DEGREE PROJECT IN MATHEMATICS,

SECOND CYCLE, 30 CREDITS

STOCKHOLM, SWEDEN 2020

Risk Factors Impact on the

P&L
KHAWLA GOUGAS

KTH ROYAL INSTITUTE OF TECHNOLOGY

SCHOOL OF ENGINEERING SCIENCES
Risk factors impact on the P&L

KHAWLA GOUGAS

Degree Projects in Financial Mathematics (30 ECTS credits)

Degree Programme in Engineering Physics
KTH Royal Institute of Technology year 2020
Supervisor at BNP Paribas CIB Paris: Frederic Prieur
Supervisor at KTH: Boualem Djehiche
Examiner at KTH: Boualem Djehiche
TRITA-SCI-GRU 2020:335
MAT-E 2020:084

Royal Institute of Technology

School of Engineering Sciences
KTH SCI
SE-100 44 Stockholm, Sweden
URL: www.kth.se/sci
 Summary
Profit and Loss (P&L) explain analysis is an income statement produced by Product Control Team for
traders to control the daily fluctuation in the value of a portfolio of trades to the root causes of the changes.
This daily income provides users with a coherent breakdown of the drivers of P&L movements between two
points in time with reference to a selected number of easily understandable pricing factors.
P&L Attribution (also called P&L explain) can be calculated in two ways, either the risk based method or
step re-evaluation method.
This paper aims at understanding both methodologies from a theoretical point of view and shows the dif-
ferences of both calculations methods and how they are interdependent in the daily work of a trader in the
sense that both methods give a rational to the P&L from different perspectives. The risk based method
involves the calculation of the trades sensitivities (also known as the Greeks) and then using them to predict
the expected change in the P&L from one period to the next by using the actual market changes in the
factors driving the transaction price over the same period and the transaction’s sensitivity to those factors.
Whereas the re-evaluation method is calculated by aggregating the impact of different valuation scenarios
and not on fixed sensitivities.

Keywords: Profit and Loss (P&L), risk factors, Greeks.

2
 Riskfaktorers inverkan på balansräkning

Sammanfattning
Analys och rapportering av balansräkning ger användarna en sammanhängande uppdelning av drivkrafterna
för P&L rörelser mellan två punkter i tid med hänvisning till ett urval av lättförståeliga prisfaktorer. P&L
attribution kan beräknas på två sätt, känslighet och scenariobaserad metoder.
Detta arbete syftar till att förstå båda metoderna ur ett teoretiskt perspektiv och visar skillnaderna och
hur de beror på varandra. Känslighetsmetoden innebär beräkning av en handels känslighet (även känd som
beräkning av grekerna) och sedan använda dem för att förutsäga den förväntade förändringen i P&L från en
period till nästa genom att använda de faktiska marknadsförändringarna i de faktorer som driver transak-
tionspriset under samma period och transaktionens känslighet för dessa faktorer. Omvärderingsmetoden
beräknas genom att olika värderingsscenariers inverkan sammanställs och inte på fasta känsligheter.

3
 Acknowledgements

I would like to thank my manager Frederic Prieur for his valuable guidance and insights leading to the
writing of this paper.

My sincere thanks also goes to the quantitative analysis team and trading application support team for
their patience and their help to respond to all my questions and help me throughout this internship to deliver
the best results.

I would also like to express my sincere gratitude to the developers who helped me set up my working
space and improve my coding skills.
A special thanks also to my school supervisor Boualem Djehiche, who provided me with the necessary sup-
port and feedback to improve my paper and work.

Finally I would like to thank all BNP Paribas teams who collaborated through data acquisition and
support during this master thesis for their valuable help.

Paris, October 2020

khawla Gougas

4
Vocabulary:

P&L The profit and loss (P&L) statement is a financial statement that summarizes the
revenues, costs, and expenses incurred during a specified period, usually a fiscal quarter
or year.

P&L explain an income statement produced by Product Control Team for traders to control the
daily fluctuation in the value of a portfolio of trades to the root causes of the changes.

SOD Start of day (business day)

COB Close of business for the trader

IR Interest rate

FX Foreign exchange rate

Greeks Trades sensitivities: "Greeks" is a term used in the options market to describe the
different dimensions of risk involved in taking an options position.

risk-based method a method of calculation of p&L explain based on trades sensitivities

Intraday result Intraday means "within the day." In the financial world, the term is shorthand
used to describe securities that trade on the markets during regular business hours.
These securities include stocks and exchange-traded funds (ETFs).

Step re-evaluation method It is a method of calculation p&L explain based on

Taylor Expansion The Taylor expansion is the standard technique used to obtain a linear
or a quadratic approximation of a function of one variable.

