W 25416
W 25416
W 25416
Gauti B. Eggertsson
Ragnar E. Juelsrud
Lawrence H. Summers
Ella Getz Wold
This working paper should not be reported as representing the views of Norges Bank. The views expressed
are those of the authors. The views expressed are those of the authors and do not necessarily reflect
those of Norges Bank or the National Bureau of Economic Research. This paper replaces an earlier
draft titled Are Negative Nominal Interest Rates Expansionary? We are grateful to compricer.se and
Christina Soderberg for providing bank level interest rate data. We are also grateful to participants
at numerous seminars and conferences, as well as Martin Floden, Artashes Karapetyan, John Shea,
Dominik Thaler, Pau Rabanal, Morten Spange and Michael Woodford for useful comments and discussions.
We thank INET for financial support.
NBER working papers are circulated for discussion and comment purposes. They have not been peer-
reviewed or been subject to the review by the NBER Board of Directors that accompanies official
NBER publications.
© 2019 by Gauti B. Eggertsson, Ragnar E. Juelsrud, Lawrence H. Summers, and Ella Getz Wold.
All rights reserved. Short sections of text, not to exceed two paragraphs, may be quoted without explicit
permission provided that full credit, including © notice, is given to the source.
Negative Nominal Interest Rates and the Bank Lending Channel
Gauti B. Eggertsson, Ragnar E. Juelsrud, Lawrence H. Summers, and Ella Getz Wold
NBER Working Paper No. 25416
January 2019, Revised in September 2020
JEL No. E3,E31,E4,E41,E42,E43,E44,E5,E51,E52,E58,E65
ABSTRACT
We investigate the bank lending channel of negative nominal policy rates from an empirical and
theoretical perspective. We find that retail household deposit rates are subject to a lower bound
(DLB). Empirically, once the DLB is met, the pass-through to lending rates and credit volumes is
substantially lower and bank equity values decline in response to further policy rate cuts. We
construct a banking sector model and use our estimate of the pass-through of negative policy rates
to lending rates as an identified moment to parameterize the model and assess the impact of
negative policy rates in general equilibrium. Using the theoretical framework, we derive a
sufficient statistic for when negative policy rates are expansionary and when they are not.
1An alternative to negative interest rates is unconventional monetary policy measures. There are several reasons, however, why it is
important to consider policy measures beyond these tools. Some of the credit policies used by the the Federal Reserve, the FDIC and
the Treasury were severely constrained by Congress following the financial crisis, as stressed by Bernanke et al. (2018). Moreover,
there remains little, if any, consensus among economists on how effective quantitative easing and forward guidance is. Plausible
estimates range from considerable effects to none (see e.g. Greenlaw et al. (2018) for a somewhat skeptical review, Swanson (2017)
for a more upbeat assessment, and Greenwood et al. (2014) for a discussion of debt management at the zero lower bound).
2The Swedish case is useful to study for a number of reasons, most importantly because we have access to high-frequent microdata on
bank-level mortgage rates. In addition, the Swedish central bank - the Riksbank - cut interest rates four times in to negative territory,
providing relatively rich variation in the data which can be used to estimate the pass-through of negative rates and how it depends on
the lower bound on the deposit rate.
2
policy rate turns sufficiently negative. This is in sharp contrast to regular circumstances, in which the two
move closely together. Because deposits are the main source of bank financing (approximately 50 % in our
data), the transmission of policy rates to banks marginal funding costs is impaired once this point is reached.
An important finding is that this collapse in pass-through does not need to happen at exactly zero. In our
empirical setting, the deposit rate remained responsive to policy rate cuts, albeit less so, until the policy rate
reached -0.25 percent.3
We proceed by considering the pass-through of negative policy rates to lending rates. To do so, we
conduct an event study around days of policy rate cuts using daily bank level mortgage rates. Under regular
circumstances, we show that the pass-through of policy rate cuts is around 80 percent within 30 days. Once
policy rates reach a level at which deposit rates are not longer responsive however, the pass-through collapses.
In fact, taking a weighted average over the different mortgage contracts offered across all banks in our sample,
the average pass-through to aggregate mortgage rates becomes slightly negative, i.e. the empirical evidence
suggests a modest increase in lending rates once the policy rate falls sufficiently below zero.
In addition to a collapse in the pass-through to lending rates, we also document an increase in the
dispersion of lending rates as the policy rate turns negative. The rise in dispersion is linked to banks financing
structures. Banks which rely more heavily on deposit financing are less likely to reduce their lending rates
once the deposit rate has stopped responding. Moving to bank-level lending volumes, we show that Swedish
banks that rely more heavily on deposit financing also have lower credit growth once the deposit rate has
reached its lower bound.
A key question in the debate surrounding negative interest rates is whether they have detrimental effects
on banks net worth. We conclude the empirical section by conducting an event study using equity values for
publicly listed Swedish banks. In contrast to policy rate cuts in normal times, policy rate cuts in negative
territory are found to negatively impact bank equity values.
Motivated by the empirical results, we construct a bank model which rationalizes the empirical findings,
and provides additional theoretical predictions for when negative policy rates are effective. The model is also
used to produce quantitative predictions on the impact of negative rates and to discuss policy interventions
which can make negative rates more effective once retail deposit rates reaches a lower bound.
Negative nominal interest rates can arise in our model because banks and households are potentially
willing to pay for storage and liquidity services provided by reserves and deposits. Taking this into account,
our model derives in place of the conventional zero lower bound two new lower bounds: a lower bound on the
policy rate (the "PLB") and a lower bound on the deposit rate banks can offer (the "DLB"). Both arise due the
the presence of paper currency, although evidence on banks cash holdings suggests that the PLB has remained
non-binding. Unlike the PLB, the available empirical evidence indicates the DLB is close to zero for small
depositors and only modestly negative for larger depositors.
In our model, policy rate cuts are perfectly transmitted to deposit rates and lending rates when the lower
bounds are non-binding, thus stimulating lending and aggregate demand. Once the DLB is reached, however,
the transmission of policy rate to the main financing source of banks breaks down – as in the data. In this
3In countries where the spread between deposit and policy rates are higher, for instance in some Eurozone countries, deposit rates
could fall by more, and hence the policy rate can go lower, before deposit rates reaches a DLB.
3
case, we show that the impact of negative policy rates on bank profits is a sufficient statistic for the qualitative
impact of negative policy rates on bank lending.
Whether negative policy rates reduce or increase bank profits depends on the balance sheet composition
of banks. In addition to lending (B) and reserves (R) on the asset side of the balance sheet, we assume that
banks hold liquid assets (A), such as government bonds, to insure against liquidity shocks. On the liability side,
in addition to deposits (D), banks finance themselves via external financing (F), which refers to all non-deposit
financing which is not subject to the DLB (e.g. bank bond issuance and larger wholesale depositors). The
sufficient statistic implies that negative policy rates expand bank lending if
𝜌𝑎 𝐴 + 𝑅 < 𝜌 𝑓 𝐹 (1)
where 𝜌 𝑎 and 𝜌 𝑓 measure the effective pass-through of the policy rate to liquid assets and external financing
respectively. Intuitively, bank profits increase (and hence lending expands) if the effective pass-through of
negative policy rates is larger to bank liabilities (𝜌 𝑓 𝐹) compared to bank assets (𝜌 𝑎 𝐴 + 𝑅). Looking at
Swedish data, we find that this condition is not satisfied, suggesting that further policy rate cuts at the DLB
was ultimately counterproductive through the bank lending channel. This is consistent with the negative
impact on bank equity values found in the data, and the lack of pass-through to lending rates. We discuss how
this sufficient statistic can explain why different empirical studies provide seemingly conflicting evidence on
the bank lending channel of negative policy rates.
After discussing our partial equilibrium results, we embed the banking model into a dynamics stochastic
general equilibrium model, which nests the standard New Keynesian model as a special case. We use our
empirical estimates of the pass-through of negative policy rates to lending rates, combined with existing
empirical evidence4 on the link between bank net worth and bank lending, as identified moments to pin down
the key parameters of the model. The model can then be used to quantitatively evaluate the effect of negative
rates on output and inflation depending on whether the DLB is binding or not, and on the initial balance sheet
composition of the bank sector. Overall, we find that the effect of policy rate cuts on output at the DLB can be
both positive and negative. In a benchmark parameterization where the balance sheet composition is set to
match the Swedish banking sector, the impact is negative and implies that a 100 basis points reduction in
the policy rate once the DLB is binding results in a 17.7 basis points contraction in output. Our quantitative
exercise suggests that for negative policy rates to have a non-trivial positive impact on the economy, other
channels – such as an asset price channel or an exchange rate channel – would have to be important.
Literature review Our paper relates to a growing empirical literature on the effects of negative interest rates
on bank outcomes. The empirical literature has focused primarily on three questions, namely the impact of
negative policy rates on deposit rates, lending rates and bank equity values.
Heider et al. (2016) document that median overnight deposit rates in the Eurozone did not fall below
zero following the introduction of negative policy rates, while Basten and Mariathasan (2018) and Hong and
Kandrac (2018) show similar findings for Switzerland and Japan, respectively. The zero lower bound on
deposit rates in the Eurozone appears to be most prevalent for household and other retail deposits. Boucinha
and Burlon (2020), Eisenshmidt and Smets (2019) and Altavilla et al. (2019) document that an increasing
4
fraction of corporate depositors are charged a negative interest rate, although the pass-through is substantially
weakened relative to when rates are positive.
In terms of bank lending, Bech and Malkhozov (2016) document how lending rates in Switzerland
increased after the introduction of negative policy rates. Basten and Mariathasan (2018) further show that the
lending margin for Swiss banks increased following negative rates, and that banks with larger reserve holdings
increased interest rates by more. Amzallag et al. (2019) find that Italian banks with higher deposit ratios
charged higher rates on fixed-rate mortgages in response to policy rates turning negative. Hong and Kandrac
(2018) document lending rate increases in Japan. Eisenshmidt and Smets (2019) on the other hand, find no
evidence that high-deposit banks in Germany increased loan rates relative to low-deposit banks. Bottero et
al. (2019) show that Italian banks with more liquid balance sheets expanded lending in response to negative
policy rates. Adolfsen and Spange (2020) document a reduction in pass-through to lending rates in Denmark
but, in contrast to our findings, show that it is relatively uniform across bank funding structures.
Ampudia and Van den Heuvel (2018) show that while policy rate cuts normally expand bank equity
values in the Eurozone, policy rate cuts in negative territory lowers bank equity values, which is consistent
with our evidence from Sweden. Moreover, the effect is larger for banks with higher deposit shares.5 Hong and
Kandrac (2018) show related evidence from Japan, documenting that Japanese banks’ equity values declined
by five percent within an hour of the announcement of negative rates by the Bank of Japan.
To summarize: Our findings on the impact of negative policy rates on deposit rates and bank equity values
are largely consistent with the existing literature for other countries. Our main contribution to the empirical
literature is to use daily interest rate data from Sweden in an event study to estimate the pass-through of
negative policy rates to lending rates. In contrast to much of the existing literature, our empirical estimates is
informative about the absolute effect of negative policy rates on lending rates, rather than a relative difference
between treated and control banks. For instance, Basten and Mariathasan (2018) study the impact on lending
margins in Switzerland using a difference in difference analysis and find that banks with higher reserve
holdings have lower pass-through relative to other banks. In contrast, we use an event study setup to estimate
the absolute pass-through of negative policy rates to lending rates. This estimate is not only of particular
policy interest, it also serves as an identified moment to match in our banking model.
The theoretical literature on negative interest rates is perhaps surprisingly somewhat smaller, given the
high stakes in the policy debate.6 Three studies are especially relevant for the theoretical analysis in this paper.
Rognlie (2015) provided a first analysis of the normative aspects of negative policy rates. A key distinction
between his paper and ours is that in his model households face only one interest rate, and the central bank
can control this interest rate directly. Since we are interested in theoretically evaluating the pass-through of
negative policy rates to other interest rates, our model differs from his by having multiple interest rates and
explicitly modeling the transmission mechanism of monetary policy through the bank sector.
Brunnermeier and Koby (2017) propose a theoretical model in which there is a reversal rate at which
point further interest rate cuts become contractionary. There are three main differences between their model
5
and ours. First, the main mechanism in their paper is not motivated by the DLB, which in our model is derived
theoretically from the households portfolio allocation problem due to the existence of money as a nominal
asset. Thus our focus is primarily on the problem of negative policy rates in the presence of paper currency,
while the reversal rate arises independently of paper currency, and can occur at either positive or negative
policy rates. Second, the reversal rate arises mainly due to maturity mismatch on the banks balance sheet,
which gives rise to capital gains/losses due to unanticipated interest rate cuts. We abstract from banks balance
sheet mismatch, as we aim to evaluate whether negative policy rates rates can substitute for regular policy
rate cuts. This consideration arises independently of whether interest rate cuts are anticipated or not. Third,
banks in our model have access to not only deposit financing and net worth, but also rely on market funding
– which is potentially subject to a different pass-through. This gives rise to the second key mechanism for
understanding the effectiveness of negative policy rates in our model, namely the banks exposure to negative
policy rates on the asset side relative to the liability side, which is the key condition for determining if negative
policy rates are effective or not in our model.
Finally, Ulate (2019) embeds a monopolistically competitive banking sector into an otherwise standard
New Keynesian model and investigates the extent to which negative policy rates are expansionary. Our
theoretical analysis differs from Ulate (2019) in that we model a richer bank balance sheet, allowing banks
to finance themselves with non-deposit funding, as well as specifying an explicit role for liquid assets. The
presence of external financing allows for substantial pass-through to banks overall funding costs, even when
the DLB is binding. Our model highlights the importance of considering the net exposure of banks to negative
policy rates, based on both assets and liabilities, rather than focusing on deposit shares alone. In Ulate (2019),
negative policy rates always reduce bank profits. In our setting, negative policy rates can both increase and
decrease bank profits depending on banks’ net exposure to negative rates. Our model can therefore reconcile
the sometimes contradicting findings of different empirical studies.
Although our results indicate that the transmission of negative interest rates through the bank sector
can have diminishing returns depending on i) the distance to the DLB and ii) banks net exposure to negative
rates, there are ways in which negative policy rates can have further expansionary effects which we do not
capture in our model. Most importantly, we do not study other transmission mechanisms such as exchange
rate effects or asset price effects. Second, if the DLB is overcome, our model predicts that negative policy
rates should be an effective way to stimulate the economy. This could happen if banks over time become more
willing to experiment with negative deposit rates, and depositors do not substitute to cash, or if there are
institutional changes which affect the DLB. In Section 4 we consider under which conditions this could happen.
An example of such policies is a direct tax on paper currency, as proposed first by Gesell (Gesell, 1916) and
discussed in detail by Goodfriend (2000) and Buiter and Panigirtzoglou (2003) or actions that increase the
storage cost of money, such as eliminating high denomination bills. Another possibility is abolishing paper
currency altogether. These policies are discussed in, among others, Agarwal and Kimball (2015), Rogoff
(2017a) and Rogoff (2017c), who also suggest more elaborate policy regimes to circumvent the zero lower
bound. The results presented here do not contradict these ideas. Rather, they suggest that given the current
institutional framework and our empirical findings, negative interest rates will not be an effective way to
stimulate aggregate demand via the bank lending channel.
6
2 Empirical analysis
Both the empirical and the theoretical section are organized around the simplified bank balance sheet
illustrated in Figure 1.
Assets Liabilities
Loans (B) -- ib
Deposits (D) -- id
Money (M) -- 0
L
Other
liquid Assets (A) -- ia Net worth (N)
The policy instrument is the interest rate the central bank pays on reserve accounts that commercial banks
hold to execute interbank transactions, 𝑖 𝑟𝑡 .7 We denote the pass-through of the policy rate to an interest rate 𝑖
by 𝜌 𝑖 , where 𝑖 refers to either liquid assets (A), deposits (D), external financing (F) or bank lending (B). The
main objective of this section is to empirically account for the pass-through of the policy rate to these other
interest rates.
We first illustrate how the policy rate is transmitted to the interest rate on liquid assets before moving on
to the liability side of the balance sheet. This part of the analysis is relatively straight forward and largely
descriptive. The main focus, however, is on the impact of negative policy rates on lending rates and volumes.
We conclude the empirical section by estimating the impact of negative policy rates on the price of equity (N).
The objective of Section 3 is to provide a theoretical framework which outlines how each component of the
balance sheet is determined under negative policy rates, using the empirical evidence to discipline the model.
7The exact implementation of negative policy rates differ across the countries that have implemented them, see Bech and Malkhozov
(2016) for a detailed overview across countries. In the case of Sweden, which we focus on, the Riksbank operates a corridor system
and the policy rate is referred to as the repo-rate. The repo-rate is essentially the interest rate banks receive for holding transaction
balances at the Riksbank. Banks can borrow from the Riksbank at 75 basis points above the policy rate and central bank reserves earn
an interest rate 75 basis points below the policy rate. Consider for example a policy rate of - 0.5 percent. In order to implement this
rate, the Riksbank sells certificates in repo transactions that pay - 0.5. As the banks are obtaining -1.25 on their reserves, they will use
the reserves to purchase these certificates. In this sense the repo-rate is essentially equivalent to the Riksbank directly paying - 0.5 on
bank reserves.
7
2.1 Liquid assets (L=M+R+A)
We start by investigating the pass-through of negative policy rates to banks’ liquid assets. Figure 2 depicts
the evolution of the repo-rate over the past 10 years, along with money market rates (STIBOR) and the interest
rate on government bonds.
As the figure reveals, there has been approximately full pass-through from the policy rate to money
market rates. The money market rate measures the rate at which commercial banks lend their excess reserves
to other banks. It is not surprising that the money market rate follows the policy rate into negative territory, as
long as the supply of reserves is sufficiently large.
Notes: This figure shows the repo-rate together with the interest rates on 1-week interbank rate (STIBOR, 1 week), 1 month
government certificates (Government certificate, 1 month) and 2 year government bonds (Government bond, 2 year). Source:
Riksbank.
What can banks do with their excess reserves? Apart from lending them to other commercial banks, they
can also buy liquid assets, such as government bonds. Viewed in this light, it is easy to see why there is also
relatively strong pass-through to liquid asset rates. According to the figure, the short term risk-free interest
rate on government bonds falls virtually one-to-one with the repo-rate and the money market rate.
Deposit financing accounts for about half of bank liabilities for large Swedish banks.8 For smaller banks
the deposit share is larger. The deposit rate is therefore especially important for evaluating banks’ marginal
costs of funds and overall profitability.
The left panel of Figure 3 depicts aggregate deposit rates in Sweden.9 Prior to the policy rate turning
negative, the aggregate deposit rate is below the policy rate and moves closely with it. As the policy rate turns
8
negative this relationship breaks down. Instead of following the policy rate into negative territory, the deposit
rate appears bounded at a level close to zero.
2
.6
1.5
1
.4
.5
0
.2
-.5
Households Corporations
0
Policy Rate Pre-Avg. -> 0.25 -> 0 -> -0.1 -> -0.25 -> -0.35 -> -0.5
Figure 3: Deposit rates (left panel) and the relative change in deposit rates (right panel).
