Deposit Convexity, Monetary

Policy, and Financial Stability

Emily Greenwald, Sam Schulhofer-Wohl and Joshua Younger

Working Paper 2315 October 2023

Research Department
https://doi.org/10.24149/wp2315
Working papers from the Federal Reserve Bank of Dallas are preliminary drafts circulated for professional comment.
The views in this paper are those of the authors and do not necessarily reflect the views of the Federal Reserve Bank
of Dallas or the Federal Reserve System. Any errors or omissions are the responsibility of the authors.
 Deposit Convexity, Monetary Policy, and Financial Stability *

Emily Greenwald†, Sam Schulhofer-Wohl‡ and Joshua Younger§

October 2023

Abstract

In principle, bank deposits can be withdrawn on demand. In practice, depositors tend to

maintain stable balances for long periods, allowing banks to fund long-dated assets.
Nevertheless, the cost of deposit funding influences banks’ capacity for maturity
transformation. Banks and researchers conventionally model the response of deposit
interest rates to market interest rates as constant, implying that deposits have nearly
constant duration. Contrary to this standard assumption, we show empirically that the
“beta” of deposit rates to market rates increases as market rates rise, causing the duration
of deposits to fall. The amount of duration risk delivered to bank balance sheets via this
channel from March 2022 to September 2023 is comparable in magnitude to the amount
of duration risk absorbed by each of the several large-scale asset purchase programs the
Federal Reserve has undertaken since 2008. Dynamic betas present a significant
challenge to bank portfolio hedgers by introducing large and dynamic risks that are difficult
to model and impractical to replicate on the asset side of the balance sheet. As a result,
deposit convexity amplifies monetary policy transmission and increases financial fragility,
mechanisms that recent banking stresses have highlighted.

JEL Codes: E430, E440, E520, G120, G210

Keywords: Banks, Depository Institutions, Interest Rates, Bank Run, Financial Markets,
Central Bank, Monetary Policy, Policy Effects

*
The views expressed herein are those of the authors and not necessarily those of the Federal Reserve Bank of Dallas,
Federal Reserve Bank of New York, or Federal Reserve System. The authors would like to thank Darrell Duffie, Stephan
Luck, Antoine Martin, Matt Plosser, and Alexi Savov for helpful conversations and comments.
†Emily Greenwald, Federal Reserve Bank of Dallas, [email protected].
‡Sam Schulhofer-Wohl, Federal Reserve Bank of Dallas, [email protected].
§Joshua Younger, Federal Reserve Bank of New York, [email protected].
1 Introduction

Banks fund long-duration assets such as loans in large part by issuing immediately available demand deposits.

Although demand deposits nominally have no duration, in practice deposit funding often behaves more like

long-term debt: Depositors typically maintain steady balances for long periods of time at relatively stable

interest rates. The transformation of nominally immediate-maturity, highly liquid deposits into stable, long-

term funding gives banks the capacity to carry long-term, fixed-rate, often illiquid assets and is at the heart

of banks’ economic function.

Recent stress in the U.S. banking system has drawn attention to gaps in risk identification and measure-

ment related to this funding model. Banking is inherently fragile (Diamond and Dybvig, 1983; Diamond,

2007). Depositors sometimes exercise their right to immediately liquidate. Moreover, depositors may move

funds more rapidly when market interest rates rise, thus exposing the bank to complex and dynamic interest

rate risks. And those risks can interact, as depositors’ response to changing market interest rates can erode

the earnings power of a banking franchise and, in the extreme, even trigger a run. An understanding of these

effects is a key input into monetary and financial stability policy—an issue made more urgent by the deposit

outflows at some banks amid the rapid rise in U.S. interest rates starting in 2022 (e.g., Board of Governors

of the Federal Reserve System, 2023).

Typical models of depositor behavior used in both economic research (e.g., Drechsler et al., 2021) and the

banking industry treat the equilibrium cost of a bank’s deposits as moving in fixed proportion to the short-

term, risk-free market interest rate—a constant “beta.” Drechsler et al. (2017) provide microfoundations for

this model. Drechsler et al. (2023) observe that if higher interest rates lead to deposit outflows, the bank’s

deposit beta will be higher, but they propose addressing this effect by modeling deposits as having a higher,

yet still constant, effective beta.

If deposits actually had a constant beta to market rates, their duration would fall somewhat with rising

market rates due to discounting effects, but not enough to eliminate the cost advantage of deposits over

long-term debt. In such an environment, banks could match the duration of their liabilities to the duration

of assets and could generally expect wider margins between interest income and interest expense and greater

profits when interest rates rise. The value of the deposit franchise would then increase with the short-term,

risk-free interest rate, a mechanism highlighted by Drechsler et al. (2021) and Luck et al. (2023). The

primary interest rate risk that banks would need to hedge would be that of falling interest rates, motivating

banks to acquire long-term assets to back their deposit liabilities.

