λr .
Q(r ) = (7)
1- λn
By inverting the function above for each random quantile, one obtains the position
number r of the bootstrap record to be used in the simulation step. The weight
function is highly dependent on the choice of parameter λ. Parameter λ must be
a number higher than 0 and smaller than 1. The smaller the parameter, the more the
weight function decreases, which means the increasing probability of choosing
newer records instead of older ones. In the case when the parameter λ is closer to 1,
160 M. Vojtková, P. Mihalech

each record has the same chance of being picked and thus the basic bootstrap is
obtained.
By the correct choice of λ one can determine the effective sample size. The correct
sample size can be checked by Kish’s effective sample size (Masuku and Singh,
2014), which calculates how many elements have the real probability of being
chosen, and is denoted as a proportion of 1 to the sum of squared weights wi :
1
nKish = . (8)
∑
n
w2
i =1 i

3. Data
The analysis was performed on anonymised data from a Slovak commercial bank.
Data consist in balances in one of the central bank accounts used for settlement of
transactions, and was modified and cleared of some specific cash flows which
occurred in a given period that might have violated the results. Additionally, some
data quality issues and extreme values identified to be of non-random character were
fixed and excluded from the analysis. The data were recorded from 23/2/2021 to
30/9/2021, with a gap (from 4/6 to 18/6) due to insufficient data quality. Outflows

Fig. 1. Cash flows used for the analysis. Cumulative daily amounts of inflows and outflows are
shown along with net cash flows.
Source: own study.
 Intraday liquidity modelling using statistical methods 161

and inflows cumulated by days are shown in Figure 1. Net cash flow in time t is
calculated as follows:

1 + ( CFt
CFt Net = CFt -Net - CFt Out ) , (9)
In

where CFt Net stands for net cash flow in time t, CFt In is inflow in time t and CFt Out
– outflow in time t.
The goal of the analysis was to elaborate a liquidity profile of cash flows by
means of a historical bootstrap simulation. For this purpose, the author used inflows
and outflows cumulated by an hour in each recorded day. The main objective was to
create an intraday liquidity profile under normal conditions and a stressed profile
which might indicate a level of outflows and inflows in the case of specific stress
situations described below. The purpose of these scenarios is the better understanding
of intraday cash flows and identifying the possible need to increase intraday funding.
It is also a proposal for banks on how to approach intraday liquidity stress testing,
which is often required by supervisor.

4. Results and discussion

4.1. Historical bootstrap simulation

Outflows CF Out and inflows CF In were both simulated separately, and the process
can be presented in the following steps:
1. Sufficiently large numbers of simulations have to be chosen in order to obtain
stable results. The number of simulations was set to 100 000, i.e. the number
when the results were sufficiently stable and calculation time in R was not insuf-
ficiently long.
2. Random sampling with replacement was performed for inflows and outflows.
Sampling was on hourly basis (e.g. flows for the interval 7:00-8:00 were simu-
lated from cumulated cash flows that occurred only during this hour). Cash flows
were then added up from 7:00 to 18{00 for all simulations to obtain total cash
inflow and total cash outflow for the entire day. Inflow for the day can be ex-
pressed in a following way (the same stands for outflow):
k
( 0, Tk )
In= ∑CF
i =1
,, i∈1,2,...,k.
In
Ini i ∈1, 2,…, k . (10)

Given that business day starts at 7:00 and ends at 18:00, in total 11 hourly cash
flows were added up. The visualised trajectories of 100 simulations are shown in
Figure 2 (outflows are shown with negative operator):
162 M. Vojtková, P. Mihalech

Fig. 2. 100 simulated inflows and outflows during the day grouped by hour.
Source: own elaboration.

3. Cash flows recorded in all bootstrap simulations distribution function of inflows

(outflows) in a given time interval (hour). Based on this function, from all the
simulations one can determine confidence level p – In p ( 0,Ti ) / Out p ( 0,Ti ). Thus,
it can be said that with p% probability, inflows (outflows) are smaller than the
given inflow (outflow). In these terms, In 0,5 ( 0,T11 ) stands for the cumulative
inflow recorded throughout the entire day in the middle of all the simulations
lined up in ascending order.
4. After the computation of all the simulations, net cash flows were calculated as the
difference between the inflows and outflows recorded on given confidence level p:
CF
=i
Net
In p ( 0, Ti ) - Out p ( 0, Ti ) , i ∈1, 2,…,11 . (11)
Inflows and outflows computed for individual hours can be calculated as the
difference between the actual and previous cash flow on given confidence level:
∆ In p (=
0,Ti ) In p ( 0,Ti ) - In p ( 0,Ti -1 ) , i ∈ 2,…,11. (12)

5. Stress scenarios supposed to simulate intraday liquidity stress were then set up on
a qualitative basis. For each scenario, different confidence levels were considered.
Base scenario
As a representation of the base scenario, the author chose median cash flows.
Both cash inflows and outflows amount to the middle simulation in terms of volume,
therefore this scenario can be expressed in the introduced terminology as In 0,5 ( 0,Tk )
 Intraday liquidity modelling using statistical methods 163

and Out 0,5 ( 0,Tk ) . This scenario represents cash flows under standard conditions.
The resulting gross cash flows are shown in Figure 3 with dashed lines, and hourly
inflows and outflows with bars.

Fig. 3. Gross cash flows simulated for base scenario.

Source: own elaboration.

Fig. 4. Net liquidity flow – standard conditions.

Source: own elaboration.

Total cumulative inflows in the base scenario reached EUR 114 million and total
cumulative outflows EUR 125 million. Total payments are one of the previously
mentioned monitoring tools introduced by the BCBS, specifically A(iii) – Total
payments, which allows to evaluate the expected amount of total payments. Another
useful indicator is cumulative net liquidity position. The largest negative net cumulative
164 M. Vojtková, P. Mihalech

position stands for the amount of liquidity sources that banks use in standard
conditions and which therefore must be always available. Net cash flows are shown
in Figure 4. standard

Fig. 5. Cash outflows profile – base scenario.

Source: own elaboration.

The net cumulative position is represented by the dashed line. The maximum
stands for the largest positive net cumulative position and basically represents the
largest amount of additional sources of liquidity available resulting from intraday
cash operations. The minimum is the largest negative net cumulative position
representing required sources of funding that banks use during the day. The largest
positive net liquidity position in the amount of EUR 30 million was reached during
the 9:00-10:00 time interval, while the largest negative liquidity position in the
amount of EUR -43 million occurred in the period 14:00-15:00. The net cumulative
position is related to tool A(i) – Daily maximum intraday liquidity usage. Another
BCBS tool that can be calculated from simulated cash flows is C(i) – Intraday
throughput, showing a profile of cash outflows recorded on an hourly basis as
a proportion to total outflows during the day. This shows which hour requires the
highest liquidity for settlement of payments. The outflows profile is shown in Figure
5; on the left y-axis is the amount of outflows in million EUR, while on the right
y-axis there are relative cumulative outflows up to a given hour. The largest outflow
in standard conditions occurs during the period 12:00-13:00 in the amount of EUR
33 million. It is worth noting that the majority of standard outflows happen up to
14:00, i.e. 91%. This means that outflows in the morning and early afternoon hours
are highest, and liquidity required in later hours is not that high.

Stress scenarios
The benefit of the base scenario is that it helps with understanding cash flows’
behaviour in standard conditions. This section defines several stress scenarios, where
cash flows should reflect the occurrence of a non-standard event. All these scenarios
 Intraday liquidity modelling using statistical methods 165

are specified on a qualitative basis by determining cash flow quantiles from a historical
bootstrap simulation. It is necessary to point out that in the case of a real stress
situation, historical data might fail to forecast the correct outcomes, therefore the
results are just a quantification of the estimate. Four stress scenarios were developed:
1. Reputation crisis,
2. Disruption in RTGS payment system,
3. Increased deposit outflows,
4. Black scenario – bank run.
The first scenario is supposed to reflect a reputation crisis. In cases when the bank
is exposed to reputation risk, e.g. some negative information about the bank’s ability
to repay its obligations is shown in the media, even if this information is false it tends
to influence clients’ behaviour and they might withdraw their money from the bank.
A decreased amount of inflows might also be expected, because clients will avoid
sending money to this bank and redirect their cash flows elsewhere. For this scenario,
simulations In 0,25 ( 0,T11 ) and Out 0,75 ( 0,T11 ) were carried out. Cash flows for each
hour will be at the level of 25% simulations with the lowest inflows, and outflows at
the level of 75% simulations with the lowest outflows (this designation is used for all
upcoming scenarios).

Fig. 6. Intraday liquidity flows in reputation stress scenario.

Source: own elaboration.

As expected, this change has quite a significant impact on net liquidity position
and total payments; net position and cash flows are shown in Figure 8. Unlike in the
base scenario, where net position is close to zero (outflows and inflows are close to
166 M. Vojtková, P. Mihalech

equal at the end of the day), in a reputation crisis scenario outflows significantly
exceed inflows. This impacts on net liquidity flows, which at the end of the day
amount to EUR -90 million. The largest net negative cumulative position is EUR
-113 million during the 14th hour, and reflects the amount of liquidity sources that
must be available in the bank to cover all the obligations due during the day. The
cumulated outflow reached EUR 165 million, and the cumulated inflow EUR
83 million.
The second scenario was labelled as a disruption of the RTGS payment system.
This scenario simulates a situation of an unexpected failure in the payment system in
the case of a hacker attack, or due to other external impacts that cause situation when
a bank is unable to accept payments from other banks. Outgoing payments will be
working without restrictions and this scenario can be defined in line with the former
designation as In 0 ( 0,T11 ) and Out 0,5 ( 0,T11 ) . In other words, the scenario simulates
a situation when inflows are zero and outflows are standard. In this case, net liquidity
outflow will be equal to cumulated outflow from the base scenario and reach
the highest negative net cumulative position at the end of the day amounting to EUR
125 million.

Fig. 7. Intraday liquidity flows in RTGS drop-out scenario.

Source: own elaboration.

The third scenario simulates increased outflows due to a higher withdrawal rate
from deposit accounts, not necessarily because the situation of mass withdrawals is
due to some bank specific crisis, but just a higher level of standard outflows caused
by the tax due date or a similar event. Nowadays, corporate clients withdraw their
 Intraday liquidity modelling using statistical methods 167

funds at a higher rate to pay off taxes and this might impact liquidity position. In
addition, the business day before a holiday might record a higher outflow rate from
retail customers. This scenario should reflect such occurrences, and scenario is
designed as In 0,5 ( 0,T11 ) and Out 0,95 ( 0,T11 ) meaning standard inflows and higher
outflows. In this case, the net liquidity position at the end of the day reached EUR
-126 million, and the largest negative net cumulative position EUR -157 million
over the period 14:00-15:00. Total outflows were EUR 240 million and inflows
remained unchanged to the base scenario in the amount of EUR 114 million.

Fig. 8. Intraday liquidity flows in increased deposit outflows scenario.

Source: own elaboration.

The fourth and the final defined scenario is labelled the black scenario. All the
aforementioned scenarios are based on the quantile value of a historical bootstrap
simulation. Given that cash inflows and outflows are strongly positively skewed,
outflows recorded at tails of distributions might be significantly higher than related
quantile value. For this reason, the author introduced conditional value-at-risk (also
referred to as expected shortfall – ES), which can be interpreted as expected loss
from the values exceeding a given quantile. The study applied this metric on cash
outflows so that conditional value at risk stands for average outflow from outflows
that exceed a given percentage of simulations. To project the study’s designation into
this methodology, Out 0,95 ( 0,T11 ) could be labelled as value at risk – VaR _ Out ( 0.95 ).
In order to calculate average outflow from simulations with higher outflows than
95% of other simulations, there would be conditional value at risk, in other words
168 M. Vojtková, P. Mihalech

ES _ Out ( 0.95 ). Expected shortfall is also a coherent risk measure as it satisfies the
sub-additivity property, unlike standard VaR (Horáková, 2015).
All the previously mentioned scenarios were intended to deal with a specific field
of increased liquidity needs during the day. However, the need for liquidity might be
significantly higher in the case of a bank run, i.e. a situation, when clients, due to
some reason – most often a reputational problem, start to withdraw all their deposits
from the bank. It should be noted that for the rest of the risks a negative outcome is
loss, in the case of inadequate liquidity needs it is bankruptcy. Therefore, a crisis
related to a bank run is one of the most severe situations banks can face, at the same
time it is also stressed that this might have no clear cause. Sometimes even an
incorrect interpretation of some steps carried out by banks can cause that clients
commence mass withdrawals of their funds. When other clients recognize this
behaviour, they tend to panic as well, and also withdraw their deposits. This triggers
off a withdrawal spiral when everyone removes their funds from the bank, and even
when the bank is otherwise financially healthy it might face severe difficulties to
withstand the crisis.

Fig. 9. Liquidity flow – black scenario (bank run)

Source: own elaboration.

In the case of a bank run, inflows are expected to decrease and outflows strongly
increase. For this scenario the author proposed In 0,1 ( 0,T11 ) and ES _ Out 0,99 ( 0,T11 )
cash flows, reflected by 10% of lowest inflows and outflows being the average from
1% simulations with the highest outflows. With these assumptions, simulated
outflows reach EUR 334 million, and inflows EUR 60 million. The net cumulative
 Intraday liquidity modelling using statistical methods 169

liquidity position at the end of the day is EUR -274 million, and the lowest recorded
during period 14:00-15:00 amounted to EUR -300 million. In comparison to other
scenarios, this time the liquidity needs are much higher. Here it is necessary to note
that the study did not have historical data with a bank run included (as it is for most
of the banks), and therefore predicting client behaviour is difficult. The usage of
historical simulation as shown here may serve as the basis for some expert judgement
adjustments to the model.

4.2. EWMA historical bootstrap simulation

As already noted, in a standard bootstrap each element has the same probability
of being chosen for simulation at any step of the simulation. Bearing in mind the fact
that the author chose records of different age, it might not be the best approach. Older
data have the same probability of being selected as the newer, even though the latter
might predict the current situation more credibly. This is valid specifically in cases
of long time series that were used as entry data for bootstrap simulation. With only
roughly 7 months of observations being available, this should not be the case,
however the author carried out the EWMA simulation for the purposes of comparison
purposes on these dataset as well.
First, one must choose the value for parameter lambda and calculate weights
Wλ ( n ) for all the records in the sample. For the purposes of the EWMA bootstrap
simulation, it was decided to apply an effective sample size of 3 months, bearing in
mind that the highest probability of being chosen is linked to the newest record, and
that the probability is exponentially decreasing when going further back into the
past. The author chose a parameter lambda equal to 0.9677, and the correctness
of the chosen value can be verified by Kish’s effective sample size. In relation to
the sample (observations from 147 days), an effective sample size with the use of
λ = 0.9677 is equal to 60, what can be considered the average amount of business
days for three months, and thus parameter λ was chosen correctly.
1 1
nKish = → = 59.95076 .
∑
n 2
w 0.0167
i=1 i

By calculating of the weight’s cumulative distribution using Kish’s effective

sample size of 60 and lambda 0.9677, one obtains the value of 0.8675. This value can
be interpreted in a way that a randomly chosen cash flow in any step of simulation
has approximately 87% probability of not being older than 3 months.
1- λr 1 - 0.9677 60
Q ( nr ) =
= → 0.8675.
=
1- λ n
1 - 0.9677147
Due to the high number of simulations, one can also say that cash flows from the
last three months make up 88% of all cash flows selected into the simulation. This
results in newest cash flows having a higher impact on the calculated net flows than
170 M. Vojtková, P. Mihalech

older ones; however, older records are not completely excluded – just their impact
is smaller. The meaning of EWMA bootstrap simulation increases in situations
when cash flows behaviour changes in time, e.g. when the bank faces a recent crisis.
This crisis is then reflected in an EWMA simulation with a higher impact than in the
basic one.

Basic scenario EWMA

First, simulate the EWMA basic scenario. All the inputs remain the same, only
the probability for each record to be chosen is different and determined by the weights
function. The cash flows simulated by the EWMA methodology are shown in Figure
10 (net flows from the basic simulation are also shown for comparison). The net
liquidity flow is in this case EUR -25 million, and the largest negative net cumulative
position is EUR 60 million during the period 14:00-15:00. The net liquidity position
is worse by EUR 14 million, in comparison to the basic bootstrap methodology. One
can conclude that there was a slight worsening of net liquidity positions in the
EWMA simulation, due to higher outflows in the last months shown in the sample.
Total inflows reached EUR 108 million (+6 m EUR) and total outflows were EUR
133 million (+8 m EUR). Changes in the outflows profile were minimal.

Fig. 10. Cash flows – EWMA basic scenario.

Source: own elaboration.

Stress scenarios EWMA and comparison to basic bootstrap

In stress scenarios, the differences between basic and EWMA simulation are
negligible, which confirms that no period of recent stress was visible in the data. Due
to slight differences between the basic and EWMA stress scenarios, the author
introduced only a net liquidity flows comparison. The biggest difference was found
 Intraday liquidity modelling using statistical methods 171

during the first few hours in the third scenario related to increased outflows. In the
first hour of a working day this difference reached EUR 43 million, and in the second
EUR 24 million. The difference in net flows at the end of the day, however, was
almost zero. The first and the second scenarios (reputation and RTGS failure,
respectively) yielded better results for the basic simulation and a slight worsening
was observed in the case of EWMA. In general, it can be concluded that there was
not a very significant difference between both approaches, which was expected –
given that the dataset was relatively small, and no period of stress occurred in the
underlying data. A comparison of all the scenarios is shown in Figure 11. The
numbering of the scenarios follows the order defined above.

Fig. 11. Comparison of basic and EWMA simulated net positions for stress scenarios.
Source: own elaboration.

For the black scenario, the results of the EWMA simulation are better than for
basic one (net outflows at the end of the day amounted to EUR -287 million for the
EWMA simulation, and EUR -287 million for the basic one).
As previously mentioned, a comparison of the basic and EWMA bootstrap serves
mostly as an example. EWMA would be most appropriate when possessing data
recorded in the past several years; in this case one might want to increase the effective
sample size to e.g. one year. From the long-term point of view, it is also necessary to
stress that cash flows have a notional value. If data were taken from several years,
the real value of older cash flows might be different from the notional one in that
time. This is important with respect to the actual trend of increased inflation. If
inflation persisted, cash flows recorded before this period might look small in
comparison to actual amounts, however, in time of their realisation they might appear
higher. This might lead to an underestimation of stress outflows in current conditions.
For this reason, in the case of the biggest dataset it might be beneficial not only to use
172 M. Vojtková, P. Mihalech

Table 2
Comparison of scenarios. Base and EWMA simulation (million EUR)
Cumulated Cumulated Net position Net position Net position
Scenario
inflow outflow end of day Max Min
Basic 113.78 -124.62 -10.84 30.43 / 9:00 -42.77 / 14:00
Basic EWMA 108.43 -133.16 -24.73 21.29 / 9:00 -59.57 / 14:00
1 sc. In_0,25/Out_0,75 82.17 -165.16 -82.99 0 / 7:00 -113.04 / 14:00
1 sc. In_0,25/Out_0,75 EWMA 76.26 -170.45 -94.20 0 / 7:00 -125.76 / 14:00
2 sc. In_0/Out_0,5 0.00 -124.62 -124.62 0 / 7:00 -124.62 / 17:00
2 sc. In_0/Out_0,5 EWMA 0.00 -133.16 -133.16 0 / 7:00 -133.16 / 17:00
3 sc. In_0,5/Out_0,95 113.78 -240.10 -126.32 0 / 7:00 -157.03 / 14:00
3 sc. In_0,5/Out_0,95 EWMA 108.18 -234.88 -126.71 0 / 7:00 -160.39 / 14:00
4 sc. In_0,1/Out_ES_0,99 60.22 -334.30 -274.07 0 / 7:00 -300.49 / 14:00
4 sc. In_0,1/Out_ES_0,99 EWMA 56.97 -317.61 -260.64 0 / 7:00 -286.83 / 14:00

Source: own elaboration.

the EWMA bootstrap but also to recalculate the cash flows to their actual fair value.
The final comparison of all the scenarios and their cumulative cash flows and net
positions is presented in Table 2.

Conclusion

Liquidity risk is one of the major banking risks, especially bearing in mind that
the results of the liquidity crisis might be severe not only for the bank itself, but
could also spread through the entire financial system. This paper focused on one
particular part of liquidity risk management in commercial banks – intraday liquidity
cash flows management. The research was based on Monitoring tools for intraday
liquidity management framework (BCBS, 2013). The Basel Committee also encouraged
banks to perform stress testing of intraday liquidity, however, no detailed approach
on how to do that was suggested. This research focused on providing a relatively
straightforward and easily repeatable solution of doing that by means of a historical
bootstrap simulation.
The approach was introduced on the anonymised data of bank inflows and
outflows occurring during the day, grouped by hour as recorded from February to
October 2021 in a commercial bank operating in the Slovak Republic. Four stressed
scenarios were suggested, however they served only as an example, and other
scenarios can be developed in the same manner. The biggest limitation of this solution
is, naturally, relying on historical data. In the case of a real stress situation, there is
no guarantee that cash flows would behave in the same way.
Cash flows were cumulated by hour, and the entire simulation was carried out on
an hourly basis, however different time intervals can also be considered as well. With
the hourly approach, stress testing could be performed at any given hour of the day
 Intraday liquidity modelling using statistical methods 173

(not just from the start of business hours at 7:00, but e.g. at 12:00). Through this
approach one could stress cash flows for only the remaining part of the day. For
example, by starting the prediction at 14:00, most of the outflows would have
occurred up to this hour (91%), and therefore the resulting net flows might not be as
severe as in a full-day simulation.
This paper also deals with cash flows in their notional amount. This information
itself does not directly reveal if the amount is high or not. All inflows and outflows
occur in the bank’s current account of obligatory reserves in the central bank. For
a determination of the severity of outflows in stress scenarios it might be beneficial
to compare their amount to the average level of the bank’s reserves in the central
bank. If this ratio is high, it means that banks rely on a bigger portion of their reserves
for intraday payments purposes and an increase of these outflows might cause
problems for the banks to cover them promptly. Vice-versa, lower percentage ratio
signals that banks use only a small portion of their reserves on intraday transactions
and their liquidity reserves are at a sufficient level.
A limitation of the research was also the fact that the author had access only to
information about cash flows but no information as to where the outflows go or from
which bank the inflows come from. Another suggestion for stress scenarios could be
a counterparty or a country-specific stress focus on countries from which the most
inflows usually come, and which carry out bigger transactions. Additionally,
information whether cash flow is related to retail, corporate or treasury operation
might be of interest and provide additional insights into cash flows structure. To
obtain this information, further research of this issue is planned in order to understand
the intraday liquidity position as well as possible.