Mark to market Mark to market (MTM) is a method of measuring the fair value of accounts
that can fluctuate over time, such as assets and liabilities.

CVA Credit Value Adjustment (CVA) is new risk measure that offers an opportunity for
banks to move beyond the system control of limits and to price
dynamically counterparty credit risk of new trades.

FSA Financial services authority

5
Contents

1 Introduction 7

2 Methodology 10
2.1 P&L explain . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
2.2 Types of P&L: . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
2.3 Types risk factors: . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
2.4 Other risk factors: . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
2.5 Risk generation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
2.5.1 First Order risks calculation . . . . . . . . . . . . . . . . . 16
2.5.2 IR Delta . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
2.5.3 IR Spread . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
2.5.4 Inflation Delta . . . . . . . . . . . . . . . . . . . . . . . . . . 18
2.5.5 Vega . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
2.6 Second Order risks calculation . . . . . . . . . . . . . . . . . . . . 18
2.6.1 IR Gamma . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
2.6.2 IR Volga . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
2.7 Calculation methodologies . . . . . . . . . . . . . . . . . . . . . . . 19

3 Step re-evaluation P&L 19

3.1 Path dependent Step re-evaluation P&L . . . . . . . . . . . . . . . 19
3.2 Path independent Step re-evaluation P&L . . . . . . . . . . . . . . 21
3.2.1 Single factor step re-evaluation P&L . . . . . . . . . . . . 21
3.2.2 Cross risk factors step re-evaluation P&L . . . . . . . . . 22

4 Risk based P&L calculation 23

5 Results 24
5.1 Step revaluation Limitations . . . . . . . . . . . . . . . . . . . . . . 24
5.2 Comparison of the P&L explain calculation methods . . . . . . 28
5.2.1 Empirical result . . . . . . . . . . . . . . . . . . . . . . . . . . 29
5.2.2 Comparison conclusion . . . . . . . . . . . . . . . . . . . . . . 30

6 Conclusion & Extension 32

7 References 34

6
1 Introduction
Financial risk management evolved throughout the past decades. The finan-
cial crisis in 2008 proved that insufficient management of financial risk can cause
negative waves throughout global markets. Thus after the crisis, a lot of regula-
tions were imposed in the capital markets to have a conform risk measurement
reference across banks and financial institutions, these references would help
regulation systems such as the CEB (Central European Bank) to track the risk
management and raise potential anomalies. In the context of industrialisation
of risk inside the banks, P&L explain methodology provides a risk management
tool across Global Markets business lines to produce comparable figures and
improve risk measurements to help manage financial risk.

Financial crises contributed to the evolution of risk management practices,

and regulatory actions. In the 1970s, research lay the intellectual foundations
for the risk management practices that were systematically implemented in the
1980s as bond trading revolutionized Wall Street. Quants developed dynamic
hedging, Value-at-Risk, and credit risk models based on the insights of financial
economics.
In parallel, the Basel I framework which is the round of deliberations by central
bankers from around the world in 1988, created a level playing field among
banks across countries. Following the 1987 stock market crash, the near failure
of Salomon Brothers, and the failure of Drexel Burnham Lambert, in 1996 the
Basel Committee on Banking Supervision published the Market Risk Amend-
ment to the Basel I Capital Accord; the amendment went into effect in 1998.
It led to a migration of bank risk management practices toward market risk
regulations. The framework was further developed in the Basel II Accord. In-
deed, the failure to measure and manage risk adequately can be viewed as a key
contributor to the 2008 global financial crisis. Subsequent innovations in risk
management practices have been dominated by regulatory innovations, includ-
ing capital and liquidity stress testing, macroprudential surcharges, resolution
regimes, and countercyclical capital requirements.
Thus one of the most important risk regulations was imposing an explication

7
of the daily P&l of the financial institutions.
The Profit and Loss (P&L) explained aims at giving a rational to the profit
and loss made between the value of the portfolio on the close of business (COB)
and the one of the start of day (STO) using these risk factors and mainly and
market risk which we can quantify. A P&L explained analysis attributes the
daily change in the value – ie the mark-to-market (MTM) – of a portfolio of
deals to its root causes. Risk managers use this knowledge of the source of
trading profits to act more effectively. For example, they will investigate if
they discover that options desk’s profits are primarily due to commodity prices
changes instead of volatility changes. Traders use this report as a diagnostic
tool to help them reconcile at end-of-day if their hand-calculated P&L estimate
doesn’t match the value produced by their trading system.

The risk computation have to be updated on a regular basis to reflect changes

in the environment in which we operate as well as in our underlying models hav-
ing an impact on the risk. This data is then used to calculate the P&L explain
and manage risk. The users range from traders to global risk managers or P&L
control teams.