Notes: This figure shows the policy rate, household deposit rate and corporate deposit rate (left panel) and the pass-through of
policy rate cuts to deposit rates (right panel). The policy rate is defined as the repo-rate. In the right panel, we show the change in
the deposit rate relative to the change in the policy rate for the last six policy rate reductions. The numbers on the x-axis denotes
the policy rate level to which the repo-rate was reduced. Source: The Riksbank, Statistics Sweden.
It is useful to take a closer look at the last six policy rate reductions. The change in the deposit rate
relative to the change in the repo-rate is illustrated in the right panel of Figure 3. The first bar captures the
average relative change in deposit rates prior to 2014. In this case, the pass-through to the aggregate deposit
rate was approximately 60 percent. For the post-2014 data, the pass-through is lower. For policy rate cuts in
positive territory, the pass-through is approximately 40 percent. For the first two cuts in negative territory, i.e.
to -0.1 percent and to -0.25 percent, the pass-through remains relatively unchanged. For the final two interest
rate cuts, however, the pass-through collapses to roughly zero. As the deposit rate reaches its lower bound
(DLB), further reductions in the policy rate do not lead to continued reductions in the deposit rate. We refer to
the period after the deposit rate has stopped responding, i.e. the last two policy rate reductions, as the period
in which the deposit bound (DLB) is binding.
Even with the DLB binding, an increase in fees could decrease the effective deposit rate. In Appendix
B, we document that the observed evolution of commission income is not consistent with fees increasing
sufficiently to compensate for the DLB on the nominal deposit rate.
For larger Swedish banks, almost half of liabilities are non-deposits. The largest remaining component
is covered bonds. The left panel of Figure 17 in Appendix B compares the interest rate on covered bonds
and other interest rates to the policy rate. As with deposit rates, the correlation between the policy rate and
covered bond rates is somewhat weaker once the policy rate turns negative.
Even if the pass-through to covered bond rates is weaker, we see from Figure 17 that the interest rate
on covered bonds with shorter maturities eventually becomes negative, suggesting a stronger pass-through
than for deposit rates. If banks respond to negative policy rates by shifting away from deposit financing, they
9
would therefore reduce their marginal financing costs. However, as we show in Figure 18 in Appendix B, this
is not the case. There is no increase in the issuance of covered bonds for Swedish banks, and the deposit share
in fact increases.
There are at least four possible explanations for why banks did not shift away from deposit financing
despite a cost advantage associated with non-deposit funding: i) maintaining a base of depositors creates some
synergies which other financing sources do not, ii) other funding sources are associated with higher liquidity
risk, iii) the room for new issuances of covered bonds may be limited by the availability of bank assets to
use for collateral, and iv) Basel III regulation makes deposit financing more attractive in terms of satisfying
new requirements. In section 3, these consideration are taken into account by assuming that banks insurance
against refinancing risk by investing in liquid assets, and that the refinancing risk associated with external
financing is larger than insured deposit-financing. Since the rate on liquid assets follows the policy rate into
negative territory, an implication of this is that external financing does not necessarily become cheaper, which
can explain the patterns observed in Figure 18.
Official bank lending rates are only available at a monthly frequency, and not at the bank level. Here, we
therefore use daily bank level data based on listed mortgage rates.10
Figure 4 plots the raw data for daily mortgage rates since 2014, with each line corresponding to one of
the eleven banks in our sample. The upper left panel depicts lending rates for mortgages with three month
fixed interest rate periods, referred to as floating rates. This is the most common mortgage contract in the
Swedish market, accounting for roughly 70 percent of the market.11 Consider first the four interest rate cuts
which occur prior to the deposit rate reaching its lower bound, i.e. the policy rate reductions all the way down
to -0.25 percent. This is the pre-DLB policy rate cuts. For these policy changes, banks respond to policy rate
cuts by consistently reducing their lending rates. This stands in stark contrast to the last two policy rate cuts,
i.e. to -0.35 and to -0.5 percent, in which there is virtually no reduction in bank lending rates. While one bank
cuts its lending rate in response to the policy rate being reduced to -0.35, the lending rate is increased again
shortly thereafter. Overall, the pass-through to lending rates appears severely weakened.
While floating rate mortgages are the most common mortgage contracts, Figure 4 also reports mortgage
contracts with fixed rate periods of one year, three years and five years. Again, there is a collapse in
pass-through for the two last policy rate reductions. In addition, the dispersion in bank lending rates increases.
10The mortgage rates are the interest rates listed by banks on mortgages with floating rates or fixed rate periods from one to
five years. Some banks also list lending rates on mortgages with fixed rate periods of ten years. However, we focus on the
mortgage contracts used by most of the banks in our sample in order to increase power. Our sample consists of the eleven
largest Swedish banks or financial institutions in terms of mortgage lending. According to The Swedish Bankers Association,
the six largest banks account for 91 percent of the mortgage market – all of which are in our sample. The largest banks as of
2016 in terms of mortgage lending are Swedbank (24 %), Handelsbanken (23 %), Nordea (15 %), SEB, (15 %), SBAB (8 %),
Lansforsakringar (6 %). The remaining 9 percent are attributed to among others Danske Bank and Skandiabanken. See the 2016
report: https://www.swedishbankers.se/media/1310/1611bolaanemarknaden_eng.pdf. All of the banks mentioned are in the sample.
11See the 2018 report on the mortgage market in Sweden: https://www.swedishbankers.se/ media/3906/1809_bolaanemarknad-
2018_en.pdf
10
Below, we investigate the underlying sources for this increase in dispersion.
3 months 1 year
3
2.5
2.5
2
2
1.5
1.5
1-1-2014 1-1-2015 1-1-2016 1-1-2014 1-1-2015 1-1-2016
repo: 0.75 0.25 0 -0.1 -0.25 -0.35 -0.50 repo: 0.75 0.25 0 -0.1 -0.25 -0.35 -0.50
3 year 5 year
4
3
3.5
2.5
3
2
2.5
1.5
Event study
To formalize the pass-through of policy rate cuts to lending rates and investigate whether it changes at the
DLB, we estimate the regression
30
X 30
X
𝑖¯𝑡𝑙 = 𝛼 + 𝛽𝑘 𝐼𝑘 + 𝐼𝑡𝐷𝐿𝐵 𝛽 𝐷𝐿𝐵
𝑘 𝐼 𝑘 + 𝜖𝑡 (2)
𝑘=−30 𝑘=−30
where 𝑖 𝑡𝑙 is the average, daily aggregate bank lending rate. We construct this variable by aggregating the
lending rate of each bank, using their total assets as weight. Furthermore, each bank’s lending rate has been
constructed by weighting each of their lending contracts by their respective market share12.
The fourteen interest rate cuts since 2009 were of different size. To account for this, we scale the lending
rate by the change in the repo-rate. Thus a value of -100 implies full pass-through of the policy rate.
The dummy variables 𝐼 𝑘 indicates the number of days 𝑘 since the repo-rate cut occurred.13 We consider
30 days prior to and following each policy rate cut. Because the lending variable is scaled, the dummy
coefficients, (the 𝛽 𝑘 ’s), have the interpretation of reflecting the degree of pass-through. For instance, a value
of 𝛽10 = −75) means there has been 75 percent pass-through of the policy rate to the lending rate 10 days after
12The floating mortgage rate receives a weight of 0.7 to reflect its large market share, whereas the mortgage contracts with fixed interest
rate periods of one year, three years and five years receives a weight of 0.3/3 each.
𝑙 = 𝛼 + P30
13Note that we could also have used the bank level lending rates, and instead run the regression 𝑖 𝑖,𝑡 𝑘=−30 𝛽 𝑘 𝐼 𝑘 + 𝜖 𝑖,𝑡 (with
standard errors clustered at the bank level and using weights to capture bank 𝑖’s market share). This leads to the same coefficient
estimates, but results in smaller confidence intervals. We therefore use the more conservative approach of not using the panel
dimension of the data in the regression.
11
the cut in the policy rate.
The third term in the regression addresses the question of whether the DLP changes the pass-through of
the policy rate to the lending rate. This term interacts the daily indicator variable 𝐼 𝑘 with a dummy variable
capturing whether the DLB is binding or not, 𝐼 𝑘𝐷𝐿 𝑃 . The coefficients 𝛽 𝐷𝐿𝐵
𝑘 ’s thus measure if there has been a
change in pass-through at the DLB. The sum 𝛽 𝐷𝐿𝐵
𝑘 + 𝛽 𝑘 measures the total pass-through of policy rate cuts
once the DLB becomes binding.
The regression results are reported in Table 1. To conserve space we only report the coefficient estimates
for every five days. Consider first the second column, which reports estimated values of 𝛽 𝑘 . In normal times,
i.e. from 2009 and until the deposit rate reaches its lower bound in late 2015, the lending rate starts falling five
days prior to the policy rate reduction. At the day of the policy rate change, however, the average lending rate
has fallen by 40 percent as much as the policy rate. Thirty days after the policy change, the pass-through has
reached 78 percent.
The third and fourth columns contain the main empirical result of the paper. The third column reports the
effect of the DLB on the pass-through of policy rate cuts. The estimated coefficient is positive and statistically
significant from the day of the policy rate cut and throughout the event window. This suggest that the DLB has
a statistically significant negative effect on the pass-through of policy rates to lending rates. Moreover, the
fourth column indicates that the point estimate of the effect of the DLB is strongly economically significant.
Policy rate cuts, once the DLB is binding, have no significant pass-through to lending rates and the point
estimates suggest that, if anything, further policy rate cuts increase lending rates.
Figure 5 shows the observed interest rates for the policy rate cuts at the DLB compared to the average
pass-through (and its associated confidence interval) when the policy rate is in positive territory. Each black
dot captures the average pass-through 𝑘 days after the repo-rate cut, and corresponds to the numbers in the
second column of Table 1. The gray area captures a 95 percent confidence interval around the estimates. The
red and blue lines depict average bank lending rates for the two final policy rate cuts, i.e. the policy changes
which occur in the post-bound period. In both cases, there is a slight increase in bank lending rates, in stark
12
contrast to the normal pass-though.14
50
0
-50
-100
Estimate 95% CI
Repo to -0.35 Repo to -0.50
One important question is whether the evolution of the listed rates used here provides an accurate picture
of the evolution of the actual rates borrowers face. For instance, Erikson and Vestin (2019) makes the point that
observed lending rates - in contrast to listed rates - fell in 2017 and 2018 and attribute this to policy rate cuts
in 2015 and 2016. In Appendix A, we aggregate the data and show that there are minimal differences between
the implied aggregate listed rate and the observed actual rates for the sample period. Moreover, we show that
the reductions in observed lending rates in 2017 and 2018 and the associated divergence between listed and
actual rates is most likely due to macroprudential regulation ultimately affecting the pool of borrowers, rather
than a (much delayed) response to policy rate cuts.
In addition to a decline in the average pass-through, there is an increase in dispersion once the deposit
rate has reached its lower bound. Moreover, this dispersion appears connected with the differential reliance of
banks on deposit financing. To illustrate this, we estimate the following regression
where pass-through is defined as before, 𝑖 denotes the bank and 𝑟 indexes one of the last two repo-rate
cuts. Data on deposit shares are from Statistics Sweden. We use deposit shares as of 2014, in order to not
capture any changes in deposit shares in response to the policy rate cuts.15 Observations are weighted by bank
14In Appendix Figure 19 we include plots similar to Figure 5 for mortgage contracts with fixed interest rate periods of 3 months, 1 year,
3 years and 5 years separately. They all show a collapse in pass-through once the deposit rate has reached its lower bound.
15This data is available for nine of the banks in the sample.
13
size. In the baseline regression, the deposit share is a continuous variable. As an alternative approach, we also
include a specification where the treatment variable is an indicator variable for whether bank 𝑖 has a deposit
share above the sample median.
The regression results are reported in Table 2. They indicate a statistically significant negative correlation
between deposit shares and pass-through, despite the low number of observations. From the first column, we
see that a ten percentage point increase in the deposit share is associated with a decrease in pass-through of
ten percentage points.16
(1) (2)
Pass-through Pass-through
Deposit share -0.957∗∗ -12.49
(-2.15) (-1.68)
Deposit share variable: continuous high or low
𝑁 18 18
Table 2: Pass-through and deposit shares.
Notes: This table shows regression results from estimating equation (3). t statistics in parentheses. * 𝑝 < 0.1, ** 𝑝 < 0.05, ***
𝑝 < 0.01.
Figure 6 illustrates the negative correlation between pass-through and deposit share, which suggests that
banks which are more reliant on deposit financing are less willing to reduce their lending rates once the DLB
is reached. Because the pass-through to deposit rates has been smaller than the pass-through to other financing
sources, banks which rely heavily on deposit financing are likely to experience smaller reductions in their total
funding costs once rates turn negative, a key feature of the model we present in section 2.
40 50 60 70 80
Deposit share (%)
16When using only an indicator variable for having low or high deposits in the second column, the regression coefficient becomes
borderline insignificant. However, the coefficient estimate suggests that on average, banks with high deposit reliance have twelve
percentage points lower pass-through than banks with low deposit reliance.
14
Identification We interpret the above event study to capture the impact of policy rate cuts on lending rates.
However, one might worry about potential confounding factors. As we use daily data it seems unlikely that
macroeconomic developments should influence our pass-through estimates. However, other monetary policy
announcements, such as quantitative easing (QE) or forward guidance, could potentially affect our findings.
The Riksbank adopted QE and negative rates simultaneously in February 2015. Quantitative easing was
subsequently increased in March 2015 and October 2015, when the policy rate was reduced further to -0.25
and -0.35 percent respectively. There was no further QE announcements related to the last policy rate cut to
-0.50 percent in February 2016. Could quantitative easing explain the breakdown in pass-through? First, we
note that the intention behind QE would be to lower lending rates. Hence, this policy would tend to work
against our results. Second, the timing of QE in Sweden does not coincide with the DLP, and hence with the
collapse in pass-through to lending rates. While quantitative easing was in place for three of the four policy
rate cuts in negative territory, the DLP was binding for the two last policy rate cuts only. The collapse in
pass-through to lending rates therefore, does not coincide with any changes to QE policies.
Another potential confounding factor is forward guidance and/or information about underlying economic
conditions. We find it plausible that the introduction of negative rates may have had an element of forward
guidance, as well as some signaling value about economic conditions. However, in order to explain the
breakdown in pass-through to lending rates, this (negative) effect must have materialized not when negative
rates were initially implemented, but rather for the two last policy rate cuts only. That is, reducing the interest
rate to -0.35 and -0.50 percent must have entailed a substantially worse signal about economic conditions
than reducing the policy rate to -0.10 and -0.25 percent. Moreover, it is unclear why this would generate a
correlation between pass-through and deposit shares. Hence, it seems more plausible that the breakdown in
pass-through is linked to the DLP becoming binding for the last two policy rate cuts.
To summarize, our daily data and the lack of confounding factors which occurred simultaneously with
the DLP becoming binding makes us confident that the event study identifies the collapse in pass-through
resulting from the deposit rate no longer responding to further policy rate cuts.
As we have documented a negative relationship between lending rate pass-through and deposit shares,
we would expect banks with higher deposit shares to also have lower lending growth once the DLB is binding.
Monthly lending volumes for household lending are available at the bank-level from Statistics Sweden, and we
focus on the period from January 2014 to December 2017. We restrict our sample to only include the banks
used in the previous analysis, and run the following regression
where Δ log(lending𝑖𝑡 ) is the 3-month percentage growth in lending rates for bank 𝑖 at the monthly date
𝑡, 𝛼𝑖 captures bank fixed effects to control for bank specific factors and 𝛿𝑡 captures time fixed effects to control
for factors common to all banks. Standard errors are clustered at the bank level and observations are weighted
by bank size.
The regression results, reported in Table 3, suggest that banks with higher deposit shares had lower
15
lending growth once the DLB became binding. The first column says that a ten percentage points increase in
the deposit share, say from 50 to 60 percent, reduces lending growth by on average 2.1 percentage points.
This compares to a mean lending growth of 2.2 percent in 2014. The second column compares banks with
above or below median deposit shares, and says that banks with above median deposit shares on average have
3.2 percentage points lower lending growth than banks with below median deposit shares in the post-bound
period. Hence, we conclude that not only are banks which rely more heavily on deposit financing likely to
increase mortgage rates in response to policy rate cuts once the DLB is reached, they also experience lower
growth in household lending volumes.
(1) (2)
Δ log(loans) Δ log(loans)
Post × deposit share -0.217**
(0.0955)
Post × deposit share -3.235**
(1.184)
N 361 361
No. of clusters 10 10
Mean of dependent variable 2.231 2.231
SD of dependent variable 7.052 7.052
Bank FE Yes Yes
Time FE Yes Yes
measure Cont. High or low
Table 3: Regression results from estimating equation (4).
Notes: * p<0.1, ** p<0.05, ***p<0.01. Post = 1 for the period when the repo-rate is below -0.25, and zero otherwise. Standard errors
clustered at the bank level. Summary statistics taken over whole sample period (2014 - 2017).
A potential concern is that the estimated coefficients in Table 3 reflect structural differences in lending
growth across banks with different deposit shares, and as such is not related to policy rate cuts at the DLB. To
ensure that this is not the case, we conduct two falsification tests where we investigate the lending response
of banks for earlier periods with policy rate reductions, including the initial cut into negative territory. The
results are reported in Table 6 in Appendix B. Reassuringly, the deposit share is only informative about the
response of bank lending to policy rate cuts once the DLB is binding.
17The four banks are Nordea, SEB, Handelsbanken and Swedbank, and the stock market data can be found at the Nordic Nasdaq. There
are three other publicly listed banks or credit companies on the Swedish stock exchange (Arion Banki, Avanza and TF Bank), but the
combined market share of these three banks is virtually zero, and so we drop them from our sample.
16
To formalize this we run the regression specified in equation (5), in which we define the excess return on
bank stock 𝑖 as Excess Return𝑖𝑡 = [log (Stock price) 𝑖,𝑡 −log (Stock price) 𝑖,𝑡−1 ] ×100 - [log (Stock price) index
𝑡 −
log (Stock price) index
𝑡−1 ] × 100.
post-zero
Excess Return𝑖𝑡 = 𝛼 + 𝛾 𝑘 Event Day 𝑘 + 𝛽 𝑘 Event Day 𝑘 × 𝐼𝑡 + 𝜖 𝑖𝑡 (5)
The 𝛽ˆ 𝑘 coefficients capture the differential effect on bank stock excess returns in the post-zero period and
are plotted in Figure 7. On the announcement day, bank stocks on average have an excess return of almost
-1.5 percentage points in the post-zero period relative to normal times.18 This negative effect is statistically
significant at the five percent level, despite our small sample size. We thus conclude that policy rate cuts into
negative territory have a detrimental impact on the excess return on Swedish bank stocks, suggesting that
market participants view them as bad news for the profitability of Swedish banks.