References

Ball, A., Denbee, E., Manning, M., and Wetherilt, A. (2011). Intraday Liquidity: Risk and Regulation.
Bank of England Financial Stability Paper, 11.
Barbe, P., and Bertail, P. (1995). The Weighted Bootstrap. Springer.
BCBS (2013). Monitoring tools for intraday liquidity management. Basel Committee on Banking
Supervision. Retrieved from: https://www.bis.org/publ/bcbs248.pdf.
Castagna, A., and Fede, F. (2013). Measuring and Managing Liquidity Risk. Wiley 1st Edition.
Cucinelli, D. (2013). The Determinants of Bank Liquidity Risk within the Context of Euro Area.
Interdisciplinary Journal of Research in Business, 2(10), 51–64.
Drehman, M., and Nikolaou, K. (2013). Funding liquidity risk: Definition and measurement. Journal of
Banking & Finance, 37(7), 2173–2182.
Efron, B. (1979). Bootstrap Methods: Another Look at the Jackknife. Springer.
Efron, B., and Tibshirani, R. (1993). An Introduction to the Bootstrap. CRC Press 1st Edition.
Farahvash, P. (2020). Asset-Liability and Liquidity Management. Wiley 1st Edition.
Fox, J., and Weisberg, S. (2018). R Companion to Applied Regression: Bootstrapping Regression
Models. R. SAGE publications, 3rd Edition.
174 M. Vojtková, P. Mihalech

Giordana, G., and Schumacher, I. (2013). Bank liquidity risk and monetary policy. Empirical evidence
on the impact of Basel III liquidity standards. International Review of Applied Economics, 27(5).
Hall, P., and Maesono, Y. (2002). A weighted bootstrap approach to bootstrap iteration. Series B:
Statistical Methodology, 62(1), 137–144.
Hesterberg, T. (2011). Bootstrap. Interdisciplinary review: Computational statistics, 3(6), 497–526.
Hong, H., Huang, J.-Zh., and Wu, D. (2014). The information content of Basel III liquidity risk measures.
Journal of Financial Stability, 15, 91–111.
Horáková, G., Páleš, M., and Slaninka, F. (2015). Teoria rizika v poistení. Wolters Kluwer, prvé vydanie.
Hull, J. (2018). Risk Management and Financial Institutions. Wiley 5th Edition.
Ippolito, F., Peydró, J.-L., Polo, A., and Sette, E. (2016). Double bank runs and liquidity risk management.
Journal of Financial Economics, 122(1), 135–154.
Kalkbrenner, M., and Willing, J. (2004). Risk management of non-maturing liabilities. Journal of Banking
& Finance, 28(7), 1547–1568.
Khan, M., Scheule, H., and Wu, E. (2017). Funding liquidity and bank risk taking. Journal of Banking
& Finance, 82, 203–216.
León, C. (2012). Estimating Financial Institutions’ Intraday Liquidity Risk: A Monte Carlo Simulation
Approach. Borradores de Economia, SSRN Electronic Journal.
Liermann, V., Li, S., and Steuer, R. (2021). Kafka – A Practical Implementation of Intraday Liquidity
Risk Management. The Digital Journey of Banking and Insurance, 3(1), 105–116.
Masuku, M., and Singh, A. (2014). Sampling Techniques and Determination of Sample Size in Applied
Statistics Research: An Overview. International Journal of Economics, Commerce and Management,
2(11),1–22.
Pagratis, S., Topaloglou, N., and Tsionas, M. (2017). System stress testing of bank liquidity risk. Journal
of International Money and Finance, 73, Part A, 22–40.
Roncalli, T. (2021). Liquidity Stress Testing in Asset Management – Part 3. Managing the Asset-Liability
Risk.
Scannella, E. (2016). Theory and Regulation of Liquidity Risk Management in Banking. International
Journal of Risk Assessment and Management, 19(1/2), 4–21.
Skoglund, J. and Chen, W. (2015). Financial Risk Management: Application in Market, Credit, Asset
and Liability Management and Firmwide Risk. Wiley.
Smolík, M. (1995). Manažment bankových aktív a pasív. ELITA.
Soprano, A. (2015). Liquidity Management: A Funding Risk Handbook. Wiley.
Received: June 2022, revised: August 2022
Acknowledgements: This research was supported by the Slovak VEGA Grant No. 1/0561/21 “The
impact of the COVID-19 crisis on business demography and employment in the Slovak Republic and
the EU”.
 Intraday liquidity modelling using statistical methods 175

APPENDIX
Results of simulations (million EUR)

Base scenario (basic bootstrap)

Cumulative Cumulative Cumulated
Hour Inflow Outflow Net position
inflow outflow outflow (%)
7:00 5,614 -12,662 -7,047 5,614 -12,662 10%
8:00 3,144 -4,330 -8,233 8,758 -16,992 14%
9:00 44,339 -5,678 30,428 53,098 -22,669 18%
10:00 4,326 -24,197 10,557 57,423 -46,866 38%
11:00 4,456 -7,045 7,968 61,879 -53,911 43%
12:00 4,977 -32,686 -19,741 66,857 -86,598 69%
13:00 4,854 -26,575 -41,461 71,711 -113,172 91%
14:00 3,498 -4,807 -42,771 75,209 -117,980 95%
15:00 26,486 -2,354 -18,639 101,695 -120,333 97%
16:00 3,307 -4,283 -19,615 105,002 -124,617 100%
17:00 8,778 0,000 -10,837 113,780 -124,617 100%

1. stress scenario – reputational crisis (basic bootstrap)

Cumulative Cumulative Cumulated
Hour Inflow Outflow Net position
inflow outflow outflow (%)
7:00 2,670 -20,515 -17,845 2,670 -20,515 12%
8:00 1,746 -6,258 -22,357 4,416 -26,773 16%
9:00 19,886 -11,842 -14,313 24,302 -38,615 23%
10:00 4,694 -38,447 -48,066 28,996 -77,062 47%
11:00 4,535 -8,463 -51,994 33,530 -85,525 52%
12:00 4,410 -35,694 -83,278 37,941 -121,218 73%
13:00 4,215 -30,886 -109,949 42,156 -152,105 92%
14:00 2,934 -6,024 -113,039 45,090 -158,129 96%
15:00 25,704 -2,257 -89,593 70,793 -160,386 97%
16:00 3,112 -4,777 -91,258 73,905 -165,163 100%
17:00 8,269 0,000 -82,989 82,174 -165,163 100%

2. stress scenario – RTGS drop-out (basic bootstrap)

Cumulative Cumulative Cumulated
Hour Inflow Outflow Net position
inflow outflow outflow (%)
7:00 0,000 -12,662 -12,662 0,000 -12,662 10%
8:00 0,000 -4,330 -16,992 0,000 -16,992 14%
9:00 0,000 -5,678 -22,669 0,000 -22,669 18%
10:00 0,000 -24,197 -46,866 0,000 -46,866 38%
11:00 0,000 -7,045 -53,911 0,000 -53,911 43%
12:00 0,000 -32,686 -86,598 0,000 -86,598 69%
13:00 0,000 -26,575 -113,172 0,000 -113,172 91%
14:00 0,000 -4,807 -117,980 0,000 -117,980 95%
15:00 0,000 -2,354 -120,333 0,000 -120,333 97%
16:00 0,000 -4,283 -124,617 0,000 -124,617 100%
17:00 0,000 0,000 -124,617 0,000 -124,617 100%
176 M. Vojtková, P. Mihalech

3. stress scenario – increased outflows (basic bootstrap)

Cumulative Cumulative Cumulated
Hour Inflow Outflow Net position
inflow outflow outflow (%)
7:00 5,614 -33,157 -27,543 5,614 -33,157 14%
8:00 3,144 -20,037 -44,435 8,758 -53,194 22%
9:00 44,339 -44,017 -44,113 53,098 -97,211 40%
10:00 4,326 -38,897 -78,684 57,423 -136,107 57%
11:00 4,456 -8,795 -83,023 61,879 -144,903 60%
12:00 4,977 -45,036 -123,082 66,857 -189,939 79%
13:00 4,854 -35,356 -153,584 71,711 -225,295 94%
14:00 3,498 -6,945 -157,031 75,209 -232,240 97%
15:00 26,486 -2,597 -133,143 101,695 -234,837 98%
16:00 3,307 -5,266 -135,101 105,002 -240,103 100%
17:00 8,778 0,000 -126,323 113,780 -240,103 100%
4. stress scenario – bank run (basic bootstrap)
Cumulative Cumulative Cumulated
Hour Inflow Outflow Net position
inflow outflow outflow (%)
7:00 1,124 -105,512 -104,388 1,124 -105,512 32%
8:00 1,431 -8,898 -111,855 2,555 -114,410 34%
9:00 7,416 -32,482 -136,921 9,971 -146,892 44%
10:00 3,815 -69,070 -202,176 13,785 -215,962 65%
11:00 3,470 -9,985 -208,692 17,255 -225,947 68%
12:00 3,382 -52,329 -257,639 20,637 -278,276 83%
13:00 3,256 -40,472 -294,855 23,893 -318,749 95%
14:00 1,895 -7,534 -300,494 25,788 -326,282 98%
15:00 23,989 -3,035 -279,541 49,777 -329,318 99%
16:00 2,675 -4,980 -281,845 52,452 -334,297 100%
17:00 7,771 0,000 -274,074 60,223 -334,297 100%
Base scenario (EWMA bootstrap)

Cumulative Cumulative Cumulated

Hour Inflow Outflow Net position
inflow outflow outflow (%)
7:00 5,865 -12,662 -6,796 5,865 -12,662 10%
8:00 2,827 -4,679 -8,648 8,692 -17,340 13%
9:00 36,469 -6,535 21,286 45,162 -23,875 18%
10:00 4,168 -29,021 -3,568 49,329 -52,897 40%
11:00 4,955 -8,355 -6,968 54,284 -61,252 46%
12:00 6,002 -31,956 -32,921 60,286 -93,208 70%
13:00 3,258 -27,184 -56,848 63,544 -120,392 90%
14:00 3,223 -5,950 -59,575 66,767 -126,342 95%
15:00 28,562 -2,220 -33,233 95,329 -128,562 97%
16:00 3,789 -4,593 -34,038 99,118 -133,155 100%
17:00 9,309 0,000 -24,729 108,426 -133,155 100%
 Intraday liquidity modelling using statistical methods 177

1. stress scenario – reputational crisis (EWMA bootstrap)

Cumulative Cumulative Cumulated
Hour Inflow Outflow Net position
inflow outflow outflow (%)
7:00 3,059 -20,515 -17,456 3,059 -20,515 12%
8:00 1,281 -7,969 -24,144 4,340 -28,484 17%
9:00 13,881 -16,240 -26,503 18,222 -44,724 26%
10:00 4,100 -35,412 -57,814 22,322 -80,136 47%
11:00 4,758 -10,263 -63,319 27,080 -90,399 53%
12:00 4,794 -34,527 -93,052 31,874 -124,926 73%
13:00 2,916 -30,961 -121,097 34,791 -155,887 91%
14:00 2,645 -7,307 -125,758 37,436 -163,194 96%
15:00 26,749 -2,421 -101,430 64,185 -165,615 97%
16:00 3,468 -4,837 -102,799 67,653 -170,452 100%
17:00 8,604 0,000 -94,195 76,257 -170,452 100%
2. stress scenario – RTGS drop-out (EWMA bootstrap)
Cumulative Cumulative Cumulated
Hour Inflow Outflow Net position
inflow outflow outflow (%)
7:00 0,000 -12,662 -12,662 0,000 -12,662 10%
8:00 0,000 -4,679 -17,340 0,000 -17,340 13%
9:00 0,000 -6,535 -23,875 0,000 -23,875 18%
10:00 0,000 -29,021 -52,897 0,000 -52,897 40%
11:00 0,000 -8,355 -61,252 0,000 -61,252 46%
12:00 0,000 -31,956 -93,208 0,000 -93,208 70%
13:00 0,000 -27,184 -120,392 0,000 -120,392 90%
14:00 0,000 -5,950 -126,342 0,000 -126,342 95%
15:00 0,000 -2,220 -128,562 0,000 -128,562 97%
16:00 0,000 -4,593 -133,155 0,000 -133,155 100%
17:00 0,000 0,000 -133,155 0,000 -133,155 100%

3. stress scenario – increased outflows (EWMA bootstrap)

Cumulative Cumulative Cumulated
Hour Inflow Outflow Net position
inflow outflow outflow (%)
7:00 5,614 -75,707 -70,093 5,614 -75,707 32%
8:00 2,827 -1,474 -68,740 8,441 -77,181 33%
9:00 36,469 -12,492 -44,763 44,910 -89,673 38%
10:00 4,168 -42,483 -83,079 49,078 -132,156 56%
11:00 4,955 -12,316 -90,439 54,033 -144,472 62%
12:00 6,002 -38,465 -122,901 60,035 -182,937 78%
13:00 3,258 -34,042 -153,685 63,293 -216,978 92%
14:00 3,223 -9,926 -160,388 66,516 -226,904 97%
15:00 28,562 -2,596 -134,423 95,078 -229,500 98%
16:00 3,789 -5,383 -136,017 98,866 -234,883 100%
17:00 9,309 0,000 -126,708 108,175 -234,883 100%
178 M. Vojtková, P. Mihalech

4. stress scenario – bank run (EWMA bootstrap)

Cumulative Cumulative Cumulated
Hour Inflow Outflow Net position
inflow outflow outflow (%)
7:00 1,615 -107,576 -105,960 1,615 -107,576 34%
8:00 1,175 -10,109 -114,895 2,790 -117,685 37%
9:00 5,536 -26,959 -136,317 8,326 -144,643 46%
10:00 3,107 -62,236 -195,447 11,433 -206,879 65%
11:00 3,209 -14,270 -206,508 14,642 -221,150 70%
12:00 3,739 -41,415 -244,184 18,381 -262,565 83%
13:00 2,248 -35,488 -277,424 20,629 -298,053 94%
14:00 1,735 -11,145 -286,834 22,364 -309,198 97%
15:00 24,204 -2,424 -265,054 46,568 -311,623 98%
16:00 2,740 -5,984 -268,299 49,308 -317,607 100%
17:00 7,661 0,000 -260,638 56,969 -317,607 100%

Source: own elaboration.

Financial Stability Paper No. 11 – June 2011

Intraday liquidity: risk and regulation

Alan Ball, Edward Denbee, Mark Manning and Anne Wetherilt

Financial Stability Paper No. 11 – June 2011

Intraday liquidity: risk and

regulation

Alan Ball, Edward Denbee, Mark Manning and Anne Wetherilt

The authors are grateful to Damian Harland at the FSA, Toby Davies, John Jackson,
Edwin Schooling Latter, Pippa Lowe, Lisa Newman, David Norcross, Ben Norman, Will Speller
and Peter Zimmerman at the Bank of England, Antoine Martin at the Federal Reserve Bank of
New York and Rodney Garratt at the University of California, Santa Barbara for useful
comments and suggestions. Any remaining errors are our own.

[email protected]
Financial Stability, Bank of England, Threadneedle Street, London, EC2R 8AH
[email protected]
Financial Stability, Bank of England, Threadneedle Street, London, EC2R 8AH
[email protected]
Financial Stability, Bank of England, Threadneedle Street, London, EC2R 8AH
[email protected]
Financial Stability, Bank of England, Threadneedle Street, London, EC2R 8AH

The views expressed in this paper are those of the authors, and are not necessarily those of
the Bank of England. This paper was finalised on 1 June 2011.

© Bank of England 2011

ISSN 1754–4262
Contents

Introduction 3

1 Why intraday liquidity risk matters 4

1.1 Intraday liquidity and payment system risks 4
1.2 Intraday liquidity risk and stress scenarios 6
1.3 How are these risks managed at present? 8

Box 1 Stress scenarios 9

2 Intraday liquidity regulation 11

2.1 Intraday liquidity regulation 11
2.2 Systems and controls and monitoring requirements 11
2.3 Buffer requirements 11

3 Intraday liquidity regulation and double duty 12

3.1 Double duty 12
3.2 Intraday liquidity regulation and double duty 14

Box 2 Intraday liquidity risk: the case of Lehman Brothers 13

4 Incentives created by intraday liquidity regulation 14

5 Ensuring a ‘good’ equilibrium 15

5.1 Implementation of liquidity-saving mechanisms 15
5.2 Throughput rules 17
5.3 Time-varying tariff 19
5.4 Collateral eligibility criteria 19
5.5 ‘Deep dive’ examination of banks’ payments behaviour 20
5.6 Proposed changes to securities settlement systems 20

Box 3 Impact of delays on liquidity in the payment system 16

Box 4 Throughput requirements: a stylised example 18

6 Conclusion 20

Annex: Payment systems and intraday liquidity 22

References 24
 Financial Stability Paper June 2011 3

Intraday liquidity: risk and regulation

Alan Ball, Edward Denbee, Mark Manning and Anne Wetherilt

Banks require access to liquidity intraday in order to settle obligations in payment and settlement
systems. The recent financial crisis has highlighted the need for banks to improve their liquidity risk
management, including the management of intraday liquidity risk. The FSA’s new liquidity regime
includes intraday liquidity as a key risk driver and requires that banks calibrate their liquid asset
buffers considering their need for liquidity intraday, both in normal and stressed circumstances. The
Bank fully supports this approach. However, this will increase the cost of intraday liquidity and so
could create incentives for banks to change their behaviour as they seek to minimise costs. If this
results in payment delays, it risks jeopardising the smooth functioning of payment and settlement
systems. There are a number of tools that authorities could use to minimise the chance of adverse
behavioural changes. Such tools include the introduction of liquidity saving mechanisms, the
strengthening of throughput rules, payment tariffs that vary through the day, setting central bank
collateral eligibility criteria for intraday liquidity and regulatory ‘deep dive’ assessments.

Introduction The introduction of real-time gross settlement (RTGS) into

CHAPS in 1996 eliminated settlement risk between participant
The provision of payment services is one of the key functions settlement banks as payments are settled finally and
of the financial system.(1) Therefore, as part of its broader irrevocably, individually and in real time.(3) However,
financial stability remit, the Bank of England is concerned participation in RTGS systems, directly or indirectly, requires
that banks adequately manage the intraday risks that arise as that banks have access to liquidity intraday. The provision of
part of their obligations to provide payment services to the this creates liquidity exposures between banks. Banks send
wider economy. To the same end, the Bank is also concerned payments but expect to receive the funds back later in the day
that the United Kingdom’s large-value payment and to meet other outgoing payment obligations. In other words,
settlement systems function smoothly and are resilient. credit risk is exchanged for liquidity risk.

This paper aims to outline how intraday liquidity risks arise Banks typically manage their payment flows so that they end
in large-value payment systems and to explain how system the day flat. That is, the value of incoming payments is
participants currently manage these risks. It then describes roughly equal to the value of outgoing ones. This means that
the policies that authorities in the United Kingdom and the liquidity exposures that arise from payment and
abroad are implementing to strengthen the resilience of settlement activity are created and extinguished within the
payment systems to intraday liquidity stress. These actions working day. While banks actively manage these exposures
on intraday liquidity risk will likely increase the cost of system they can nevertheless be very large: almost all banks in
participation, which may change the incentive structure for CHAPS regularly have intraday liquidity exposures in excess of
system participants. We present a series of policy tools that £1 billion to individual counterparties.(4) For larger banks these
authorities could use to mitigate any potential unintended exposures are regularly greater than £3 billion. In CREST
behavioural changes that could arise as a result of bilateral liquidity exposures can exceed £6 billion (although
forthcoming liquidity regulation. these are typically secured with highly liquid assets).(5)

The large-value payment and settlement systems, CREST (1) See King (2009) and Bank of England (2010).
and CHAPS, play a vital role in the United Kingdom’s (2) Based on 2010 UK GDP at market prices of £1.456 trillion from ONS.
(3) In the Ernest Sykes Memorial Lecture in 1989 Governor Robin Leigh-Pemberton spoke
financial system. On average, in 2010, over £730 billion of of the risk advantages of payment systems which could reduce the duration of
transactions were settled every day across the two systems. settlement counterparty exposures. RTGS eliminates these exposures entirely. This
was followed by a discussion paper that explored the risks of the payment and
This equates to the United Kingdom’s 2010 nominal GDP settlement landscape (Bank of England (1989)).
(4) RTGS has ensured that these liquidity exposures are never translated into credit
settled every two days.(2) exposures.
(5) Source: Bank of England payments database.
4 Financial Stability Paper June 2011