Despite the conventional approach to modeling deposit betas, there is ample evidence that the ratio

between deposit rates and the short-term, risk-free rate increases as the risk-free rate rises—a dynamic beta.

1
If a bank does not raise deposit rates more rapidly as the risk-free rate reaches higher levels, the bank

can expect to lose deposits. Lost deposit funding is generally replaced with wholesale sources that tend to

be closer to market rates, and therefore more expensive. Thus, monetary tightening reduces the duration

of deposits by more than the simple effect of discounting. In a sense, commercial banks have sold their

depositors options on interest rates, making deposits highly convex. Owing to both practical and accounting

considerations, these duration dynamics are difficult to replicate on the asset side of the balance sheet or with

derivatives, leaving typical banks with an unhedged exposure that exhibits significant negative convexity.

In this paper, we develop a simple analytical model of deposits’ duration and convexity. Through the

lens of the model, we show that deposit convexity has three important consequences for monetary policy

and financial stability.

First, deposit convexity amplifies the bank lending channel of monetary policy transmission (e.g., Bernanke

and Blinder, 1988; Drechsler et al., 2017). That amplification has two sources. All else equal, increasing

deposit betas raise the expected interest expense of incremental funding more quickly than static deposit

betas. Also, because deposit convexity means that rising interest rates shorten the effective duration of

deposits, rising interest rates leave banks holding more net duration on their balance sheets. Thus, rising

interest rates more rapidly reduce banks’ incentive to lend. Similar to the effects of a central bank reducing

holdings of long-duration assets and leaving the private sector to hold more duration (Li and Wei, 2013), this

effect erodes banks’ capacity to hold other long-duration assets, adding to the headwinds to credit supply.

We estimate that the increase in interest rates from December 2021 through July 2023 increased the duration

risk on bank balance sheets by the equivalent of $4.5 trillion in 10-year Treasury securities, of which roughly

60% was due to discounting effects and 40% to dynamic betas. The effect of dynamic betas on the duration

risk held by banks is larger than the amount of duration in any of the Federal Reserve’s pre-Covid-19 asset

purchase programs and comparable to the amount of duration in the market functioning purchases in March

and April 2020 or the subsequent asset purchases from May 2020 through March 2022.1

Second, deposit convexity can amplify the transmission of central bank asset runoff. Since June 2022,

the Federal Reserve has not only raised interest rates but also allowed some of its assets to mature. When

the central bank reduces its asset holdings, its liabilities also fall. The Federal Reserve’s liabilities include

both bank reserves and non-reserve liabilities. To the extent that the reduction in liabilities comes from

bank reserves—a liability of the central bank and an asset of commercial banks—bank deposits may also

fall as commercial banks have fewer assets to fund. Commercial banks then lose deposits at the same time
1 Deposits are both a liability of banks and an asset of depositors. A change in the aggregate duration of deposits therefore

does not change the amount of duration on the consolidated balance sheet of the private sector, including both banks and
non-banks. However, the bank lending channel is predicated on a special role for banks in intermediation, such that a change
in the distribution of duration risk between banks and non-banks has macroeconomic consequences.

2
that convexity causes the remaining deposits’ duration to fall, further reducing commercial banks’ capacity

to hold duration. While this channel’s effect in the current U.S. tightening cycle has been modest to date,

as most of the reduction in Federal Reserve liabilities since 2022 has come from non-reserve liabilities, its

effect bears monitoring should reserves fall more notably.

Finally, dynamic betas affect the role of interest rates in the transition from a stable to a run equilibrium

in banking. Drechsler et al. (2023) argue that increases in the value of the deposit franchise relative to

banking assets can generate fragility, as depositors are increasingly incentivized to withdraw their funding in

a coordinated way. We find that modeling interest rate risk with static betas tends to overstate the duration

of deposits when rates are low and understate duration when rates are high. Importantly, relying on this

overstated duration of deposits increases run risk in the transition away from the effective lower bound (ELB)

on interest rates by allowing for the fair value of fixed-income assets to deteriorate more quickly than the fair

value of the deposit franchise actually increases. This represents a novel channel through which maintaining

interest rates at the lower bound increases financial fragility.

The paper proceeds as follows. Section 2 describes standard approaches that banks use to model deposit

betas, which typically ignore the dynamics in favor of assuming a constant beta. Section 3 provides empirical

evidence on the dynamics of deposit betas. Section 4 lays out a simple model of dynamic deposit betas.