-4 -3 -2 -1 0 1 2 3 4
3 Theoretical analysis
We now move on to construct a model of bank lending with negative policy rates based on the preceding
empirical analysis. We first consider a partial equilibrium model where all interest rates are exogenous. This
setting allows us to clarify a number of issues, including the existence of a policy rate bound (PLB) and how
the presence of a deposit rate bound (DLB) affects the transmission of monetary policy also in absence of
a lower bound on the policy rate. We then embed the partial equilibrium model into a general equilibrium
framework in order to close the model and endogenize all interest rates. To conserve space, much of the
derivation of the model is relegated to Appendix C.
18The significant negative effect on the announcement day is robust to replacing 𝐼 𝑝𝑜𝑠𝑡−𝑧𝑒𝑟 𝑜 with 𝐼 𝐷𝐿𝐵 , but the results become more
noisy as we then only have 4 banks and 2 policy rate reductions in the post-period. Moreover, simply looking at the excess returns for
every policy rate cut seems to suggest that the negative excess returns materializes once the policy rate becomes negative.
17
3.1 Bank lending and negative policy rates in partial equilibrium
Consider a perfectly competitive risk-neutral bank that maximizes the discounted present value of
dividends
∞
X
max 𝐸 𝑡 𝛿 𝑡 𝐷 𝐼𝑉𝑡 (6)
𝑡=0
where 𝐷 𝐼𝑉𝑡 is dividends and 0 < 𝛿 < 1 is a discount factor. The bank maximizes the path of dividends subject
to the following flow budget constraint:
𝐷 𝐼𝑉𝑡 + 𝐵𝑡 + 𝑅𝑡 + 𝐴𝑡 + 𝑀𝑡 − 𝐷 𝑡 − 𝐹𝑡 (7)
𝑏
= (1 + 𝑖 𝑡−1 + (1 + 𝑖 𝑟𝑡−1 )𝑅𝑡−1
)𝐵𝑡−1 𝑎
+ (1 + 𝑖 𝑡−1 ) 𝐴𝑡−1
𝑑 𝑓
+𝑀𝑡−1 − (1 + 𝑖 𝑡−1 )𝐷 𝑡−1 − (1 + 𝑖 𝑡−1 )𝐹𝑡−1
−𝐶 (𝐹𝑡 , 𝐿 𝑡 , 𝑁𝑡 ) − Ψ(𝑅𝑡 , 𝑀𝑡 ) − Γ(𝐵𝑡 , 𝑁𝑡 ) − 𝑆(𝑀𝑡 )
following the notation in Figure 1, i.e. 𝐵𝑡 is loans the bank extends with interest 𝑖 𝑡𝑏 , 𝑅𝑡 is reserves with interest
𝑖 𝑟𝑡 , 𝑀𝑡 is paper currency which pays a zero return and 𝐴𝑡 is "other" liquid assets with interest 𝑖 𝑡𝑎 . To finance its
asset holdings, the bank raises funds via deposit-like financing 𝐷 𝑡 at an interest cost of 𝑖 𝑡𝑑 , and via other funds
𝑓
𝐹𝑡 at an interest cost of 𝑖 𝑡 . Throughout the analysis, 𝐷 𝑡 will capture funding that is subject to a lower bound,
whereas 𝐹𝑡 captures all other funding which in principle is not subject to a (strict) lower bound.
The net worth of the bank, 𝑁𝑡 , is defined as:
𝑓
𝑁𝑡 ≡ (1 + 𝑖 𝑡𝑏 )𝐵𝑡 + (1 + 𝑖 𝑟𝑡 )𝑅𝑡 + (1 + 𝑖 𝑡𝑎 ) 𝐴𝑡 + 𝑀𝑡 − (1 + 𝑖 𝑡𝑑 )𝐷 𝑡 − (1 + 𝑖 𝑡 )𝐹𝑡 (8)
while liquid assets are defined as
𝐿 𝑡 ≡ 𝑅𝑡 + 𝑀𝑡 + 𝐴𝑡 (9)
The bank can hold non-loan assets only in positive quantities, i.e. we impose the non-negativity constraints
𝐴𝑡 ≥ 0, 𝑅𝑡 ≥ 0, 𝑀𝑡 ≥ 0. (10)
Bank intermediation costs are captured by the four functions 𝐶 (.),Ψ(.), Γ(.) and 𝑆(.) corresponding
to different aspects of the bank’s intermediation process. Since the model is solved using a log-linear
approximation, only assumptions about the elasticity of each function with respect to their arguments are
needed to characterize the dynamics.
In line with the banking literature, e.g. Freixas and Rochet (2008), we assume that banks hold liquid assets
to insure against liquidity risk. Liquidity risk arises because loans 𝐵𝑡 are illiquid while external financing 𝐹𝑡
is subject to refinancing risk. The cost of liquidity risk is captured by the function 𝐶 (𝐿 𝑡 , 𝐹𝑡 , 𝑁𝑡 ).19 To capture
the costs associated with liquidity risk and banks’ insurance motive in a reduced-form way, we assume that
𝜕𝐶 𝜕𝐶 𝜕𝐶
> 0, < 0 and ≤ 0. Liquid assets decreases the costs associated with liquidity risk up until a
𝜕𝐹 𝜕𝑁 ∗ 𝜕𝐿
satiation point 𝐿 (𝑁𝑡 , 𝐹𝑡 ) and has zero impact for 𝐿 > 𝐿 ∗ . Away from this satiation point, the elasticity of 𝐶
with respect to 𝐹 is 𝛾 𝑓 ≥ 0 and the negative of the elasticities of 𝐶 with respect to 𝐿 and 𝑁 are denoted by
19See Freixas and Rochet (2008) for micro foundations of a cost function that captures liquidity risk.
18
𝛾𝑙 ≥ 0 and 𝛾𝑛 ≥ 0.20 We assume that the cost associated with liquidity risk is decreasing in net worth.21
While all liquid assets reduce liquidity risk, reserves and paper currency also contribute to reducing
banks operational costs due their special "money role," e.g. in settling inter-bank transactions and servicing
the public’s demand for cash. We capture this by the function Ψ(.). For simplicity, money and reserves are
perfect substitutes for the purpose of liquidity provision.22 Define the monetary base 𝑀 𝐵𝑡 ≡ 𝑅𝑡 + 𝑀𝑡 . We
assume that Ψ 0 (𝑀 𝐵𝑡 ) ≤ 0 up to a satiation point - denoted 𝑀 𝐵∗ - and Ψ 0 (𝑀 𝐵𝑡 ) = 0 for 𝑀 𝐵𝑡 > 𝑀 𝐵∗ . The
negative of the elasticity of the banks intermediation function with respect to 𝑀 𝐵𝑡 , when 𝑀 𝐵𝑡 < 𝑀 𝐵∗ , is
denoted 𝛾𝑚𝑏 ≥ 0.23
Holding money involves a cost captured by 𝑆(𝑀𝑡 ), with 𝑆 0 (𝑀𝑡 ) > 0. A simple interpretation of this cost
is that it involves the storage costs of money. More generally it should be thought of as the additional cost of
using paper currency relative to reserves in financial intermediation. The elasticity of the storage costs with
respect to M is denoted 𝛾𝑚𝑠 .24. As holding currency is costly, while holding reserves is not, banks choose
not to hold currency whenever the interest on reserves is positive. It may, however, choose to increase its
currency holdings once the reserve rate becomes negative. In this case the bank needs to consider storage
costs of holding cash as discussed in section 1.25
𝜕Γ 𝜕Γ 𝜕2Γ
The function Γ(𝐵𝑡 , 𝑁𝑡 ) captures the cost of lending. We assume that
> 0, < 0 and < 0.
𝐵 𝜕𝑁 𝜕𝐵𝜕𝑁
There is a variety of ways to micro-found these assumptions. The simplest is through regulatory capital
requirements. 26 One simple interpretation of why intermediation costs are strictly increasing in lending is
that they arise due to borrower default, see Curdia and Woodford (2011), e.g. due to capacity constraints in
loan monitoring. The elasticity of this cost function with respect to lending is denoted by 𝜈, while the negative
of the elasticity of lending with respect to net worth is 𝜄.27
We assume that the bank pays out a fixed fraction 𝜔 of its 𝑡 − 1 net worth in dividends.28 The flow
20Thus 𝛾 𝑓 ≡ 𝜕𝐶 𝐹 𝜕𝐶 𝑁 𝜕𝐶 𝐿
𝜕𝐹 𝐶 , 𝛾 𝑁 ≡ − 𝜕𝑁 𝐶 ,𝛾 𝐿 ≡ − 𝜕𝐿 𝐶 .
21This assumption is only important for Proposition 5 in the appendix.
22This simplification is not as restrictive as it may seem, as any relative disadvantage of cash to reserves can be captured by a storage
cost function S().
𝜕Ψ 𝑀 𝐵 .
23Thus 𝛾𝑚𝑏 ≡ 𝜕𝑀 𝐵 Ψ
¯
24Thus 𝛾𝑚𝑠 = 𝜕𝑀 𝑀¯ .
𝜕𝑆
𝑆
25Assuming that banks do not hold cash is a harmless abstraction as vault cash is a trivial component of banks balances. As currency is
an asset with a zero return, however, explicitly accounting for it is critical when thinking about the central banks ability to charge
negative interest rates on reserve balances since banks can always substitute reserves for cash – giving rise to the PLB
26The simplest interpretation of the role of net worth is that it comes about due to a capital requirement, often modeled as
𝑁𝑡
≥𝜅
𝐵𝑡
where 𝜅 is a parameter. A constraint of this form would be a limiting case of the smooth function assumed here, one in which the cost
of intermediation goes to infinity if the constraint is breached. The dependence of lending costs on net worth is micro-founded in
Holmstrom and Tirole (1997) and Gertler and Kiyotaki (2010), as well as documented empirically, for example, in Jimenez, Ongena,
Peydro, and Saurina (2012).
𝜕Γ 𝐵 and
27Thus, 𝜈 ≡ 𝜕𝐵 Γ
𝜄 ≡ − 𝜕𝑁 𝑁
𝜕Γ
Γ .
28A common specification in the literature, see e.g. Curdia and Woodford (2011), assumes that dividends are fully paid out within each
period. In that case, net worth is always zero. The generalization considered here is important for our application since negative
19
constraint can then be simplified to29
1
𝑁𝑡 = (1 − 𝜔)𝑁𝑡−1 + 𝑍𝑡 (11)
1 + 𝑖 𝑡𝑑
where 𝑍𝑡 is the profit of the bank, i.e.
𝑓
𝑖 𝑡𝑏 − 𝑖 𝑡𝑑 𝑖 𝑟𝑡 − 𝑖 𝑡𝑑 𝑖 𝑡𝑎 − 𝑖 𝑡𝑑 𝑖 𝑡𝑑 𝑖 𝑡 − 𝑖 𝑡𝑑
𝑍𝑡 ≡ 𝐵𝑡 + 𝑅𝑡 + 𝐴𝑡 − 𝑀𝑡 − 𝐹𝑡 −𝐶 (𝐹𝑡 , 𝑅𝑡 , 𝐴𝑡 , 𝑁𝑡 )−Γ (𝐵𝑡 , 𝑁𝑡 )−Ψ(𝑅𝑡 , 𝑀𝑡 )−𝑆(𝑀𝑡 )
1 + 𝑖 𝑡𝑑 1 + 𝑖 𝑡𝑑 1 + 𝑖 𝑡𝑑 1 + 𝑖 𝑡𝑑 1 + 𝑖 𝑡𝑑
(12)
Equation (11) says that net worth today depends on the retained earnings from last period plus the current
profits of the bank.
Denoting the value of the bank at time 𝑡 by 𝑉 (𝑁𝑡−1 ), the bank’s problem is
𝑉 (𝑁𝑡−1 ) = max (𝜔𝑁𝑡−1 + 𝛾𝑉 (𝑁𝑡 ))
𝐵𝑡 ,𝑅𝑡 , 𝐴𝑡 , 𝑀𝑡 ,𝐹𝑡
s.t. (10) and (11). The full details of the solution to the bank’s problem is included in Appendix C. In what
follows, we characterize the key components.
The first order condition that determines optimal bank lending is:
𝑖 𝑡𝑏 − 𝑖 𝑡𝑑
= Γ𝐵 (𝐵𝑡 , 𝑁𝑡 ) (13)
1 + 𝑖 𝑡𝑑 | {z }
| {z }
Marginal cost of
Marginal benefit of
lending
lending
This condition is the key equation of the model because it determines the lending activities of the banking
sector. The left hand side denotes the marginal benefit of lending, given by the spread between the rate the
bank obtains from extending loans (which it takes as given in partial equilibrium), and the interest on deposits,
𝑖 𝑡𝑑 .
Importantly, despite the presence of another source of external financing, the deposit rate captures
the marginal cost of funding for banks. The reason is that external financing carries refinancing risk. In
equilibrium, at any interior solution, the bank is indifferent between 𝐷 𝑡 and 𝐹𝑡 . Put differently, the bank
𝑓 𝜕𝐶
equates the marginal cost of deposits 𝑖 𝑡𝑑 with the marginal cost of external financing, i.e. 𝑖 𝑡 + . Hence, a
𝜕𝐹
𝑓
reduction in 𝑖 𝑡 relative to 𝑖 𝑡𝑑 leads to a shift in financing from deposits 𝐷 𝑡 to external financing 𝐹𝑡 , while
keeping the marginal cost of funding fixed at 𝑖 𝑡𝑑 .30 The right hand side reflects the marginal cost of lending
which is increasing in B.31
policy rates can have a negative effect on net worth, and via this channel, lending.
29To obtain this characterization use the expression for net worth 8 and solve for 𝐷 𝑡 and substitute for 𝐷 𝑡 in the flow constraint.
30Note that, even though marginal costs of the two funding sources are equalized, this does not mean that the presence of external
financing does not affect bank outcomes. If banks shift from 𝐷 to 𝐹, for instance when the policy rate is negative and the rate on 𝐷 is
at the DLB, the overall profitability of the bank is affected.
31An alternative to the set-up considered here would be to assume that banks engage in monopolistic competition in lending markets, as
in for instance Ulate (2019). In such a setting, loans would not be priced according to marginal costs but rather set as a mark-up over
the interest rate on liquid assets plus any interaction between lending and net worth, as for instance captured by a capital constraint. If
the feedback from net worth to lending is small, this would imply that lending rates could fall when the policy rate is reduced, even
though the deposit rate is at the DLB. Note, however, that such a scenario is not consistent with the empirical evidence in section 2.
The empirical evidence in section 2 is, however, consistent with monopolistic competition and a strong feedback from net worth to
lending rates. For lending and ultimately aggregate demand, such a case is isomorphic to what we consider here.
20
Before moving on to analyzing the effect of changes to the policy rate, it is worth asking: Is there a bound
on how low the policy rate can go? The next section addresses this question.
The policy rate bound ("PLB") arises in the model because the banks have the option of exchanging
reserves for paper currency. The PLB thus depends heavily on the cost for the banks of holding paper currency
relatively to reserves, captured by the function 𝑆(.). To understand how it emerges, it is helpful to consider the
following specification
𝑆(𝑀𝑡 ) = 𝛼 𝑚 𝑀𝑡 . (14)
Suppose that the DLB is zero. In this case Proposition 1 defines the policy rate bound.
Proposition 1. (The Policy Rate Bound) If the cost of using paper currency is given by equation (14), the
DLB is zero, then the lower bound on the policy rate is
𝑖 𝑟𝑡 ≥ 𝑖 𝑃𝐿𝐵 ≡ −𝛼 𝑚 (15)
The proof of this proposition follows directly from the first order conditions of the banking problem,
and shown in Appendix C.2. Since banks value the services central bank reserves provide, the central bank
can charge a negative interest rate on reserves. However, money also provides these services, but at a higher
relative cost captured by 𝑆(𝑀𝑡 ) = 𝛼 𝑚 𝑀𝑡 . If the central bank charges a too high price for the service provided
by reserves, the banks withdraw their reserves in favor of paper currency, using it to insure against liquidity risk
and settle interbank transactions outside of the central bank. The cost function in equation (14) is proportional,
so that if the interest rate charged on reserves (e.g. -2 percent) is lower than the negative of the marginal cost
of storing cash captured by −𝛼 𝑚 (e.g. -1.5 percent), the bank will withdraw all reserves. If the marginal
storage cost is increasing, then as the central bank lowers the reserve rate, banks convert reserves into cash. If
the cost of holding cash increases without a bound, there is in principle no PLB.
Empirically, there has been somewhat limited substitution from reserves to cash, see Figure 20 in the
appendix. It is therefore to assume that the PLB is below rates seen so far. We therefore proceed under the
assumption that the PLB does is non-binding.
We now consider the effect of policy rate cuts, under the assumption that the PLB is not binding, and
𝑓
keeping the lending rate, 𝑖 𝑡𝑏 fixed. Let us suppose that 𝑖 𝑡𝑑 , 𝑖 𝑡𝑎 and 𝑖 𝑡 , are functions of the policy rate 𝑖 𝑟𝑡 , with
reduced form pass-through coefficients 𝜌 𝑑 , 𝜌 𝑎 and 𝜌 𝑓 respectively. If 𝜌 𝑖 = 1 there is full pass-through to the
interest rate 𝑖, while 𝜌 𝑖 = 0 indicates zero pass-through. For instance, if the deposit rate is not responding to
further policy rate cuts due to the DLB, then 𝜌 𝑑 = 0. In general equilibrium, the pass-through of the policy
rate to all rates is determined endogenously. Yet, most of the insights can be illustrated in partial equilibrium.
We use a log-linear approximation of the model around a steady state in which there is full pass-through
so that 𝚤¯𝑟 = 𝚤¯𝑑 = 𝚤¯ 𝑓 = 𝚤¯𝑎 , using bars to denote steady state values. The full set of log-linearized equations are
21
listed in Appendix C. The following proposition highlights a useful observation.
Proposition 2. (Full pass-through and liquidity satiation) If there is full pass-through in steady state, i.e.
𝚤¯𝑟 = 𝚤¯𝑑 = 𝚤¯ 𝑓 = 𝚤¯𝑎 , then the bank is satiated in liquid assets (𝐿 = 𝐿 ∗ ) and holds no paper currency (𝑀 = 0).
The proof of this proposition follows directly from the first order condition of the banks problem,
i.e. equations (54)-(56) in Appendix C. Together, these equations imply that if 𝑖 𝑓 = 𝑖 𝑎 = 𝑖 𝑑 = 𝑖 𝑟 , then
𝐶 𝐿 = 𝐶𝐹 = Ψ𝑅 = 0, i.e. the bank is satiated in liquid assets. The steady state values for external financing
𝐹, other liquid assets 𝐴 and central bank reserves 𝑅 are then implicitly defined by equations (54)-(56) in
Appendix C. Because paper currency and reserves serve the same role, but money generates storage costs,
banks choose to hold no paper currency.
We now move on to analyze the impact of negative policy rates on bank lending in partial equilibrium,
and how it depends on the transmission of the policy rate 𝑖 𝑟𝑡 to other interest rates, i.e. the pass-through
coefficients 𝜌 𝑖 for 𝑖 ∈ {𝑑, 𝑎, 𝑓 }.