Furthermore, settlement banks may extend uncollateralised a situation where settlement banks willingly provide liquidity
intraday credit to their customers who access the system to the system as a whole, thereby allowing payments to be
indirectly. Such members may be heavily reliant on their settled throughout the day with minimal delays.
settlement bank to provide intraday liquidity. These credit
exposures can also be very large. The Bank estimates that The Bank supports the FSA’s approach of including intraday
some individual intraday credit lines can be in excess of 10% of liquidity risk as a key risk driver in their wider liquidity
a settlement bank’s core tier one capital. During a crisis, regulations and so making banks hold liquid assets to support
settlement banks may be reluctant to provide liquidity, given their intraday liquidity needs over and above those for balance
that this exposes them to credit risk, putting an extra burden sheet resilience. Banks are currently required to calculate their
on the indirect member. intraday liquidity needs but only to hold a proportion of the
requisite assets. The intention is that the buffers will gradually
Some observers have pointed to the general smooth increase to cover all elements of liquidity risk. This will
functioning of the payment and settlement systems during increase the cost of providing intraday liquidity which may, in
the financial crisis as a sign that intraday liquidity risks are turn, affect the willingness of banks to provide liquidity in the
well-managed. However, intraday liquidity played an UK payment and settlement systems. As such it warrants
important role when individual institutions were experiencing careful consideration.
liquidity stress. The widespread allowance of double duty
The paper is organised as follows: Section 1 explains why
(prudential asset buffers being used to support intraday
intraday liquidity risk matters, and how it is currently
liquidity needs) and authorities’ reluctance to impose
managed. Section 2 outlines various proposals for managing
additional liquidity costs on banks has meant individual
intraday liquidity risk and discusses their key features: systems
institutions were more exposed to liquidity stress. Since the
and controls, monitoring and buffer requirements. Section 3
crisis, authorities have begun to focus on intraday liquidity. In
discusses the risks posed by double duty and how intraday
September 2008 the Basel Committee on Banking Supervision
liquidity regulations can address them. Section 4 then
(BCBS) published ‘Principles for Sound Liquidity Risk
considers how intraday liquidity regulation might change
Management and Supervision’ which provides guidance to
incentives to provide liquidity in the payment system. In light
banks and supervisors on liquidity risk. Principle 8 explicitly
of this discussion, Section 5 offers some thoughts on how to
tackles banks’ management of intraday liquidity risk. And the
ensure that implementing intraday regulation can be best
Basel III liquidity framework published in December 2010 says
managed so as to minimise the risk of disruption to the
that the Basel Committee is currently reviewing how to tackle
payment system. These include the use of liquidity savings
intraday liquidity risk.(1)
mechanisms and throughput rules. Section 6 concludes.
In the United Kingdom, the Financial Services Authority (FSA)
has included intraday liquidity as a key risk driver in its new
1 Why intraday liquidity risk matters
liquidity regime. The aim of the authorities is to ensure that
1.1 Intraday liquidity and payment system risks
banks are able to meet their payment and settlement
Most large value payment systems around the world settle
obligations on a timely basis in both normal financial
payments using RTGS. This means that payments are settled
conditions and under stress. This will require banks to (a) have
individually on a gross basis requiring that banks have
adequate systems and controls to manage intraday risks,
sufficient liquidity on their settlement accounts for each
(b) report data on intraday liquidity usage and risks and
payment. Whilst banks can re-use liquidity from incoming
(c) have an adequate pool of high quality liquid assets to
payments to fund outgoing payments, timing mismatches
support intraday needs in both normal and stressed
between incoming and outgoing payments can lead to
conditions.
significant liquidity needs.(2) Moreover, these intraday
positions can be much larger than banks’ end of day net
While strengthening the management of intraday liquidity risk,
positions.
these requirements will raise the opportunity cost of providing
liquidity in the payment system. Previously, banks’ regulatory Settlement banks meet these intraday liquidity needs using
liquid asset requirements were not calibrated to include their own funds. These can be in the form of reserve account
intraday liquidity risk. However, these liquid assets were being balances, intraday borrowing secured with eligible collateral
used as collateral, intraday, in the large-value payment and from the central bank or borrowing in the interbank money
settlement systems. In essence, the liquid asset buffer was markets.(3) Indirect participants, who access systems via a
being used to support intraday liquidity needs, but regulatory
requirements were not calibrated to take into account these (1) Basel Committee on Banking Supervision (2010).
(2) See Manning et al (2009) for a more in-depth discussion of the economics of
needs (a practice referred to as double duty). For UK large-value payment systems.
settlement banks this meant that the cost of intraday liquidity (3) Settlement banks may use other sources of funding for their intraday liquidity needs
such as committed and uncommitted intraday credit lines but it is not thought that
was, in effect, negligible. Amongst other things, this has led to these are significant.
 Financial Stability Paper June 2011 5

correspondent bank, need to have cash pre-placed with their There are further costs associated with delaying payments.
settlement bank or have access to an intraday credit line, Individual settlement banks may incur financial penalties
which may or may not be secured.(1) For further explanation of associated with not settling time-critical payments on time;
how intraday liquidity needs arise and are met in an RTGS delaying customer payments may adversely affect the
payment system see the Annex. relationship between settlement bank and client; and making
payments later may increase a bank’s vulnerability to
Typically, payments sent in large-value payment systems are operational risk, in turn raising operational risk in the system
not intraday time critical: they do not have to be settled by a as a whole.
particular time during the day. However, there are some
exceptions, such as CLS pay-ins,(2) CCP margin payments(3) Delays of payments until the end of the day increase the likely
and retail system clearings, which have to be settled by a impact of an operational problem, in particular the potential
specific point in the day.(4) Anecdotal evidence suggests that for payments to remain unsettled and cause liquidity
settlement banks in the United Kingdom consider dislocation. For example, consider a bank that suffers a
approximately 4% of total payments by value and 5% by system failure so it is unable to make payments for the rest of
volume to be time critical. Settlement banks typically manage the day. The impact of this operational issue will depend upon
their payments by first releasing small and time-critical the value and volume of payments that are unsettled at the
payments, before submitting their larger payments. They time of the failure and the net liquidity position that the bank
report that a large number of payments are known in advance has with respect to the other banks in the system. Both are
(eg as a result of securities transactions), which facilitates the likely to be higher if a bank has delayed its payments (ie it is
timing of payments and aids liquidity management. holding more of other banks’ liquidity). The impact of the
Settlement banks may also instruct their large clients to either operational stress is therefore greater.(9)
pre-position funds or give advance notice of large payments
that need to be made on their behalf. Furthermore, if payments relate to the settlement of a credit
exposure, then delay can increase the amount of credit risk in
In other words, through careful planning of payment flows, a the financial system.
settlement bank can reduce the need for intraday liquidity.
This, however, is not without costs. As settlement banks aim As a result, each settlement bank faces a trade-off between
to match incoming and outgoing payments to reduce their the cost of liquidity on the one hand and the cost of delaying
intraday liquidity usage, they may find it necessary to delay payments on the other. Moreover, the cost of delay is a cost to
payments to other settlement banks. In the extreme, banks both the individual bank and to the system as a whole, and
may adopt a ‘receipt-reactive’ strategy, only making outgoing hence can be thought of as an externality. Economists have
payments using liquidity received from incoming payments.(5) investigated the incentives that banks have to delay payments.
Bech and Soramäki (2002) use simulations to show the
Liquidity provision in a payment system has much in common existence of the trade-off in a pure RTGS system. Galbiati and
with the Volunteer’s Dilemma.(6) This is a game in which Soramäki (2008) show that the demand for costly liquidity
everyone benefits from someone else contributing. But every increases as the cost of delay increases. In other words, banks
participant would prefer to free ride on the contributions from make payments earlier when it becomes more expensive to
others. Liquidity provision is similar; once one bank has made delay. Bech and Garratt (2003) use a game-theoretic
a payment, other banks can use that liquidity to fund their framework to investigate how bank behaviour depends on the
outgoing payments at no cost to themselves. In this type of intraday credit policy of the central bank. They conclude that
game the strategic solution can result in the underprovision of
liquidity.(7)
(1) A correspondent bank is a direct participant of a payment or settlement system that
provides access to the system to other banks, making payments on their behalf.
However, participation in a payment system is a repeated (2) CLS is a multi-currency cash settlement system. It requires participant banks to make
payments to it, ‘pay-ins’, five times a day. It then uses this liquidity to settle
game. It is generally believed that repeated games tend to transactions and make ‘pay-outs’ to participants. The failure to make a CLS pay-in on
result in co-operative solutions.(8) The relatively smooth flow time is seen by banks to have serious consequences.
(3) Central counterparties (CCPs) take margin to ensure members’ performance on
of payments throughout the day in CHAPS can be viewed as potential obligations to it or cover market movements on unsettled transactions. If a
member fails to make a margin payment on time then the CCP may declare that
evidence of long-term co-operative behaviour. If a bank member to be in default and close out the member’s outstanding positions.
unilaterally delays payments, it will indeed reduce its own (4) The CPSS report ‘New developments in large-value payment systems’ contains a
discussion of time criticality.
intraday liquidity needs (see Box 3). But this means that other (5) Johnson, McAndrews and Soramäki (2004) suggests receipt-reactive gross settlement
settlement banks will receive payments later, and in turn, will as a liquidity saving device.
(6) See Diekmann (1985).
either have to wait for matching funds or use their own (7) Weesie and Franzen (1998) show that cost sharing can increase the probability that a
public good is produced. However, there remains a non-zero probability that the
liquidity to fund outgoing payments. Hence, delay by one public good is not produced. In the context of a payment system, we interpret this to
settlement bank can lead to delays by all other settlement imply that strategic behaviour by banks can lead to the underprovision of liquidity.
(8) For example, see Friedman (1971).
banks and reduced turnover in the payment system as a whole. (9) See Bech (2008).
6 Financial Stability Paper June 2011

in collateralised and fee-based regimes both early settlement Chart 2 Timing of payments in CHAPS
and delay are possible equilibria. However, in a fee-based
Value (right-hand scale)
system (where the central bank charges a fee for the use of Volume (left-hand scale)
intraday liquidity) the outcome also depends upon the Number of payments £ billions
6,000 14
probability of counterparties receiving a payment request. The
result is that in some circumstances delay may be socially 5,000
12

efficient.
10
4,000

In practice, the liquidity/delay trade-off differs from country to 8

3,000
country. For example, in the Fedwire system banks pay a
6
charge for unsecured borrowing intraday from the Federal
2,000
Reserve based upon the value and time for which they 4

borrowed.(1) In contrast, in the United Kingdom, intraday 1,000 2

liquidity is provided free of charge, but is secured against
eligible collateral. The cost of this intraday borrowing is the 0 0

opportunity cost of holding such collateral. Prior to the FSA’s 05:00 07:00 09:00 11:00 13:00 15:00
Time
new liquidity regime, the liquid asset buffers of UK banks were Source: Bank of England.
not calibrated to include intraday liquidity risk. Yet, it was this
asset pool that is used to support intraday liquidity needs in
the payment and settlement systems. Hence, as assets were individual settlement banks’ exposure to liquidity risk. This is
not held specifically for intraday risk, for UK banks the intraday explained in more detail in the next sub-section.
liquidity cost of participation was effectively zero. This is
slowly changing as the FSA implements its new liquidity rules. 1.2 Intraday liquidity risk and stress scenarios
So far, we have focused on the system impact of payments
Fedwire and CHAPS exhibit very different payment timing being delayed. But the intraday element of the FSA’s liquidity
profiles. There may be a number of reasons why this is the regulation is primarily concerned with ensuring that individual
case. Armentier et al (2008) and Becher et al (2008) both firms can fulfil their obligations in the payment and settlement
suggest that the policies that govern the provision of daylight systems. As explained in Section 1.1 intraday liquidity needs
overdrafts, and therefore the cost of liquidity, are significant arise due to the timing mismatches between incoming and
factors.(2) Chart 1 shows that in Fedwire payments tend to be outgoing payments. Both settlement banks and their clients
bunched towards the end of the day, whereas Chart 2 shows are vulnerable to any stress that causes this mismatch to
that payments in CHAPS are distributed more evenly increase, resulting in an increase in the amount of intraday
throughout the day. liquidity required to continue making payments in a timely
fashion.

Chart 1 Timing of payments in Fedwire

There are four distinct categories of stress scenario that
US$ billions
160 capture the main intraday liquidity risks in a payment or
settlement system. The first three involve changes in banks’
140
payment profiles, which in turn affect their intraday liquidity
120
needs. The fourth involves a change in the value of assets that
100 a bank has to support its intraday liquidity needs. The four
80
types of stress are:

60
(i) a credit or liquidity shock affecting the bank directly,
40 reducing counterparties’ willingness to make payments to
20
it in a timely fashion;
(ii) an operational, credit or liquidity shock affecting the
0
ability of a major counterparty in the payment system to
21:00 00:00 03:00 06:00 09:00 12:00 15:00 18:00
Time make payments to the settlement bank as expected;
Source: Federal Reserve Bank of New York.

To conclude, for the payment system to operate smoothly, (1) The Federal Reserve have announced that they intend to change their intraday
liquidity policy, increasing the cost of uncollateralised daylight overdrafts and
banks need to be willing and able to make payments in a providing collateralised daylight overdrafts at zero fee. See
timely manner throughout the entire business day. Doing so www.federalreserve.gov/newsevents/press/other/20100930a.htm.
(2) There may also be institutional reasons why the payment profile in CHAPS and
reduces operational risk in the system. But it also reduces Fedwire differs. The cost of liquidity is only a partial explanation.
 Financial Stability Paper June 2011 7

(iii) a credit or liquidity shock affecting a major customer or stops making payments to the stricken counterparty, it will
group of customers of the settlement bank, preventing need to use more of its own liquidity to make its payments to
them from receiving payments as expected; and the other banks in the system.
(iv) market conditions change which mean that a given pool
of assets generates less intraday liquidity. Customer or correspondent stress
Banks that offer correspondent banking services normally
Indirect participants may experience the stresses above, but advance intraday credit to their financial institution customers,
they may also be vulnerable to different manifestations of for which they may or may not charge. As a result, customer
those stresses. If an indirect participant is suffering a stress, payments are typically made using the settlement bank’s
this could lead its correspondent bank to cut unsecured liquidity. This leaves the settlement bank vulnerable to any
intraday credit lines and request prefunding or shock that hits one or more of its large customers. This could
collateralisation. An indirect participant may be less occur:
vulnerable to a customer stress, but could be vulnerable to a
stress afflicting its correspondent bank meaning that it is • if the settlement bank sent out payments on behalf of its
unable to submit payments in a timely fashion. customers expecting that counterparties would send
payments for the customer in return;
Below, we briefly explain how these stresses may manifest • the customer bank suffered from, or was perceived to be
themselves in the payment and settlement systems. suffering from, liquidity or credit issues; and
• other banks responded by delaying payments to the
Bank stress stricken bank.
Banks rely upon incoming funds from counterparties to fund
This ultimately hits the settlement bank’s liquidity position, as
outgoing payments. This recycling of funds allows intraday
it does not receive payments that it was expecting.
liquidity needs to be substantially less than the gross value of
payments. If a bank is heavily reliant on incoming payments
The settlement bank then has two choices. It can either
then it may be vulnerable to a liquidity stress should its
choose to stop making payments on behalf of its customer,
counterparties decide to delay or stop making payments to it.
which could potentially exacerbate the stress and increase the
This could occur:
probability of default, or it can continue to make payments for
its customer, but at the cost of using more intraday liquidity.
• if the bank suffered, or was perceived to be suffering from,
a liquidity or credit shock that made other banks in the
A bank may also find that when a customer is facing a stress
payment system question its viability; and
there is an increase in the volume and value of payments that
• if other banks responded by demanding that they receive
it needs to make. Again the settlement bank is faced with the
payments before they are prepared to send any.
choice of delaying payments or using more intraday liquidity
to meet its customer’s demands.
The effect of this is for the stricken bank’s incoming payments
to be delayed and the timing mismatch between incoming and An indirect system participant may be vulnerable if its
outgoing payments to increase. This would in turn lead to an correspondent bank is experiencing problems. If the
increase in demand for intraday liquidity by the stricken bank. correspondent is suffering from liquidity problems, it may be
unable to extend the same amount of unsecured intraday
Counterparty stress credit as normal. This could create an extra liquidity burden
During the day, settlement banks build up net debit and net for the indirect participant or could result in payment delays.
credit positions with respect to their counterparties. These
individual bilateral positions can leave settlement banks Correspondent banking relationships also expose banks to
vulnerable to a stress that impacts their counterparties’ ability credit risk.(1) Intraday credit lines that settlement banks extend
to send payments. The following would constitute a to their customers are typically uncollateralised. This exposes
counterparty stress scenario: the settlement bank to the risk of its customer defaulting
intraday. Should a default occur after the settlement bank has
• if, at a given point during the day, a settlement bank has a sent payments on its customer’s behalf, the settlement bank
large net debit position with respect to another settlement could find itself liable for the full value of the intraday credit
bank, ie it has sent more payments to that bank than it has line that it has extended. Conversely, some customer banks
received; and will systematically have credit balances on account with their
• if that counterparty were unable to send payments. correspondent. These credit balances are typically unsecured

The settlement bank would be short of the liquidity that it was

(1) Harrison et al (2005) provide a more detailed explanation of the risk implications of
expecting to receive. As such, even if the bank immediately correspondent banking.
8 Financial Stability Paper June 2011

and can be seen as a form of intraday lending from the A common feature of settlement banks’ intraday liquidity
customer to the correspondent bank. This exposes the management tools is the use of bilateral limits. These limits
customer bank to the potential default of its correspondent. are used to manage the risk of a counterparty being unable or
unwilling to make payments. To do so, banks place a cap on
Customer stress in a securities settlement system the amount of liquidity that they are prepared to provide to
In a securities settlement system, such as CREST, there is also any individual counterparty, ie they limit the largest bilateral
potential for customer stresses. When a settlement bank net debit position. These limits vary across counterparties and
settles a securities transaction it can do so using its own throughout the day. They are typically operational limits, and
liquidity or liquidity generated by one of its customers.(1) If a are managed dynamically during the day.
bank systematically relies upon the liquidity provided by a
customer in order to fund its intraday liquidity needs then it The CHAPS system rules provide further risk mitigation in the
could be highly vulnerable to shocks which affect that form of throughput guidelines. The rules state that a bank
customer. For example, the customer bank could fail and so must send, on average over the month, at least 50% of
no longer participate in the system. Alternatively, the payments by value by noon and 75% by 14:30. This helps to
customer bank may suffer operational issues that mean it is limit the liquidity exposures that banks have with each other
unable to generate liquidity either by selling securities or via and acts as a co-ordinating device which reduces the liquidity
self-collateralising repo. Both of these shocks would require needs of the system as a whole. It also acts to limit the scope
that the settlement bank raise extra liquidity in order to fulfil for any bank to free-ride on the liquidity provision of other
its normal payment and settlement obligations. banks.

Market stress Settlement banks that make payments on behalf of customers

A bank may set aside an amount of assets that it considers to need to manage the associated liquidity and credit risk. They
be sufficient to cover its intraday liquidity needs in both can do this by limiting the amount of intraday credit that they
normal and stressed conditions. However, during a extend to their customers. Once a customer has used all of its
market-wide stress the credit and liquidity quality of a credit, the settlement bank will stop making payments on its
particular asset could fall. This could lead to an increase in the behalf until it receives incoming payments. These limits are
haircut applied or to the asset becoming ineligible for central typically unsecured and unadvised (the client is unaware of its
bank liquidity facilities. For indirect participants, variation in intraday limit). This gives settlement banks the freedom to cut
an asset’s value could lead its correspondent bank to apply the credit that they are prepared to extend if they believe
higher haircuts. In either case, the bank would have less there is an increased chance of default. Some settlement
intraday liquidity than it expected. banks may also ask their customers to collateralise their
intraday credit lines, protecting them against the risk of an
Contagion intraday default.
Each of these stresses may cause the settlement bank to
require more intraday liquidity in order to make its payments Beyond these risk controls there are other, less formal,
in a timely fashion. If the bank did not have access to this mechanisms which allow banks to manage intraday liquidity
liquidity, it would have no choice but to delay payments. In risk effectively. Firstly, in the United Kingdom the small
addition to exacerbating perceptions of weakness, delaying community of settlement banks helps to create a co-operative
payments would subsequently cause the bank’s counterparties system in which banks can co-ordinate payment flows. This is
to require more liquidity to make their payments. If other especially useful in a crisis when liquidity managers and
banks also had insufficient liquidity to make payments, then payment system operators can manage liquidity risk by direct
they too would have no choice but to delay. This could result communication with each other. Secondly, many of the
in substantial delays to payment flows and even potential system participants have built up years of experience
gridlock. A simple illustrative example is given in Box 1. managing payment flows. This enables them to identify
unusual behaviour quickly and respond accordingly, thus
1.3 How are these risks managed at present? limiting the impact of intraday liquidity stress should it occur.
Settlement banks have developed a range of different tools to
manage these intraday liquidity risks. Many settlement banks The design of payment and settlement systems can help to
use internal schedulers with functionality that enables them to reduce intraday liquidity needs and aid intraday liquidity
manage their payment flows smoothly throughout the day. management. For example, a number of RTGS payment
These schedulers allow banks to limit the impact of intraday
liquidity stress and respond quickly should a stress occur. For (1) In CREST, liquidity can be generated in a number of ways. The primary mechanism is
the self-collateralising repo (SCR). When a security that is eligible for the Bank of
example, schedulers may allow banks to stop making England’s intraday liquidity facilities is purchased it is automatically repoed to the
payments to a particular counterparty, or take manual control Bank in return for intraday liquidity to the value of the security minus a haircut. The
SCR mechanism allows the liquidity generated by the SCR to be simultaneously used
of large value payments, releasing them individually. to fund the purchase of the security.
 Financial Stability Paper June 2011 9

Box 1 Bank A’s intraday liquidity needs are equal to the largest net
debit position that it incurred during the day. This is the
Stress scenarios
minimum amount of central bank reserves needed in order to
make all its payments without delay. In this example, Bank A’s
Intraday liquidity stresses
largest net debit position, and thus liquidity need, was
This box contains a stylised example of the different types of
£1.53 billion. This is indicated by the dashed red line.
intraday liquidity stresses that a bank could be exposed to and
which may cause an increase in its intraday liquidity needs. It
Bank stress
uses a single day of real payment data taken from the RTGS
In our example, Bank A is the stricken bank. Banks B and C
system for payments between three banks, labelled A, B and C.
both delay all their payments to Bank A by 30 minutes.
Each bank has obligations to make payments to each of the
Bank A makes payments at the same rate in order to convince
other banks. Payments are made using a combination of
the other banks that it is capable of continued participation in
incoming payments and intraday liquidity in the form of
the system.
central bank reserves.

Chart B shows how Bank A’s payment activity and liquidity

The data have been slightly adjusted so that each bank sends
needs evolve across the day. The delay in payments to Bank A
the same value as it receives. This box is meant to be
is shown by the rightward shift of the pink, payments received
illustrative of how these risks can materialise and not to give a
line. These delays lead to an increase in the largest net debit
quantitative assessment of how large they are.
position, and thus liquidity needs. As a result of this stress,
Bank A’s intraday liquidity needs increase by 40%, from
Chart A shows how Bank A’s payment profile and intraday
£1.53 billion to £2.15 billion.
liquidity needs evolve during the payment system day. The
blue line shows the cumulative amount that Bank A has sent
to Banks B and C since the start of the day. The pink line Chart B Bank A’s payments sent, received and net
shows the cumulative value of payments that Bank A has position after it suffers a bank stress
received from Banks B and C since the start of the day. The Bank A sent Bank A net
Bank A received Largest net debit
orange line shows Bank A’s net position. £ billions
10

8
Chart A Bank A’s payments sent, received and net
position 6

Bank A sent Bank A net

Bank A received Largest net debit 4
£ billions
10
2
8 +
0
6 –
2
4
4
06:00 08:00 10:00 12:00 14:00 16:00
2 Time
+ Sources: Bank of England and authors’ calculations.
0
– Counterparty stress
2
In our example, Bank C suffers a stress at around 9.30 am. This
4 prevents Bank C from sending any payments to Bank A until
06:00 08:00 10:00 12:00 14:00 16:00
Time
the end of the day. Bank A responds immediately by stopping
Sources: Bank of England and authors’ calculations.
making payments to Bank C, again until the end of the day.
Bank A continues to send and receive its payments to and from
Bank B as normal.
When the blue line is above the pink line it indicates that
Bank A has sent more payments than it has received and so
Chart C shows how Bank A’s liquidity needs evolve during the
has used its own liquidity to fund its payments. Conversely,
day. After the stress Bank A only sends and receives payments
when the blue line is below the pink line it indicates that
to Bank B. Bank A’s largest net debit position, its intraday
Bank A has received more payments than it has sent.
liquidity need, increases as a result of the liquidity stress.
Bank A was relying on incoming payments from Bank C to fund
10 Financial Stability Paper June 2011

some of its payments to Bank B. As these funds did not make payments at the normal rate. Bank A grants sufficient
materialise, Bank A now requires more funds. This is indicated liquidity to allow it to do so. This stress begins at 9.30 am.
by the dashed red line which has increased 45% from
£1.53 billion to £2.23 billion. Chart D shows how Bank A’s payment activity and liquidity
needs evolve across the day. Compared to the base case, a
Chart C Bank A’s payments sent, received and net
proportion of the payments have been delayed for 90 minutes.
position after Bank C suffers a stress This shifts the red payments received line to the right, resulting
Bank A sent Bank A net
in an increase in the timing mismatch between incoming and
Bank A received Largest net debit outgoing payments. Bank A’s intraday liquidity needs increase
£ billions
10 by 6% from £1.53 billion to £1.62 billion.
8

6 Chart D Bank A’s payments sent, received and net

position after one of its customers suffers a stress
4
Bank A sent Bank A net
Bank A received Largest net debit
2 £ billions
10
+
0
8
–
2
6

4
06:00 08:00 10:00 12:00 14:00 16:00 4
Time
Sources: Bank of England and authors’ calculations. 2
+
Customer stress 0
In this example one of Bank A’s customers suffers a stress. This –
2
customer is responsible for 10% of the payments that Bank A
sends and receives. Not having a direct relationship with 4
06:00 08:00 10:00 12:00 14:00 16:00
Bank A’s customers the other settlement banks delay all Time
payments for 90 minutes whilst the customer continues to Sources: Bank of England and authors’ calculations.
 Financial Stability Paper June 2011 11

systems have introduced liquidity saving mechanisms such as 2.2 Systems and controls and monitoring requirements
an offsetting algorithm (see Section 5.1) to improve the The ‘Principles for Sound Liquidity Risk Management and
liquidity efficiency of the system. Supervision’ state that banks should:

Together, these tools helped the payment and settlement i) have the capacity to measure expected inflows and
systems function smoothly during the worst days of the recent outflows, anticipate the intraday timing and forecast the
financial crisis. These tools are, however, not perfect risk range of potential net funding shortfalls that might arise
mitigants and could introduce another set of risks, for example during the day;
reliance on individuals’ experience to manage payment flows. ii) have the capacity to monitor intraday liquidity positions
They also do not tackle the risks posed to the customer banks. against expected activities and available resources;
Prior to the recent financial crisis, regulators did not focus on iii) arrange to acquire sufficient intraday funding to meet their
intraday liquidity risk and there were no standard monitoring intraday objectives;
or liquidity management measures in place. In the next iv) have the ability to manage and mobilise collateral as
sections we explain how new intraday liquidity regulations in necessary to obtain intraday funds;
the United Kingdom and abroad are expected to meet some of v) have a robust capability to manage the timing of their
these shortcomings. liquidity outflows; and
vi) be prepared to deal with unexpected disruptions to their
2 Intraday liquidity regulation intraday flows.