Section 5 uses the model to analyze how dynamic betas affect the bank lending channel of monetary policy,

section 6 analyzes the effects of central bank balance sheet expansion and contraction, and section 7 examines

how dynamic betas can contribute to bank runs. Section 8 concludes.

2 Bank deposit modeling

Like economic researchers, bankers make assumptions about deposits’ sensitivity to market interest rates.

While banks have discretion to set the rate of interest on their deposits, in practice they must respond

to market forces and must model how those responses will unfold in order to manage risk and maximize

profits. Banks’ deposit beta modeling can range from standardized assumptions to multivariate regression

models and may also include management judgment. Almost all publicly available deposit beta techniques

employed in practice are linear. The following describes banks’ modeling practices based on the authors’

personal experience observing bank practices and working with vendor models of interest rate risk (see, e.g.,

Greenwald and Gray, 2013).

Default assumptions: Smaller banks often use vendor-supplied interest rate risk models that include

vendor-recommended or default assumptions for deposit betas. These models typically apply a constant

deposit beta to all scenarios regardless of the magnitude or direction of the change in interest rates. Some

3
vendors derive default betas from empirical models. Others derive them from information published by

regulators. For example, in 1995 the federal bank regulatory agencies proposed a standardized measure of

banks’ interest rate exposures (Office of the Comptroller of the Currency et al., 1995). The proposal included

a weighted-average duration maturity ladder for all non-maturity deposits for a 100-basis-point change in

rate. Bank regulatory agencies ultimately did not adopt the standardized measure, but some vendor models

continue to leverage it to generate default deposit betas. In models of this type, deposit volumes assumed

to mature within one year are assigned a 100 percent beta and remaining volumes are assigned a 0 percent

beta, resulting in a constant overall beta regardless of the level of market interest rates.

Linear regressions: Vendor models and customized deposit modeling studies predominantly use univariate

linear regressions of a time series of deposit rates on market rates to estimate a deposit beta. Models can

better account for the effect of market rates on deposit pricing by controlling for other relevant factors,

such as deposit volumes. In practice, banks that attempt to control for such other factors typically use

multivariate linear regression, thus continuing to assume that the marginal effect of market rates on deposit

rates does not change with the level of the market rate. Some models include lags of deposit rates, which

allows for a dynamic impulse-response relationship between market rates and deposit rates; however, the

model coefficients are still assumed to be independent of the level of market rates.

3 Empirical evidence on deposit betas

This section uses evidence on bank executives’ self-reported estimates of deposit betas in different interest

rate environments as well as banks’ reports of deposit volumes and interest expenses to show how deposit

betas vary with risk-free short-term interest rates in practice.

3.1 Bank executives’ self-reported deposit betas

The Federal Reserve’s Senior Financial Officer Survey (SFOS) periodically questions roughly 80 banks,

across a range of sizes and business models, about their funding.2 The November 2022 and May 2023

surveys included questions on realized deposit betas since the start of the hiking cycle (March 2023) as well

as expectations for deposit betas in the next six months.

The survey questions segmented deposits into three types, mirroring those used in liquidity regulations

(Bank for International Settlements, 2013): (1) retail deposits held by individuals, (2) wholesale operational

deposits held by institutions and which are generated by clearing, custody, and cash management, and (3)

non-operational wholesale deposits for balances in excess of those operational deposits. The surveys asked:
2 See https://www.federalreserve.gov/data/sfos/sfos.htm.

4
 • November 2022:
– “For each of the deposit categories below, please indicate your bank’s cumulative deposit beta
from March 2022 through the end of October 2022. For reference, the target range for the federal
funds rate over this period increased 300 basis points.”
– “Looking ahead six months to April 2023, for each of the deposit categories below, please select
your expectations for your bank’s cumulative deposit beta since March 2022.”
• May 2023:
– “For each of the deposit categories below, please indicate your bank’s cumulative deposit beta
from March 2022 through the end of April 2023. For reference, the target range for the federal
funds rate over this period increased 475 basis points.”
– “Looking ahead to November 2023, please select your expectations for your bank’s cumulative
deposit beta since March 2022 for each of the deposit categories below.”

For context, the effective federal funds rate (EFFR) averaged 1.49% from the beginning of March 2022 to

the end of October 2022 while 6-month spot-starting interest rate swaps tied to EFFR averaged 4.64% over

the November 2022 survey month. That suggests the forward-looking cumulative betas for the November

2022 survey apply to a 14-month period with an average EFFR of 2.84%. Similarly, for the May 2023 survey,

EFFR averaged 2.74% from March 2022 to April 2023 and 6-month swaps averaged 4.94% over the survey

month, suggesting the cumulative betas apply to a 20-month period with an average EFFR of 3.32%.