An important thing to mention about this income statement "P&L explana-

tion" is that despite trying to explain fully the P&L it would always envolve
an unexplained factor that should theoreticly tend to zero but it it’s not null it
doesn’t mean that our report is incorrect. In fact, the P&L number is always
considered correct in a P&L explained report, although it certainly makes risk
managers uncomfortable if the value in the ‘unexplained’ column on the report
is not zero. Therefore, when you do find the cause of the unexplained P&L,
which is usually done manually after the fact, the total P&L number never
changes due to any reallocation, even a reallocation from ‘unexplained’ to one
of the explanatory columns.

Besides the explanatory benefit of the P&L explain, it can also catch errors
on the current trades since it puts the trade’s P&L in context with other trades
and with the same trade done the previous day. Various mistakes in booking
will be detected by putting a trade in context with its peers. For example, if a

8
new trade was erroneously entered with 10 or 100 times more notional, then it
will have an abnormally large P&L compared to similar trades. In context with
previous day: suppose a piece of market data was incorrectly used in the P&L
calculation today. Then the P&L between today and yesterday would show
a much higher or lower jump than expected, alerting the reader to a possible
problem.

The importance of P&L fluctuation explanation in the financial markets mo-

tivated me to work on this subject and try to understand the P&L explain
theory and calculation methods.
This project took place at BNP Paribas Arbitrage which is a subsidiary of the
French international banking group specialised in the management of the trad-
ing activity on derivative products. BNP Paribas is the largest French banking
group and the largest bank in the Eurozone. It became one of the five largest
banks in the world following the 2008 financial crisis.

9
2 Methodology
To achieve the goal mentioned in the introduction, I started gathering different
documentation sources to have a theoretical background vis-à-vis the P&L ex-
plain, then I started training data on both models to compute the P&L explain
to finally conclude the differences and also interdependence of both method.

Figure 1: Methodology of work

2.1 P&L explain

The P&L explain aims at justifying the P&L from a set of observable market

10
parameters. Being able to find a rational to the P&L from the risk figures is
a way to validate that the risk positions are correct. Risk, Actual P&L and
P&L explained are interdependent legs. Risk is the starting point, both from
a timing perspective as well as in causality and deserves primary attention.
Risk production involves the generation and reporting of the Greeks which rep-
resent the sensitivities that the value of a portfolio of positions in risky deriva-
tives has to change in the underlying variables and the parameters on which
the derivative value is dependent. These underlying variables and parameters
are the risk factors for that portfolio.
All risk factors are not created equal even for any one given portfolio of posi-
tions. Some risk factors disproportionately affect the change in the value of the
portfolio depending on the type of derivative instruments in the portfolio and
the current state and value of all variables.

Literature on the P&L explain is very limited since the concept is very new
in the banking system and recently imposed by the Financial authority to have
a detailed view on the profit and loss to track that the P&L adjustments are
regulated. So I constructed the theoretical part from a set of references.
First of all since the P&L discussed in this paper is focused on the interest rate
products, I used the model theory used in [1] and [2] enriched with Piterbarg
papers [9] and [10] that relate the models to risk management and discusses risk
factors more in details. For the P&L explain risk factors, Acuity and derivatives
LLC paper [7] discuss different risk effects that constitute the P&L attribution
from quantified to non quantified ones (as seen in Figure 3).
Finally the other papers provide a broader vision on the P&L risk management
for different products from equity to fixed income ones.

2.2 Types of P&L:

• Official P&L: It is the P&L reported directly by the bank. It is computed

using one of the following resources:
Central close markets: Data obtained directly from exchanges when
available.
Multi Cut Off (MCO): It uses either market close data or snapshot of
live data time consistent with market close.

11
 • P&L Predict: The P&L as it can be explained at Close Of Business
(COB) by traders in various regions. (It is approximated since the official
P&L may rely on some market data published later by another region).
• Official P&L Explain: It is the P&L as it can be explained when all
the market data is available. It is generally computed on T + 1.

2.3 Types risk factors:

• Market Effect: This is the core of the P&L explanation process. It aims
at explaining the impact on the P&L of the changes in the market param-
eters happening during the day. From the various P&L explain steps, the
market effect is the only one which can be explained from the market risk
sensitivities such as volatility, movement of spot, movement of interest rate.

• Environment Effect: Environment effects refers to all the non-trading

related parameters (excluded from the VaR) which are updated either
overnight or at COB and have a P&L impact. This includes updates in
static data, changes in the CSA agreements (collateral agreements), model
changes, analytics upgrades.