2.1 Intraday liquidity regulation The supervisor would be expected to monitor that banks
In the ‘Principles for Sound Liquidity Risk Management and adhere to these principles. And there are plans to introduce
Supervision’, the Basel Committee on Banking Supervision has internationally agreed monitoring metrics for intraday risk in
identified the issue of the management of intraday liquidity risk due course.
and collateral. In particular, principle 8 states that: ‘A bank
should actively manage its intraday liquidity positions and risks The benefits of introducing monitoring requirements are
to meet payment and settlement obligations on a timely basis threefold. They encourage individual banks to take a more
under both normal and stressed conditions and thus contribute systematic and disciplined approach to intraday liquidity
to the smooth functioning of the payment and settlement management; they allow greater transparency to the
systems.’ Beyond this there are references to intraday liquidity supervisor and payment system overseer; and they permit the
management in a number of other principles.(1) supervisor to carry out cross-sectional analyses of banks’
resilience to intraday liquidity shocks as well as monitoring
The Principles advocate an approach that places the emphasis how a bank’s resilience evolves over time.
on individual banks to have adequate systems in place to
measure and manage their intraday liquidity risk. The role of 2.3 Buffer requirements
the supervisor is to ensure that banks adhere to these An intraday liquidity buffer aims to ensure that banks always
Principles. The Principles highlight the importance of intraday have sufficient intraday liquidity to fund their intraday
liquidity management both due to its impact on the bank itself positions, and that intraday liquidity needs do not reduce
and the impact on the bank’s counterparties. banks’ resilience to other liquidity shocks.

In the United Kingdom, the FSA’s regulatory approach is The FSA’s new liquidity regime will require banks to hold a
consistent with that advocated in the Principles. The buffer of highly liquid assets calibrated to ensure that they
responsibility for ensuring that systems and controls are have sufficient intraday liquidity to make payments in a timely
adequate to measure and manage intraday liquidity risks rests manner in both normal and stressed conditions. Intraday
with the individual bank and the FSA’s role is to ensure that the liquidity needs will be calculated in addition to the liquid assets
systems and controls are robust. In addition, however, the FSA needed by a bank for balance sheet resilience purposes.(2)
requires banks to calculate how much intraday liquidity they Intraday liquidity risk is considered one of the key risk drivers
would need in both normal and stressed conditions. In time, that determine a bank’s liquid asset buffer requirement.(3)
banks will be required to hold sufficient liquid assets to cover
these needs. This will be in addition to the assets that banks (1) For example, Principle 9 on collateral says that ‘systems should be capable of
monitoring shifts between intraday and overnight or term collateral usage’ and
will be required to hold to meet their balance sheet liquidity Principle 10 on stress testing says that ‘tests should consider the implication of the
scenarios across different time horizons, including on an intraday basis’.
buffer (hereafter ‘prudential asset buffer’) requirements. This (2) The Basel Committee on Banking Supervision has agreed international standards for
section describes the two main approaches available to calculating the liquid asset buffers that banks will be required to hold for balance
sheet resilience purposes. The standards relate to two ratios: a Liquidity Coverage
supervisors to ensure that banks measure and manage their Ratio and a Net Stable Funding Ratio. See Basel Committee of Banking Supervision
(2010).
intraday liquidity risk — monitoring tools and buffer (3) See FSA handbook BIPRU 12.5.14 R,
requirements — and discusses their costs and benefits. http://fsahandbook.info/FSA/html/handbook/BIPRU/12/5.
12 Financial Stability Paper June 2011

The principle for a settlement bank’s intraday liquidity buffer However, the practice of double duty is not without risks.
calculation is explained in the FSA handbook.(1) Banks are Conceptually, there is one significant risk associated with using
required to calculate how much intraday liquidity they need in the same assets for two separate purposes: when the assets
normal times to fund their participation in payment and are being used for purpose A they are not available for purpose
settlement systems and then estimate how this could change B and vice versa. In practice this manifests itself in two ways:
under liquidity stress. The stresses referred to are an
unforeseen, name-specific liquidity stress lasting for two weeks (a) Balance sheet resilience risk: if a bank’s prudential asset
and an unforeseen, market-wide liquidity stress lasting three buffer is serving the purpose of providing intraday
months. liquidity, then it cannot be as effective as a buffer against
a run on deposits.
The handbook also says that banks should take into (b) Intraday liquidity risk: if a bank suffers a prolonged balance
consideration both their own liquidity needs and those of sheet liquidity stress, this uses up the bank’s prudential
customers. And should consider the impact of other asset buffer meaning the bank has insufficient funds
participants withholding some or all of their payments and available to operate effectively in the payment systems.
customers increasing the value and volume of payments.
Balance sheet resilience risk
Banks are required to estimate the size of their own buffers as If a bank suffers from a balance sheet stress it will need to
part of the Individual Liquidity Adequacy Assessment (ILAA) liquidate its prudential asset buffer in order to pay its
process. A bank’s ILAA is then reviewed by the FSA who will wholesale and retail depositors. This means the bank will need
come to a conclusion about the appropriate level of liquidity to sell or encumber some assets to generate cash, which it
protection. then pays out through the payment systems. If the assets that
the bank had earmarked to support this stress have been used
3 Intraday liquidity regulation and double intraday, then the bank may not have immediate access to
duty them and so be unable to make timely payments to its
depositors. Thus, the assets are not always available for the
This section explains the practice of double duty and why it ‘run’ they were designed to protect against. If the bank delays
may bring risks to participants in payment and settlement payments to its depositors then other market participants may
systems (Section 3.1). It then considers what policy actions question the bank’s ability to pay. This could further
supervisors can take to limit these risks and examine how exacerbate the balance sheet stress.
these actions need to be accompanied by measures to
mitigate any possible adverse effects on payment flows This is the scenario that Lehman Brothers faced in the run-up
(Section 3.2). Box 2 uses the Lehman Brothers bankruptcy as to its bankruptcy. During this period, Lehman’s correspondent
an example of the role double duty can have in a liquidity banks in various systems demanded that it collateralise or
crisis. prefund its intraday liquidity usage. These assets were taken
from the bank’s liquid asset pool and so were no longer
3.1 Double duty available to pay out its depositors. Despite appearing to be
Central banks and supervisors encouraged the introduction of liquid, by 15 September 2008 Lehman had insufficient
RTGS into payment systems to minimise settlement risk for unencumbered liquid assets to support its projected cash
high-value payments.(2) But they recognised that there was a outflow. Box 2 outlines in more detail the role intraday
significant liquidity cost to using it. In RTGS systems, timing liquidity risk played in Lehman’s collapse.
mismatches between outgoing and incoming payments lead
banks to use liquidity intraday. This intraday liquidity need can In normal times, or at the beginning of a period of liquidity
be many times the bank’s end of day liquidity need. To reduce stress, the probability of this risk crystallising is relatively low.
this cost, many authorities decided to allow banks to use their Banks’ prudential asset buffers are typically many times larger
prudential asset buffers, held for wider liquidity resilience, to than their intraday liquidity needs. However, as a balance
support payments activity intraday without making banks sheet stress progresses, the prudential asset buffer will shrink,
include intraday liquidity risk in the calibration of the buffer: a increasing the probability of this risk materialising. As such, as
practice known as ‘double duty’. in the case of Lehman Brothers, a bank’s intraday liquidity
needs may increase and take up an ever increasing proportion
Double duty reduces the cost of making payments in RTGS. In of the bank’s liquid assets.
fact, if a bank’s intraday liquidity usage is less than the amount
of liquid assets it is required to hold by its supervisor, then the
opportunity cost is zero. This reduces the incentives for banks
to delay payments and thus facilitates the smooth functioning (1) www.fsa.gov.uk/pages/handbook.
of payment and settlement systems. (2) See Committee on Payment and Settlement Systems (1997).
 Financial Stability Paper June 2011 13

Box 2 Lehman’s correspondent banks request

Intraday liquidity risk: the case of Lehman collateralisation and prefunding
In the run-up to Lehman’s bankruptcy, its correspondent banks
Brothers became less willing to advance intraday credit to allow
Lehman to participate in payment and settlement systems.
On 15 September 2008, Lehman Brothers Holdings Inc. filed Throughout 2008 Lehman’s correspondents started to apply a
for Chapter 11 bankruptcy.(1) Evidence presented in the number of different conditions to reduce their exposure to
Valukas report suggests that a lack of liquidity turned out to be Lehman. At first, haircuts were increased on tri-party repo
one of the key elements that precipitated the firm’s collapse: collateral to protect the correspondent bank. This was
on 14 September 2008, Lehman no longer had sufficient followed by banks asking for explicit collateralisation of
available liquidity to fund its daily operations. A significant positions. When market conditions deteriorated, Lehman’s
factor in Lehman’s shortage of liquidity was the provision of collateral was revalued and it was required to post extra
collateral to its clearing banks to fund its intraday liquidity collateral. Some banks preferred to ask Lehman for cash
positions. Lehman was using an increasing portion of the asset deposits to prefund their intraday liquidity needs.
pool to cover intraday liquidity risk. Ultimately this meant
Lehman’s pool of liquid assets was unable to meet the In total, by 12 September 2008, Lehman’s correspondent
outflows it was calibrated for. This is precisely the risk posed banks had received over US$16 billion in collateralisation and
by double duty. prefunding.

Lehman’s correspondent banking relationships Lehman runs out of liquidity

Across the world Lehman used a number of different In Lehman’s last days its liquid asset pool contained more and
correspondent banks to provide it with payment and more assets that had been pledged intraday to its
settlement services. In the United States, the most significant correspondent banks. According to Lehman’s own
was JP Morgan who acted as their agent in tri-party repo post-mortem analysis, its liquid asset pool decreased from
arrangements.(2) Citibank provided Lehman with access to US$41 billion prior to its earnings announcement on
CLS, the multicurrency foreign exchange settlement system. 9 September 2008 to about US$25 billion on 12 September.
HSBC provided Lehman with settlement services in the The stressed market conditions of the time meant that
United Kingdom, acting as their settlement bank in CREST. approximately US$7 billion of these assets were not
Lehman also had correspondent banking relationships with immediately liquid. Of the remaining US$18 billion of assets,
Bank of America, Bank of New York Mellon and Standard Bank. US$16 billion were required by Lehman’s correspondent banks.
In each case, Lehman’s correspondent provided it with intraday This left only US$2 billion of truly liquid assets available to
credit to allow it to settle trades and make payments. meet outflows. On 15 September, Lehman was forecast to
have a net cash outflow of US$4.5 billion. With insufficient
Lehman’s liquidity pool liquid assets to cover this shortfall Lehman filed for Chapter 11
Lehman reported that it had a liquidity pool ‘primarily bankruptcy protection.(5)
intended to cover expected cash outflows for twelve months in
a stressed liquidity environment.’ The pool was designed to The risks posed by double duty finally came to bear. Lehman
cover a range of cash outflows, while assuming that the firm had not calibrated its liquid asset pool to include intraday
could not issue any unsecured debt or generate liquidity liquidity risk, but in a crisis intraday liquidity turned out to be a
outside that pool of assets (eg by selling less liquid assets). significant drain on its asset pool. Lehman found that it was
Lehman reported that at the end of the first quarter of 2008, significantly less liquid than both it and the market believed it
the liquidity pool consisted of US$34 billion. At the end of the to be. It was this lack of liquidity that was the final straw that
second quarter it was US$45 billion and at the end of the third broke Lehman Brothers.
quarter US$42 billion.(3) The size of the asset pool gave market
participants comfort that Lehman had low vulnerability to
liquidity risk.(4) But this asset pool increasingly contained
assets that Lehman had pledged or transferred to its
correspondent banks intraday. Although these assets were (1) This box draws heavily on Volume 4 of the Chapter 11 proceedings report prepared by
viewed as unencumbered overnight, a failure to pledge these the Examiner, Anton R. Valukas. The full report is available at:
http://lehmanreport.jenner.com.
assets at the start of the next day would have likely had (2) A triparty repo is a repo in which a third party (eg a custodian bank, a clearing house
or a CSD) is responsible for the management of collateral during the life of the
negative implications for the willingness of these banks to clear transaction.
for Lehman. In other words, the assets were not available to (3) Valukas, A (2010), page 1,409.
(4) Valukas, A (2010), page 1,415.
meet other outflows. (5) Valukas, A (2010), page 1,454.
14 Financial Stability Paper June 2011

Intraday liquidity risk the bank’s supervisor. However, it is not always

Prudential asset buffers are typically calibrated to ensure that straightforward to understand ex ante if a stress is an intraday
banks hold sufficient liquid assets to withstand a prolonged or an overnight stress.
liquidity stress. For example, the BCBS’ proposed Liquidity
Coverage Ratio aims to ensure that a bank has ‘an adequate A further option, which is the essence of the FSA’s intraday
level of unencumbered, high quality liquid assets that can be liquidity regulation, is for banks to calibrate their liquidity
converted into cash to meet its liquidity needs for a 30-day buffers to include intraday liquidity needs as a separate risk
calendar time horizon under a significantly severe liquidity driver. This does not prohibit banks from using the rest of their
stress scenario specified by supervisors’.(1) prudential asset buffer intraday, as the portion assessed for
each risk driver is fungible and available to meet liquidity
As the 30-day balance sheet stress progresses, the bank’s requirements of any origin. However, a properly calibrated
resilience to an intraday liquidity stress will fall with the size of buffer would make it highly unlikely that banks would fail to
the prudential asset buffer. Eventually the bank may have meet a regulatory stress scenario that is based on liquidity
insufficient liquidity to be able to conduct its normal outflows that are overnight or longer on account of needing to
payments and settlement business in a timely manner and so continue participation in the payment and settlement
be forced to delay payments to its counterparties. At the systems. Banks could still hold as much of their liquidity
extreme the bank will have to behave ‘receipt reactively’ all buffer in the payment systems as they required. If a bank
day, relying entirely on incoming payments to fund its habitually used its prudential asset buffer it would indicate
outgoing ones. In these circumstances, the bank’s that the buffer was insufficient and should be recalibrated. In
counterparties will be unlikely to be willing to send payments effect, the FSA has decided to limit the risks of double duty
to it. The stricken bank may then be unable to continue to described above without completely abolishing it.
participate in the payment and settlement systems.
The introduction of an intraday liquid asset buffer to deal with
An alternative way to express this is to say that a bank needs the risks of double duty increases the cost of intraday liquidity.
to continue to operate in the payment and settlement systems The implications of this are discussed in Section 4.
to avoid default. If this requires the bank to use a fixed
amount of collateral each day to generate intraday liquidity, it 4 Incentives created by intraday liquidity
can only use the remainder of its collateral pool to meet regulation
outflows. Therefore, in practice, if that pool is sized on the
basis of potential outflows from other sources and does not There are many benefits to introducing more rigorous controls
include an intraday component, then the bank will have less around banks’ monitoring and management of intraday
resilience to liquidity stresses than the supervisory authorities liquidity risk. However, doing so may change the incentives for
intended. banks that participate in the payment and settlement systems.
This could have a negative impact on system stability. When
This argues in favour of authorities taking action to tackle the implementing intraday liquidity regulations, authorities need
risks posed by double duty. to pay particular attention to the incentives created and
ensure that they have adequate tools available for dealing with
3.2 Intraday liquidity regulation and double duty any unintended systemic consequences.
There are a number of ways that authorities can tackle the
risks posed by double duty. The strictest way would be an Systems and monitoring
outright ban on prudential assets being used intraday in the A focus on banks’ internal intraday liquidity risk management
payment and settlement systems under any circumstances. systems is unlikely to create adverse incentives in themselves.
However, this may not be desirable or, indeed, feasible. When If intraday liquidity monitoring extends to the use of defined
a liquidity stress occurs, the payouts need to be made in the metrics about liquidity usage, the risk of such adverse
payment and settlement systems. It seems reasonable that a behavioural change is more likely.
bank should use its prudential asset buffer to support this.
After all this is why the bank has such a buffer. Furthermore, In particular:
were a severe intraday liquidity stress to occur, it seems • if banks know exactly which dimensions the supervisor will
inconceivable that a supervisor would prefer that the bank use to assess them, then this may create incentives to alter
protected its prudential asset buffer by failing settlement of a payment flows to manipulate the metrics;
time critical payment rather than used those assets to fill a • if a bank is unsure how it compares to its peers it may
temporary liquidity shortfall. attempt to manage its liquidity more tightly, even when
unnecessary, for fear of attracting regulatory attention.
One alternative approach is to allow the use of prudential
assets only in stressed circumstances with the agreement of (1) Basel Committee on Banking Supervision (2010), page 3.
 Financial Stability Paper June 2011 15

This could result in delayed payments and increased Furthermore, it is possible that the implementation of new
operational risk. regulation will cause disruption in the payment systems as
banks delay payments in an attempt to reduce liquidity usage.
Buffers Authorities have a role in both influencing the new equilibrium
Introducing intraday liquidity buffers should make banks more and ensuring that the transition happens smoothly. Section 5
resilient to any potential liquidity stress. However, their discusses tools that could help guide this process.
introduction also makes intraday liquidity more costly. This is
because of the opportunity cost of banks changing their 5 Ensuring a ‘good’ equilibrium
portfolio allocations to hold more high-quality, low-return
assets in the place of lower-quality, higher-return assets. A number of alternative policy and system-design measures
Banks in a payment system may have arrived at an could, in principle, be taken to mitigate the risk of a ‘bad’
equilibrium that depends upon the collective trade-off equilibrium emerging. These fall into two broad categories:
between the cost of liquidity and the cost of delay. When that
trade-off is changed to make liquidity more costly, banks may i) measures that lower the system-wide demand for liquidity
not be willing to provide liquidity at the same level. In by improving coordination, and hence liquidity-recycling,
particular, banks that believe they provide more than their fair in the payment system; and
share of liquidity in the near costless regime, may want to cut ii) measures that lower the opportunity cost of generating
back their liquidity provision. intraday liquidity.

The increase in liquidity cost would be borne by most banks Some of these options may build on existing design features of
participating in a payment system, although not necessarily large value payment and settlement systems; others may
evenly. If the buffer is calibrated to take account of both a require that entirely new processes or arrangements be
bank’s normal intraday liquidity usage and its exposure to introduced.
intraday liquidity risk then those banks that habitually use
more liquidity or are more vulnerable to intraday liquidity The principal alternative options falling into these categories
stresses would be expected to hold a larger buffer. are listed in Table A and elaborated in the remainder of this
section. Either set of measures may also be complemented by
It is reasonable to expect that banks will use historical data on ‘deep dive’ examination of banks’ intraday payments
payment activity to calibrate their buffers. Historical liquidity behaviour to inform the appropriate calibration of banks’
usage can give an indication of the amount of liquidity intraday buffers and to ensure banks are not attempting to
required to fulfil normal obligations. And ‘what-if?’ analyses of free ride on the liquidity provision of others to reduce their
various stress scenarios can illustrate the potential impact of own liquidity needs.
liquidity stresses on a bank’s intraday liquidity needs.
Table A Policy options to mitigate the risk of payment delays
However, basing the buffer on historical data creates
Measures that improve liquidity-recycling Measures that lower the opportunity cost
incentives for banks to reduce their liquidity usage in order to in the payment system of generating intraday liquidity
benefit from a lower buffer come recalibration. A standard
Implementation of liquidity-saving Setting the central-bank eligible criteria
result in the payments economics literature is that costly mechanisms for collateral sufficiently wide: including,
foreign-currency collateral, where
intraday liquidity introduces incentives for banks to delay Implementation and enforcement appropriate
payments, waiting for incoming payments before sending of binding throughput guidelines

outgoing ones.(1) It should be apparent, however, that if all Creating incentives for early submission
of payments through variable tariffs
banks try to delay payments then the result is that no-one
benefits from reduced liquidity usage and there is an increase
in systemic operational risk: an outcome that adversely affects 5.1 Implementation of liquidity-saving mechanisms
everyone. Box 3 discusses the potential impact of banks Altering the design of the RTGS system to directly improve
delaying payments. liquidity efficiency is a first way to mitigate the risk of
settlement delays. In particular, the introduction of so-called
As explained in Section 1.1, there is evidence of co-operative liquidity-saving mechanisms (LSMs) can allow a bank to
behaviour in the United Kingdom payment and settlement significantly reduce the amount of liquidity required to meet a
systems. Indeed, this is the expected theoretical outcome in a given quantum of payments while minimising the delay in
repeated game. Therefore, it is likely that banks will find a settlement.(2) In fact, LSMs can incentivise earlier submission
co-operative equilibrium in a world with costly liquidity. But, of payment instructions compared to plain-RTGS. In recent
it is not clear what that equilibrium will be and whether it will
be optimal from a system stability perspective. (1) See Manning et al (2009) and Bech and Garrett (2003).
(2) See Norman (2010) for more explanation of liquidity-saving mechanisms.
16 Financial Stability Paper June 2011

Box 3 Table 1 Descriptive statistics when Bank A delays

Impact of delays on liquidity in the payment Base case Bank A delays
system System throughput 52.7% 51.1%
Average time of settlement 11:56 am 12:03 am
This box aims to illustrate the possible impact of banks Total intraday liquidity usage £3.82 billion £4.14 billion
delaying payments in response to an intraday liquidity buffer
requirement. As in Box 1, the charts reflect payments between All banks delay all payments by 30 minutes
three banks, labelled A, B and C. The exercise is meant to be In a second scenario, Banks B and C respond to Bank A’s
illustrative and not to reflect the quantitative potential impact payment delays by also delaying all their payments by
of banks delaying payments. 30 minutes. The result is that there is no change in the relative
timing of payments, but all payments are made later in the
Bank A delays all of its payments by 30 minutes day. This is shown in Chart B.
Bank A sends and receives payments to and from Banks B
and C. Bank A unilaterally decides to delay payments so as to
reduce its intraday liquidity requirements. This results in an Chart B Bank A’s payments sent, received and net
increase in operational risk for Bank A as payments are more position if all banks delay all payments by 30 minutes
concentrated towards the end of the day. Bank A sent Bank A net
Bank A received Largest net debit
£ billions
10
Chart A shows that Bank A is able to reduce its intraday
liquidity needs by unilaterally delaying payments to Banks B 8

and C. The blue, payments sent line shifts to the right and
6
consequently Bank A’s largest net debit position falls. Bank A’s
intraday liquidity needs reduce by 16% from £1.39 billion to 4

£1.17 billion.
2
+
0
Chart A Bank A’s payments sent, received and net
–
position if it delays all payments by 30 minutes
2
Bank A sent Bank A net
Bank A received Largest net debit 4
£ billions
10 06:00 08:00 10:00 12:00 14:00 16:00
Time

Sources: Bank of England and authors’ calculations.