A log-linear approximation of (11) and (13) around the steady state yields
1 1 + 𝚤¯𝑏 𝜄 ˆ
𝐵ˆ 𝑡 = ( 𝑏 𝑑 )(ˆ𝚤 𝑡𝑏 − 𝚤ˆ𝑡𝑑 ) + 𝑁𝑡 (16)
𝜈 − 1 𝚤¯ − 𝚤¯ 𝜈−1
1 + 𝚤¯𝑑 𝐵¯
𝑁ˆ 𝑡 = {(1 − 𝜔) 𝑁ˆ 𝑡−1 + 𝚤ˆ𝑡𝑏 + Ω𝜌 𝚤ˆ𝑟𝑡 } (17)
Θ 𝑁¯
where 𝑁ˆ 𝑡 ≡ log 𝑁𝑡 /log 𝑁,
¯ 𝐵ˆ 𝑡 ≡ log 𝐵𝑡 /log 𝐵,
¯ 𝚤ˆ𝑡𝑑 ≡ log(1 + 𝑖 𝑡𝑑 )/(1 + 𝚤¯𝑑 ), 𝚤ˆ𝑡𝑏 ≡ log(1 + 𝑖 𝑡𝑏 )/(1 + 𝚤¯𝑏 ), and
¯
Θ ≡ 1 − 𝜈𝜄 (¯𝚤 𝑏 − 𝚤¯𝑑 ) 𝑁𝐵¯ > 0. The key coefficient is Ω𝜌 , discussed in detail below.
In the remaining analysis, we make the following assumption
𝜄 (1 − 𝜔)(1 + 𝚤¯𝑑 ) 𝑁¯
< (18)
𝜈 𝚤¯𝑏 − 𝚤¯𝑑 𝐵¯
which says that the feedback effect from changes to net worth to lending is not too strong, which is a necessary
condition for the approximate solution to have a unique bounded solution. Observe that it implies that Θ > 0.
An approximated partial equilibrium can be defined as a collection of processes for { 𝑁ˆ 𝑡 , 𝐵ˆ 𝑡 } which solve
𝑓
equations (16) and (17) given an exogenous sequence for {ˆ𝚤 𝑟𝑡 , 𝚤ˆ𝑡𝑑 , 𝚤ˆ𝑡 , 𝚤ˆ𝑡𝑎 , 𝚤ˆ𝑡𝑏 }. An important observation is that
there is no first-order impact of changes in non-loan asset holdings on the bank’s lending decision determined
by equation (16). Yet, as will be clear below, the steady state values of these variables are important via their
impact on net worth through Ω𝜌 .
The first term in equation (16) captures that bank lending is an increasing function of the spread between
borrowing and deposit rates. Since 𝑖 𝑡𝑑 = 𝜌 𝑑 𝑖 𝑟𝑡 , a lower policy rate translates into lower funding costs for the
banks in proportion to the coefficient 𝜌 𝑑 . If 𝜌 𝑑 = 0, lower policy rates are not translated into lower deposit
rates and thus the major financing cost channel of bank lending is unaffected by the decline in the policy rate.
The second term shows that lending is increasing in net worth. This follows directly from the assumption that
intermediation costs depend negatively on net worth. The strength of this force depends on both elasticities
𝜄 and 𝜈 .i.e. by how much intermediation costs change with higher net worth and increased lending. To
understand the total effect of a policy rate cut on bank lending, we thus need to understand the effect on net
worth.
22
Equation (17) summarizes the evolution of net worth as a function of past net worth, the lending rate and
the policy rate. Net worth depends positively on its own lagged value as well as the lending rate. What about
the effect of the policy rate on net worth? The key coefficient is Ω𝜌 , defined as
𝐴¯ 𝑎 𝑅¯ 𝐹¯ 1 + 𝚤¯𝑏 𝐵¯ 1
Ω𝜌 ≡ (𝜌 − 𝜌 𝑑 ) + (1 − 𝜌 𝑑 ) − (𝜌 𝑓 − 𝜌 𝑑 ) − 𝜌 𝑑 ( − )(19)
𝑁 ¯ 𝑁¯ 𝑁¯ ¯
1 + 𝚤¯ 𝑁 1 + 𝚤¯𝑑
𝑑
This coefficient summarizes how changes in policy rates are translated into net worth through each
𝐴¯
component of the bank’s balance sheet. Consider the term 𝑁¯
(𝜌 𝑎 − 𝜌 𝑑 ), for instance. If the policy rate cut
changes the interest on deposits by more than the interest on liquid assets, i.e. 𝜌 𝑎 < 𝜌 𝑑 , then a policy rate cut
increases the net interest margin on liquid asset holdings, which in turn contributes to increasing the net worth
of the bank. The second term reflects this balance sheet effect due to reserves, the third term due to external
financing and the final term due to bank lending.
Using equations (16) and (17), it is straightforward to show that the effect of a policy rate cut on lending
at time 𝑡 is
𝜕 𝐵ˆ 𝑡 𝑑 1 1 + 𝚤¯𝑏 1 + 𝚤¯𝑑 𝜄
= −𝜌 ( ) + Ω𝜌 (20)
𝜕ˆ𝚤 𝑟𝑡 𝜈 − 1 𝚤¯𝑏 − 𝚤¯𝑑 Θ 𝜈−1
| {z } | {z }
Marginal funding cost channel Bank capital channel
which then continues to affect future borrowing at time 𝑡 + 𝑗 via the effect on net worth in future periods.
In our model, policy rate cuts can therefore stimulate lending via two channels. The first channel arises
because lower policy rates lower the marginal funding cost of banks. This impact relies on the pass-through to
deposit rates 𝜌 𝑑 . Once there is no pass-through to deposit rates, this mechanism shuts down, as long as the
bank remains indifferent between deposits and other sources of financing. However, policy rate cuts can still
increase lending via a bank capital channel captured by Ω𝜌 . If lower policy rates expand bank profitability,
higher net worth will induce banks to lend more even if the deposit rate is unchanged.
The next proposition considers the case when there is full pass-through to all interest rates.
Proposition 3. (Policy rate changes with full pass-through). Suppose that there is full pass-through to all
𝜕 𝐵ˆ𝑡
interest rates so that 𝜌 𝑎 = 𝜌 𝑑 = 𝜌 𝑓 = 1. Then a reduction in 𝚤ˆ𝑟𝑡 increases bank lending, i.e. 𝑟 < 0
𝜕 𝑖ˆ𝑡
1 1+¯𝚤 𝑏 𝐵¯
To prove this proposition, note that for this special case, Ω𝜌 = 1+¯𝚤 𝑑
− 1+¯𝚤 𝑑 𝑁¯
< 0. Hence, when there is
full pass through, the effect of policy rate cuts on bank lending is unambiguous.
This conclusion of the last proposition changes when there is no pass-through to the deposit rate, i.e.
𝜌𝑑 = 0. This corresponds to the case in which the DLB is binding, i.e. once policy rate reductions move
far enough into negative territory. In this case the marginal financing cost of the bank is no longer affected.
Whether or not a policy rate cut stimulates lending then depends on how net worth is affected.
Proposition 4. (Policy rate cut when DLB is binding). Suppose there is no pass-through to deposit rates so
that 𝜌 𝑑 = 0. Then a reduction in 𝚤ˆ𝑟𝑡 increases lending if
𝜌 𝑎 𝐴¯ + 𝑅¯ < 𝜌 𝑓 𝐹¯ (21)
23
and decreases lending otherwise.
In this section, we outline the remaining components of the model. We assume that there are "patient"
households which save and "impatient" households which borrow, intermediated by banks. Nominal frictions
enter via staggered price setting by firms as in Calvo (1983). The way households and firms are modeled
follows closely Benigno et al. (2014). Accordingly, the exposition of this aspect of the model is brief. Most
derivations, such as first order conditions, are relegated to the appendix.
24
Households
Borrowers (𝑏) make up a fraction 𝜒 of households, while savers (𝑠) make up the remaining share 1 − 𝜒.32
Households have at time 0 a utility function of the form
∞
" ! #
(𝑁𝑡 ) 1+𝜂
𝑗 𝑗
𝑗
X
𝑗 𝑡 𝑗 𝑀𝑡
U𝑡 = E0 𝛽 𝑈 (𝐶𝑡 ) + 𝑚 − 𝜁𝑡 with 𝑗 = 𝑠 or 𝑏 (22)
𝑃𝑡 1+𝜂
𝑡=0
where 𝐸 𝑡 is the expectation operator, 𝛽 𝑗 is the discount factor with 0 ≤ 𝛽 𝑏 ≤ 𝛽 𝑠 < 1, 𝜁𝑡 is a preference shock
𝑗
and 𝐶𝑡 is aggregate consumption
𝜃
ˆ 1 𝜃−1
𝑗 𝑗 𝜃−1
𝐶𝑡 ≡ 𝐶𝑡 (𝑖) 𝜃 𝑑𝑖
0
𝑗
where 𝐶𝑡 (𝑖) is the consumption of good of variety i, 𝜃 > 1 is the intratemporal elasticity of substitution
𝑗
between goods, 𝑁𝑡 is hours worked and 𝜂 ≥ 0 is a parameter. To facilitate aggregation, the utility function!
𝑗
𝑗 𝑗 𝑀𝑡
for consumption is assumed to be 𝑈 (𝐶𝑡 ) = 1 − exp(−𝑞𝐶𝑡 ) where 𝑞 is a parameter. The function 𝑚
𝑃𝑡
represents the utility of holding real money balances, which increases in real money balances up to a satiation
point 𝑀𝑡
𝑃𝑡 = 𝑚 ∗ , after which 𝑚 0 = 0.
The households’ budget constraint is:
𝑗 𝑗 𝑗 𝑗 𝑗 𝑗 𝑗
𝑀𝑡 − 𝐵𝑡 = 𝑊𝑡 𝑁𝑡 − 𝐵𝑡−1 1 + 𝑖 𝑡−1 + 𝑀𝑡−1
𝑗
𝑀 𝑗 𝑗 𝑗 𝑗 𝑗
−𝑃𝑡 𝑆 ℎ ( 𝑡 ) − 𝑃𝑡 𝐶𝑡 + Ψ𝑡 + 𝜓𝑡 − 𝑇𝑡 − 𝐹𝑡 (23)
𝑃𝑡
𝑗
where 𝐵𝑡 denotes a one period banking contract 𝑗. 𝐵𝑡𝑏 > 0 is debt with interest rate 𝑖 𝑡𝑏 , while 𝐵𝑡𝑠 < 0 is bank
deposits with interest rate 𝑖 𝑡𝑑 . 𝑆 ℎ ( 𝑀𝑃𝑡−1
𝑡
) captures the storage costs of holding money, measured in units of
𝑗 𝑗
the consumption good, Ψ𝑡 is firms profits, and 𝜓𝑡 is bank dividends. Firm profits are distributed to both
𝑗
household types based on their population shares while only savers receive bank dividends. 𝐹𝑡 represents
exogenous pension payments of each agent. This pension is the source of external funds on the bank’s balance
sheet.
The household maximize (22) subject to (23). This leads to the standard consumption Euler equations
for each household type, an optimal labor supply condition and transversality conditions, all of which are
relatively standard and reported in the appendix. In addition, the households rule for optimal holdings of
money gives rise to the lower bound on the deposit rate.
The deposit lower bound Consider the optimal money holdings of the saver. If 𝑚 𝑡 ≤ 𝑚¯ then money
demand is given by:
𝑀𝑡𝑠
𝑚0
𝑃𝑡 𝑖 𝑡𝑑 0 𝑀𝑡
𝑠
= − 𝑆 ( ) (24)
𝑈 0 (𝐶𝑡 )
𝑠
1 + 𝑖 𝑡𝑑 𝑃𝑡
25
while if 𝑚 𝑡 ≥ 𝑚¯ the household is satiated in money and
𝑀𝑠 𝑖𝑑
𝑆0( 𝑡 ) = 𝑡 𝑑 (25)
𝑃𝑡 1 + 𝑖𝑡
Equation (25) generates the bound on deposit rates.
𝑀𝑡𝑠
The simplest formulation of a storage costs is that it is proportional to money holdings, i.e. 𝑆( 𝑃𝑡 ) = 𝛼𝐻 𝑀𝑡
𝑃𝑡
for some 𝛼 𝐻 > 0. In this case condition (25) implies
𝛼𝐻
𝑖 𝑡𝑑 = ≡ 𝑖 𝐷𝐿𝐵
1 − 𝛼𝐻
which represents the effective lower bound on deposit rates for small depositors.
A straight forward interpretation of why banks have not been willing to impose negative rates on deposits
is that the marginal storage cost 𝛼 𝐻 is small for small depositors, i.e. 𝛼 𝐻 ≈ 0, and that for regular depositors
cash provides a close substitute to deposits. We maintain this assumption for the remainder of the paper, and
therefore set 𝛼 𝐻 = 0 which implies a DLB for the deposit rate at zero, i.e. 𝑖 𝑡𝑑 ≥ 𝑖 𝐷𝐿𝐵 = 0. Towards the end of
the paper we discuss policies that may relax this bound.
Firms
Each good 𝑖 is produced by a firm 𝑖. Production is linear in labor, i.e. 𝑌𝑡 (𝑖) = 𝑁𝑡 (𝑖), where 𝑁𝑡 (𝑖) is a
𝜒 𝑠 1−𝜒
Cobb-Douglas composite 𝑁𝑡 (𝑖) = 𝑁𝑡𝑏 (𝑖) 𝑁𝑡 (𝑖) as in Benigno et al. (2014). This implies that the
two household types receive a fixed share of income. The preference specification implies that firms face a
−𝜃
𝑃𝑡 (𝑖)
downward-sloping demand function 𝑌𝑡 (𝑖) = 𝑌𝑡 . In each period, a fraction 𝛼 of firms are not able
𝑃𝑡
to reset their price as in Calvo (1983). Thus, the likelihood that a price set in period 𝑡 applies in period 𝑇 > 𝑡
is 𝛼𝑇 −𝑡 and in the absence of price adjustments, prices are assumed to be indexed to the inflation target Π. A
firm that resets its price chooses the price that maximizes the present value of discounted profits in the event
that the price remains fixed. That is, each firm 𝑖 choose 𝑃𝑡 (𝑖) to maximize
∞
𝑡 𝑃0 (𝑖) 𝑊𝑡
X
𝑡
E0 (𝛼𝛽) 𝜆 𝑡 Π 𝑌𝑡 (𝑖) − 𝑌𝑡 (𝑖) (26)
𝑃𝑡 𝑃𝑡
𝑡=0
where 𝜆𝑇 ≡ 𝑞 𝜒 exp −𝑞𝐶𝑡𝑏 + (1 − 𝜒) exp −𝑞𝐶𝑡𝑠 , which is the weighted marginal utility of consumption
and 𝛽 ≡ 𝜒𝛽 𝑏 + (1 − 𝜒) 𝛽 𝑠 .33 This maximization problem leads to the standard New Keynesian Phillips Curve.
The banks problem has already been outlined in the partial equilibrium section and the first order
conditions (53) - (58) from the appendix has to hold in equilibrium. In general equilibrium, however, the
interest rates and the price level are endogenously determined.
In general equilibrium, the source of external funding for the banks are the pension funds that invest 𝐹𝑡
𝑓
and receive an interest rate 𝑖 𝑡 as well as government bonds 𝐴𝑡 that pay an interest 𝑖 𝑡𝑎 . These pension funds are
𝑗
financed by a lump sum pension fee, 𝐹𝑡 , in the households budget constraint. 𝐹𝑡 should be interpreted as
a stand-in for large investors, like pension funds, or large retail depositors which can possibly be charged a
33Recall that the firm is owned by both types of households according to their respective population shares.
26
negative interest rate, since storing assets in terms of paper currency might be prohibitively costly for this
group. Liquid assets in the form of government bonds, 𝐴𝑡 , are held by the banks as well as the households
through the pension funds.34
The government
The government sets monetary policy according to the Taylor type policy rule
Π𝑡 𝑦𝑡
𝑖 𝑡 = 𝑟 𝑡𝑛 ( ) 𝜙 𝜋 ( ) 𝜙 𝜋 (27)
Π̄ 𝑦¯
where 𝑟 𝑡𝑛 is the natural rate of interest which corresponds to the interest rate in the case of flexible prices, and
𝜙 𝜋 and 𝜙 𝑦 are coefficients. Π̄ is the inflation target of the central bank and 𝑦¯ is the natural level of output
which is constant in the model and equal to steady state output.
We assume that when interest on reserves is above the deposit bound, then 𝑖 𝑟𝑡 = 𝑖 𝑡𝑑 . However, once the
interest on reserves is below the DLB, then i𝑡𝑑 = 𝑖 𝐷𝐿𝐵 . Thus we impose the following constraint
𝑖 𝑡𝑑 = max{𝑖 𝐷𝐿𝐵 , 𝑖 𝑟𝑡 } (28)
which implies full pass-through to deposit rates, unless the effective lower bound is binding.
The government budget constraint is
𝑔 𝑔 𝑔
𝐴𝑡 + 𝑀𝑡 + 𝑅𝑡 = (1 + 𝑖 𝑡 ) 𝐴𝑡−1 + 𝑀𝑡−1 + (1 + 𝑖 𝑟𝑡−1 )𝑅𝑡−1 + 𝐺 − 𝑇𝑡
Total government liabilities 𝐿𝐵𝑡 are assumed to be fixed and follow an exogenous process, i.e,
𝑔
𝐿𝐵𝑡 = 𝐴𝑡 + 𝑀𝑡 + 𝑅𝑡 (29)
Taxes adjust so that the government budget constraint is satisfied, and are equally distributed across the two
agents in steady state. For simplicity, in response to shocks, only the tax on the saver is varied, so that
𝑇𝑡 = 𝑇¯ 𝑏 + 𝑇𝑡𝑠 (30)
An equilibrium is defined as a collection of stochastic processes for the endogenous variables that solve
the household problem, the firm problem, the bank problem, and monetary and fiscal policy follow the policy
regimes specified in the previous section. The equilibrium is defined in the appendix. The model is solved
via a log-linear approximation around the steady state, while explicitly taking of the DLB. From now on, we
rewrite the model in real terms. Lower case letters refer to the real value of their nominal (big case letters)
counterparts.
Before moving on to the numerical simulations, we characterize the log-linear approximated model
briefly. We show that our model can be written in the form of the standard three-equation New Keynesian
model, with two important differences: First, it is the deposit rate which enters the dynamic IS equation rather
34Observe that if the interest on government bonds goes negative, then individual household will substitute out of government bonds in
favor of deposits that carry zero interest.
27
than the policy rate. Second, the natural rate of interest is now endogenous and depends on credit market
outcomes.