• Time Effect: From a high level point of view, time effect can be assimi-
lated to the cost of running a trading book. This document takes a broad
definition of the time effect which includes all the P&L effects associated to
the switch from COBt−1 to Start of Day SODt . Depending on the position
this could generate a positive P&L (in this case the book is often said to
have a “positive carry”) or a negative P&L (“negative carry”).

• Non-Trading P&L: Non-trading P&L actually refers to the P&L ele-

ments which do not make part of the daily traders’ activity but still have
an overall impact on the P&L and potentially the cash balance. As such,

12
 and in order to match the official P&L a P&L explain will have to receive a
feed of those non-trading P&L effects which can typically be fees, brokers
fees etc. By their nature, those effects cannot usually be computed from
front office system and hence must be fed from official P&L systems.

• Liquidity Risk: Liquidity risk is normally characterised in two forms. As-

set liquidity risk represents the risk that a transaction cannot be executed
at market prices, perhaps due to the size of the position and/or relative
illiquidity of the underlying. Funding liquidity risk refers to the inability to
fund contractual payments or collateral requirements, potentially forcing
an early liquidation of assets and crystallisation of losses.

Note That this paper will only cover the market effect that can be quan-
tified through the market risk sensitivities, most of other risk factors are
quantified for specific financial products once the trade happens and gen-
erates fixed numbers that middle officers take into account in their P&L
explain analysis. Whereas time effect, needs a lot of assumptions on the
opening market structure to conclude the close of business one such as if
forwards are realised or not at the end of the day, if not we are dealing
with a carry market that needs more analysis and doesn’t enter the scope
of this paper in terms of quantification.

13
 Figure 2: P&L explained effects on BNP Data january 2020

This graph based on BNP Paribas CIB data (time series of P&L explain
effects) shows the contribution of different risk factors to the explanation of the
P&L during january month 2020. We can see that Time effect and Market effect
contribute the most since they’re related to the every trades price fluctuation.
2.4 Other risk factors:

14
 Figure 3: Risk, Actual P&L and P&L explained [7]

This figure shows the risk factors that contribute in the P&L explanation
(P&L attribution), and that the aim of both methodologies is to have a num-
ber that converge to the actual or official P&L that is calculated differently
based on close of business P&L of different business lines.

There are events (other than risk factors market events) that have Mark to
Market impact; P&L due to these may be attributed by revaluing affected po-
sitions to obtain the isolated P&L impact of the event.

15
 • CVA P&L explain
P&L due to the exposure changers to a counter party’s changing default
risk.

• New trade P&L explain

P&L due to new trades done that day. New Trade P&L is a critically im-
portant metric for P&L control purposes yet the FSA (Financial Services
authority)found that more than half of firms it surveyed cannot distinguish
between new trade P&L and P&L due to cancels/amends.

• Trade Life cycle Events P&L explain

P&L due to trade lifecycle events e.g. amendments, novations(the act of
substituting a valid existing contract with a replacement contract), un-
winds(close a position) etc. And P&L due to underlying lifecycle events
e.g. option exercises, rate resets, credit events etc.

• Current day expected Cash

Cash flows payable on the valuation day drop off the valuation of the trade,
however remain a part of the MTM (mark to market) of the trade and need
to be added back.

• fees, commissions and trader credits

P&L Costs due to fees, sales commissions and trader credits earned from
the trade.

2.5 Risk generation

2.5.1 First Order risks calculation

2.5.2 IR Delta

The sensitivity of the valuation function to a change in the outright level of the
interest rate for a given currency rCCY which represents the interest rate of a

16
currency CCY.
The value of IR.Delta (interest rate Delta) is as follows:

∂V (Pmkt (tSOD ))
IR.Delta(tSOD , CCY ) = .
∂rCCY
with V (Pmkt (tSOD ) the valuation funtion of our portfolio at the Start of Day
(SOD) that depends on the different market parameters (time t, interest rate
r, stock price if there is an equity s, dividends curve, repo curve...) other pa-
rameters might be in included for complex contracts and exotic options.
If our function is differentiable we would apply the derivative otherwise we
would use finite differentiation method. The continuity of the price of our port-
folio depends on the type of options, stocks, indexes that it includes and the
complexity of our pricing function. Here are the practical steps to compute this
sensitivity:

• Compute an initial PV (present value) for the considered trade or portfolio

of trades.
• Perturb the first tenor of the yield curve and rebuild the pricing market
keeping other parameters constant.
• Reprice the considered trade and calculate the difference of the PVs that
represents the sensitivity to the first tenor repeat the same for all tenors.
2.5.3 IR Spread

For a given currency, IR basis spread represents the sensitivity of the valuation
function V to a change in the spread between the curve used to compute the
IR Delta (called the yield curve) given basis curve (for example OIS/IBOR ba-
sis: the London Interbank Offered Rate (LIBOR) and the Overnight Indexed
Swap (OIS) rate.) The sensitivity of the valuation function to a change in the
spread between the curve used to compute IR Delta and a given basis curve is
as follows (Inflation delta):

∂V (Pmkt (tSOD ))
IR.Basis.Spread(tSOD , CCY, Basis) = .
∂SpdBasis
CCY

17
2.5.4 Inflation Delta

The sensitivity of the valuation function to a change in the level of an inflation

index I.