8

6 Table 2 shows that this is a situation in which no one benefits,

but everyone suffers. The total liquidity usage across the
4 system is unchanged, as is the distribution of that liquidity
usage. However, by delaying all payments by 30 minutes the
2
banks have increased operational risk in the system: noon
+
throughput has fallen by 17 percentage points and the average
0
time of settlement is now 27 minutes later.(1)
–
2
06:00 08:00 10:00 12:00 14:00 16:00
Time Table 2 Descriptive statistics when all banks delays
Sources: Bank of England and authors’ calculations.
Base case All banks delay
However, as illustrated in Table 1, there is a systemic cost to System throughput 52.7% 35.7%
these delays. Firstly other banks intraday liquidity needs Average time of settlement 11:56 am 12:23 pm
increase as a result: Bank B’s increase by 3%, whereas Bank C’s Total intraday liquidity usage £3.82 billion £3.82 billion
increase by 41%. The aggregate effect is to increase the
amount of liquidity needed to settle all payments from
£3.82 billion to £4.14 billion. Secondly, as shown in Table 1,
noon throughput falls from 52.7% to 51.1% and the average
time at which payments were settled increases by 7 minutes.
These two measures indicate that the system is more
vulnerable to operational risk. (1) The delay in the average time of settlement is not quite 30 minutes as payments that
were delayed until after 4 pm are assumed to have been settled at 4 pm.
 Financial Stability Paper June 2011 17

years, such design features have been introduced in a number and McLafferty and Denbee (2011)), though these do not
of countries’ large-value payment systems (CPSS (2005) and accrue evenly to all participants. McLafferty and Denbee also
Bech et al (2008)).(1) suggest that it is dividing payments into a priority and a
non-priority channel that produces the liquidity savings. The
In most cases, a system design that incorporates sophistication of the offsetting algorithm serves to reduce
liquidity-saving features comprises the following key elements: settlement delay for a given liquidity saving.

• a central queue; The Bank of England has acknowledged these benefits and has
• an offsetting algorithm; and commenced development of an LSM for CHAPS payments.
• liquidity-reservation functionality. The intention is for it to be implemented in 2013.

A central queue allows payment orders to wait for settlement 5.2 Throughput rules
where they are ‘visible’ to an offsetting algorithm. An Throughput rules establish the minimum proportion of a
offsetting algorithm finds pairs or groups of payments that can participant’s daily settlement flow (measured in either volume
be settled simultaneously, with reduced liquidity usage. or value terms) that must be settled by a particular time.
Liquidity-reservation functionality allows banks to set aside Several countries’ large-value payment systems have such
liquidity for time-critical payments to ensure that they can be rules in place, including those in Canada, Hong Kong and the
settled without delay. It also acts to constrain the liquidity United Kingdom. As mentioned earlier, in the United Kingdom,
available to non-priority payments allowing them to queue members of the CHAPS Clearing Company have mutually
and take advantage of the liquidity benefits of the offsetting agreed to submit, on average, 50% of daily settlement flow by
algorithm. noon, and 75% by 2.30 pm, each calendar month. As described
in Becher et al (2008), enforcement of the throughput
In a system in which all three elements are present guidelines in CHAPS currently relies on peer pressure rather
participants channel payments according to their priority: than financial or regulatory sanctions. Persistent breaches
‘urgent’ payments are submitted for immediate settlement, result in a participant being called to give evidence to a
while non-priority payments are submitted to the central committee of its peers, the so-called ‘Star Chamber’.
queue awaiting ‘conditional release.’
Buckle and Campbell (2003) demonstrate that throughput
In TARGET2 and similar systems, payments are released for rules can promote efficient liquidity recycling.(4) The authors
settlement subject to an algorithm that works through the show that, as long as the penalty for non-compliance with the
queue and identifies pairs or groups of offsetting payments throughput rule exceeds the private benefit of delaying
that can be settled simultaneously, subject to the balance on payments, banks will comply and the system-wide liquidity
the participant’s settlement account not falling below a requirement for a given quantum of payments will fall
threshold level.(2) Consistent with the RTGS philosophy, (ie liquidity turnover will increase). The authors also consider
offsetting payments are, from a legal standpoint, still settled the optimal design of throughput rules; in particular, the
on a gross basis with intraday finality under such optimal number of throughput requirements within the
arrangements. Hence there is no reintroduction of credit risk. payments day. They demonstrate that there will, in theory,
However, the liquidity requirement reduces to the net always be a positive (albeit diminishing) efficiency benefit to
difference between the values of the offsetting payments. implementing an additional throughput requirement: ‘in the
limit, all payments could be made with an infinitesimal
In principle, centralised queuing combined with an offsetting amount of collateral recycled infinitely frequently during the
algorithm should encourage early submission of non-priority day.’ This is illustrated in Box 4.
payment instructions. Willison (2005) finds that the incentive
to delay payments declines upon the introduction of such (1) Ten years ago, in a sample of a dozen of the largest payment systems in the world,
approximately 3% of payments by value were settled in systems using
features: if every bank submits payments early, there is a liquidity-saving features; by 2005 this proportion had risen to 32%. See Bech et al
greater likelihood that the algorithm will identify offsetting (2008). This trend has continued since, most notably with the introduction of
TARGET2 in the euro area.
payments. (2) For more detail on liquidity-saving mechanisms in theory and in practice, see
Manning et al (2009) and McAndrews and Trundle (2001).
(3) Willison (2005), Jurgilas and Martin (2010) and Martin and McAndrews (2008)
Theoretical and empirical work to date generally concludes examine the effectiveness of such measures from a theoretical standpoint. Galbiati
and Soramaki (2010) apply agent-based modelling techniques and conclude that
that introducing such measures would indeed deliver liquidity when the opportunity cost of liquidity is high a system that allows payments to be
channelled to an offsetting stream improves efficiency, with better co-ordination of
savings.(3) Norman (2010) quotes estimated savings of 20% settlements.
and 15%, respectively, in the Korean and Japanese systems, (4) There is, however, an inherent source of inefficiency in throughput requirements.
That is, they enforce the same payment timing on all banks, irrespective of possibly
where offsetting algorithms were recently introduced. differing liquidity costs and differing customer preferences. For instance, if one bank’s
Preliminary estimates of prospective system-wide savings in liquidity cost is higher than the others, it is efficient for that bank to submit its
payments later than banks that face lower liquidity costs. The throughput guideline
CHAPS are of the order of 30% (Ercevik and Jackson (2009) ignores these bank-specific factors.
18 Financial Stability Paper June 2011

Box 4 Chart B Banks’ sending profiles with behavioural change

Throughput requirements: a stylised example to meet the throughput guidelines.
Bank A Bank C
Bank B Throughput rules
This box illustrates how enhanced throughput guidelines could Per cent
100
be used to encourage banks to submit payments earlier. They 90
could also act as a co-ordinating device which may result in a 80
reduction in overall liquidity usage. As in the previous boxes, 70
the data used here are adjusted real data for payments 60
between three banks on a single day. The results are
50
illustrative of the possible impact of more throughput
40
guidelines and are not meant to provide quantitative estimates
30
of the scale of the impact.
20

10
Chart A presents the banks’ baseline sending profiles, where
0
the y-axis captures the percentage of a day’s payments sent.
06:00 08:00 10:00 12:00 14:00 16:00
Time
Sources: Bank of England and authors’ calculations.
Chart A Banks’ base-line sending profiles
Bank A Bank C
Bank B Throughput rules
Per cent Table 1 Impact of throughput on banks’ liquidity needs
100

90 Liquidity needs without Liquidity needs with Percentage

throughput rules throughput rules change
80
Bank A £1.53 billion £1.48 billion -3%
70
Bank B £1.34 billion £1.49 billion +11%
60
Bank C £945 million £659 million -30%
50
System £3.82 billion £3.63 billion -5%
40

30

20

10

0
06:00 08:00 10:00 12:00 14:00 16:00
Time
Sources: Bank of England and authors’ calculations.

Let us now assume that multiple throughput requirements are

introduced: 10% by 8 am; 20% by 9 am; 30% by 10 am up to
80% by 3 pm. As is clear from Chart A, in order to comply
with the early throughput guidelines, Bank B would need to
change its sending profile considerably. Banks A and C would
also have to make some changes to their sending profile.

Chart B shows the profile with behavioural changes to meet

the guidelines.

In this example, the introduction of throughput rules leads to a

decline in liquidity needs for all three banks, reflecting the
improved liquidity recycling early in the day. Bank B has to
commit additional liquidity to the system to meet the earlier
throughput requirements and so uses more liquidity. Banks A
and C, however, enjoy savings of 3% and 30%, respectively,
since they are able to recycle incoming flows from Bank B.

Across the system liquidity needs fall by 5% and earlier

settlement reduces the potential impact of operational risk.
 Financial Stability Paper June 2011 19

The authors recognise, however, that in practice there will be a Table B SIC tariff structure
finite optimal number of throughput requirements: first,
payments are not infinitely divisible; and, second, since the Time Transaction Transaction fees (CHF)(a)
value (CHF)
arrival of payment instructions is stochastic and uncertain, at Delivery Settlement
some point participants will fail to comply simply because they
End-of-day processing — 8 am 0.007 0.00
have no payments to send within the specified time-frame.
8 am–11 am <100,000 0.01 0.02
>100,000 0.10 0.15
This exposes a crucial implementation challenge. In CHAPS,
while most participants tend to meet both the noon and 11 am–2 pm <100,000 0.10 0.20

2.30 pm throughput guidelines and on average system >100,000 0.30 0.70

throughput is above 50% at noon, some participants still 2 pm — end of day <100,000 0.40 0.60
persistently miss the target.(1) These banks must demonstrate >100,000 1.00 2.00

to the CHAPS ‘Star Chamber’ that their failure to meet the Source: SIX Interbank Clearing.
guideline is not due to their withholding liquidity, but rather (a) Fee for sending banks. The recipient bank pays a 0.02 CHF fixed-fee regardless of payment size.
the absence of payment instructions to submit to the system.

This illustrates the difficulties in designing an optimal Calibration of the tariff is important, but challenging. To
enforcement framework for throughput rules. Arguably, a impact effectively on banks’ incentives, the tariff must be
stronger enforcement mechanism than peer pressure would be calibrated to alter the bank’s optimal trade-off between
preferable, to ensure that the cost of non-compliance exceeds liquidity cost and delay cost. James and Willison (2004)
the cost of payments delay. Peer pressure is also unlikely to be estimate the opportunity cost of posting liquidity in CHAPS as
credible when a bank feels that its capital is at risk, ie it is being equal to the difference between a bank’s unsecured and
unlikely to prevent a bank from delaying payments to reduce secured costs of borrowing, which (at that time) was equal to
credit risk to other participants experiencing solvency 7 basis points per annum. This equates to a daily cost of
concerns. around 0.02 basis points. With a delay cost of zero, a
£1 million payment should attract a tariff of the order of £2;
Even if enforcement problems could be resolved, there would with a positive delay cost, the required tariff would decline.
remain scope for banks to delay payments strategically within
a throughput period, potentially leading to spikes in Theoretically, there may be a case for a linear, rather than
throughput just ahead of the target time. Indeed, as evident in exponential tariff, since in the presence of stochastic
Chart 2 (Section 1.1), this pattern is currently observed in operational shocks, the delay cost faced by a bank itself rises
CHAPS. Any strengthening of throughput guidelines should exponentially. As such, an exponentially increasing tariff could
perhaps therefore also include more extensive monitoring of impose an excessive penalty (particularly where a bank may
banks’ payments behaviour, to actively discourage such receive a material proportion of customer instructions late in
strategic behaviour (see Section 5.5, below). the day).

5.3 Time-varying tariff 5.4 Collateral eligibility criteria

An alternative to hard submission deadlines enforced through The overall cost of providing intraday liquidity to the system
penalties is to incentivise early submission through pricing. A depends on both the quantum required to settle payments on
progressive tariff, which increases through the day, might be a timely basis and the unit opportunity cost of generating
expected to give banks good incentives to submit payments liquidity. The policy measures described above all consider the
early. And in contrast to the blunter instrument of throughput former, but in theory steps could also be taken that operate on
guidelines, each bank can continue to optimise its sending the latter. In a collateralised system, one such measure might
profile based on its own liquidity and delay costs be to lower the opportunity cost of generating liquidity by
(incorporating the increase in the tariff).(2) modifying the eligibility criteria of collateral accepted for
intraday liquidity. By doing so, central banks would enable
The Swiss payment system, SIC, is perhaps the leading example settlement banks to post assets that they are already holding
of a system with a progressive tariff (see Table B). In the Swiss on their balance sheets, and for which they have no alternative
case, the tariff increases exponentially over time.(3) The Swiss intraday use.
tariff also rises much more steeply for high-value payments.
This would seem appropriate: first, banks face a sharper
incentive to delay high-value payments due to the higher
(1) See Bank of England (2009).
associated liquidity demand; and second, the system-wide (2) See Ota (2011).
liquidity impact of the delay of a high-value payment will be (3) Serbia has introduced a progressive tariff, which also increases exponentially through
the payments day. Brazil’s STR system also has a time varying tariff, with payments
significantly greater than that of a low-value payment. settled before 9 am charged half the normal fee.
20 Financial Stability Paper June 2011

When considering the appropriate set of eligible collateral, • improved payments efficiency (better co-ordination of
central banks may find there are a number of constraints to payments with others in the system); and/or
providing funds against less liquid securities. First, it may • free-riding on other banks’ liquidity (which would be
provide incentives for banks to readjust their portfolio evident from, on average, later submission times, and —
composition towards riskier assets, although on an intraday combining information from the payment system and
time-frame there may be few opportunities for this (perhaps bank’s internal scheduler — delays between the timing of
only the extension of credit by settlement banks to their the arrival of payment instructions and the timing of
customer banks). In this regard, more intensive monitoring settlement in the system).
could assist.
If such an examination revealed that the bank had indeed been
Another, perhaps more material, concern is that it could engaging in free-riding, then the regulator could insist on
introduce tensions between monetary policy and payment maintaining the intraday liquidity buffer at the pre-existing
systems objectives. If the central bank has wider eligibility level, thereby preserving system-wide liquidity and facilitating
criteria for intraday liquidity than for access to the operational good recycling.
standing facilities (OSFs) this could cause difficulties at the
end of the day. For example, a settlement bank that has If the threat of such a ‘deep dive’ examination is credible —
borrowed intraday liquidity from the Bank via intraday repo which it should be if the process is well articulated and the
may find that it is short of cash at the end of the day and so be authorities have the appropriate powers to request
unable to unwind the repo. Furthermore, the intraday repo information and the tools to conduct the analysis — the
may have been secured with assets that are ineligible for the ex ante incentive to delay will be minimised.
OSFs. This would mean that the repo could not be rolled
without the bank replacing the collateral with higher-quality 5.6 Proposed changes to securities settlement
assets. The bank may not have such assets available and so be systems
unable to borrow from the central bank, interfering with the There are a number of proposed system enhancements that
functioning of monetary policy implementation. will reduce liquidity risk in securities settlement systems. EUI,
the operator of the CREST service, has announced the
5.5 ‘Deep dive’ examination of banks’ payments introduction of a ‘Term DBV’ product in June 2011.(2) This will
behaviour be in addition to the existing overnight delivery-by-value
A possible complement to mechanisms to improve (DBV) mechanism in CREST. When a firm enters a DBV
coordination of payments may be intensive analysis of banks’ transaction it unwinds every morning (ie the cash is returned
payments behaviour as captured in their liquidity reports to to the cash lender and the securities are returned to the cash
the prudential regulator. borrower) and then it re-winds every evening. The Term DBV
product will allow repos to be transacted and only unwound at
In the United Kingdom’s case, the framework for such analysis the end of the term. This will lead to a reduction in the need
could be the FSA’s Individual Liquidity Adequacy Assessment for intraday liquidity for the cash borrowing bank.
(ILAA). In the presence of an additional component of the
liquidity buffer for intraday liquidity, a bank’s may wish to There are also plans to move from a supply driven liquidity
delay payments to reduce intraday liquidity usage and so generation model, whereby intraday liquidity is automatically
reduce the size of the intraday element of its liquid asset created from the central bank when eligible securities are
buffer upon recalibration.(1) purchased, to a demand driven one where liquidity is created
when it is needed to settle a transaction. This should limit the
As such, if a bank were to submit its liquidity report to the amount of liquidity that is generated. To the extent that this
prudential regulator, arguing for a sizable reduction in its incentivises more prudent liquidity management this will
intraday liquidity buffer, then, a ‘deep dive’ examination of reduce the potential for intraday liquidity stresses to
that bank’s payment system performance could be carried out, crystallise.
drawing on information both from transaction records within
the payment system, and information sought directly from the These enhancements show how the design of security
bank on the timing of receipt of client instructions. Such settlement systems can play a significant role in reducing
analysis would enable the regulator to gauge whether the liquidity risk and influencing participant behaviour.
reduction in the required buffer reflected:

• a change in the bank’s underlying business (ie the

magnitude and timing of customer or proprietary flows);
and/or (1) Since, once in place, the incremental opportunity cost to the bank of drawing on the
buffer to facilitate settlement of payments early in the day is zero.
(2) See Euroclear UK and Ireland (2010).
 Financial Stability Paper June 2011 21

6 Conclusion

This paper has sought to clarify the main features of intraday

liquidity usage and the intraday liquidity risks that arise in
payment and settlement systems. In particular, the paper has
highlighted how the practice of double duty, where prudential
asset buffers are used as collateral in the payment system
without being calibrated to include intraday liquidity risk, may
create undesirable risks, both to individual participants and the
system as a whole.

The Bank supports the FSA’s approach of including intraday

liquidity risk as a key risk driver in their wider liquidity
regulations and so making banks hold liquid assets to support
their intraday liquidity needs over and above those for balance
sheet resilience. However, while critical in mitigating intraday
liquidity risk, increasing the size of a bank’s liquid asset buffer
to include this risk might create incentives for banks to reduce
the provision of intraday liquidity in large-value payment
systems.

This paper has shown that such a change in behaviour could

have the undesirable effect of increasing operational risk, and
potentially credit risk, in these systems. The paper has
therefore outlined a menu of options for mitigating this risk,
some relating to changes in the payment system, others aimed
at lowering the opportunity cost of providing liquidity.
Together with regulatory vigilance, they might ensure that the
full benefits of intraday liquidity regulation are achieved,
without undermining the resilience of large-value payment
systems.
22 Financial Stability Paper June 2011

Annex Intraday liquidity sources

Payment systems and intraday liquidity In both cases Bank X has to have access to intraday liquidity in
order to fund its payments. In case 2, Bank Y was able to make
Intraday liquidity all of its payments without access to intraday liquidity beyond
In an RTGS payment system, the settlement of payments the payments that it received from Bank X.
requires that banks have access to intraday liquidity. To
illustrate, two banks, Bank X and Bank Y, each have to make Bank X can obtain intraday liquidity from a number of
two payments of one unit to the other bank. The order of the sources:(1)
settlement of these payments determines how much liquidity
each bank needs and how much is required in aggregate. i) Central bank reserves. It can start the day with cash on its
central bank reserve account which can be used for making
Case 1 payments.
Bank X makes its first payment, then Bank Y makes both of its
payments and finally Bank X makes its second payment. ii) Collateral that can be pledged/or repoed to the central
bank in return for intraday liquidity. Most Central Banks
offer settlement banks free collateralised intraday liquidity.
Bank X Bank Y
1
Balance -1 Balance +1 iii) Overnight money markets. Bank X could enter the
overnight unsecured money market and borrow some
funds. If these funds are received before it needs to make
Bank X Bank Y the first payment, it can use these funds for intraday
1
Balance 0 Balance 0 liquidity.

Indirect participants
Bank X Bank Y Many banks do not participant directly in large value payment
1
Balance +1 Balance -1 and settlement systems but use the services of a
correspondent bank instead. A correspondent bank is a direct
participant of the system and offers to settle payments on
Bank X Bank Y behalf of another bank.
1
Balance 0 Balance 0

The result is that both banks use 1 unit of liquidity. Payment system

Case 2
Bank X makes both of its payments and then Bank Y makes
both of its payments.
Bank X Bank Y

Bank X Bank Y
1
Balance -1 Balance +1

Customer Customer Customer Customer

Bank A Bank B Bank C Bank D
Bank X Bank Y
1
Balance -2 Balance +2

Bank A sends a payment to Bank D:

Bank X Bank Y Bank A will send an order to Bank X to ask it to make a
1
Balance -1 Balance +1 payment on its behalf. Bank X will debit Bank A’s account and
send the message to the payment system. This will debit
Bank X’s settlement account and credit Bank Y’s account.
Bank X Bank Y When Bank Y receives the payment it will in turn credit
1
Balance 0 Balance 0 Bank D’s account.