The household and firm problems, together with the monetary policy rule, can be summarized as follows:
𝜋ˆ 𝑡 = 𝜅 𝑦ˆ 𝑡 + 𝛽𝐸 𝑡 𝜋ˆ 𝑡+1 (32)
28
where 𝑧ˆ𝑡 is bank’s profits given by
𝑛 𝛽 𝑏 𝑏¯ 𝑟¯ 𝑎¯ 𝑓¯ 𝛽 𝑏 𝑏¯ 𝑟¯ 𝑎¯ 𝑓¯ 𝑓 Γ̄
𝑧ˆ𝑡 = −{ 𝑠 + + − }ˆ𝚤 𝑡𝑑 + 𝑠 𝚤ˆ𝑡𝑏 + 𝚤ˆ𝑟𝑡 + 𝚤ˆ𝑡𝑎 − 𝚤ˆ𝑡 + 𝜄 𝑛ˆ 𝑡 (38)
Λ 𝛽 Λ̄ Λ̄ Λ̄ Λ̄ 𝛽 Λ̄ Λ̄ Λ̄ Λ̄ Λ̄
𝑓
1+𝑖 𝑎 1+𝑖
𝚤ˆ𝑡𝑎 ≡ log 1+¯𝚤𝑡𝑎 , 𝚤ˆ𝑡 ≡ log 1+¯𝚤𝑡𝑓 , 𝑧ˆ𝑡 ≡ 𝑧𝑡𝑛−¯ 𝑧¯ and 𝑛ˆ 𝑡 ≡ log 𝑛𝑛¯𝑡 . The profit equation is expressed in terms of
𝑓
1 1 1 𝑏 𝑦¯ 𝑦¯
𝑏ˆ 𝑡 = 𝑏 𝑏ˆ 𝑡−1 − 𝑏 𝜋ˆ 𝑡 + 𝑏 𝚤ˆ𝑡−1 − 𝜒 𝑦ˆ 𝑡 + 𝜒 𝑐ˆ𝑡𝑏 (39)
𝛽 𝛽 𝛽 𝑏¯ 𝑏¯
𝑐ˆ𝑡𝑏 = 𝐸 𝑡 𝑐ˆ𝑡+1
𝑏
− 𝜎(ˆ𝚤 𝑡𝑏 − 𝐸 𝑡 𝜋𝑡+1 − 𝜁ˆ𝑡 + 𝐸 𝑡 𝜁ˆ𝑡+1 ) (40)
We assume full pass-through from the policy rate to liquid assets and external financing away from the
DLB, but allow for partial pass-through at the DLB, i.e.:
𝚤ˆ𝑡𝑎 = 𝜌 𝑎 𝚤ˆ𝑟𝑡 if 𝚤ˆ𝑟𝑡 < 𝑖 𝐷𝐿𝐵 and 𝚤ˆ𝑡𝑎 = 𝚤ˆ𝑟𝑡 if 𝚤ˆ𝑟𝑡 > 𝚤¯𝐷𝐿𝐵 (41)
𝑓 𝑓
𝚤ˆ𝑡 = 𝜌 𝑓 𝚤ˆ𝑟𝑡 if 𝚤ˆ𝑟𝑡 < 𝑖 𝐷𝐿𝐵 and 𝚤ˆ𝑡 = 𝚤ˆ𝑟𝑡 if 𝚤ˆ𝑟𝑡 > 𝚤¯𝐷𝐿𝐵 (42)
𝑓
An approximate equilibrium is a collection of stochastic processes for prices { 𝜋ˆ 𝑡 , 𝚤ˆ𝑟𝑡 , 𝚤ˆ𝑡𝑑 , 𝚤ˆ𝑡 , 𝚤ˆ𝑡𝑎 , 𝚤ˆ𝑡𝑏 , 𝑟ˆ𝑡𝑛 } and
quantities { 𝑦ˆ 𝑡 , 𝑐ˆ𝑡𝑏 , 𝑏ˆ 𝑡 , 𝑛ˆ 𝑡 , 𝑧ˆ𝑡 } that solve equations (31) - (40) given an exogenous process { 𝜁ˆ𝑡 }.
Monetary policy affects aggregate demand in equation (31) through two channels. If the DLB is not
binding, the most direct one is via a reduction in the deposit rate, stimulating spending. The second is that
policy rate endogenously changes the natural rate of interest through the credit spread in equation (35).
To illustrate the key mechanism analytically, we make three simplifying assumptions to clarify a point
that holds more generally, namely that condition (21) for the net balance sheet exposure to negative policy
rates is once again the key determinant of whether negative policy rates are expansionary or not once the DLB
is binding.
Consider the case in which 𝜔 = 1, and 𝑏ˆ 𝑡 = 0 for 𝑡 > 0. 35 Finally, suppose that prices are fixed so that
𝜅 = 0. Consider now the effect of a one-time reduction in the policy rate at time 0, while the deposit rate is
unchanged (with the pass-through to liquid assets and external financing given by 𝜌 𝑎 and 𝜌 𝑓 ).
Solving (31) through (40) yields:
𝑓¯
( 𝑟𝑛¯¯ + 𝜌 𝑎 𝑎𝑛¯¯ − 𝜌 𝑓 𝑛¯ )
𝑏ˆ 0 = Υˆ𝚤 𝑟𝑡 (43)
𝛽 𝑏 𝑏¯ 1 𝛽𝑠
𝛽 𝑠 𝑛¯ + Θ{ 𝜈−1
𝜄 Υ+ 𝜄 𝛽 𝑠 −𝛽 𝑏 }
1
𝚤ˆ0𝑏 = − 𝑏ˆ 0 (44)
Υ
35The simplest way of thinking about the second assumption, is that in response to a shock in period 0 there are government lump sum
redistributions to bring debt back to steady state in period 1.
29
𝑦ˆ 0 = −𝜎 𝜒ˆ𝚤 0𝑏 (45)
where Υ ≡ 𝜎 𝜒{(1 − 𝜒) 𝑏𝑦¯¯ > 0 and the parameter Θ > 0 is defined as in the partial equilibrium model. Thus,
just as in the partial equilibrium model, policy rate cuts once the DLB is binding reduce lending according
to equation (44) if the condition in equation (21) is violated. In general equilibrium, however, as shown in
equation (44) this increases borrowing rates, in line with our empirical evidence from Sweden. Moreover, as
shown in equation (45), the result is a contraction in aggregate demand. While these results can be shown
more generally, the more fundamental question is what is the quantitative implication of negative policy rates.
An inspection of the equations above reveals that the model is fully parameterized via the assignment of
¯ 𝑓¯ 𝑟¯ 𝑎¯ 𝑦¯ Γ̄
the banks balance sheet parameters ( Λ̄𝑛¯ , Λ̄𝑏 , , , , , ),
Λ̄ Λ̄ Λ̄ 𝑏¯ 𝑏¯
the interest rate pass-through parameters 𝜌 𝑎 , 𝜌 𝑓 ,
the structural parameters, (𝜎, 𝜅, 𝛽 𝑠 , 𝛽 𝑏 , 𝜒, 𝜈, 𝜄, 𝜔), the parameters of the policy function (𝜙 𝜋 , 𝜙 𝑦 , Π̄), along
with the stochastic process for 𝜉𝑡 . We now turn to the parameterization and quantitative predictions of the
model.
The analysis so far suggests that a reduction in the policy rate, once the DLB is reached, contracts lending
if the transmission of negative policy rates to the asset side is larger than to the liability side, or
𝑟¯ 𝑎¯ 𝑓¯
+ 𝜌𝑎 > 𝜌 𝑓 (46)
Λ̄ Λ̄ Λ̄
|{z} |{z} |{z}
0.05 0.42 0.21
where the relevant components of the balance sheet are expressed as a ratio of total assets.
The values reported under the curly brackets in equation (46) come from Swedish data and suggest that –
in the Swedish case – banks exposure to negative policy rates is such that policy rate cuts at the DLB are
expected to reduce bank profits. The condition is calibrated using balance sheet data from 2014 to capture the
state of the banking system just prior to negative rates being implemented. The empirical counterpart of 𝑎¯ is
total assets less reserves and bank lending, 𝑓¯ is total liabilities less deposits and equity, 𝑟¯ is reserve balances
with the central bank and cash.
𝑛¯ ¯
As shown in the previous section, the approximated model also requires assigning values to Λ̄
and Λ̄𝑏 .
The values assigned to these ratios come from the same data source and are reported, along with the other
Γ̄
relevant balance sheet ratios in Table 4. In addition to these ratios, the ratio Λ̄
appears in the approximated
𝛽𝑠
−1
Γ̄ 𝑏¯ 𝛽𝑏
model. From the optimal lending conditions, the steady state of the model implies that Λ̄
= Λ̄ 𝜈 .
The pass-through to other liquid assets at the DLB (𝜌 𝑎 ) is assumed to be one, reflecting the strong pass-
through of negative rates to reserve-like assets such as money market instruments and short-run government
30
Banking Sector Moments Value Empirical Counterpart
𝑎
Liquid asset ratio = 0.42 Liquid assets to total assets of 42%
Λ
𝑟
Reserve ratio = 0.05 Reserves to total assets of 5 %
Λ
𝑓
External financing ratio = 0.53 External financing to total assets of 53 %
Λ
𝑏
Loans to total assets = 0.53 Loans to total assets of 53 %
Λ
𝑛
Net worth to total assets = 0.05 Net worth to total assets of 5 %
Λ
𝑧
Profit to net worth = 0.173 Profit to net worth of 17.3 %
𝑛
𝑏
Loans to GDP = 1.4 Loans to GDP of 140 %
𝑦
Pass-through coefficient to liquid assets 𝜌𝑎 = 1 Pass-through of negative rates of 100 %
Pass-through coefficient to external financing 𝜌 𝑓 = 0.4 Pass-through of external financing of 40 %
Table 4: Banking sector parameters.
Notes: This table shows key banking sector moments and pass-through coefficients necessary to calibrate the log-linear
approximated model, outlined in Appendix C. Banking sector moments are based on the 2014 balance sheet of the Swedish
banking system. Pass-through coefficients are computed based on Figure 3 and Figure 17.
bonds documented in Figure 17. The pass-through to external financing at the DLB (𝜌 𝑓 ) is assumed to be 0.4.
𝑓
This number is obtained by first computing 𝑖 𝑡 as the weighted average of the money market rate, weighting
the money market rates and covered bond rates with weights computed from the banks balance sheets.36
Numerical values are assigned to the remaining structural parameters in three steps. First, values are
chosen from the existing literature in cases where they are relatively standard. Second, values for the novel
parameters of our model are chosen by matching certain features of the data. Finally, given the other parameters,
some structural values can be assigned exploiting steady state relationships.
We start by setting the relatively standard parameters equal to values used in the existing literature,
following especially Benigno et al. (2014) closely. The values chosen for 𝜎, 𝜅, 𝜒, 𝛽 𝑠 , 𝛽 𝑏 , 𝜙 𝜋 , 𝜙 𝑦 , Π̄ are reported
in Table 5. The values for the policy rule are standard, see e.g. Gali (2008). Steady state saving and borrowing
interest rates are pinned down by 𝛽 𝑠 , 𝛽 𝑏 and Π̄. The risk-free rate is assumed to be 1.5 percent in steady state
based on US data. The borrowing rate is pinned down based on Mehra and Prescott (2008), who report a
31
Parameter Value Source/Target
Intertemporal elasticity of substitution 𝜎 = 0.66 Smets and Wouters (2003)
Share of borrowers 𝜒 = 0.61 Justiniano et al. (2015)
Steady-state gross inflation rate Π̄ = 1.005 Match annual inflation target of 2
Discount factor, saver 𝛽 𝑠 = 0.9963 Annual real savings rate of 1.5
Discount factor, borrower 𝛽 𝑏 = 0.991 Annual real borrowing rate of 3.5 (Mehra and Prescott, 2008)
Slope of AS equation 𝜅 = 0.02 Eggertsson and Woodford (2003)
Taylor coefficient on inflation gap 𝜙Π = 1.5 Gali (2008)
Taylor coefficient on output gap 𝜙𝑌 = 0.5/4 Gali (2008)
Marginal cost of lending 𝜈 = 4.9 Match pass-through of negative rates from Table 1.
Elasticity of lending costs wrt. net worth 𝜄 = 2.55 𝜄 = (𝜈 − 1) × 1.89/2.89, based on MAG (2010) and model equations
Payout ratio 𝜔 = 0.17 Generate profit to net worth ratio of 17.3 %
Shock Value Source/Target
Preference shock 13.37 % temporary decrease in 𝜁𝑡 Generate a 4 drop in output on impact
Persistence of preference shock 𝜌 = 0.88 Duration of lower bound of 12 quarters
Table 5: Calibration
spread of 2 percent in US data. The values for 𝜅, 𝜎 and 𝜒 reported in Table 5 are relatively conventional
and values in this range can be found in a large number of sources, see for instance Benigno et al. (2014) or
Justiniano et al. (2015).
The two key parameters that are specific to the model are 𝜄 and 𝜈. We assign values to these parameters
based on our own estimate of the pass-through of negative policy rates to borrowing rates at the DLB from
Table 1 and empirical estimates from MAG (2010). MAG (2010) is a meta study released by the Bank of
International Settlements that considers the effect of a reduction in bank equity on lending, using a large
number of models. A summary finding is that a one percentage point increase in bank target capital leads to a
1.89 percentage point reduction in lending.
Using these two pieces of evidence, 𝜄 and 𝜈 are pinned down in two steps. First, defining the bank capital
𝑛𝑡
ratio as Υ𝑡 = 𝑏𝑡 ,
and taking lending rates as given, condition (13) can be used to relate lending and bank
𝜄 𝑑 𝑏ˆ 𝑡
capital to yield 𝑏ˆ 𝑡 = Υ̂𝑡 . MAG (2010) provides an empirical estimate for = 1.89 from which it
𝜈−𝜄−1 𝑑 Ψ̂𝑡
1.89
follows that 𝜄 = × (𝜈 − 1). In the second step, we use the pass-through of policy rate cuts to borrowing
2.89
rates at the DLB estimated in Table 1 as an identified moment (Nakamura and Steinsson, 2018) to set 𝜈.
Specifically, the full model is solved and 𝜈 is chosen so that in general equilibrium - if the DLB is binding - a
100 basis points reduction in the policy rate increases borrowing rates by 5.9 basis points on impact.
A remaining parameter is 𝜔. To choose this parameter we exploit that in steady state 𝑧¯/𝑛¯ = 𝛽 𝑠 + 𝜔 − 1.
Using data on 𝑧¯/𝑛¯ from Table 5 yields 𝜔 = 0.173.
Finally, in order to conduct numerical experiments in the model, we further need to specify a shock process
for 𝜉𝑡 . In the numerical experiment considered next, we assume that 𝜉𝑡 follows a first order autoregressive
process with persistence 𝜌. At time 0 there is an initial shock. The size of the shock, and its persistence, is
picked to generate a 4.5 percent drop in output and a duration of the lower bound of twelve quarters assuming
that the PLB is binding at zero. The output drop is calibrated to match the average decline in real output in
Switzerland, the Euro Area, Sweden and Denmark in the aftermath of the financial crisis.
32
3.4 Simulation results
Notes: This figure shows the impulse response functions of inflation, output, the policy rate and the natural rate of interest in
response to a preference shock. The red solid line corresponds to a model where the central bank follows the Taylor rule into
negative territory, but in which the deposit rate is subject to a lower bound. The black solid line corresponds to a model where the
central bank follows the Taylor rule in positive territory, but where the policy rate is subject to a zero lower bound. The dashed
black line corresponds to a model in which there are no bounds on any interest rate.
Figure 8 plots the impulse responses of inflation, output, the natural rate and the policy rate to an
innovation in 𝜉ˆ0 . The solid black line shows the evolution of these variables in response to the shock, assuming
the policy rate and the deposit rate are constrained at zero. The shock directly reduces the natural rate to
approximately five percent. Since the policy rate does not follow the natural rate into negative territory, the
result is an output contraction of -4.5 percent and a drop in inflation from the target rate of 2 percent to slightly
above zero.
The dashed black line shows the response of the economy if there are no lower bounds. In this case, the
central bank fully accommodates the shock by cutting interest rates to -5 percent. This policy rate reduction is
sufficient to offset the effect of the initial shock, leading to no drop in inflation and output.
The red line illustrates the case when we impose a DLB on the deposit rate but not on the policy rate. In
response to the shock, the central bank cuts the policy rate to roughly -12 %, which is implied by the Taylor
rule. In this case, given the balance sheet structure of the banking sector, output contracts by by roughly two
additional percentage points.
33
Figure 9: Response of model under the baseline calibration to a preference shock.
Notes: This figure shows the impulse response functions of the deposit rate, borrowing rate and bank net worth, in response to a
preference shock. The red solid line corresponds to a model where the central bank follows the Taylor rule into negative territory,
but in which the deposit rate is subject to a lower bound. The black solid line corresponds to a model where the central bank
follows the Taylor rule in positive territory, but where the policy rate is subject to a zero lower bound. The dashed black line
corresponds to a model in which there are no bounds on any interest rate.
Figure 9 sheds light on the underlying mechanism. Due to the DLB, the negative policy rate is not
transmitted to deposit rates. As a result, there is limited pass-through to banks financing costs and so no major
reduction in lending rates. Further, because banks in our economy are net holders of assets with a negative
rate, the introduction of negative rates yields a reduction in bank profits. This translates into lower bank net
worth and higher intermediation costs, thus increasing the interest rate spread – indeed the parameter 𝜈 was
chosen so as match the estimated increase in spreads in the data. As a result, output falls by an additional
amount when the central bank goes negative. Going into negative territory, once the DLB is binding, is thus
contractionary. The overall effect of negative rates in the simulation is only illustrative, and is not meant to
represent actual policy by any central bank in recent past, since it is assumed that the policy rate follows a
Taylor rule, even once rate turn negative, which implies very negative rates – negative enough so that the PLB
might be an issue. Subsection 3.4.2 gives a better sense for the likely quantitative effect of negative rates for
more modest policy rate cuts into negative territory, and highlight how this effect depends on parameter values
assumed.
34
Figure 10: Effect of negative policy rates and net exposure of the banking system
Notes: This figure shows the on-impact difference in borrowing rates (left) and output (right) between two models - one model
where the central bank does not impose a negative policy rate and one where the central bank follows a standard Taylor rule also
below zero. The output and borrowing rate difference is scaled by absolute value of the average policy rate below zero. On the
x-axis, we redo the exercise for different values of equation (46). We fix 𝑅, 𝐴 and 𝐹 and solve the model for different values
of 𝜌 𝑓 ∈ (0.3, 1). The black dashed line is the results using our baseline calibration. The red solid line is a calibration where
𝜈 = 51.5, which generates a pass-through at the DLB equal to the coefficient estimate in Table 1 plus 1 standard deviation. The
blue horizontal lines show the effects in the case when the DLB is non-binding.
In Table 1 we reported the pass-through of policy rate cuts to lending rates at the DLB. Pass-through can
also be computed in the model. The horizontal blue line in the left panel of Figure 10 shows the pass-through
of interest rate cuts when the DLB is not binding. As the figure reveals it is approximately -100, that is,
borrowing rates fall approximately as much as the repo-rate in response to policy rate cuts in the quarter in
which they occur.