∂V (Pmkt (tSOD ))
IR.Delta(tSOD , I) = .
∂I
2.5.5 Vega

The sensitivity of the valuation function to a change in the ATM volatility of

a given volatility market σ.

∂V (Pmkt (tSOD ))
Ind.V ega(tSOD , σ) = .
∂σ
The sensitivity of the valuation function to a change in the forward spread
level S0 (ZWC).

∂V (Pmkt (tSOD ))
ZW C.Spd(tSOD , S0 ) = .
∂S0

2.6 Second Order risks calculation

2.6.1 IR Gamma

The change in the IR Delta for a change in the absolut level of interest rate for
a currency.
∂ 2 V (Pmkt (tSOD ))
IR.Gamma(tSOD , CCY ) = .
∂ 2 rCCY
2.6.2 IR Volga

Volatility gamma represents the change in the IR Vega risk for a change in the
ATM volatility level.
∂ 2 V (Pmkt (tSOD ))
IR.V olga(tSOD , σ) = .
∂ 2σ
• Compute an IR Vega Risk on the unperturbed market.
• Apply a parallel shift to the ATM volatility.
• Compute an IR Vega on the shifted market.

18
 • Compute the difference between initial IR Vega and the one post pertur-
bation.

2.7 Calculation methodologies

• Risk based method: Computes P&L by multiplying risk of a portfolio

by the change in the underlying market moves.
• Step re-evaluation method: It is based on a sequence of PVs (Present
Value) providing a graduate path between two markets which allows to
allocate the P&L associated to the various risk factors to the change of a
set of parameters. In other words, in each step between two markets we
are going to perturb one risk parameter such as the spot and calculate our
first P&L step, then we are going to perturb the volatility and so on. This
would give us the effect of each risk parameter on our P&L calculation.

3 Step re-evaluation P&L

We assume at any point t time, the value V (Pmkt(t) ) of our portfolio of trades
is a function of a set of valuation parameters, These parameters may include
future prices, volatility for Market observable data or fixing Leg or payment
one for pricing parameters and so on..
 
P1
Pmkt(t) =  ...  ,
Pn
then we will apply a set of perturbations delta
 
δ1
δ =  ...  ,
δn
in order to obtain the closing parameters Pmkt(tC OB) .

3.1 Path dependent Step re-evaluation P&L

19
 The principle of the path dependent step re-evaluation is to sequentially
shift the market parameters from their initial value (SOD market: Start of Day
market) to their final value (COB market: close of business).
That’s why we need to define delta
 
δ1
 ... 
 
δ 
 
δSRI =  i  ,
0
.
 .. 
0
which corresponds to the i-th perturbation of the market and in this case, the
P &LSRi associated to this step is

SRi = V (Pmkt(t) + δSRi ) − V (Pmkt(t) − δSRi ),

we then have
N
X
SRi = V (Pmkt(tCOB ) ) − V (Pmkt(tSOD ) ).
i=1

The advantage of this methodology is that it is able to allocate the realized

P&L to the change in each market factor. It is also reasonably cheap to com-
pute as for a step re-evaluation P&L of N steps, there are N + 1 valuations to
perform.
On the downside the P&L allocated to each of those steps is actually depen-
dent on the ordering on the vector,thus the notion of path dependent step
re-evaluation P&L.

Let’s take the example of an at-the-money (ATM) option being pushed out
of-the-money (OTM) by the changes in the rates during the day. Assuming the
Vega step is performed before the interest rate step, the P&L allocated to the
Vega step can be quite significant whereas if the Vega step had been performed
after the interest rate step, then its P&L would be much smaller.

20
3.2 Path independent Step re-evaluation P&L

In the path independent method we want to allocate the P&L associated to

the change of each market parameter independently.
For each step of the P&L calculation process, we will perturb the market, com-
pute the PV of our portfolio on the the perturbed market.
We are performing the valuation steps in line with the Taylor expansion of a
multi-variate function. Thus we will have to introduce an unexplained term.