The result is that Bank X uses two units of liquidity whereas (1) There are a number of other sources of intraday liquidity that a bank could use such
as intraday credit lines or incoming payments from ancillary systems. For simplicity
Bank Y uses none. we have only included the most commonly used.
 Financial Stability Paper June 2011 23

Bank A sends a payment to Bank B:

Bank A will send an order to Bank X to ask it to make a
payment on its behalf. As Bank B is also a customer of Bank X
the payment does not need to be sent to the payment system.
Bank X will debit Bank A’s account and credit Bank B’s account.
The payment will settle across the books of Bank X. This type
of payment is known as an internalised payment. Bank X
requires no liquidity to settle such a payment, but Bank A
does.
Journal of Financial Market Infrastructures Volume 1/Number 1, Fall 2012 (75–107)

Estimating the intraday liquidity risk of

financial institutions: a Monte Carlo
simulation approach
Carlos León
Financial Infrastructure Oversight Department, Payment Systems and Banking
Operation Division (PSBOD), Banco de la República, Carrera 7 # 14–78,
Bogotá, Colombia; email: [email protected]

The recent financial crisis has shown that liquidity risk is far more important
and intricate than regulators had previously acknowledged. The shift from bank-
based to market-based financial systems and from deferred net settlement systems
to liquidity-demanding real-time gross settlement of payments explains some of
the shortcomings of traditional liquidity risk management. Although liquidity reg-
ulations do exist, they are still in an early stage of development and discussion.
Moreover, not all connotations of liquidity are being equally addressed. Unlike
market and funding liquidity, intraday liquidity has been absent from financial
regulation, and has appeared only recently, after the crisis. This paper addresses
the measurement of large-value payment system intraday liquidity risk. Based on
the generation of bivariate Poisson random numbers for simulating the minute-
by-minute arrival of received and executed payments, each financial institution’s
intraday payments’ time-varying volume and degree of synchrony (ie, timing) is
modeled. Modeling the uncertainty of intraday payments allows us to oversee par-
ticipants’ intraday behavior, to assess their ability to fulfill intraday payments at
a certain confidence level, to identify participants that are nonresilient to changes
in payment timing mismatches, and to estimate intraday liquidity buffers. These
results are useful for financial authorities and institutions given the increased
importance of liquidity risk as a source of systemic risk, and the recent regulatory
amendments.

The opinions and statements in this paper are the sole responsibility of the author and do not
necessarily represent those of Banco de la República or its Board of Directors. Results are illustrative;
they may not be used to infer credit quality or to make any type of assessment for any financial
institution. The author is indebted to Clara Machado for the numerous and vital discussions that
supported the design of the model and the final version of the paper. Comments and suggestions
were provided by Fernando Tenjo, Joaquín Bernal, Freddy Cepeda and Fabio Ortega. Valuable
comments and suggestions from an anonymous referee significantly enhanced the original version
of this paper. Large-value payment system data was processed with assistance from Freddy Cepeda
and Fabio Ortega. Any remaining errors are the author’s own.

75
76 C. León

1 INTRODUCTION
It is widely accepted that liquidity risk mismanagement played a key role in the
recent global financial crisis. The literature recommends improving liquidity risk
management by imposing and monitoring liquidity requirements on systemically
important banks and broker dealers (French et al (2010)), or designing liquidity
buffers in order to mitigate systemic risk (Capel (2011); International Monetary Fund
(2010b); Borio (2009); and Tirole (2008)).
Although liquidity regulations and tools do exist, they are still in an early stage
of development and discussion (International Monetary Fund (2010a) and Tucker
(2009)). Moreover, not all connotations of liquidity are equally addressed by risk
literature or regulation. Liquidity has focused on market and funding liquidity, where
both correspond to the ability to generate cash from balance sheet liabilities and asset
positions, respectively, whereas liquidity risk has traditionally focused on measuring
mismatches between them (eg, short-term liabilities and liquid assets).1
As acknowledged by Ball et al (2011), prior to the recent financial crisis regulators
did not focus on intraday liquidity risk and there were no standard monitoring or
liquidity management measures in place. Authorities have only begun to focus on
intraday liquidity since the crisis. Two main structural shifts may explain the new
emphasis on intraday liquidity:
(1) the shift from bank-based to market-based financial systems;

(2) the evolution of payment systems from deferred net settlement systems to
liquidity-demanding real-time gross settlement (RTGS) systems.
It is important to acknowledge that these structural shifts have not resulted from
shocks. They are the result of a continuous and protracted evolution of financial
markets. However, because prudential regulation has ignored these structural shifts for
decades, the regulatory challenge is substantial: an intraday liquidity risk management
framework must be designed.
Among the four typical stages of risk management (ie, identifying, assessing, mon-
itoring and mitigating risk) this paper focuses on the second stage. The approach pre-
sented in this paper for assessing the intraday liquidity risk of large-value payment
systems (LVPSs) is based on the generation of bivariate Poisson random numbers for
simulating the minute-by-minute arrival of received (incoming) and executed (out-
going) payments. In this sense, following Ball et al (2011), the identified source of

1 For instance, Tirole (2008) refers to funding liquidity as how much can be raised on the liability side

of the balance sheet, while market liquidity refers to the asset side, with prudential measurements of
liquidity usually consisting of measuring some mismatch between short-term liabilities and liquid
assets.

Journal of Financial Market Infrastructures Volume 1/Number 1, Fall 2012

 Estimating the intraday liquidity risk of financial institutions 77

intraday liquidity risk modeled here is the timing mismatch between incoming and
outgoing payments, which may lead to significant intraday liquidity needs.
This Monte Carlo simulation procedure is capable of modeling the intraday-
dependency governing the joint arrival of both types of payments, along with their
value. Thus, the simulation procedure is able to capture the degree of synchrony (ie,
the timing) attained by each financial institution when receiving and executing pay-
ments (ie, the virtuous circle of coordinated actions by settlement agents), where such
synchrony and the volume of payments varies throughout the day.
The main outcome of the proposed procedure is estimating a risk measure or metric
such as an intraday liquidity value-at-risk (VaR) that is able to answer a rather specific
question: what are an institution’s maximum intraday liquidity needs at a defined
confidence level? An answer to this question may be suitable for:

(1) overseeing participants’ intraday behavior;

(2) assessing their ability to fulfill intraday payments at a certain confidence level;

(3) identifying participants who are nonresilient to changes in timing mismatches

of payments;

(4) estimating intraday liquidity buffers.

Regarding the most recent amendments to financial regulation and the increasing
importance of liquidity risk as a source of systemic risk, results are useful for financial
authorities, institutions and market infrastructures tackling the challenge of managing
intraday liquidity risk.
This paper is structured as follows. Section 2 briefly addresses the increasing rel-
evance of intraday liquidity risk management. Section 3 describes the intuition and
procedure behind the proposed approach to simulating the minute-by-minute liq-
uidity (Appendix A contains a comprehensive technical explanation of the Monte
Carlo simulation algorithm). Section 4 presents preliminary results and analysis for
a set of institutions participating in Colombia’s LVPS. Section 5 presents some final
comments on the approach, its usefulness and the challenges ahead.

2 THE INCREASING RELEVANCE OF LIQUIDITY AND INTRADAY

LIQUIDITY RISK MANAGEMENT
The recent financial crisis has highlighted the need to improve liquidity risk man-
agement, including the management of intraday liquidity risk (Ball et al (2011)).
Liquidity is by no means a new concept. It is, however, still an elusive one. It com-
prises several dissimilar connotations, such as market, funding or intraday liquidity.
Although these connotations of liquidity allow for a fairly clear theoretical distinction,

Research Paper www.risk.net/journal

78 C. León

in practice they are entangled, particularly under stress scenarios. In this sense, all
connotations of liquidity should be equally addressed by prudential regulation.
Notwithstanding the importance of properly addressing liquidity risk and its con-
notations, related regulation is scarce when compared with solvency regulation. The
contemporary momentum of liquidity and intraday liquidity regulation has arisen
from the recent global financial crisis, which exposed structural shifts in financial
markets that have increased the relevance of designing a proper prudential regulatory
framework.
Two such structural shifts are commonly acknowledged in the literature. First,
the shift from bank-based to market-based financial systems, and second, the shift
from deferred net settlement systems to liquidity-demanding RTGS of payments.
As explained below, the former has increased the importance of all connotations of
liquidity risk, whereas the latter has emphasized the relevance of intraday liquidity
risk.

2.1 The relevance of liquidity risk in market-based financial

systems
The underdevelopment of liquidity regulation is due to the traditional focus on sol-
vency, which is the legacy of banking runs dominating systemic risk origins since the
outbreak of the Great Depression. Consequently, liquidity risk has evaded prudential
regulation.
The best example of the absence of liquidity risk management is the regulatory
approach by the Basel Committee on Banking Supervision. As documented by Good-
hart (2008), the Basel Committee on Banking Supervision’s original goal was to
reverse the downward trend in capital and liquidity for the main international com-
mercial banks. However, for reasons yet to be discovered, this downward trend was
reversed for capital, not for liquidity. Thus, according to Eichengreen (2008), some of
the Basel Accord’s critics argue that its focus on capital adequacy (ie, lack of liquidity
requirements) encouraged regulators to neglect the importance of liquidity in their
supervisory activities.
Additionally, not only does the Basel Committee’s regulation disregard liquidity
risk management, but it is dedicated to banks only, for which it measures mismatches
between short-term liabilities (eg, deposits) and liquid assets (eg, loans and invest-
ment portfolios). In today’s financial markets, with many functions that previously
defined banks’ traditional domain increasingly being performed by securities markets
and nonbank market participants (Kambhu et al (2007)),2 focusing on solvency and

2 This scheme, in which markets and nonbank participants involve in credit intermediation activities

traditionally performed (and regulated to be performed) by banks, is commonly referred to as the

“parallel banking system” or “shadow banking system” (Krugman (2009) and Acharya et al (2009)).

Journal of Financial Market Infrastructures Volume 1/Number 1, Fall 2012

 Estimating the intraday liquidity risk of financial institutions 79

ignoring liquidity is highly unsafe from a prudential point of view. Therefore, the
structural shift from bank-based to market-based systems and the evolution of finan-
cial infrastructures, where market and asset liquidity has become as important as the
solvency of banks, explains to some extent the increasing relevance of liquidity risk
management.
Hence, as a consequence of the nature of the global financial crisis, Ackermann
(2008) draws two key conclusions. First, in our capital-based financial system, which
has developed as a result of disintermediation and credit risk transfer, liquidity is far
more important than in a purely bank-based financial system. Second, better liquidity
management, rather than higher capital buffers, is likely to provide the right answer.3

2.2 The relevance of intraday liquidity risk in RTGS payment

systems
A second structural change that has brought about the increasing relevance of liquid-
ity risk is the evolution of financial markets from deferred net settlement systems to
liquidity-demanding RTGS of payments.4 As recognized by the Committee on Pay-
ment and Settlement Systems (1997), this structural shift was encouraged by banking
authorities in an attempt to limit settlement and systemic risk in the interbank settle-
ment process, and to contribute to the reduction of the settlement risk in securities and
foreign exchange transactions. However, as pointed out by Bernal (2009) and Bech
and Soramäki (2002), in RTGS systems the reduction in settlement risk is traded off
against increased liquidity requirements so that the payment system becomes more
reliant on the virtuous circle of coordinated actions by participating settlement agents
and, therefore, increased liquidity risk.5 Following Ball et al (2011), this means that,
whereas RTGS systems financial institutions can reuse liquidity from incoming pay-
ments to fund outgoing payments, timing mismatches between incoming and outgoing
payments can lead to significant (intraday) liquidity needs.

3 Ackerman (2008) goes further, stating that higher capital requirements may have an adverse effect:

if they are too onerous, more activities will be pushed to unregulated parts of the financial system.
4 A deferred net settlement system effects the settlement of obligations or transfers between or among

counterparties on a net basis at some later time. An RTGS system consists of the continuous (real-
time) settlement of funds or securities transfers individually on an order-by-order basis (without
netting); the processing of instructions is carried out on an individual basis at the time they are
received rather than at some later time (Committee on Payment and Settlement Systems (2003)).
Bech (2008) documents that the number of central banks that implemented RGTS systems increased
from 3 in 1985 to 93 at the end of 2006.According to World Bank (2011), from a total of 142 countries
surveyed as of December 2010, 116 (83%) have an RTGS system for their LVPSs.
5 Such increasing demand for intraday liquidity is also documented by Bech (2008), Rochet (2008)

and Committee on Payment and Settlement Systems (1997).

Research Paper www.risk.net/journal

80 C. León

Consequently, as documented by Leinonen and Soramäki (2004), interest in intra-

day liquidity results from payment system development and shrinking delivery times.
Before the 1990s, operations were strictly on the daily level and intraday liquidity had
no significance. As the speed of processing payments has been increased and central
banks have converted to RTGS, intraday liquidity has received increasing emphasis.
For instance, the emergence of intraday liquidity risk is rather clear in the evolution
of Basel Committee on Banking Supervision’s approach to liquidity risk. What the
Basel Committee on Banking Supervision (2000) regarded as the “Principles for the
Assessment of Liquidity Management in Banking Organizations” referred to short-
term liquidity management in a day-to-day basis for banks reliant on short-term
funding, and in a one-to-three-months-ahead basis for other banks that are not reliant
on short-term funding. Intraday liquidity risk was mentioned but was not regarded as
being decisive.
More recently, the Basel Committee on Banking Supervision (2008) explicitly
included the management of intraday liquidity risk as a principle on its own (Princi-
ple 8), where its purpose is meeting payment and settlement obligations on a timely
basis under both normal and stressed conditions in order to contribute to the smooth
functioning of payment and settlement systems.6 Furthermore, even more recently,
the document entitled Basel III (Basel Committee on Banking Supervision (2010))
acknowledges that intraday liquidity needs may not be covered by Basel III’s new
liquidity coverage ratio,7 and states that the Basel Committee on Banking Supervision
is reviewing if (and how) intraday liquidity should be addressed.
In the report “Principles for financial market infrastructures” (Committee on Pay-
ment and Settlement Systems and the International Organization of Securities Com-
missions (2012)), Principle 7 is dedicated to liquidity risk management by financial
market infrastructures, where the main objective is that such infrastructures maintain
sufficient liquid resources to effectuate same-day and, where appropriate, intraday
and multiday settlement of payment obligations with a high degree of confidence
(Committee on Payment and Settlement Systems and International Organization of
Securities Commissions (2012)).

6 The Basel Committee on Banking Supervision (2000) document only referred to intraday liquid-
ity four times, with only one mention related to liquidity management (Principle 5). The Basel
Committee on Banking Supervision (2008) document makes about sixty references to intraday liq-
uidity (Principles 3, 5, 9 and 10), and devotes Principle 8 to recognizing its importance within a
bank’s broader liquidity management strategy and its contribution to systemic risk via the smooth
functioning of the payment system.
7 The liquidity coverage ratio is a standard measure that aims to ensure that a bank maintains an

adequate level of unencumbered, high-quality liquid assets that can be converted into cash to meet
its liquidity needs for a thirty calendar day time horizon under a significantly severe liquidity stress
scenario specified by supervisors (Basel Committee on Banking Supervision (2010)).

Journal of Financial Market Infrastructures Volume 1/Number 1, Fall 2012

 Estimating the intraday liquidity risk of financial institutions 81

Another example of the increasing focus of regulation on intraday liquidity risk is

the UK Financial Services Authority’s (FSA’s) novel liquidity adequacy regulation
(FSA (2012)), which considers intraday liquidity to be a key risk driver in its new
liquidity regime. As in the principles of Basel Committee on Banking Supervision
(2008), the FSA’s aim is to ensure that firms are able to meet their payment and settle-
ment obligations on a timely basis in both normal and stressed conditions, emphasizing
that this is important for the firm, the firm’s counterparties and the smooth functioning
of payment and settlement systems as a whole. It is worth noting that, unlike the Basel
Committee on Banking Supervision, the FSA’s liquidity regulation is not intended for
banks only, and refers to “firms”, with this term comprising banks, building societies
and some types of investment firms. According to JP Morgan (2010), this new regu-
latory regime includes for the first time all FSA-regulated broker-dealer firms under
formal liquidity resource supervision.
In the author’s view, concurring with Hervo (2008), structural developments in the
financial industry have led to a clear trend in shortening time horizon of liquidity risk
and liquidity management. As Hervo (2008) quotes regarding contemporary financial
markets, “my short-term is intraday, my medium-term is overnight and my long-term
is one week”.
Although payment and settlement systems have received relatively little attention
from financial market researchers (Leionen and Soramäki (2004)), intraday liquidity
literature has gained momentum, especially with works by Bech (2008), Leionen
(2007) and Koponen and Soramäki (1998). In the Colombian case, LVPS intraday
liquidity has been addressed by Bernal et al (2011), Bernal (2009) and Bernal and
Merlano (2005), while some models based on LVPS payments data for payments
simulation, connectedness assessment and the identification of systemic importance
have been developed recently (Cepeda (2008); Machado et al (2009); León et al
(2011); and León and Machado (2011)).
However, the available literature does not address intraday liquidity risk explicitly,
and lacks risk measures or metrics (such as an intraday liquidity VaR or expected
shortfall) that are able to answer a rather specific question: what are an institution’s
maximum intraday liquidity needs at a defined confidence level? The next two sections
deal with how to address this type of question.

3 MONTE CARLO INTRADAY LIQUIDITY SIMULATION

Monte Carlo simulation methods are suitable for addressing problems of almost any
degree of complexity, and can easily address factors that most other approaches have
difficulty with. They become more useful as the complexity and/or dimensionality of
the problem increases (Dowd (2005)). Therefore, as the case in hand comprises several
factors to be simultaneously modeled in order to deal with a financial institution’s

Research Paper www.risk.net/journal

82 C. León

intraday liquidity uncertainty, Monte Carlo provides a compelling approach. The next
two subsections contain an explanation of the intuition behind using the Monte Carlo
simulation approach to deal with intraday liquidity uncertainty and on the designed
procedure, respectively.

3.1 Dealing with intraday liquidity uncertainty

According to Principle 8 of Basel Committee on Banking Supervision (2008), a strat-
egy for achieving intraday liquidity management objectives involves six elements.
Elements 1, 2 and 6 are as follows.8

Financial institutions should have the capacity to measure expected daily gross
liquidity inflows and outflows, anticipate the intraday timing of these flows
where possible, and forecast the range of potential net funding shortfalls that
might arise at different points during the day.

Financial institutions should have the capacity to monitor intraday liquidity

positions against expected activities and available resources.

Financial institutions should be prepared to deal with unexpected disruptions

to their intraday liquidity flows.

Additionally, according to Basel Committee on Banking Supervision (2008, Prin-

ciples 10 and 11), liquidity stress tests and contingency plans should observe the
following elements.

Stress tests should consider the implication of the scenarios across different
time horizons, including on an intraday basis.

Financial institutions’ stress tests should consider how the behavior of counter-
parties would affect the timing of cash flows, including on an intraday basis.

The size of financial institutions’ liquidity cushions should also reflect the
potential for intraday liquidity risks.

Likewise, Committee on Payment and Settlement Systems and International Organi-

zation of Securities Commissions (2012) regards intraday liquidity as key for overall
liquidity risk management by financial market infrastructures. For instance, these
principles encompass the following considerations.

8 The reader should be aware that these principles from Basel Committee on Banking Supervision

(2008) are limited to banks. Taking into account the importance of the nonbanking institutions in
financial markets, the author avoids limiting the application of these principles to banks; the author
refers to “financial institutions” instead of banks. Please note that all emphasis is the author’s.

Journal of Financial Market Infrastructures Volume 1/Number 1, Fall 2012

 Estimating the intraday liquidity risk of financial institutions 83

A financial market infrastructure should maintain sufficient liquid resources …

to effect same-day and, where appropriate, intraday and multiday settlement
of payment obligations with a high degree of confidence under a wide range of
potential stress scenarios.

A financial market infrastructure should have effective operational and analyt-
ical tools to identify, measure, and monitor its settlement and funding flows on
an ongoing and timely basis, including its use of intraday liquidity.
A common approach suitable for tackling the quantitative demands imposed by these
elements is Monte Carlo simulation.All elements, particularly the italicized parts, con-
verge to modeling financial institutions’ intraday payments uncertainty (ie, expected
liquidity, potential shortfalls, liquidity scenarios, etc), whereas dealing with uncer-
tainty is the whole point of building a Monte Carlo model (Hubbard (2009)).9
Traditional Monte Carlo methods in finance are aimed at repeatedly simulating the
random process governing the returns of an asset or instrument, where the govern-
ing process is the geometric Brownian motion, along with some other variations to
this customary process.10 Such typical application of Monte Carlo is rather straight-
forward and flexible since the random process is easily simulated (ie, there is only
one stochastic process for each asset), even when considering the dependence across
different assets.
However, simulating intraday liquidity is more intricate. In order to measure
expected intraday liquidity inflows and outflows (Basel Committee on Banking Super-
vision (2008, Principle 8)), two different stochastic processes are to be simulated: one
governing the inflows (arrival of received payments), and the other governing the out-
flows (arrival of executed payments). Because of the type of behavior to be modeled
(arrival of payments), geometric Brownian motion is unsuitable, and an arrival-type
process has to be used. Each process has to be generated with a Poisson process.
Furthermore, since the degree of synchrony between the arrival of received and
executed payments is critical for modeling the virtuous circle of coordinated actions
by agents typical of RTGS systems (Bernal (2009)), the simulation of the random
processes has to capture executed and received payments’ dependence: each process
has to be generated from a bivariate Poisson process. Paraphrasing Ball et al (2011),
this would allow for modeling the timing mismatches between incoming and outgoing

9 Note that the term “uncertainty” is not used in the Knightian sense (Knight (1921)), where uncer-
tainty is related to cases in which quantifying probabilities is not possible. In this paper the term
“uncertainty” is as in Hubbard (2009), where it corresponds to the lack of complete certainty about
the true outcome, with uncertainty being measurable (contrary to Knight’s use of the term) by the
assignment of probabilities to various outcomes.
10 Some of the most popular variations are jump-diffusion models (Merton (1976) and León (2009))

and fractional Brownian motion (Mandelbrot and Van Ness (1968) and León and Reveiz (2011)).