The x axis captures the exposure of the banking sector to negative rates, according to Proposition 4. A
simple way of changing the exposure is to vary 𝜌 𝑓 , the pass-through of negative rates to external financing. In
the absence of the DLB, the exposure of the banking system to negative rates is irrelevant. The dashed black
line shows the pass-through to borrowing rates once the DLB is binding. The red dot denotes a pass-through
of 5.9, in line with our empirical estimates. The vertical line highlights the empirical exposure to negative
rates using Swedish bank balance sheets. As 𝜌 𝑓 varies, the pass-through to borrowing rates varies. The
shaded yellow area captures variations in the pass-through resulting from different values of {𝜈, 𝜄}.
35
3.4.2 Output effect of negative policy rates
To quantitatively summarize how the DLB reduces the efficiency of policy rate cuts, and how sensitive
the conclusions are to parameter values, it is useful to compute a interest rate cut multiplier, 𝑀0 , defined as
𝑦ˆ ∗ − 𝑦ˆ PLB=0 𝑦ˆ ∗ − 𝑦ˆ PLB=0
𝑀0 = 𝑟 1∗ 𝑟1PLB=0 = 1 𝑟 ∗1| (49)
|ˆ𝚤 1 − 𝚤ˆ1 | |ˆ𝚤 1
where 𝑦ˆ PLB=0
1 denotes the output on impact if the policy rate is at a lower bound of 0 and 𝚤ˆ1𝑟 zero = 0 is the
corresponding policy rate at the bound. These numbers correspond to the initial point of the black line in
∗
Figure 10. 𝚤 1𝑟 is a policy intervention and the corresponding output response is 𝑦ˆ ∗1 for a particular scenario.
Consider, for example, the interest rate multiplier if there is no bound on any interest rate – the blue line in
Figure 10. In that case, the multiplier is 0.9 and has the interpretation that a 100 basis point cut in the policy
rate leads to a 0.9 percent increase in output. Moreover, it is independent of the balance sheet composition of
the banking sector.
Once the DLB is reached, however, the balance sheet composition of banks is fundamental as illustrated
by the condition in Proposition 4. The dashed line in Figure 10 shows the interest rate multiplier for different
net exposure measures. The vertical line illustrates the benchmark parameterization, in which case the
multiplier is - 0.177. Put differently, a 100 basis point interest rate cut reduces output by 0.177 percent. The
figure also illustrates two extreme cases for the choice of 𝜄 and 𝜈. The vertical horizontal line shows the case
of 𝜄 = 0, in which there is no output effect of negative rates, while the dark red line shows what we consider
relatively high values of 𝜈. The figure illustrates that the sign of the interest rate cut multiplier can change,
depending on the banking sectors net exposure to negative rates.
37We note, however, that most existing studies focuses on the relative differences in bank lending under negative policy rates according
to different bank characteristics, while our results are also informative about the level of the aggregate lending response.
36
yet binding.
The presence or absence of a zero lower bound on deposit rates is not sufficient to predict whether the
bank lending channel will be fully operational. This is illustrated by comparing the findings in our paper to
the findings in Basten and Mariathasan (2018). They investigate the impact on bank lending of negative policy
rates in Switzerland. When the Swiss National Bank went below zero, deposit rates were already close to the
zero lower bound. However, Swiss banks according to their summary statistics38 had a relatively low share of
liquid assets relative to total assets, and a higher share of debt and interbank funding, suggesting a negative
net exposure to negative rates.39 In our data, the average net exposure to negative interest rates is positive,
potentially explaining why the overall impact on bank net worth in Sweden and Switzerland differs. The
empirical findings in both this paper and in Basten and Mariathasan (2018) are therefore consistent with our
model.
37
4.2 Eliminating or lowering the lower bound
The critical friction in our model is the zero-lower bound on deposit rates. Without this friction, negative
interest rates would be similar to conventional monetary policy. Consistent with our findings, Altavilla et al.
(2019) find that banks which were able to impose negative deposit rates also expanded lending in response to
interest rate cuts into negative territory. Unfortunately, however, so far only a minority of banks have been
able to do so – at lest for a meaningful fraction of their deposit customers.
The critical question is, therefore, how to remove the zero lower bound on deposit rates more broadly.
One way of doing so in the context of our model is if the government takes action to increase the cost of
holding paper money. There are several ways to do this. The oldest example is a tax on currency, as outlined by
Gesell (n.d.). Gesell’s idea would show up as a direct reduction in the bound on the deposit rate in our model,
thus giving the central bank more room to lower the interest rate on reserves - and the funding costs of banks.
Another possibility is to ban higher denomination bills, a proposal discussed in among others Rogoff
(2017a,c). To the extent that this would increase the storage cost of money, this too should reduce the bound
on the banks’ deposit rate.
An even more radical idea, which would require some extensions to our model, is to let the reserve
currency and the paper currency trade at different values, rather than on par as we have assumed. This proposal
would imply an exchange rate between electronic money and paper money. Agarwal and Kimball (2015),
Rogoff (2017c) and Rogoff (2017b) discuss a concrete proposal, where a key pillar – but perhaps also a
challenge to implementability – is that the reserve currency is the economy-wide unit of account by which
taxes are paid, and accordingly what matters for firms price setting. If such an institutional arrangement is
achieved, then there is nothing that prevents a negative interest rate on the reserve currency while cash in
circulation is traded at a different price, given by an arbitrage condition. We do not attempt to incorporate this
extension to our model but note that it seems relatively straightforward, and has the potential of solving the
DLB problem.
Indeed, the take-away from the paper should not be that negative nominal rates are always non-
expansionary, simply that they are predicted to be less stimulative than normal interest rate cuts under the
current institutional arrangement. The prevalence of the low interest rates going forward gives all the more
reason to contemplate departures from the current framework, such as those mentioned briefly here and
discussed in more detail by some of the authors cited above.
5 Conclusion
Since 2014, several countries have experimented with negative policy rates. In this paper, we have
documented that negative central bank rates have not been transmitted to aggregate deposit rates, which remain
stuck at levels close to zero. Using bank level data from Sweden, we documented a disconnect between the
policy rate and lending rates, once the policy rate became sufficiently negative. We further showed that this
disconnect was partially explained by reliance on deposit financing. Banks which rely more heavily on deposit
financing were less likely to reduce their lending rates in response to policy rate cuts once the deposit rate
had reached its lower bound. Consistent with this, we found that Swedish banks with high deposit shares
38
experienced lower credit growth after the deposit rate had become unresponsive. Furthermore, we documented
negative excess returns on Swedish bank stocks surrounding the announcement of a negative policy rate,
which significantly differed from the response in positive territory.
Motivated by our empirical findings, we developed a New Keynesian model with savers, borrowers, and a
bank sector. By including money storage costs and central bank reserves, we captured the disconnect between
the policy rate and the deposit rate at the lower bound. In this framework, we highlighted how the strength
of the bank lending channel of negative policy rates depends crucially on whether (part) of banks financing
costs are subject to a binding lower bound and if so, the mix of assets and liabilities with different degrees of
pass-through on the banks balance sheet.
Given the long-term decline in interest rates, the need for unconventional monetary policy is likely to
remain high in the future. Our findings highlight conditions for when negative policy rates can work and when
they are likely to be ineffective. The question remains, however, are there other unconventional monetary pools
that can be more effective? While the existing literature has made some progress in evaluating these measures,
the question of how monetary policy should optimally be implemented in a low interest rate environment
remains a question which should be high on the research agenda.
39
References
Adolfsen, Jakob Feveile and Morten Spange, “Modest pass-through of monetary policy to retail rates but no reversal,” Danmarks
Nationalbank Working Papers, 2020, (154), 1–47.
Agarwal, Ruchir and Miles Kimball, Breaking through the zero lower bound, International Monetary Fund, 2015.
Altavilla, Carlo, Lorenzo Burlon, Mariassunta Giannetti, and Sarah Holton, “Is there a zero lower bound? The effects of
negative policy rates on banks and firms,” 2019.
Ampudia, Miguel and Skander Van den Heuvel, “Monetary policy and bank equity values in a time of low interest rates,” ECB
Working Paper, 2018.
Amzallag, Adrien, Alessandro Calza, Dimitris Georgarakos, and Jo ao Sousa, “Monetary policy transmission to mortgages in a
negative interest rate environment,” ECB Working Paper Series, No. 2243, 2019.
Basten, Christoph and Mike Mariathasan, “How Banks Respond to Negative Interest Rates: Evidence from the Swiss Exemption
Threshold,” Technical Report 2018.
Bech, Morten and Aytek Malkhozov, “How have central banks implemented negative policy rates?,” BIS Quarterly Review, 2016,
2016 (March).
Benigno, Pierpaolo, Gauti B Eggertsson, and Federica Romei, “Dynamic debt deleveraging and optimal monetary policy,”
Technical Report, National Bureau of Economic Research 2014.
Bernanke, Ben, Timothy Geithner, and Henry Paulson, “What We Need to Fight the Next Financial Crisis,” Technical Report,
The New York Times 2018.
Bottero, Margherita, Ms Camelia Minoiu, José-Luis Peydró, Andrea Polo, Mr Andrea F Presbitero, and Enrico Sette,
Negative monetary policy rates and portfolio rebalancing: evidence from credit register data, International Monetary Fund, 2019.
Boucinha, Miguel and Lorenzo Burlon, “Negative rates and the transmission of monetary policy,” ECB Economic Bulletin, 2020,
2020 (3).
Brunnermeier, Markus K and Yann Koby, “The Reversal Interest Rate: An Effective Lower Bound on Monetary Policy,”
Preparation). Print, 2017.
Buiter, Willem H and Nikolaos Panigirtzoglou, “Overcoming the zero bound on nominal interest rates with negative interest on
currency: Gesell’s solution,” The Economic Journal, 2003, 113 (490), 723–746.
Caballero, Ricardo J. and Emmanuel Farhi, “The safety trap,” The Review of Economic Studies, 2017, p. rdx013.
Calvo, Guillermo A, “Staggered prices in a utility-maximizing framework,” Journal of monetary Economics, 1983, 12 (3), 383–398.
Curdia, Vasco and Michael Woodford, “The central-bank balance sheet as an instrument of monetary policy,” Journal of Monetary
Economics, 2011, 58 (1), 54–79.
Eggertsson, Gauti B and Michael Woodford, “The Zero Bound on Interest Rates and Optimal Monetary Policy,” Brookings Papers
on Economic Activity, 2003, 1.
and , “Optimal monetary and fiscal policy in a liquidity trap,” in “NBER International Seminar on Macroeconomics 2004”
The MIT Press 2006, pp. 75–144.
and Neil R Mehrotra, “A model of secular stagnation,” Technical Report, National Bureau of Economic Research 2014.
Eisenshmidt, Jens and Frank Smets, “Negative interest rates: Lessons from the euro area,” Series on Central Banking Analysis and
Economic Policies no. 26, 2019.
Erikson, H and D Vestin, “Pass-through at mildly negative policy rates: The swedish case,” Staff memo, Sveriges Riksbank, 2019.
Freixas, Xavier and Jean-Charles Rochet, Microeconomics of banking, MIT press, 2008.
Gali, Jordi, “Monetary Policy, Inflation, and the Business Cycle: An Introduction to the New Keynesian Framework,” 2008.
Gesell, S, “The Natural Economic Order, translated by Philip Pye, 2002,” 1916.
Gesell, Silvio, Die nat? rliche Wirtschaftsordnung durch Freiland und Freigeld.
Goodfriend, Marvin, “Overcoming the zero bound on interest rate policy,” Journal of Money, Credit and Banking, 2000,
pp. 1007–1035.
Greenlaw, David, James D Hamilton, Ethan Harris, and Kenneth D West, “A Skeptical View of the Impact of the Fed Balance
Sheet,” Technical Report, National Bureau of Economic Research 2018.
Greenwood, Robin, Samuel Hanson, Joshua S Rudolph, and Lawrence H Summers, Government debt management at the zero
lower bound, Hutchins Center on Fiscal & Monetary Policy at Brookings Institution, 2014.
40
Groot, Oliver De and Alexander Haas, “The signalling channel of negative interest rates,” 2020.
Hameed, Allaudeen and Andrew Kenan Rose, “Exchange rate behavior with negative interest rates: Some early negative
observations,” 2016.
Heider, Florian, Farzad Saidi, and Glenn Schepens, “Life Below Zero: Bank Lending Under Negative Policy Rates,” Working
paper, 2016.
Hong, Gee Hee and John Kandrac, “Pushed past the limit? How Japanese banks reacted to negative interest rates,” IMF Working
Paper 18/131, 2018.
Justiniano, Alejandro, Giorgio E Primiceri, and Andrea Tambalotti, “Household leveraging and deleveraging,” Review of
Economic Dynamics, 2015, 18 (1), 3–20.
Kiley, Michael T and John M Roberts, “Monetary policy in a low interest rate world,” Brookings Paper on Economic Activity, 2017.
Krugman, Paul R, “It’s Baaack: Japan’s Slump and the Return of the Liquidity Trap,” Brookings Papers on Economic Activity, 1998,
1998 (2), 137–205.
MAG, “Assessing the macroeconomic impact of the transition to stronger capital and liquidity requirements,” December, URL:
http://www. bis. org/publ/othp12. pdf, 2010.
Mehra, Rajnish and Edward C Prescott, “The equity premium: ABCs,” Handbook of the equity risk premium, 2008, pp. 1–36.
Nakamura, Emi and Jón Steinsson, “Identification in macroeconomics,” Journal of Economic Perspectives, 2018, 32 (3), 59–86.
Rognlie, Matthew, “What lower bound? Monetary policy with negative interest rates,” Unpublished manuscript, Department of
Economics, Harvard University (November 23), 2015.
Rogoff, Kenneth, “Dealing with Monetary Paralysis at the Zero Bound,” The Journal of Economic Perspectives, 2017, 31 (3), 47–66.
, “Monetary Policy in a Low Interest Rate World,” Journal of Policy Modeling, 2017.
Rogoff, Kenneth S, The Curse of Cash: How Large-Denomination Bills Aid Crime and Tax Evasion and Constrain Monetary Policy,
Princeton University Press, 2017.
Smets, Frank and Raf Wouters, “An estimated dynamic stochastic general equilibrium model of the euro area,” Journal of the
European economic association, 2003, 1 (5), 1123–1175.
Summers, Lawrence H, “US economic prospects: Secular stagnation, hysteresis, and the zero lower bound,” Business Economics,
2014, 49 (2), 65–73.
Swanson, Eric T, “Measuring the effects of Federal Reserve forward guidance and asset purchases on financial markets,” Technical
Report, National Bureau of Economic Research 2017.
Ulate, Mauricio, “Going negative at the zero lower bound: The effects of negative nominal interest rates,” in “in” Federal Reserve
Bank of San Francisco 2019.
41
A Listed vs. transaction-based rates
An important question is whether the listed rates used in the main body of the text is a valid proxy for
actual interest rates. One approach to gauge whether this is the case, is to compare the implied aggregate
listed rates with observed aggregate transaction-based rates.
Figure 11 depicts a weighted average of the bank level floating mortgage rates aggregated to the monthly
level, along with the official aggregate rates that are based upon transaction data. First note that the listed rates
are somewhat higher than the transaction rates. This is not surprising, as bank customers can negotiate a lower
rate based on loan characteristics. 40 The dashed vertical lines represent repo-rate reductions. Importantly,
around these policy changes, the difference between listed rates and aggregate rates is stable, suggesting that
the breakdown in pass-through documented in the event study is not an artifact of the listed rates.
3
LTV LTI
2
0
1
0
-.25
-1
Figure 11: Listed rates and aggregate rates (left panel) and cumulative changes (right panel).
Notes: This figure shows listed and aggregate rates (left panel) and the cumulative changes in listed and aggregate rates (right
panel). Cumulative change calculated since November 2015. Dashed vertical lines correspond to repo-rate reduction dates. Solid
vertical lines in the right panel correspond to June 2016 and March 2018, when mortgage market regulation was introduced.
Erikson and Vestin (2019) point out that there was some reduction in aggregate rates in 2017 and 2018,
and suggest that this might reflect a delayed pass-through from the last repo-rate reduction. While we cannot
rule out that there was some long-term pass-through, we find it likely that the reduction in aggregate rates was
connected to the introduction of new mortgage market regulation.
We follow Erikson and Vestin (2019) and plot the cumulative change in interest rates since November
2015 in the right panel of Figure 11. The last repo-rate reduction is captured by the dashed vertical line. Note
that there is no pass-through to either listed rates or official transaction rates in the months following the
repo-rate cut. Over time however, there is a divergence between listed rates and aggregate rates. As a result,
by the end of 2018 – almost three years after the last repo-rate reduction – there appears to have been full
pass-through aggregate rates, while listed rates remain unchanged. What explains the discrepancy?
As documented in the event study, normally full pass-through – or close to full pass-through – is achieved
within a period of thirty days. Full pass-through within a period of two/three years would therefore entail a
40For example, the Swedish bank Nordea writes alongside their listed rate that “The mortgage rate we offer is an individual offer. Your
mortgage rate can therefore turn out lower than the rate we list here.” In Swedish, from Nordeas website 5/10/2019: Bolanerantan vi
erbjuder dig ar individuell. (...) Din bolaneranta kan darfor bli lagre an den ranta vi visar har.
42
severe delay of the effectiveness of monetary policy. An alternative explanation for why aggregate transaction
rates fell is unrelated to the interest rate policy and rather triggered by change in composition of Swedish
borrowers. A possible explanation is that during this period the Swedish Financial Authority implemented
new amortization requirement that encouraged encouraged lower loan-to-value (LTV) and loan-to-income
(LTI) ratios, both of which would explain why transaction rates during this period started deviating more
and more from listed rates. These regulation were implemented in June 2016 and March 2018, and led to a
reduction in average LTV and LTI ratios for new borrowers, the dates are denoted by the two vertical lines in
figure 11. As shown in Figure 11, the new policies introduced seem to coincide almost perfectly with the
divergence between listed rates and aggregate rates. Moreover, we have studied the effect of a similar policy
change in 2010 to evaluate whether the same pattern of divergence was observed then. As shown in Figure 15
in Appendix A, there was a quantitatively similar divergence following the new regulation in 2010. Hence,
our assessment is that this empirical evidence is highly suggestive of that the reduction in aggregate lending
rates two/three years after the last repo-rate cut was driven by regulatory measures, not by the repo-rate cuts
themselves.
6%
10%
13% 47%
24%
Figure 12: Decomposition of liabilities (as of September 2015) for large Swedish banks. Source: The Riksbank
43
Switzerland Japan Denmark
3.00
.6
6
2.00
4
.4
2
1.00
.2
0
0.00
-2
-1.00
-.2
2008 2010 2012 2014 2016 2018
2008 2010 2012 2014 2016 2018 2008 2010 2012 2014 2016 2018
Households Corporations
Households Policy Rate Households Policy Rate Policy Rate
5
4
4
3
3
2
2
1
1
0
0
2008 2010 2012 2014 2016 2018 2008 2010 2012 2014 2016 2018
Figure 13: Aggregate deposit rates for Switzerland, Japan, Denmark, the Euro Area and Germany. The policy rates are defined as
SARON (Switzerland), the Uncollaterized Overnight Call Rate (Japan), the Certificates of Deposit Rate (Denmark) and the Deposit
Rate (Euro Area and Germany). The red vertical lines mark the month in which policy rates became negative. Source: the Swiss
National Bank (SNB), Bank of Japan, the Danish National Bank (DNB), and the European Central Bank (ECB).