3.2.1 Single factor step re-evaluation P&L

A simple form of bump and reset explain will capture the P&L associated to
each risk factors independently.
V (Pmkt(tC OB) ) = V (Pmkt(tS OD) ) + ∇T δ.

We then introduce the UnExplained term:

V (Pmkt(tC OB) ) = V (Pmkt(tS OD) ) + ∇T δ + U nExplained,

with:

∂V (Pmkt(t) )
∇i = .
∂Pi
The P&L associated to each SRi∗ step is:

SRi∗ = V (Pmkt(t) + δSR

∗
i
) − V (Pmkt(t) ).

The computational cost of this method grows linearly with the number N of
steps.

Remark:

The UnExplained term is a critical metric that regulators and product con-
trol within a bank alike pay attention to.

21
P&L explained is used to test the hypothesis that the risk factors identified for
a risky position are sufficient to materially explain the value change expected
from the risky position.
Such that if position sensitivities to those risk factors are calculated, then the
value change observed over a day can be attributed to the market price change
of those risk factors, with the magnitude of the estimated as a sum product of
the risk factor sensitivities and the corresponding daily risk factor price change.
Any residual P&L left unexplained (P&L Unexplained) would be expected to
be small if the identified risk factors are indeed sufficient to materially explain
the expected value change of the position AND if the models used to calculate
sensitivities to these risk factors are correct.
P&L Unexplained is thus a critical metric that when large may highlight in-
stances where the risk factors classified for a risky position are incomplete or
the models used for sensitivities calculations are incorrect or inconsistent.

This expression is equivalent to pricing our 20Y interest rate swap with our
shifted delta which the first order risk factor calculated as below minus of the
present value of our IRS (interest rate rate swap) with a non shifted delta:
1- Compute an initial PV for the considered trade or portfolio of trades
2- Perturb the first tenor of the yield curve and rebuild the pricing market
keeping all other pricing parameters in Pm kt(tSOD ) constant.
3- Re-price the considered trade or portfolio of trade on the perturbed market.
PV difference between (1) and (3) is the sensitivity (or risk) to the first tenor
4- Repeat the same sequence for all other tenors in the curve. Note that it is
not recommended to reset the perturbation done on (2) before doing the next
perturbation.

3.2.2 Cross risk factors step re-evaluation P&L

Generally portfolios may exhibit a significant dynamic and thus the previous
method won’t be valid. This abnormality would manifest in a large Unexplained
term.
In this case we need to push Taylor expansion to the second order.

22
 1
V (Pmkt(tC OB) ) = V (Pmkt(tS OD) ) + ∇T δ + δ T AV δ + U nExplained
2

With AV is the Hessian matrix:

∂ 2 V (Pmkt(t) )
AVi,j =
∂Pi ∂Pj
with i, j ∈ J1, N K.
∗
Practically, we define the new steps SRi,i of the bump-and-reset explain cap-
turing the second order and cross risk P&L:

SRi,j = V (Pmkt (t) + δSRi,j ) − V (Pmkt(t) ) − SRi − SRj .

with:
 
0
 ... 
 
 
δ 
 
δSRi,j =  i 
0
.
 .. 
 
δj 
0
. The computation cost needs an extra N2 valuations to capture the cross risk

P&L so the number of steps is now quadratic.

4 Risk based P&L calculation

Risk factor sensitivities from the prior close (T-1) are used to attribute the
P&L expected from changes from the prior market close in the market price of
the risk factors.
Once risks have been computed, the P&L can be computed from the definition
of the taylor expansion of the valuation function V, generally speaking by mul-
tiplying the risk by the change in the underlying market parameters.

23
The valuation function of a given trades or portfolio of trade can be evaluated
at tCOB starting from its value at tSOD .

For a linear portfolio a first order theta-delta only approximation is suf-

ficient. However, most derivative instruments are non-linear and have some
curvature e.g. interest rate swaps are quasi-linear in the floating leg and have
some gamma. So more conservatively, a second order theta-delta-gamma Tay-
lor approximation is more likely. And where the positions include optionality
then a second-order theta-delta-gamma-vega Taylor approximation is chosen.
Below shows a second-order theta-delta-rho-gamma-vega-volga-vanna Taylor
approximation that would be appropriate for a portfolio of complex options on
a single factor.
1
V (Pmkt(tCOB ) ) = θ∆t + δ∆S + γ∆S 2 + · · · .
2

with:
θ representing the rate of change between the option price and time, or time
sensitivity - sometimes known as an option’s time decay.
δ representing the rate of change between the option’s price and a 1 unit change
in the underlying asset’s price.
γ representing the rate of change between an option’s delta and the underlying
asset’s price.
Method’s limitations:
• It is not very accurate since some risk factors can be missing and this the
P&L won’t be fully explained.
• The necessary volume of data to store can be unmanageable.