Research Paper www.risk.net/journal

84 C. León

payments that lead to an increase in the amount of intraday liquidity required to

continue making payments in a timely fashion.
Additionally, the size (ie, monetary value) of the payments has to be modeled along
with the frequency of arrival, where the size of payments may not be distributed as
a normal variable and where samples may not be large enough to make (parametric)
distributional assumptions. Finally, since the behavior of the arrivals, their dependence
and their monetary value are not constant throughout the day, the simulation’s factors
or parameters have to be time-dependent.
Altogether, these considerations demand a Monte Carlo simulation model that is
significantly different from its standard implementation in finance. The next subsec-
tion addresses the procedure behind an implementation of the Monte Carlo model
that is suitable for this paper’s purposes. Appendix A contains a comprehensive tech-
nical explanation on the Monte Carlo simulation algorithm, with an emphasis on the
simulation of bivariate Poisson random variables for the intraday arrival of executed
and received payments.

3.2 Implementation
The model could be concisely described as the joint and time-dependent simulation of
a bivariate Poisson processes for intraday executed and received payments, and their
monetary value. The core of the model is the Monte Carlo simulation of bivariate
Poisson random variables for the intraday arrival of executed and received payments,
while the simulation of their monetary value using bootstrap historical simulation is
subordinated to their arrival. The implementation of the model proposed here is done
in Matlab. It consists of an algorithm executing the procedure depicted in Figure 1
on the facing page.
The algorithm consists of five main inputs: two databases, for LVPS payments and
opening balances, which contain information for all participating financial institutions
during one day; and three manual inputs, which select the financial institutions to
analyze, define the intraday time frames to use, and the number of simulations to
generate.
After reading the inputs, the algorithm selects the first financial institution (eg,
bank A) and the first time frame to use (eg, from 07:00 to 08:00). According to this
selection, the LVPS orders and opening balances databases are filtered. Afterward,
the algorithm classifies both types of payments (ie, executed and received) for the
selected institution and time frame.
Afterward (part (a) of Figure 1 on the facing page), the Monte Carlo simulation of
the payments’ arrival starts by estimating the intensity of the executed and received
processes (
E and
R ), along with their correlation (.E;R/ ). After estimating the
parameters required for the simulation of the bivariate Poisson process for the intraday

Journal of Financial Market Infrastructures Volume 1/Number 1, Fall 2012

 Estimating the intraday liquidity risk of financial institutions 85

FIGURE 1 Algorithm’s procedure (flow chart).

Start
CUD
opening Payments
balance database

Financial Institutions
(x = 1,2,…,X)
x=1
Intraday timeframes
(n = 1,2,...,N)

Number of simulations n=1

(q = 1,2,...,Q)

Received Type Executed

Received Executed
payments payments

Estimate payments'
Estimate received correlation (πR,E) Estimate executed
payments' intensity (λ R) payments' intensity (λE)

(a)

q=1

Simulate bivariate Poisson

process (Arrival)

Received Executed
Type

Arrival Arrival
Yes Yes
(b)
Simulate monetary value Simulate monetary value
of received payments of executed payments

Simulated received Simulate executed

payments payments

Received Executed
Simulated intraday net
payments (received−executed)

Simulated intraday net balance

(net payments+opening balance) No
q>Q q=q+1
Yes
No
n>N n=n+1
Yes No
x>X x=x+1
Yes
End

(a) Monte Carlo simulation of bivariate Poisson processes for the intraday arrival of payments. (b) Bootstrapped
historical simulation of received and executed payments’ monetary value. Source: author’s own design.

Research Paper www.risk.net/journal

86 C. León

arrival of payments, the algorithm generates the first of the simulations to make
for this financial institution, for the selected time frame. Based on the algorithm
designed by Yahav and Shmueli (2011) for simulating bivariate Poisson processes,
the algorithm yields a minute-by-minute two-dimensional vector where the simulated
joint-occurrence of executed and received payments is registered.11
Subsequently, after simulating the first path of arrivals for the selected financial
institution and time frame, the algorithm employs the bootstrapped historical simu-
lation method for generating the monetary value of each of the arrivals previously
simulated (part (b) of Figure 1 on the preceding page). Thus, each time the algorithm
generates the arrival of an executed (received) order, the algorithm employs a uniform
random number generator to resample – with replacement – from the historical record
of monetary values of executed (received) payments that the selected financial insti-
tution made during the selected time frame. This process yields a minute-by-minute
two-dimensional vector where the simulated value of executed and received orders is
registered.
Next, the monetary value of received and executed payment orders is subtracted.
The result is the simulated intraday net payments. If the opening balance is added, the
result is the simulated intraday net balance for the selected financial institution and
time frame. Both results are the main building blocks of the model: a single simula-
tion of the minute-by-minute liquidity balance (with and without opening balance)
for a selected financial institution and time frame. In order to make all the simula-
tions defined by the user, and to cover all financial institutions and time frames, the
algorithm performs three loops.12

4 PRELIMINARY RESULTS
Based on a day of transactions from the February 2012 LVPS database, this section
applies the proposed model to simulate the intraday liquidity of two selected financial
institutions.13 The financial institutions selected belong to the top-ten systemically
important institutions according to León and Machado (2011), and they correspond

11 Note that the occurrence of executed or received payments per minute is not limited to a binary
outcome (ie, 0 or 1 payments). During a minute several occurrences may take place, with each
minute potentially containing several payments. Appendix A contains a comprehensive technical
explanation of the Monte Carlo simulation algorithm.
12 This is a technical drawback of the proposed model. In the case of Colombia’s LVPS, where more

than 100 financial institutions participate directly in the LVPS, with thirteen time frames (ie, hourly,
from 07:00 to 20:00), with 1000 simulations, the whole procedure for a single day consists of about
2 million registers.
13 Due to disclosure issues, the opening balance and payments data in this section was multiplied by

a factor of 1˙0:1 in order to ensure financial institutions’ anonymity without sacrificing congruence
and comparability.

Journal of Financial Market Infrastructures Volume 1/Number 1, Fall 2012

 Estimating the intraday liquidity risk of financial institutions 87

FIGURE 2 Hourly intraday payments for selected institutions CBF1 and BDF1.

×108 (a) ×108 (b)

2.0 2.0
Payment (US$ ×108)

Payment (US$ ×108)

1.5 1.5
1.0 Received Received
1.0
0.5 0.5
0 0
−0.5 −0.5
−1.0 Executed −1.0
Executed
−1.5 −1.5
−2.0 −2.0

08:00 12:00 16:00 20:00 08:00 12:00 16:00 20:00

(a) CBF1 opening balance: US$750.0 million. Total executed/received: US$498.7/538.1 million. (b) BDF1 opening
balance: US$0.3 million. Total executed/received: US$872.8/871.3 million. Received and executed payments have
positive and negative signs, respectively. Triangles along the x-axis identify the presence of CSD and LVPS liquidity
saving mechanisms. Source: author’s calculations, data from the LVPS.

to a commercial banking firm (henceforth referred to as CBF1) and a broker-dealer

firm (BDF1). The selected time frame corresponds to an hour-by-hour14 breakdown
of the Colombian LVPS working hours (ie, 07:00–20:00). The selected number of
simulations for all calculations is 1000, but figures display 100 simulations for com-
prehensibility purposes.
Figure 2 displays the observed minute-by-minute intraday payments for selected
institutions CBF1 and BDF1. Executed and received payments appear with negative
and positive signs, respectively. It is clear that modeling the intraday liquidity as a
non-time-varying process would be inconvenient. The intraday liquidity of financial
institutions is heavily dependent on the schedule or timeline designed by the adminis-
trator of the LVPS and by all other infrastructures that use the LVPS as their settlement
system, and on the design of the LVPS, which may be deferred net settlement systems
or RTGS, and may also include liquidity saving mechanisms and other types of rules
that affect decision-making by the system’s participants.

14 An hourly division of the intraday was used in order to attain data-abundant time frames for a
representative number of institutions for the whole intraday. Given that highly active institutions
along the whole intraday are scarce, using “long” windows reduces the estimation error of the
parameters and maximizes the number of institutions to work with. Although the chosen length
of time frames is convenient, its soundness is worth examining, as pinpointed by the anonymous
referee. Appendix C displays three different assumptions for the length of the windows that divide
the intraday time frame (ie, one hour, thirty minutes, fifteen minutes), where there is evidence of
a nonnegligible difference between the assumptions, but where it is evident that the cross-section
analysis and the conclusions remain.

Research Paper www.risk.net/journal

88 C. León

Figure 2 on the preceding page confirms that the pattern of intraday payments is
determined in most part by the liquidity saving mechanisms of the Central Bank’s
Central Securities Depository (CSD) and the LVPS (triangles along the x-axis); the
former consists of optimization algorithms running at 11:50, 14:20, 15:30, 16:15,
16:55 and 17:45, while the latter runs at 16:00.15 The CSD’s optimization algorithms
are key to the intraday liquidity of financial institutions since the central government
local bond market (ie, the TES market) is the most active and liquid in the Colombian
financial system, where CSD TES-related payments account for about 50% of LVPS
total payments (Banco de la República (2011)).
It is also clear that, for the selected date, both institutions display a distinctive pattern
of intraday liquidity. Beyond any particularity arising from the choice of the date, these
results arise from their characteristic business and regulatory framework. For instance,
commercial banks have to comply with reserve requirements, whereas broker-dealer
firms do not have to. Commercial banks trade bonds and foreign exchange on their
own account only, whereas broker-dealer firms trade on their own account and on
behalf of clients, profiting from brokerage via commissions; broker-dealer firms are
allowed to trade stocks, whereas commercial banks are not.
It is important to stress that such characteristics may explain two distinctive fea-
tures of the selected financial institutions. First, the intraday liquidity pattern is more
concentrated at the end of the day for BDF1. This pattern results from

the lack of reserve requirement and the corresponding low level of opening
balance,

its reliance on the liquidity arriving from the virtuous circle of coordinated
actions by other settlement agents,

the prominence of liquidity saving mechanisms provided by the CDS for TES-
related transactions.

On the other hand, CBF1, which holds high levels of opening balance (ie, around 2700
times that of BDF1) due to reserve requirements, may execute payments earlier.16

15 For an introduction to the design and functioning of the Colombian RTGS payment system and its
timeline, please refer to Bernal et al (2011), Banco de la República (2011) and Bernal and Merlano
(2005).
16 As in 52% of the countries using an RTGS system (World Bank (2011)), local participants may

use all their reserve requirements balance for intraday settlement purposes. Reserve requirements
in Colombia are based on the daily averaging of reserve requirements within a two-week reserve
maintenance period. According to Gray (2011), averaging reserve requirements is an effective way
of enhancing liquidity management, but the reserve maintenance period needs to be at least two
weeks long.

Journal of Financial Market Infrastructures Volume 1/Number 1, Fall 2012

 Estimating the intraday liquidity risk of financial institutions 89

FIGURE 3 Estimated intensities (
E and
R ) and correlation (.E;R/ ) parameters.

(a) (b)

Intensity (arrivals per minute)

Intensity (arrivals per minute)

2.5 2.5

2.0 2.0

or correlation
or correlation

1.5 1.5

1.0 1.0

0.5 0.5

0 0
7–8 11–12 15–16 19–20 7–8 11–12 15–16 19–20

(a) CBF1. (b) BDF1. Light gray bars: intensity of executed. Dark gray bars: intensity of received. Dashed line: cor-
relation. Triangles along the x-axis identify the presence of CSD and LVPS liquidity saving mechanisms. Source:
author’s calculations and data from the LVPS.

Second, the size of payments is also distinctive, with payments executed and received
by the BDF being about 1.8 times those of the CBF.
The estimated parameters for the Monte Carlo simulation of bivariate Poisson
process for the intraday arrival of payments for both selected financial institutions are
displayed in Figure 3.17
Based on the estimated parameters, Figure 4 on the next page exhibits 100 extracts
from simulating 1000 times the minute-by-minute intraday liquidity of the two
selected financial institutions with hourly time frames.18 This figure corresponds to
the simulated intraday net balance; that is, it considers the opening balance of each
institution.19

17 The intensities (
and
) were estimated using the standard maximum likelihood estimate
E R
for a Poisson distribution, whereas the correlation (.E;R/ ) was estimated as a standard correlation
coefficient; as previously explained, each parameter was estimated for its corresponding time frame.
18 To achieve a fair approximation of the correlation of the simulated bivariate Poisson series to the

target correlation is the mainstay of the bivariate Poisson process and the model. As presented in
Appendix B, the mean of the correlation of the simulated bivariate Poisson series replicates the target
correlation, whereas the simulated correlation of each series disperses around the target correlation
as expected.
19 Under certain circumstances it would be useful not to consider the opening balance. For example,

to analyze the ability of an institution to face executed payments with received payments (ie, the
virtuous circle of liquidity). Other analysis may be available by excluding some intraday funding
sources, for example. This would allow for an analysis of the reliance of an institution on Central
Bank or on the monetary market’s intraday liquidity. In forthcoming papers the author expects to
implement such variations to the model.

Research Paper www.risk.net/journal

90 C. León

FIGURE 4 Observed and simulated intraday net balance paths for CBF1 and BDF1.

CBF1

Net balance (US$ million)

750.0

BDF1
0

07:00 09:00 11:00 13:00 15:00 17:00 19:00 21:00

CBF1 opening balance: US$750.0 million. BDF1 opening balance: US$0.3 million. Gray lines: simulated. Black lines:
observed. Source: author’s calculations.

Taking into account the fact that the Colombian LVPS is an RTGS system where no
intraday overdraft is allowed, it is interesting to discover that intraday liquidity paths
simulated for the CBF1 remain positive for any scenario. Thus, under the model and
assumptions proposed here, the CBF would not exhaust its liquidity, and would be
able to fulfill its intraday payments without delays or queuing. The rationale behind
this finding is the existence of the reserve requirement for commercial banks, which
serves as an important source of liquidity for this type of financial institution, as in
Bernal et al (2011).
Meanwhile, because the opening balance of BDF1 is significantly lower than that
for CBF1 (about 0.04% of CBF1’s), BDF1’s simulated paths where its liquidity is
exhausted are representative. This also corresponds with findings by Bernal et al
(2011) regarding the reliance of nonbanking institutions (ie, with no reserve require-
ments) on the virtuous circle of intraday liquidity, along with the presence of sig-
nificantly higher turnover ratios for BDFs when compared with CBFs; for the two
selected financial institutions, the turnover ratio (ie, the ratio of payments and opening
balance) reached 0.7 and 3181.2 for CBF1 and BDF1, respectively.
Since the simulated minute-by-minute balance of received and executed orders
is available, it is possible to estimate a VaR type of measure of intraday liquidity
risk.20 This measure would answer the following question: what are an institution’s
maximum intraday liquidity needs at a defined confidence level? The answer to this

20Estimating other risk measures (eg, expected shortfall) is straightforward under the model pro-
posed here.

Journal of Financial Market Infrastructures Volume 1/Number 1, Fall 2012

Research Paper

TABLE 1 CBF1 and BDF1: intraday liquidity VaR (US$, millions).

Estimating the intraday liquidity risk of financial institutions

Scenarios (99% VaR)
‚ …„ ƒ
Intraday 17:00
(upper bound) 15:30 (lower bound)
‚ …„ ƒ ‚ …„ ƒ ‚ …„ ƒ
Opening Opening Opening
Opening Executed Turnover Net balance Net balance Net balance
Institution balance payments ratio balance utilization (%) balance utilization (%) balance utilization (%)

CBF1 750.0 498.7 0.7 570.6 23.9 606.3 19.2 678.5 9.5
BDF1 0.3 872.8 3181 126.7 46 294 94.1 34 389 34.8 12 782

Source: author’s calculations.

www.risk.net/journal

91
92 C. León

question is displayed in Table 1 on the preceding page for three different intraday
scenarios:21 the maximum intraday liquidity needs within the day, by the end of the
day (ie, 17:00), and at a significant moment within the day (here, for example, 15:30).
A 99% confidence level and 1000 simulations are used for all calculations.
The first scenario (ie, within the day) corresponds to the “upper bound”. Bech and
Soramäki (2002) define the upper bound as the amount of liquidity required to settle
all payments immediately. Under this bound the liquidity need is maximized, as in
an RTGS payment system. This is the most conservative (ie, liquidity demanding)
intraday scenario.
The second scenario (ie, by the end of the day) corresponds to the “lower bound”.
Bech and Soramäki (2002) define the lower bound as the liquidity required by the sys-
tem if all payments are to be settled collectively at the close of the day, as in a deferred
net settlement system. The author’s choice of the “by the end of the day” minute for
the Colombian case corresponds to the time where the settlement of securities and
cash has reached about 95–98% and 85–90%, respectively.
Finally, the choice of 15:30 as a significant moment within the day for the Colom-
bian LVPS corresponds to the middle of both institutions’ executed payments most
active time frame (ie, 15:00–16:00). During this hour the payments executed by
both institutions correspond to 37.5% of their executed payments, the highest of the
intraday, where the accumulated executed payments reach 67.8% and 79.8% of each
institutions’ total, for CBF1 and BDF1, respectively. Furthermore, the 15:30 time is
half an hour before the closing of the access to the monetary (Lombard) liquidity
window of the Central Bank.
However, as previously mentioned, because the Colombian LVPS relies on an
RTGS system where overdrafts are not allowed, all paths below the zero net balance
level (ie, negative net balances) are simply unfeasible. Yet simulating those paths
allows the resilience of a financial institution facing an unexpected and extreme change
in its intraday liquidity patterns to be estimated. The results from the simulation may
help to identify nonresilient institutions, since institutions that are heavily reliant on
incoming payments may be vulnerable to a liquidity stress should their counterparties
decide to delay or stop making payments to it (Ball et al (2011)).
A financial institution displaying net balance paths significantly below zero could
be considered as nonresilient and a potential source of delays and interruptions for
the functioning of the LVPS under extreme but plausible circumstances. The overall
resilience of such an institution would depend, for instance, on its stock of eligible
and unencumbered collateral for accessing central bank liquidity facilities, or for
accessing the monetary market.

21 Please note that these scenarios correspond to the time horizon in a typical VaR model.

Journal of Financial Market Infrastructures Volume 1/Number 1, Fall 2012

 Estimating the intraday liquidity risk of financial institutions 93

The previously presented VaR-type measure of intraday liquidity risk is replicated

for a wider set of CBFs and BDFs.22 Based on the same database and assumptions
(eg, 99% confidence interval, 1000 simulations, three different intraday scenarios)
Table 2 on the next page exhibits the 99% net balance VaR and the percentage of
opening balance that would have been used to surmount such scenario for an average
CBF and BDF.23
The mean VaR by type of institution confirms the findings for CBF1 and BDF1.
The average CBF holds a significant amount of funds at the beginning of the day
(about 49.5% of the average executed payments), which allows it to withstand a 99%
confidence level adverse setting at any moment within the day. Even at the most
severe scenario (ie, the upper bound) the average CBF holds a positive net balance,
and requires a fraction of its opening balance (94.1%). As before, such a significant
amount of funding at the beginning of the day results from reserve requirements for
CBFs.
On the other hand, the average BDF, which is not covered by reserve requirements,
holds a significantly lower opening balance at about 0.04% of the opening balance of
the average CBF, and equivalent to 0.5% of the average executed payments. Because
of this low level of funds at the beginning of the day, the average BDF would be
unable to fulfill its intraday payments at a 99% confidence level adverse setting, not
even at the less adverse scenario (ie, the lower bound).24 The resilience of the average
BDF would depend mainly on its stock of eligible and unencumbered collateral for
accessing central bank’s liquidity facilities or for accessing the monetary market.
Results for the selected individual institutions (ie, CBF1 and BDF1) and for an
average CBF and BDF concur with results obtained by Bernal and Merlano (2005),
Machado et al (2009) and León et al (2011). For instance, based on the comparison
of the upper bound and the available balances of financial institutions, Bernal and
Merlano (2005) found that liquidity shortages existed for BDFs, even at the aggregated
level, whereas CBFs’ required reserve balance held at the central bank was enough
to settle the totality of obligations in the payments system. Likewise, based on a mix
of network theory and simulation of payments, Machado et al (2009) and León et al
(2011) found that BDFs are prone to exhausting their liquidity and queuing payments

22 The LVPS database for the selected date comprised nineteen CBFs and twenty-six BDFs, among
other types of financial institutions. The institutions included in the set used in this exercise were
eleven CBFs and eight BDFs (see Table 2 on the next page); the criterion for inclusion was a
threshold of at least ten payments per hour on average within the day.
23 The results of Table 2 on the next page correspond to nonweighted averages. When using weighted

averages the intraday liquidity requirements increase for both types of financial institutions, but the
analysis remains.
24 A similar conclusion is obtained by Bernal and Merlano (2005) regarding delays due to insufficient

intraday funds by BDFs and other nonbanking firms in the Colombian market.

Research Paper www.risk.net/journal

 94
Journal of Financial Market Infrastructures

C. León
TABLE 2 Average CBF and BDF: intraday liquidity VaR (US$, millions).

Scenarios (average 99% VaR)

‚ …„ ƒ
Intraday 17:00
(upper bound) 15:30 (lower bound)
‚ …„ ƒ ‚ …„ ƒ ‚ …„ ƒ
Average Average Average Opening Opening Opening
Type of opening executed turnover Net balance Net balance Net balance
institution balance payments ratio balance utilization (%) balance utilization (%) balance utilization (%)

CBFs 305.8 618.3 2.0 18.1 94.1 59.2 80.6 92.4 69.8
BDFs 1.3 257.2 199.6 58.9 4671.3 43.3 3462.3 23.8 1945.6
Volume 1/Number 1, Fall 2012

Nonweighted averages for eleven CBFs and eight BDFs. Source: author’s calculations.
 Estimating the intraday liquidity risk of financial institutions 95

when the LVPS network faces an attack (ie, failure to pay by an overly connected
selected institution).
However, as previously mentioned, the approach presented here improves the mea-
surement of intraday liquidity risk by allowing for estimating standard metrics such as
VaR or expected shortfall. This is an important contribution since it provides a known
framework (eg, VaR) for designing high degree of confidence stress scenarios such as
those suggested by Committee on Payment and Settlement Systems and International
Organization of Securities Commissions (2012) for financial market infrastructures.25
The results of the model are important for financial authorities.Authorities in charge
of prudential regulation, supervision and oversight may use this type of intraday
liquidity VaR in order to assess the resilience of financial institutions or financial
market infrastructures when confronting intraday liquidity shocks. This information
is key for authorities since, as emphasized by Kodres (2009), failure or insolvency
are not the only sources of systemic shocks, but mere failure-to-pay or nonpayment
of transactions can gridlock the entire financial system.
Furthermore, as acknowledged by Committee on Payment and Settlement Systems
(2005), a higher-than-optimal degree of systemic risk in key payment and settlement
systems may result from negative externalities, with RTGS-related negative externali-
ties coming in the form of insufficient incentives to consider the full impact on others of
delaying outgoing payments. In this sense, the model’s results are an approximation to

the impact of changing timing mismatches on an institution’s intraday liquidity,

its ability to avoid delaying outgoing payments,

its share of systemic risk in the LVPS.
Additionally, taking into account recent amendments to financial regulation (eg, from
the Committee on Payment and Settlement Systems and International Organization
of Securities Commissions, Basel Committee on Banking Supervision and the FSA),
this model may be a starting point for assessing financial institutions’ liquidity and
intraday liquidity adequacy. Accordingly, being able to contrast financial institutions’
real-time observed intraday liquidity with the model’s resulting intraday liquidity
uncertainty may be valuable input for an overseer trying to identify abnormal
intraday liquidity stances.