UK
FR
IT
AU
US
RO
BE
ES
TU
DE
PO
CZ
NL
LU
AS
0 20 40 60 80 100
Fraction of respondents that would withdraw money from savings account (%)
Figure 14: Fraction of households who would withdraw money from their savings account if they were levied a negative interest
rate. Solid line represent unweighted average of 76.4 . Source: ING (2015)
44
3
LTV cap
2
1
0
1-2011 1-2012 1-2013
Figure 15: Listed rates and aggregate rates – 3 months. Cumulated change since March 2010.
1
.8
% of total assets
.4 .2
0 .6
Figure 16: Actual and Counterfactual Commission Income as a Share of Assets (). The counterfactual commission income is
calculated as the amount of commission income that would be necessary to make up for the bound on the nominal deposit rate, all
𝐷𝑒 𝑝𝑜𝑠𝑖𝑡 𝑠 𝑐𝑓
else equal. The counterfactual commission income is given by actual commission income plus 𝐴𝑠𝑠𝑒𝑡 𝑠𝑡 𝑡 𝑖 𝑡 − 𝑖 𝑡 , where 𝑖 𝑡 is the
𝑐𝑓
average aggregate deposit rate and 𝑖 𝑡 is a counterfactual deposit rate calculated under the assumption that the markdown to the
repo-rate is constant and equal to the pre-zero markdown. Source: Statistics Sweden and own calculations.
45
3
2
1.5
2
1
1
.5
0
0
-1
-.5
2012m1 2014m1 2016m1 2018m1
2012m1 2014m1 2016m1 2018m1
Covered bond, 2Y Covered bond, 5Y
Governement bond, 5Y STIBOR, 3M Est. funding cost Counterfactual
Repo rate Repo rate
Figure 17: Other interest rates (left panel) and an estimate of average bank funding costs (right panel).
Notes: This figure shows other interest rates (left panel) and an estimate of the average bank funding cost (right panel). The
estimated average funding cost is computed by taking the weighted average of the assumed interest rates of the different funding
sources of the bank. Certificates are assumed to have the same interest rate as 2Y covered bonds, while unsecured debt are
assumed to have the same interest rate as 2Y covered bonds plus a 2 percent constant risk-premium. The counterfactual series
corresponds to the case when the spread between the repo-rate and the estimated funding cost remains fixed at pre-negative levels.
Weights are based on the liability structure of large Swedish banks, see Figure 12. Source: The Riksbank
25,000
.6 .58
20,000
Total Issuance, Mill. EUR
.54
10,000
.52
Figure 18: Left panel: Issuance of covered bonds, Swedish banks. Right panel: Deposit share, Swedish banks. Vertical lines
correspond to the date negative interest rates were implemented. Source: Association of Swedish Covered Bond Issuers, The Riksbank
and Statistics Sweden
46
Actual and predicted pass-through (%) Actual and predicted pass-through (%)
50
50
0
0
-50
-50
-100
-150
-100
-30 -20 -10 0 10 20 30 -30 -20 -10 0 10 20 30
Actual and predicted pass-through (%) Actual and predicted pass-through (%)
100
100
50
50
0
0
-50
-50
-100
-100
Figure 19: Event study different mortgage rates. Upper left: 3 months. Upper right: 1 year. Lower left: 3 years. Lower
right: 5 years.
Cash reported by MFI excluding ESCB - Euro Area Cash & CB credit SEK (mill SEK)
90000
20000
-0.10
-0.20
-0.10 -0.50
80000
15000
-0.30
-0.40
70000
10000
60000
5000
50000
40000
2008m1 2010m1 2012m1 2014m1 2016m1 2018m1 2008m1 2010m1 2012m1 2014m1 2016m1 2018m1
Figure 20: Left panel: Cash held by Euro Area Banks. Variable name: Cash reported by MFIs excluding ESCB - Euro Area.
Millions of Euro. Source: ECB. Right panel: Cash held by Swedish banks. Variable name: Cash and credit balances at central banks
in SEK. Million SEK. Source: Statistics Sweden. The dashed, red lines refer to the interest rate cuts in negative territory.
47
(1) (2) (3) (4)
Δ log(loans) Δ log(loans) Δ log(loans) Δ log(loans)
Post × deposit share 0.0189
(0.0894)
Post × deposit share -0.0646
(0.0959)
Post × deposit share 0.731
(1.476)
Post × deposit share -0.763
(1.040)
N 1227 1227 1227 1227
No. of clusters 10 10 10 10
Mean of dependent variable 2.587 2.587 2.587 2.587
SD of dependent variable 7.713 7.713 7.713 7.713
Bank FE Yes Yes Yes Yes
Time FE Yes Yes Yes Yes
measure Cont. Cont. High or low High or low
placeboperiod repo-rate ∈ [1,0) repo-rate ∈ (0,-0.25] repo-rate ∈ [1,0) repo-rate ∈ (0,-0.25]
Table 6: Falsification test of lending volume regression. We consider two different “post-periods”, according to the
level of the repo-rate. The first period is defined as the period from 18th of December 2012 to the 17th of February 2015,
when the repo-rate is in the interval between 1 and 0 percent. The second period is the period from the 17th of February
2015 to the 7th of July 2015 when the repo-rate is in negative territory but the pass-through to deposit rates are above its
ZLB.
48
𝛾 𝑓 > 1 + 𝛾𝑙 (50)
which implies a lower bound on the elasticity of liquidity costs with respect to external financing. This
condition ensures that the maximization problem of the bank is well behaved.
While all liquid asset reduce liquidity risk, reserves and paper currency contribute to reducing banks
operational costs due their special "money role," e.g. in settling inter-bank transaction. Bank inter-mediation
costs are decreasing in R and M up to a satiation point. The following functional form captures this assumption
(
𝜆 𝑅 (𝑀𝑡 + 𝑅𝑡 ) −𝛾𝑅 if𝑀𝑡 + 𝑅𝑡 < 𝑅 ∗
Ψ(𝑀𝑡 , 𝑅𝑡 ) =
Ψ̄ if 𝑀𝑡 + 𝑅𝑡 ≥ 𝑅 ∗
where 𝛾 𝑅 > 0 measures the elasticity of bank transaction costs with respect to R+M. For simplicity, cash and
reserves play the same role in facilitating transactions. The difference between the two assets is that bank
reserves pay an interest of 𝑖 𝑟𝑡 , while paper currency pays zero interest. This implies that the bank does not
choose to hold currency while the interest on reserves is positive. It may, however, choose to do so once
the reserve rate becomes negative. In this case the bank needs to consider storage costs of holding cash.
The assumption that the bank does not hold cash at positive interest rates is a harmless abstraction as vault
cash is a trivial component of banks balances. As money is an asset with a zero return, however, it requires
careful consideration when thinking about the central banks ability to charge negative interest rates on reserve
balances, as under the current institutional framework, banks can always substitute reserves for cash.
Holding money entails a storage cost, which is convex in money if 𝑀𝑡 > 0:
𝑆(𝑀𝑡 ) = 𝜆 𝑀 𝑆 (𝑀0 + 𝑀𝑡 ) 𝛾𝑀 𝑆 (51)
where 𝛾𝑚𝑠 > 1. The parameter 𝑀0 represents a fixed cost. Once rates turn negative, equation (51) determines
how much cash the banks will hold.
The function Γ(𝐵𝑡 , 𝑁𝑡 ) captures the cost of lending. It can be rationalized as being due to default, which
becomes more pronounced the higher the lending due to finite monitoring resources, see e.g. Curdia and
Woodford (2011) for a discussion of this interpretation. Both the total cost of lending and the marginal cost of
lending are assumed to decrease in net worth. There is a variety of ways to underpin this assumption. The
simplest is regulatory requirements, which stipulate that bank lending can only exceed a certain fraction of
bank’s net worth.41
The intermediation function takes the form
Γ(𝐵, 𝑁) = 𝜆 𝐵 𝐵 𝜈 𝑁 − 𝜄
where 1 ≤ 𝜈 < 𝜈 measures the elasticity of intermediation costs to lending, and 0 < 𝜄 < 𝜄 measures the
41The simplest interpretation of the role of net worth is that it comes about due to a capital requirement, often modeled as
𝑁𝑡
≥𝜅
𝐵𝑡
where 𝜅 is a parameter. A constraint of this form would be a limiting case of the smooth function assumed here, one in which the cost
of intermediation goes to infinity if the constraint is breached. More generally, tying deposits up in lending is typically considered
costly due to the illiquidity of these assets – but illiquidity is of greater concern the lower the net worth of the bank. The dependence of
lending costs on net worth is micro founded in Holmstrom and Tirole (1997) and Gertler and Kiyotaki (2010), as well as documented
empirically, for example, in Jimenez, Ongena, Peydro, and Saurina (2012).
49
elasticity with respect to net worth, and
𝑖 𝑏 − 𝑖 𝑑 𝑁¯ 1 𝑁¯
𝑣¯ ≡ ( ) , ¯
𝜄 ≡ (52)
1 + 𝑖 𝑑 Γ̄ 1 + 𝑖 𝑑 Γ̄
The upper bounds on 𝜈¯ and 𝜄¯ ensure the existence of a bounded solution to the banks problem.42
Provided that 𝑉 0 (𝑁𝑡 ) > 0 the first order conditions are:
𝑖 𝑡𝑏 − 𝑖 𝑡𝑑
𝐵𝑡 : = Γ𝐵 (𝐵𝑡 , 𝑁𝑡 ) (53)
1 + 𝑖 𝑡𝑑
𝑖 𝑟𝑡 − 𝑖 𝑡𝑑
𝑅𝑡 : = 𝐶 𝐿 (𝐹𝑡 , 𝐿 𝑡 , 𝑁𝑡 ) + Ψ𝑅 (𝑅𝑡 , 𝑀𝑡 ) (54)
1 + 𝑖 𝑡𝑑
𝑓
𝑖 𝑡 − 𝑖 𝑡𝑑
𝐹𝑡 : = −𝐶𝐹 (𝐹𝑡 , 𝐿 𝑡 , 𝑁𝑡 ) (55)
1 + 𝑖 𝑡𝑑
𝑖 𝑡𝑎 − 𝑖 𝑡𝑑
𝐴𝑡 : = 𝐶 𝐿 (𝐹𝑡 , 𝐿 𝑡 , 𝑁𝑡 ) (56)
1 + 𝑖 𝑡𝑑
𝑑
𝑁𝑡 : 𝛿𝜔 − (1 − 𝜔)𝛿𝐸 𝑡 (1 + 𝑖 𝑡+1 )𝜙𝑡+1 + 𝜙𝑡 + Γ𝑁 (1 + 𝑖 𝑡𝑑 ) = 0 (57)
−𝑖 𝑡𝑑
𝑀𝑡 : = 𝑆 𝑀 + 𝐶 𝐿 + Ψ𝑀 − 𝜓 𝑡 (58)
1 + 𝑖 𝑡𝑑
𝑀𝑡 ≥ 0, 𝜓𝑡 ≥ 0, 𝜓𝑡 𝑀𝑡 = 0 (59)
where 𝜙𝑡 is the Lagrange multiplier of equation (11) and denotes the marginal value of net worth at time 𝑡 to
the bank, and the Envelope has been used to substitute out for the value function in the first order condition for
𝑁𝑡 . Condition (58) is the money demand equation, and 𝜓𝑡 is the Lagrange multiplier on 𝑀𝑡 ≥ 0. Condition
(59) is a Kuhn-Tucker complementary slackness condition. A partial equilibrium is defined by an exogenous set
𝑓
of interest rates {𝑖 𝑡𝑑 , 𝑖 𝑡𝑏 , 𝑖 𝑟𝑡 , 𝑖 𝑡 , 𝑖 𝑡𝑎 } taken as given by the banks, and values for {𝐵𝑡 , 𝑀𝑡 , 𝑁𝑡 , 𝐹𝑡 , 𝑅𝑡 , 𝐴𝑡 , 𝜙𝑡 , 𝜓𝑡 }
that solve equations (53)-(59), together with the flow budget constraint in equation (11).
Proof: Combine (54) and (58), noting that Ψ𝑅 = Ψ𝑀 , and assuming the DLB binding so that 𝑖 𝑑 = 0 to yield
𝜓𝑡 = 𝑆 𝑀 + 𝑖 𝑟𝑡 = 𝛼 𝑀 + 𝑖 𝑟𝑡 0 (61)
where the last inequality follows from the complementary slackness condition (59).
42The ratio 𝑁
Γ is net worth as a fraction of the bank lending intermediation costs. Empirically this is a large number and this restriction
does not impose a tight bound on the functional form assumed.
50
not increase their reliance on non-deposit financing despite substantially larger pass-through to these interest
rates. Below we explain this in the context of the model. Let us first summarize a useful observation.
Proposition 5. If 𝜌 𝑑 = 0 and 𝑖 𝑟𝑡 < 0, the bank is not satiated in liquidity and 𝐿 𝑡 < 𝐿 ∗ .
The proposition follows directly from equations (54)-(56). An immediate observation from inspecting
(55) and (56) is that if there is greater pass-through of policy rates to the interest rate on assets and external
𝑓
financing, then there is a positive spread between the 𝑖 𝑡𝑎 − 𝑖 𝑡𝑑 as well as 𝑖 𝑡 − 𝑖 𝑡𝑑 . This, in turn, implies that the
bank will no longer be satiated in the liquid asset, 𝐶 𝐿 > 0 so that 𝐿 𝑡 < 𝐿 ∗ .
The same is not true for reserves as seen from equation (54). It is possible that the bank is satiated in
reserves to settle interbank transactions, i.e. 𝑅𝑡 ≥ 𝑅 ∗ so that 𝜒𝑅 = 0, but that there is still a spread between 𝑖 𝑟𝑡
and 𝑖 𝑡𝑑 because reserves also serve a role to reduce liquidity risk via the function 𝐶 (.). Let us now discuss
the implications of negative policy rates on external financing, reserves and other assets in the approximated
equilibrium.
External financing Using the result from Proposition 5 that 𝐿 𝑡 < 𝐿 ∗ , a log-linear approximation of the
banks asset demand in equation (56), can be combined with a log-linear approximation of its demand for
external financing in equation (55) to yield:
𝐹¯ 𝐿¯
(1 + 𝛾𝑙 ) (𝜌 𝑑 − 𝜌 𝑓 ) 𝐶
¯ + 𝛾𝑓 𝐶¯
(𝜌 𝑎 − 𝜌𝑑 ) 𝛾 𝑓 𝛾𝑛
𝐹ˆ𝑡 = 𝚤ˆ𝑟𝑡 + 𝑁ˆ 𝑡 (62)
𝛾 2𝑓 − 𝛾 𝑓 (1 + 𝛾𝑙 ) 𝛾 2𝑓 − 𝛾 𝑓 (1 + 𝛾𝑙 )
where the denominator is positive due to equation (50) and the solution for 𝑁ˆ 𝑡 is given by equation (17).
Equation (62) has a useful interpretation. If there is full pass-through, then external financing only
increases and decreases with the bank’s net worth. To the extent that negative rates have negative effects on
banks equity, then, this reduces the banks reliance on external financing.
Consider the implication of a collapse in pass-through to the deposit rate, i.e. 𝜌 𝑑 = 0. Then the first term
says that banks will increase their reliance on external financing if
𝛾 𝑓 𝐿¯ 𝜌𝑓
< 𝑎
1 + 𝛾𝑙 𝐹¯ 𝜌
Hence, even if the policy rate is fully passed through to the external financing rate, the bank might still
choose not to rely more heavily on this funding source. To interpret this conditions, observe that the interest
rate the bank pays for external financing is only one part of the cost to the bank. The other is that it needs to
hold more liquid assets due to the higher liquidity risk which this source of funding generates. If there is
incomplete pass-through to the external financing, while there is full pass-through to the liquid assets, then the
bank may reduce its reliance on external financing once rates turn negative.
Since the pass-through in Sweden to liquid asset was stronger than to the bonds the banks could issue,
this might rationalize why they did not rely more strongly on external financing in response to negative rates.
51
1 − 𝜌𝑎
𝑅ˆ𝑡 = 𝚤ˆ𝑟
𝜒 𝑡
(63)
𝛾𝑅 𝛾𝑅 + 1𝑅
This suggests that if there is incomplete pass-through to liquid assets, then banks reduces their reserves
when the policy rate is lowered. The reason is that reserves pay a lower interest rate than the alternative liquid
assets.
In partial equilibrium, 𝑅ˆ𝑡 measures the reserve holdings of a single bank. Once the model is integrated
into general equilibrium however, the government sets both the interest on reserves, 𝚤ˆ𝑟𝑡 , and the quantity 𝑅ˆ𝑡 .
This deserves further discussion.
Reserves measure the money balances a given bank holds on an account at the central bank. What can
it do with reserves? One option is to exchange them for paper currency from the central bank. We have so
far made the assumption that the storage cost of money is large enough so that the banks choose not to hold
any cash, but we will revisit this assumption shortly. Another option is to use reserves to buy a liquid assets.
An important observation, however, is that while this reduces the reserves of the bank, it leaves aggregate
reserves unchanged.43
Reserves and liquid assets given a negative reserve rate A natural question is what happens if the central
bank chooses a negative interest on reserves when 𝜌𝑎 < 1. As equation (63) suggests, this means that any
single bank (in partial equilibrium) tries to reduce its reserve holdings. Yet, in general equilibrium, total
reserves are pinned down by the government. How is an equilibrium ensured?
Equation (63) suggests that the only way 𝑅𝑡 ≥ 𝑅 ∗ can be consistent with general equilibrium is that there
is full pass-through to the liquid assets, i.e. 𝜌 𝑎 = 1. In this case, all banks remain fully satiated in reserves for
the purpose of interbank transactions once the rate turns negative. Yet, as suggested by Proposition 5, once
rates turn negative, banks are not satiated in liquid assets. Since reserve and other liquid asset are perfect
substitutes, the partial equilibrium model we have specified will only pin down the quantity of total liquid
asset, not its composition. Solving the log-linear approximation of equations (54) and (56) together then yields
𝐶¯ 1 𝛾𝑓 𝛾𝑛 ˆ
𝐿ˆ 𝑡 = 𝚤ˆ𝑟𝑡 + 𝐹ˆ𝑡 − 𝑁𝑡
¯𝐿 (1 + 𝛾𝑙 )𝛾𝑙 1 + 𝛾𝑙 1 + 𝛾𝑙
where the solution for 𝐹ˆ𝑡 is given by equation (62) and 𝑁ˆ 𝑡 by equation (17). This relationship suggests that
negative rates do not have clear predictions for what happens to aggregate liquid assets, it depends on the
effect it has on external financing and net worth. Meanwhile, from the perspective of the banks, the split of
liquid assets between reserves and other assets is indeterminate, and ultimately depends on the choices made
by the central bank which can choose both 𝚤ˆ𝑟𝑡 and 𝑅ˆ𝑡 .