5 Results
5.1 Step revaluation Limitations

The P&L computation of a given portfolio (ie. trading book) consists in com-
puting its present value in a given market context. The step revaluation ap-
proach consists in:

24
 • Classifying the differences between two P&L computations at D-1 and D.
• Computing some intermediary P&L values associated to this classification
and then allocate the P&L contribution for each category of difference.
Then we can choose a given level of details for the classification and cate-
gorize all the differences between D-1 and D. The P&L contribution for each
category can be computed in two ways:
In a cumulative mode (path dependent): i.e. starting from the D-1 P&L com-
putation, we import step by step each category of difference and for each its
P&L contribution. After the final step, there may be an unexplained effect
when comparing the final importation step and the D P&L value if the classi-
fication does not cover all potential differences. Otherwise there should not be
any unexplained effect in theory.

Figure 4: Cumulative step revaluation P&L calculation result based January 2020 Bnp data

In a star mode (path independent): i.e. for each category, the P&L contri-
bution is always computed from the D-1 P&L computation. We may also start
in a cumulative way and switch to the star mode. This approach may result in

25
an unexplained effect even for a complete classification.

Figure 5: Star step revaluation P&L calculation result

There are difficulties for the stepped revaluation method, since in order to
price the contract we generally use Monte Carlo simulation. This method does
not provide the true present value of a contract but only an approximated value.
As the target of this document is not to present the Monte Carlo simulation,
we will consider all the underlying mathematical concepts as known, but we
will try to describe their consequences in a way that can be understood by a
large audience. Then we can consider that, for most of the stocks books with
Monte Carlo calculations, the computed P&L value is in fact the real one plus
a “brownian noise” (i.e. a trust interval). And the problem is that this noise
depends of the pricing context: pricing a contract alone or with some others but
also technical differences in the market data representation (let’s think about
a null pillar on a dividend curve and the same curve without the pillar). Note
that PDE (partial differencial equaltions) computations are not fully exact also

26
(e.g. grid discretization) but the error does not depend of the pricing context
and then there are no painful consequences for the step revaluation.

Figure 6: Noisy step revaluation P&L calculation result

For the step revaluation effect computation, this brownian noise is polluting a
lot the results because the effect contributions are computed as premium differ-
ences and then the noise can be multiplied by two in the worst case. Then even
for a cumulative step revaluation computation with a complete classification,
there can be some unexplained P&L while the theory says there is no.

27
 Figure 7: Weekly P&L 5M graph convergeance

5.2 Comparison of the P&L explain calculation methods

To present one of the most risk based method limitations, let’s take an ex-
treme shock scenario. Since we focus on the underlying, we can neglect the
Theta, Vega and cross-gammas for our P&L prediction. Then we take the
delta-gamma expansion of the P&L with s the perturbation of spot we have.
1
V (Pmkt(tCOB ) ) − V (Pmkt(tCOB ) ) = δs + γs2
2
Let the valuations function be a simple valuation of a Call option for example.
We can replace gamma by her finite difference approximation, which gives:
1 δc+s − δc 2
V (C + s) − V (C) = δs + ∗ s
2 S
s
V (C + s) − V (C) = δs + ∗ δc+s − δc
2
This is a first order method. For large shocks, the approximation error (hence
the unexplained PnL) becomes material, so the method diverges.

28
5.2.1 Empirical result

Let’s consider a 1 year maturity Call option with 2 percent strike, with shocks
from 10bps to 3 percent. The delta-gamma (second order Taylor expansion)
prediction method diverges as the shock increases to 2 percent.

Figure 8: Empirical result of P&L calculation result

We can see that as the shock increases our delta-gamma P&L increases to
diverge so the unexplained factor diverges too. To generalise, Taylor expansion
method converges only for large orders, so we need high order sensitivities to
be stocked on a daily basis in order to compute our P&L explain. It is very
expensive and needs large data-base, that’s why the re-evaluation method tend
to be more practical since it shocks every risk factor separately with (1bp basis
point) for example and evaluates the impact on the P&L of our contract.