25 The principles issued by Committee on Payment and Settlement Systems and International Orga-
nization of Securities Commissions suggest using a high degree of confidence for determining
adequate levels of liquidity for financial infrastructures, where the default of the participant and its
affiliates that would generate the largest aggregate liquidity obligation is the suggested metric for
such stress scenarios. However, it is difficult to assess the degree of confidence of such an event
happening. In this sense, the approach proposed here may provide an intraday liquidity risk metric
that works under well-known parameters, such as a VaR with a high degree of confidence.

Research Paper www.risk.net/journal

96 C. León

5 FINAL REMARKS
As the most recent financial crisis has revealed, nonbanking institutions are as impor-
tant as banking institutions nowadays, and liquidity is as important as solvency, where
financial infrastructures such as the LVPS play a key role for financial stability. This
evolution of financial systems, resulting from the shift to market-based financial sys-
tems and to RTGS of payments, has been protracted but ignored to some extent,
especially by prudential regulation.
As mentioned, prior to the recent financial crisis, regulators did not focus on intraday
liquidity risk and there were no standard monitoring or liquidity management mea-
sures in place (Ball et al (2011)). Regulation is working hard in order to catch up with
risks arising from increasingly important nonbanking institutions, financial infras-
tructures and liquidity. Regarding the latter, it is clear that regulators (for example,
the Committee on Payment and Settlement Systems and International Organization
of Securities Commissions, Basel Committee on Banking Supervision and the FSA)
are updating their regulatory framework in order to enhance liquidity risk manage-
ment for financial institutions and infrastructures, where intraday liquidity is one of
the major concerns and challenges. These efforts parallel the documented trend in
shortening time horizons of liquidity risk and liquidity management, with intraday
liquidity as the new standard for what is considered to be short term.
The proposed model addresses a key question for intraday liquidity risk man-
agement: what are an institution’s maximum intraday liquidity needs at a defined
confidence level? The chosen approach allows the minute-by-minute liquidity of any
financial institution to be modeled, where the main risk factors to be modeled are the
arrival of executed and received payments (ie, their intensity), their synchrony (ie,
their timing or correlation) and their size (ie, their monetary value). As mentioned,
following Ball et al (2011), the identified source of intraday liquidity risk modeled
here is the timing mismatch between incoming and outgoing payments, which may
lead to significant intraday liquidity needs.
Besides answering that key question, the model may be suitable for quantitatively
supporting analysis regarding four main issues:

overseeing participants’ intraday behavior;

assessing their ability to fulfill intraday payments at a certain confidence level;

identifying participants who are nonresilient to changes in timing mismatches
of payments;

estimating intraday liquidity buffers.26

26 As documented and discussed by Ball et al (2011), introducing intraday liquidity buffers should

make institutions more resilient to any potential liquidity stress. However, their introduction also
makes intraday liquidity more costly, and may result in incentives to delay payments.

Journal of Financial Market Infrastructures Volume 1/Number 1, Fall 2012

 Estimating the intraday liquidity risk of financial institutions 97

These four issues, as demonstrated by the most recent financial crisis, are critical for
mitigating liquidity and systemic risk, for financial institutions and financial market
infrastructures.
Finally, as previously stated, the model’s results are an approximation to the main
negative externality arising from a financial institution within an RTGS-based LVPS:
an institution’s insufficient regard for the potential costs or losses that others would
incur in the event of its failure to fulfill its payments in a timely manner. In this
sense the model assesses the impact of varying timing mismatches on an institution’s
intraday liquidity, its ability to avoid delaying outgoing payments and its contribution
to systemic risk.
While the advantages of the model are rather apparent, some drawbacks are worth
mentioning. First, as with any other model, its outcomes should be analyzed and used
with care; they intend to provide a fair explanation of reality based on their assump-
tions, and they are by no means a substitute for sound judgment, or the sole metric to
use to measure intraday liquidity risk. Second, the author considers this model a novel
approximation to a long-ignored problem, but recognizes that its usefulness depends
on the ability of financial authorities to articulate the measurement of intraday liquid-
ity risk with the other stages of risk management (ie, monitoring and mitigating risk),
and to understand the business line of each type of institution. Third, the methodology
is demanding in terms of computational resources.
The author also recognizes several challenges ahead.
The first is to develop an appropriate backtesting method.
The second is to run the model for a long period (eg, a month), which may discard
any particularities and biases in the selection of a specific date and would allow for
a comprehensive characterization of the intraday patterns of financial institutions.
Despite the fact that results concur with other related models and papers that use
longer periods, it is a well-known fact that the daily averaging of reserve requirements
within the two-week reserve maintenance period results in cyclic patterns of opening
balances for credit institutions under local regulation.
The third refers to analyzing the effects of excluding some major liquidity sources
from the estimation of the model’s parameters. The author’s first choice would be
to exclude the systemically most important financial institution, or each institution’s
highest-liquidity-contributing counterparty. This variant would permit the estimation
of the change in intraday payment synchrony and uncertainty due to the absence or
failure to pay by a relevant counterparty.
The fourth challenge consists of some technical enhancements to the algorithm:

implementing a time-dependent Poisson process, where the progressive one-

hour-window execution of the algorithm proposed here is replaced by time-
dependent intensity rates and correlations;

Research Paper www.risk.net/journal

98 C. León

allowing for further intraday liquidity uncertainty from monetary value vary-
ing payments, where the bootstrap procedure is replaced by a time-varying
parametrical assumption of the size of the payments;

capturing and modeling the dependence between received and executed pay-
ments across all participants, and not only the dependence between received
and executed payments for a single institution.27

The author expects to address these challenges in forthcoming extensions of this

paper, especially those corresponding to the algorithm’s technical enhancements.

APPENDIX A. MONTE CARLO SIMULATION OF INTRADAY

PAYMENTS BASED ON BIVARIATE POISSON
PROCESSES AND BOOTSTRAP HISTORICAL
SIMULATION
The model could be described as the joint and time-dependent simulation of a bivariate
Poisson processes for intraday executed and received payments, and their monetary
value. The core of the model is the Monte Carlo simulation of bivariate Poisson random
variables for the intraday arrival of executed and received payments, while simulations
of their monetary value using bootstrap historical simulation is subordinated to their
arrival. Both simulation procedures are addressed below.

A.1 Monte Carlo simulation of bivariate Poisson processes for the

intraday arrival of payments
The bivariate Poisson generation is based on the algorithm proposed by Yahav and
Shmueli (2011).28 Their algorithm is based on the NORTA (NORmal To Anything)
procedure for generating multivariate Poisson (labeled “P”) data with a target corre-
lation structure (˘P ) and arrival rates (
1 ;
2 ; : : : ;
x ), which is based on simulating
data from a multivariate normal (N ) distribution and converting it into an arbitrary
continuous distribution with specific correlation matrix. Letting ˚.x/ be the normal

27 The fourth challenge is particularly demanding. It requires shifting from bivariate to multivariate
Poisson processes, where the dimension of the problem would escalate from independently gener-
ating two joint series of length q for each participant (ie, received and executed payments) to jointly
generating N  2 series of length q for the whole system, where N and q stand for the number of
participants and the number of simulations, respectively. The most appealing feature of such a shift
is its ability to explicitly model institutions’ connectedness (via the dependence between received
and executed payments across institutions), whereas the model proposed here does it implicitly.
28 This section is based on Yahav and Shmueli (2011). Several references were omitted in order to

preserve readability.

Journal of Financial Market Infrastructures Volume 1/Number 1, Fall 2012

 Estimating the intraday liquidity risk of financial institutions 99

cumulative distribution function and letting ˝.x/ be any target cumulative distribu-
tion function, the generalized NORTA procedure is as follows.

(a) Generate a q-dimensional normal vector XN with mean
D 0, variance  D 1,

and a correlation matrix ˘N .29

(b) For each value XN , i 2 1; 2; : : : ; q, calculate the normal cumulative distribution

function (CDF) (˚.x/):
Z x
1 2
˚.x/ D p e
u =2 du (A.1)

1 2 2

(c) For each ˚.x/, calculate the target cumulative distribution function (˝.x/):

Xyi D ˝
1 .˚.XNi // (A.2)

(d) The resulting vector XY is then a q-dimensional vector, distributed according

to the target cumulative distribution function (˝.x/), with correlation ˘N .

Despite the simplicity of the NORTA procedure, generating bivariate or multi-

variate probability distributions when the target distribution is a discrete probability
distribution (ie, a random variable can assume only a certain clearly separated values)
is more complicated. Such is the case for the Poisson distribution.
The Poisson distribution describes the number of times an event occurs during a
specified time, space, area or volume interval. It is a discrete probability distribution
since it is formed by counting (Lind et al (2006)), and is based on two assumptions:

(1) the probability of an event occurring is proportional to the length of the interval;

(2) the probability of an event occurring in an interval is independent of its occur-

rence in other intervals.

The Poisson distribution is defined by a single parameter (
) that expresses the
probability of a number of events occurring in a fixed interval of time (ie, the mean
number of occurrences in a particular interval), where
is commonly referred to
as the intensity of the process. The Poisson cumulative distribution function ( .x/)
corresponds to:
Xx
e


i
 .x/ D (A.3)
iŠ
iD0

29 Generating normal multivariate random numbers (ie, with correlation matrix ˘ ) is straightfor-
N
ward by means of the Cholesky decomposition. The interested reader is referred to Cuthbertson and
Nitzsche (2004) and León and Reveiz (2011).

Research Paper www.risk.net/journal

100 C. León

With sufficiently large
, the normal distribution is a fair approximation to the Poisson
distribution, where
is its mean and variance. Conveniently, as the Poisson distribu-
tion converges asymptotically to a normal distribution, attaining multivariate Poisson
distributed random variables with correlation ˘P is straightforward since the depend-
ence between both distributions would also converge asymptotically (˘N  ˘P ).
However, as pointed out byYahav and Shmueli (2011), when the normal distribution
is not a fair approximation to the Poisson distribution (ie, when
is small), the
convergence in correlation ceases to exist (˘N ¤ ˘P ). The main consequence of
such lack of convergence in distribution is that the feasible correlation between two
random Poisson variables is no longer in the traditional range [1; 1], but in a narrower
range [ P > 1;  P 6 1].
Furthermore, the smaller the intensity of any of the Poisson processes, the nar-
rower the correlation range, and the more difficult it is to obtain a target correlation.
As demonstrated by Shin and Pasupathy (2010), as any intensity rates
1 and
2
approximate to zero (
1 ;
2 ! 0), the minimum feasible correlation approximates to
zero ( P ! 0). As exhibited in Figure 2 on page 87 and Figure 3 on page 89, this is
the case at the beginning and the end of the day, when payments are rather scarce.30
Therefore, in order to attain bivariate Poisson random variables with intensity rates

1 and
2 , this paper follows the functional approximation developed by Yahav and
Shmueli (2011) to estimate the relationship between the desired Poisson correlation
(P ) and the actual (normal) correlation (N ). The procedure is as follows.
(a) Let U be a vector of uniform randomly distributed variables and compute the
correlation mapping [ P > 1;  P 6 1], where:
)
 P D corr.

11
.U /; 

1
2
.1  U //
(A.4)
 P D corr.

11
.U /; 
1

2
.U //
(b) Compute the coefficients of the exponential function estimated by Yahav and
Shmueli (2011):31 9
P  P >
aD >
>
P  P =

> (A.5)
P C a > >
b D log ;
a

30 Since the minimum feasible correlation approximates to zero ( P ! 0) when intensity rates
approximate to zero (
1 ;
2 ! 0), the implemented algorithm includes an instruction to round any
intensity below 0.0167 (ie, one arrival per hour) to zero, and to simulate the two Poisson processes
as noncorrelated ( P D 0).
31 Based on simulations, Yahav and Shmueli (2011) find that the relationship between the desired

correlation (P ) and the actual correlation (N ) is best approximated by an exponential function
P D a  ebN  a

Journal of Financial Market Infrastructures Volume 1/Number 1, Fall 2012

 Estimating the intraday liquidity risk of financial institutions 101

(c) Based on the previously computed coefficients, compute the correlation

required to generate bivariate normal random variables (N ) that approximate
the target correlation (P ):


1 P C a
N D log (A.6)
b a

(d) Generate bivariate normal random variables with 1 D 2 D 0, 1 D 2 D 1

and correlation N .

(e) Based on the bivariate normal random variables (XN  N.0; N /), follow
the NORTA procedure with the Poisson distribution as the target CDF, with
intensity rates
1 and
2 .

XPi D 
1 .˚.XNi // (A.7)

The resulting vector XP is then a two-dimensional vector, distributed according to

the Poisson cumulative distribution function ( .x/), with intensity rates
1 and
2 ,
and correlation P . For the problem at hand, this procedure yields two vectors:
(1) a minute-by-minute vector of executed payments;

(2) a minute-by-minute vector of received payments, where several executed or

received payments per minute may occur; that is, according to the intensity
rates there may be more than one arrival per minute.
Both vectors contain the minute-by-minute arrival of payments (their occurrence, not
their value), where both vectors approximate the target correlation corresponding to
the estimated synchrony between received and executed payments.32
Due to the time-variant characteristics of the LVPS intraday payments, which dis-
play time frames with distinctive intensity rates and correlations, the aforementioned
procedure is not applied to the whole day; instead, it is applied to one-hour win-
dows, which allows for capturing the intraday seasonality of executed and received
payments.

32 In order to attain comparability between simulations, the algorithm is instructed to always obtain
the same monetary value of received (respectively, executed) payments as in the observed data. Such
a restriction excludes the effect of payments size, and permits focusing on the analysis of changes in
payments’ synchrony. This does not mean that the same monetary values are used from simulation
to simulation, but that the same nonparametrical distribution (ie, the observed payments) was used
to generate different monetary values from simulation to simulation. Despite the convenience of
the chosen approach to simulating the value of the payments, it may be useful to allow for further
intraday liquidity uncertainty arising from value varying payments. This additional source of risk
may be modeled by making a time-varying parametrical assumption of the size of the payments.

Research Paper www.risk.net/journal

102 C. León

An alternative to this progressive one-hour-window execution of the algorithm is to

design a time-dependent Poisson process, where intensity rates and correlations are
functions of time.33 Despite being a more elegant alternative, two main issues stand
against it. First, such a procedure may further complicate the algorithm, which is
already burdensome and time-consuming. Second, the design of the time-dependent
parameters is by no means trivial, and may further introduce model risk into the
approach.

A.2 Bootstrapped historical simulation of received and executed

payment value
The previous subsection presented the method for simulating the occurrence or arrival
of both received and executed payments, but not their monetary value. The author’s
choice for simulating the monetary value of payments is based on bootstrapped his-
torical simulation.
Based on Dowd (2005), the author designs a bootstrap procedure (resampling with
replacement)34 to simulate the monetary value of received and executed payments
once they occur as a result of the arrival simulation method previously described. Each
time a received (respectively, executed) payment arrives the model draws a sample
from the received (respectively, executed) historical records, and takes its monetary
value as the received (respectively, executed) payment’s value for that occurrence.
Compared with other methods available for simulating the monetary value of the
payments, the bootstrap avoids unreliable assumptions such as normality of the data
set or the existence of large samples (Dowd (2005)). Regarding the normality of
payments, it is clear that they do not converge to a Gaussian distribution (Figure A.1
on the facing page), while it is also common to find intraday periods characterized by
small samples to work with (Figure 2 on page 87 and Figure 3 on page 89).35

33 The anonymous referee suggested this clever and elegant alternative. Despite being somewhat
impractical for the moment (ie, the current algorithm is already time-consuming and computationally
demanding), this enhancement is being considered for a newer and more efficient version of the
algorithm.
34 The bootstrap procedure consists of sampling from a data set of size n. Each sampling requires the

generation of uniform random numbers between 1 and n to randomly draw observations from the
data set; drawn observations are returned to the data set (ie, observations are replaced into the data
set). Since Monte Carlo may be broadly defined as a method that provides approximate solutions by
performing statistical sampling experiments on a computer (Fishman (1995)) or a random number
generator that is useful for forecasting, estimation, and risk analysis (Mun (2006)), the bootstrap
procedure may be considered as involving a Monte Carlo procedure within. Therefore, the author
regards the whole method presented here as an implementation of a Monte Carlo simulation.
35 Small samples of payment orders are a significant problem for fairly inactive financial institutions,

which tend to make payments infrequently.

Journal of Financial Market Infrastructures Volume 1/Number 1, Fall 2012

 Estimating the intraday liquidity risk of financial institutions 103

FIGURE A.1 Distribution of LVPS payments (selected day).

5000
4500
4000
3500

Frequency
3000
2500
2000
1500
1000
500
0
0 1 2 3 4 5 6 7 8 9
US$ × 107

Source: author’s calculations, data from the LVPS.

APPENDIX B
Achieving a fair approximation of the correlation of the simulated bivariate Poisson
series to the target correlation is the main goal of the bivariate Poisson process and
the model. Figure B.1 on the next page demonstrates that the mean of the correlation
of the simulated bivariate Poisson series replicates the target correlation, while the
simulated correlation of each series disperses around the target correlation.
It is worth mentioning that, since the minimum feasible correlation approximates
to zero ( P ! 0) when intensity rates approximate to zero (
1 ;
2 ! 0), the imple-
mented algorithm includes an instruction to round any intensity below 0.0167 (ie, one
arrival per hour) to zero, and to simulate the two Poisson processes as noncorrelated
( P D 0). The author regards this as a safe practice because of the theoretical sup-
port behind such an assumption (Yahav and Shmueli (2011) and Shin and Pasupathy
(2010)), and because, during low-intensity periods (eg, 07:00–09:00, 19:00–20:00),
the frequency of the payments is nonsignificant relative to the rest of the intervals.

APPENDIX C
In order to assess the soundness of the selected length of the windows that divide the
intraday time frame, Table C.1 on the next page displays the VaR for each institution
(ie, CBF1 and BDF1) and type of institution (ie, CBFs and BFDs), where the metric
corresponds to the 99% confidence VaR as a percentage of total executed payments.
It is rather apparent that the cross-section analysis is the same for any of the
three assumptions considered, with the sole exception of the 15:30 scenario when

Research Paper www.risk.net/journal

104 C. León

FIGURE B.1 Algorithm’s correlation fit.

(a) (b)
Observed Observed
Mean of simulations Mean of simulations
Correlations Correlations

1.0 1.0

Correlation
Correlation

0.5 0.5

0 0

−0.5 −0.5
7–8 11–12 15–16 19–20 7–8 11–12 15–16 19–20

(a) CBF1. (b) BDF1. Source: author’s calculations.

TABLE C.1 CBF1, BDF1, CBFs and BDFs: intraday liquidity risk (99% VaR, as percentage
of total executed payments).

One-hour Thirty- Fifteen-

window minute minute
(default) windows windows Average

Intraday (upper bound)

CBF1 35.6 35.4 42.4 37.8
BDF1 15.3 21.0 20.5 18.9
CBFs 44.6 41.3 48.3 44.8
BDFs 33.1 33.0 20.5 28.9

15:30
CBF1 28.3 22.1 14.1 21.5
BDF1 12.0 17.8 19.1 16.3
CBFs 36.2 30.1 35.9 34.0
BDFs 21.8 20.7 20.0 20.8

17:00 (lower bound)

CBF1 15.5 15.0 12.1 14.2
BDF1 4.5 7.0 8.2 6.5
CBFs 29.2 20.7 28.6 26.1
BDFs 10.4 8.0 7.4 8.6

Nonweighted averages for eleven CBFs and eight BDFs. Source: author’s calculations.

Journal of Financial Market Infrastructures Volume 1/Number 1, Fall 2012

 Estimating the intraday liquidity risk of financial institutions 105

comparing CBF1 and BDF1. However, since the monetary value of the payments
made by BDF1 is significantly higher than the value of the payments made by CBF1
(ie, 1.8 times), and the opening balance is significantly higher for CBF1 (ie, 2700
times the BDF1s), the analysis and conclusions of the paper are valid regardless of
the chosen assumption.
Nevertheless, any implementation and analysis resulting from the proposed
approach should be aware of the trade-off between a more precise characterization of
the intraday process by a more granular breakdown of the time frame, and the avail-
ability of abundant observed arrivals to estimate the parameters of the simulation for
a meaningful and diverse sample of institutions.

REFERENCES
Acharya, V. V., Philippon, T., Richardson, M., and Roubini, N. (2009). Financial crisis of
2007–2009: causes and remedies. In Restoring Financial Stability, Acharya, V. V., and
Richardson, M. (eds). Wiley.
Ackermann, J. (2008). The subprime crisis and its consequences. Journal of Financial
Stability 4(4), 329–337.
Ball, A., Denbee, E., Manning, M., and Wetherilt, A. (2011). Intraday liquidity: risk and
regulation. Financial Stability Paper 11, Bank of England (June).
Banco de la República (2011). Reporte de Sistemas de Pago. Report (June).
Basel Committee on Banking Supervision (2000). Sound practices for managing liquidity in
banking organizations. Bank for International Settlements (February). URL: www.bis.org/
publ/bcbs69.pdf.
Basel Committee on Banking Supervision (2008). Principles for sound liquidity risk man-
agement and supervision (draft for consultation). Bank for International Settlements
(June). URL: www.bis.org/publ/bcbs138.pdf.
Basel Committee on Banking Supervision (2010). Basel III: international framework for liq-
uidity risk measurement, standards and monitoring. Bank for International Settlements,
December. URL: www.bis.org/publ/bcbs188.pdf.
Bech, M. (2008). Intraday liquidity management: a tale of games banks play. FRBNY Eco-
nomic Policy Review, Federal Reserve Bank of New York (September), 7–23.
Bech, M., and Soramäki, K. (2002). Liquidity, gridlocks, and bank failures in large value
payment systems. In E-money and Payment Systems Review, Pringle, R., and Robinson,
M. (eds). Central Banking Publications, London.
Bernal, J. (2009). Money markets, monetary policy and payment systems: policy issues for
central banks in times of turbulence. Journal of Payments Strategy and Systems 3(4),
301–310.
Bernal, J., and Merlano, C. (2005). Payment system policies and intraday liquidity in a real
time gross settlement system: the Colombian case. Payments and Securities Clearance
and Settlement Systems Research Series 3, Centro de Estudios Monetarios Latinoamer-
icanos and World Bank.
Bernal, J., Cepeda, F., and Ortega, F. (2011). Cuantificación de la contribución de las
fuentes de liquidez en el sistema de pagos de alto valor en Colombia: una aproximación
preliminar. Borradores de Economía 683, Banco de la República.

Research Paper www.risk.net/journal