A key prediction of the model, however, is that for the equilibrium to be consistent with large excess
reserves, there has to be full pass-through to liquid assets. Full pass-through of the policy rate to liquid assets
43Consider a bank A which uses 100 dollars of reserves to buy assets from a seller that holds an account at bank B. The payment takes
place by bank A drawing down 100 dollars of its reserve balances at the central bank, and depositing them at bank B which in turn
increases deposits for the seller of the liquid asset. Thus aggregate reserves are unchanged, the reduction of bank A’s reserves is met
by an increase by bank B. Thus, absent changing the reserves into paper currency, it represents a "closed system" in which total
reserves are determined by the central bank.
52
is consistent with the experience in Sweden. There the negative policy rate was passed through to other liquid
assets, such as interest on government debt. The pass-through to liquid assets was often used as evidence for
the success of the policy. While this element of the policy intervention was successful in terms of creating
more fiscal space for government finances, it is not an indication of success from the perspective of the bank
lending channel. Since the negative rates show up on the asset side of the banks balances sheet, rather than on
the liability side, they are a net negative for the banks.
Reserves and money So far we have not considered the implication of banks’ option to convert reserves
into paper currency. For simplicity, we assumed that the bank held no currency because it served the same
role as reserves, which in contrast to cash paid an interest, and moreover had no storage cost.
Once interest on reserves turn negative, however, converting reserves into cash can be an attractive option
for the bank. Consider an equilibrium in which interest rates on reserves are sufficiently negative so that
𝑀𝑡 > 0 and assume that there is no pass-through to deposit rates, so that 𝜌 𝑑 = 0. In this case 𝜓𝑡 = 0 in
equation (59) and the log-linear approximation of the banks demand for money is
1 𝑀∗ 𝑟 𝜒¯ 𝑀 ∗ ˆ
𝑀ˆ 𝑡 = − 𝚤ˆ𝑡 + 𝛾 𝑅 (𝛾 𝑅 + 1) 𝑅𝑡
𝛾 𝑀 (𝛾 𝑀 − 1) 𝑆¯ 𝑅¯ 𝑆¯
which suggests that as the reserve rate turns negative, banks increase cash holdings. since cash offers a
gross return of zero, although entailing a storage cost. Overall, the experience so far has been that banks have
not to a significant extent moved into cash, indicative of a large value of 𝛾 𝑀 .
An important assumption is the storage cost was convex, i.e. 𝛾𝑚𝑠 > 1. One might argue, however, that
there are constant or increasing returns to building storage facilities for money, at least when cash holding are
sufficiently large. Consider, for example, the assumption of constant returns, in which 𝛾 𝑀 = 1, when cash
holding become sufficienly large, so that cost of storage is proportional to the supply of money the bank holds
for any further cash holding, i.e. 𝜆 𝑀 𝑆 𝑀𝑡 . In this case, provided that the bank holds money, the interest rate
on reserves is bounded by 𝑖 𝑟𝑡 ≥ −𝜆 𝑀 𝑆 as in the example we considered in the main text, since if it charges
more negative interest on reserves than 𝜆 𝑀 𝑆 , banks will convert reserves into paper currency and interbank
transactions will be settled outside of the central bank.
The assumption that money is as effective as reserves in settling interbank transactions might seem a bit
extreme. Nevertheless, it seems reasonable to think that if the central bank charges very negative rates, banks
will find it in their interest to exchange reserves for currency, and instead construct an alternative payment
system that solves the same problem at a lower cost.
Definition C.1 (Non-linear equilibrium). A non-linear equilibrium of our model is a sequence of 23 endogenous
n o∞
variables 𝑦 𝑡 , 𝜋𝑡 , 𝑐 𝑡𝑏 , 𝑐 𝑡𝑠 , 𝑚 𝑡𝑏 , 𝑚 𝑡𝑠 , 𝑏 𝑡𝑏 , 𝑏 𝑡 , 𝑟 𝑡 , 𝑎 𝑡 , 𝑚 𝑡 , 𝜓𝑡 , 𝑓𝑡 , 𝑙𝑡 , 𝑛𝑡 , 𝑧𝑡 , 𝐹𝑡 , 𝐾𝑡 , 𝜆𝑡 , Δ 𝑡 , 𝑊, 𝑇𝑡 , 𝑓 and 5 endogenous prices
n o∞ 𝑡=0
𝑓 𝑔
𝑖 𝑡𝑏 , 𝑖 𝑡𝑑 , 𝑖 𝑡 , 𝑖 𝑡 𝑖𝑟𝑡 such that equations (64) - (91) holds.
𝑡=0
53
From the household problems there is a consumption Euler equation for each the borrower (64) and the
saver (65), respectively, the budget constraint of the borrower (66), the money demand equations for each
type of agent (67) and (68). The aggregate resource constraint (82) implies all production is consumed. 𝑏 𝑡𝑏 is
defined as aggregate borrowing per borrower, while 𝑏 𝑡 ≡ 𝜒𝑏 𝑡𝑏 is aggregate borrowing (81).
The banks problem give rise to (69)-(76). Relative to the text, the price level is now endogenous.
𝑢 0 (𝑐 𝑡𝑏 )𝜁𝑡 = 𝛽𝑏 (1 + 𝑖 𝑡𝑏 )𝐸 𝑡 𝑢 0 (𝑐 𝑡+1
𝑏 −1
)Π𝑡+1 𝜁𝑡+1 (64)
0
𝑢 (𝐶𝑡𝑠 )𝜁𝑡 = 𝛽𝑠 (1 + 𝑖 𝑡𝑑 )𝐸 𝑡 𝑢 0 (𝑐 𝑡+1
𝑠 −1
)Π𝑡+1 𝜁𝑡+1 (65)
𝑏 𝑡𝑏 = (1 + 𝑖 𝑡−1
𝑏
)Π𝑡−1 𝑏 𝑡−1
𝑏
− 𝑦 𝑡 + 𝑐 𝑡𝑏 − (1 − 𝛾 𝑏 )Π𝑡−1 𝑚 𝑡−1
𝑠
+ 𝑚 𝑡𝑠 (66)
Ω0 𝑚 𝑡𝑠
𝑖 𝑡𝑑
= (67)
𝑢 0 (𝑐 𝑡𝑠 ) 1 + 𝑖 𝑡𝑑
Ω0 𝑚 𝑡𝑠
𝑖 𝑡𝑏
= (68)
𝑢 0 (𝑐 𝑡𝑏 ) 1 + 𝑖 𝑡𝑏
𝑖 𝑡𝑏 − 𝑖 𝑡𝑑
= Γ𝑏 (𝑏 𝑡 , 𝑛𝑡 ) (69)
1 + 𝑖 𝑡𝑑
𝑖 𝑟𝑡 − 𝑖 𝑡𝑑
= 𝐶𝑙 ( 𝑓𝑡 , 𝑙 𝑡 , 𝑛𝑡 ) + 𝜒𝑚 (𝑟 𝑡 , 𝑚 𝑡 ) (70)
1 + 𝑖 𝑡𝑑
𝑖 𝑡𝑎 − 𝑖 𝑡𝑑
= 𝐶𝑙 ( 𝑓𝑡 , 𝑙 𝑡 , 𝑛𝑡 ) (71)
1 + 𝑖 𝑡𝑑
𝑓
𝑖 𝑡 − 𝑖 𝑟𝑑
= −𝐶 𝑓 ( 𝑓𝑡 , 𝑙 𝑡 , 𝑛𝑡 ) (72)
1 + 𝑖 𝑡𝑑
𝑖 𝑡𝑑
− = 𝑆 𝑚 (𝑚 𝑡 ) + 𝐶𝑙 ( 𝑓 𝑡, 𝑙 𝑡 , 𝑛𝑡 ) + 𝜒𝑚 (𝑟 𝑡 , 𝑚 𝑡 ) + 𝜓𝑡 (73)
1 + 𝑖 𝑡𝑑
𝜓𝑡 𝑀𝑡 = 0 (74)
𝑛𝑡 = 1 + 𝑖 𝑡𝑑 𝑧 𝑡 + (1 − 𝜔) 𝜋𝑡−1 𝑛𝑡−1 (75)
𝑔 𝑓
𝑖 𝑡𝑏 − 𝑖 𝑡𝑑 𝑖 𝑟𝑡 − 𝑖 𝑡𝑑 𝑖 𝑡 − 𝑖 𝑡𝑑 𝑖 𝑡𝑑 𝑖 𝑡 − 𝑖 𝑡𝑑
𝑧𝑡 = 𝑏𝑡 + 𝑟𝑡 + 𝑎𝑡 − 𝑚𝑡 − 𝑓𝑡 − 𝐶 ( 𝑓𝑡 , 𝑙 𝑡 , 𝑛𝑡 ) − Γ (𝑏 𝑡 , 𝑛𝑡 ) − 𝜒(𝑟 𝑡 , 𝑚 𝑡 )
1 + 𝑖 𝑡𝑑 1 + 𝑖 𝑡𝑑 1 + 𝑖 𝑡𝑑 1 + 𝑖 𝑡𝑑 1 + 𝑖 𝑡𝑑
(76)
𝜙 𝜙
𝑖 𝑟𝑡 = 𝑟 𝑡𝑛 Π𝑡 Π 𝑦 𝑡 𝑌 (77)
𝑓
𝑖𝑡 = 𝜌 𝑓 𝑖 𝑟𝑡 (78)
𝑔
𝑖𝑡 = 𝜌 𝑎 𝑖 𝑟𝑡 (79)
𝑖 𝑡𝑑 = 𝑚𝑎𝑥 −𝛾 , 𝑠
𝑖 𝑟𝑡 (80)
𝜒𝑏 𝑡𝑏 = 𝑏 𝑡 (81)
𝑦 𝑡 = 𝜒𝑐 𝑡𝑏 + (1 − 𝜒)𝑐 𝑡𝑠 (82)
(83)
54
𝑎 𝑡 + 𝑚 𝑡 + 𝑟 𝑡 = (1 + 𝑖 𝑡𝑎 )𝑎 𝑡−1 + 𝑚 𝑡−1 + (1 + 𝑖 𝑟𝑡−1 )𝑟 𝑡−1 + 𝐺 − 𝑇𝑡 (84)
𝑎 𝑡 + 𝑚 𝑡 + 𝑟 𝑡 = 𝑊¯ (85)
𝑓𝑡 = 𝑓¯ (86)
𝜃−1 1
𝜃−1
Π𝑡
©1 − 𝛼 Π ª 𝐹𝑡
® = (87)
1−𝛼 ® 𝐾𝑡
« ¬ " 𝜃−1 #
Π𝑡
𝐹𝑡 = 𝜆 𝑡 𝑦 𝑡 + 𝛼𝛽𝐸 𝑡 𝐹𝑡+1 (88)
Π
𝜂 1+𝜂
" 𝜃−1 #
𝜆𝑡 Δ 𝑡 𝑦𝑡 Π𝑡
𝐾𝑡 = 𝜇 + 𝛼𝛽𝐸 𝑡 𝐾𝑡+1 (89)
𝑞 exp{−𝑞𝑦 𝑡 } Π
𝜆 𝑡 = 𝑞 𝜒 exp{−𝑞𝑐 𝑡𝑏 } + (1 − 𝜒) exp{−𝑞𝑐 𝑡𝑠 } (90)
𝜃−1 𝜃
𝜃−1
Π𝑡
©1 − 𝛼
𝜃
Π𝑡 Π ª
Δ𝑡 = 𝛼 Δ 𝑡−1 + (1 − 𝛼) ® (91)
Π 1−𝛼 ®
« ¬
𝑦𝜂
𝜇× =1 (92)
𝑞 exp{−𝑞𝑌 }
The consumption Euler equations (65) and (65) of the saver and borrowers imply that the the inverse of
the discount factor of each agent is equal to the real borrowing and savings rate
1 + 𝚤¯𝑏
(𝛽 𝑏 ) −1 = (93)
Π̄
and
1 + 𝚤¯𝑠
(𝛽 𝑠 ) −1 = (94)
Π̄
In steady-state, we assume 𝜌 𝑓 = 1, 𝜌 𝑎 = 1 which implies that 𝚤¯ 𝑓 = 𝚤¯𝑔 = 𝚤¯𝑟 where 𝑖 𝑟𝑡 = 𝑟 𝑛 Π 𝜙Π 𝑦 𝜙𝑌
The budget constraint (66), together with that in steady state implies
𝑐¯𝑏 1 − 𝛽𝑏 1 ¯
=1− 𝑏 (95)
𝑦¯ 𝛽𝑏 𝜒
55
which when combined with the resource constraint (82) implies the steady state consumption of the saver
as
𝑐¯𝑠 1 1 − 𝛽𝑏 ¯
=1+ 𝑏 (96)
𝑦¯ 1 − 𝜒 𝛽𝑏
𝑏¯
We directly choose the steady-state output debt 𝑦¯ from the data, in which case the two equations above
𝑐¯ 𝑏 𝑐¯ 𝑠
pin down 𝑦¯ and 𝑦¯ for a given 𝜒. Neither ratio, however, enters the linear approximation of the model shown
in next subsection.
The first-order condition for lending (53) implies that
𝑖¯𝑏 − 𝑖¯𝑑 𝛽𝑏 Γ̄
= 1 − = 𝜈 (97)
1 + 𝑖¯𝑏 𝛽 𝑠
𝑏¯
𝑏¯
which, given 𝑦¯ and 𝜈 pins down
𝛽 𝑏
Γ̄ 1 − 𝛽 𝑠
= ¯
(98)
𝑦¯ 𝜈 𝑏𝑦¯
Steady state profit is then
𝛽𝑏 ¯ ¯
𝑧¯ = (1 − ) 𝑏 − 𝐶 − Γ̄ − 𝜒¯ (99)
𝛽𝑠
Steady state net worth is
𝑛¯ = (1 + 𝑖 𝑑 )( 𝑧¯ + (1 − 𝜔)𝑛𝜋 −1 ) (100)
or
𝑧¯ 𝛽𝑠 + 𝜔 − 1
= 𝛽 𝑠 𝜋 −1 − (1 − 𝜔)𝜋 −1 = (101)
𝑛¯ 𝜋
Finally, we directly pick the steady state values of 𝑟,¯ 𝑎¯ and 𝑓¯ from the data. These do not affect the steady
state values of the variables described above, but are important for dynamics.
𝑐ˆ𝑡𝑏 = 𝐸 𝑡 𝑐ˆ𝑡+1
𝑏
− 𝜎(ˆ𝚤 𝑡𝑏 − 𝐸 𝑡 𝜋ˆ 𝑡+1 + 𝜁ˆ𝑡 − 𝐸 𝑡 𝜁ˆ𝑡+1 ) (102)
𝑐ˆ𝑡𝑠 = 𝐸 𝑡 𝑐ˆ𝑡+1
𝑠
− 𝜎(ˆ𝚤 𝑡𝑑 − 𝐸 𝑡 𝜋ˆ 𝑡+1 + 𝜁ˆ𝑡 − 𝐸 𝑡 𝜁ˆ𝑡+1 ) (103)
56
Budget constraint of borrower (66) yields
1 1 1 𝑏 𝑦¯ 𝑦¯
𝑏ˆ 𝑡 = 𝑏 𝑏ˆ 𝑡−1 − 𝑏 𝜋ˆ 𝑡 + 𝑏 𝚤ˆ𝑡−1 − 𝜒 𝑦ˆ 𝑡 + 𝜒 𝑐ˆ𝑡𝑏 (104)
𝛽 𝛽 𝛽 𝑏¯ 𝑏¯
Lending condition (69) yields
𝛽𝑠 − 𝛽𝑏 ˆ 𝛽𝑠 − 𝛽𝑏
𝚤ˆ𝑡𝑏 − 𝚤ˆ𝑡𝑑 = (𝜈 − 1) 𝑏 𝑡 − 𝜄 𝑛ˆ 𝑡 (105)
𝛽𝑠 𝛽𝑠
Net worth condition yields
1−𝜔 𝛽𝑏 1 − 𝜔
𝛽 𝑠 𝑛ˆ 𝑡 = 𝛽 𝑠 𝚤ˆ𝑡𝑑 + 𝑧ˆ𝑡 + 𝑛ˆ 𝑡−1 − 𝑠 𝜋ˆ 𝑡 (106)
Π̄ 𝛽 Π̄
The definition of profits yields
𝑛 𝛽 𝑏 𝑏¯ 𝑟¯ 𝑎¯ 𝑓¯ 𝛽 𝑏 𝑏¯ 𝑟¯ 𝑎¯ 𝑓¯ 𝑓 Γ̄
𝑧ˆ𝑡 = −{ 𝑠
+ + − }ˆ𝚤 𝑡𝑑 + 𝑠 𝚤ˆ𝑡𝑏 + 𝚤ˆ𝑟𝑡 + 𝚤ˆ𝑡𝑎 − 𝚤ˆ𝑡 + 𝜄 𝑛ˆ 𝑡 (107)
Λ 𝛽 Λ̄ Λ̄ Λ̄ Λ̄ 𝛽 Λ̄ Λ̄ Λ̄ Λ̄ Λ̄
The resource constraint yields
Policy rule
Deposit bound
𝚤ˆ𝑡𝑎 = 𝜌 𝑎 𝚤ˆ𝑟𝑡 if 𝚤ˆ𝑟𝑡 < 𝑖 𝑒𝑙𝑏 and 𝚤ˆ𝑡𝑎 = 𝚤ˆ𝑟𝑡 if 𝚤ˆ𝑟𝑡 > 𝑖 𝑒𝑙𝑏 (112)
𝑓 𝑓
𝚤ˆ𝑡 = 𝜌 𝑓 𝚤ˆ𝑟𝑡 if 𝚤ˆ𝑟𝑡 < 𝑖 𝑒𝑙𝑏 and 𝚤ˆ𝑡 = 𝚤ˆ𝑟𝑡 if 𝚤ˆ𝑟𝑡 > 𝑖 𝑒𝑙𝑏 (113)
1 (1−𝛼) (1−𝛼𝛽)
Coefficient are defined as: 𝜎 ≡ 𝑞¯ 𝑦¯ , 𝜅≡ 𝛼( 𝜂+𝜎 −1 )
, 𝛽 ≡ 𝜒𝛽 𝑏 + (1 − 𝜒) 𝛽 𝑠 , 𝑖 𝑒𝑙𝑏 ≡ − log(1 + 𝚤¯𝑑 ). Observe
¯ ¯
that once one chooses parameters for (𝜎, 𝜅, 𝛽.𝛽 𝑠 , 𝛽 𝑏 , 𝜒, 𝜈, 𝜄) and pins from the data ( 𝑏𝑦¯¯ , Λ̄𝑏 , Λ̄𝑟¯ , Λ̄𝑎¯ , Λ̄𝑓 ), then 𝜔
Γ̄
and Λ̄
are not free parameters but implied by the model equations.
The analytic characterization in the text for the IS equation is obtained by (1) combining the two
consumption Euler equations with the resource constraint and (2) defining the natural rate of interest, 𝑟ˆ𝑡𝑛 , as
57
the interest rate consistent with zero deviation of output from steady state and inflation on target in the absence
of a lower bound.
58