Figure 9: Divergence of Taylor expansion method for small orders

29
5.2.2 Comparison conclusion

So as we have seen this scenario showcases the divergence of our method for a
simple contract and helps us understand one of the method’s limitations. At
the industrial level, on a big scale portfolios this method still can be used under
the convergence condition because it partially helps explain our P&L despite
the accuracy problem. For the re-evaluation method, it is an exact method as
the theory behind is very precise and doesn’t include a lot of approximations.
Since it doesn’t have to stock all the sensitivities, it is actually faster to com-
pute. The most important limitation of this method is that it is expensive to
rerun during the day, because it should import all the changes in the market
data and shock our risk factors successively to see the impact on the P&L,
that’s why it is generally computed at the end of our business day.
So both method present advantages and disadvantages, but we choose the
adapted method depending on our financial products and the risk factors they
may include.
Step re-evaluation Risk based
Pros -Exact methodology. -Provide link between risk factors
and P&L
-Faster than risk-based calculation -Fast intraday predict if we as-
wise. sume and risk factors are avail-
able
Cons -Expensive to re-run for intraday re- -Not all P&L effects can be com-
sult. puted risk-based
-Only provides an allocation of the -Accuracy cannot be guaranteed
P&L with no risk and market moves.

We need to mention that the previous analysis is based on the market effects
as it analyses the P&L explained by the Greeks. In real markets life, other
parameters may play a big role in the P&L result fluctuation such as time be-
tween the close of business and start of day and other environmental effect that
doesn’t enter in the scope of this thesis analysis.
I chose quantitative parameters to make the study very concrete using mathe-

30
matical equations.

31
6 Conclusion & Extension
Unlike risk sensitivities based explain P&L, re-evaluation method provide
an aggregate measure of the P&L impact due to a risk factor’s market price
change; without differentiating between and attributing to the varying rates of
value change embedded in the relationship between portfolio value and the risk
factor. It is important to note this distinction between risk based explain and
re-evaluation methods’ aggregate measures as they are sometimes used inter-
changeably and sometimes netted together.
Re-evaluation methods may isolate the P&L impact of the change in value of
an interest rate swaption to a change in the forward curve but have nothing to
say about how much of that change was due to first order effect of that change
i.e. delta and how much was due to second order effects i.e. gamma and so on.
This is analogous to estimating the velocity of a car driven at varying rates of
acceleration between points 1 and 2 simply by dividing the distance covered
by the time taken but without attribution to the segments of the journey with
different foot-to-pedal action.

32
 In addition to helping to enrich and evolve the bank’s models, the P&L ex-
plained processes described above help build a rich data set on P&L and its
linkages into risk, VAR(Value At Risk) and CVA(credit value adjustment).

A data set that can be mined through analysis for insight on P&L drivers,
strategy and trade performance across all components of P&L and much more.
Unexplained P&L and VAR breaches also provide information about opera-
tional problems and the root cause analysis to close these out can help build
a very rich picture of the efficacy of a bank’s operational infrastructure and
better target remediation efforts. However, a 2010 Ernst Young survey
of 14 global financial institutions found that on average their survey re-
spondents’product controllers spent only 13% of their time on the analysis and
explain of P&L; but spent 46% of the time on the production of P&L, and an-
other 22% on reporting and sign off. This distribution of effort is worrying but
the relative amount of time taken up by production is likely a good indicator of
the weaknesses that often exist and persist in the infrastructure and processes.

Hopefully the degree of focus that regulators have paid to the P&L explained
production processes over the last 2 years may encourage progress in that sense
that would increase the banks profit and rebalance the efforts.

33
7 References
[1] Björk T.(1998). Arbitrage Theory in Continuous Time. Oxford University Press, Oxford.

[2] Björk T. and Christensen, B. (1999). Interest Rate Dynamics and Consistent Forward Rate Curves.
Mathematical Finance, 9(4), 323–348.

[3] Gregory J., Credit Risk and Credit Value Adjustment: A Continuing Challenge for Global Financial
Markets, 2nd Edition. Published by Wiley, 2012

[4] Filipović, D. (2001). Consistency Problems for Heath-Jarrow-Morton Interest Rate Models. Springer
Lecture Notes in Mathematics, Vol. 1760. Springer Verlag., Berlin, Heidelberg.

[5] Hull, J. (2003). Options, Futures, and Other Derivatives (5th edn). Prentice Hall, Englewood Cliffs,
N.J.

[6] Sundaresan, S. (2009). Fixed Income Markets and Their Derivatives (3rd edn).Academic Press.

[7] Acuity derivatives LLC, Why Profit & Loss Attribution or judging Weatherman.New York, NY 10004
USA

[9] Piterbarg, V.V. and Andersen, L.B.G. Interest Rate Modeling, Foundations and Vanilla Models, vol-
ume I. Atlantic Financial Press, 2010.

[10] Piterbarg, V.V. and Andersen, L.B.G. Interest Rate Modeling, Products and Risk Management,
volume III. Atlantic Financial Press, 2010.

34
TRITA SCI-GRU 2020:335

www.kth.se