Because the backward-looking and forward-looking questions apply to different rate levels, we can use

the survey results to test whether depository institutions expect dynamic betas.

Figure 1 summarizes the responses for domestic banks. They show a clear expectation of increases in

deposit betas through April 2023. This shift in expectations is most pronounced for retail deposits, which

are both the largest component of deposit funding at most banks and thought to be among the most stable.

How do the survey responses square with our description of bankers’ standard modeling? Typical risk

monitoring and management policies may not explicitly recognize a dynamic beta, but banks regularly review

model assumptions. If these reviews are sufficiently frequent, they can produce dynamic betas. The SFOS

responses suggest that many banks recalibrate their models in this way. As we discuss below, however,

implementing dynamic betas via regular review of static assumptions can introduce fragility relative to a

dynamic model.

3.2 Banks’ realized deposit interest expenses across rate environments

To get a more granular sense of deposit beta behavior, we turn to the Call Reports (Federal Deposit Insurance

Corporation, 2023a,b,c). All insured depository institutions in the United States disclose their banking assets,

liabilities, income and expenses on these reports on a quarterly basis. We calculate banks’ deposit interest

5
rates by examining their interest-bearing domestic and total liabilities, as well as interest expense in total

and that associated with depository liabilities.

From these data, we calculate annualized aggregate rates paid for all reporting commercial banks with

a domestic parent for two complementary measures: (1) interest expense attributable to domestic deposits

as a percentage of total interest-bearing domestic deposit balances (RD ) and (2) total interest expense

as a percentage of the book value of total assets (RF ). RD reflects the behavior of domestic deposits in

isolation—in a sense, the pure deposit beta—while RF reflects the total cost of funds. As we will describe

later, both are relevant in measuring interest rate risk in the banking book.

Figure 2 shows how bank deposit interest rates and total funding costs have evolved over three recent

monetary policy tightening cycles: 2004–2007, 2015–2019 and the cycle that began in 2022 (for data through

June 2023). The relationship of deposit rates and funding costs to the EFFR is visibly nonlinear, with the

slope increasing as the EFFR rises. Regression results reported in Table 1 demonstrate that the nonlinearity

is statistically significant, rejecting the assumption of a constant beta.

The empirical findings raise the question of why betas are dynamic. Although many observers assume

that banks simply “raise rates” across all their deposit products to compete for funding, history suggests

that a changing mix of deposit products is a key driver. In particular, as shown in the left panel of Figure

3, depositors have tended to migrate into time deposits from other products as market rates have risen.

For example, from March 2022 to September 2023, aggregate time deposit balances among domestically

chartered commercial banks increased by nearly $700 billion while other balances fell by more than $1.7

trillion3 . Time deposit rates, shown in the right panel of Figure 3, also move more closely with policy rates

than do the rates on other deposits. Call Report data show that time deposits have therefore accounted for

a disproportionate fraction of increased commercial bank interest expense during the past few hiking cycles:

roughly 52% (1.4 times ex ante market share), 34% (1.6x) and 24% (2.6x) in 2004–2006, 2015–2018, and

2022–2023 (through the second quarter of 2023), respectively.

4 A simple model of interest rate risk for a deposit franchise

The mechanism of funding costs rising due to a changing mix of deposits motivates our simple model of the

interest rate risk on banks’ deposits.

Consider a bank whose liabilities at time t consist of non-maturity demand deposits Dt and wholesale

funding Wt . Suppose that a fraction β(Rt ) of the deposits come from depositors who care only about interest

earnings and will leave immediately if they are not paid the market rate Rt , while a fraction 1 − β(Rt ) come
3 Data provided by FRED, tickers DPSDCBW027NBOG and LTDDCBW027NBOG.

6
from depositors who do not respond to market interest rates because they value other services the bank

provides. The bank price discriminates between the two groups, paying interest Rt to the rate-sensitive

depositors and no interest to the rest. We allow β to be a function of Rt because of the observation that

the deposit mix tends to change as market rates rise. We assume 0 < β(Rt ) < 1 and β ′ (Rt ) ≥ 0. However,

we assume β does not depend directly on time. Both types of depositors also leave the bank for exogenous

reasons at rate α per unit time, at which point the bank replaces their deposits with wholesale funding, which

also pays Rt . (In practice, a bank might incur marketing costs to obtain new price-insensitive depositors,

might pay price-insensitive depositors an interest rate that depends on when they became customers, might

have some fixed-rate time deposits, and would have some equity capital; allowing for these details would not

change the basic implications of our model.) We further assume a constant balance sheet size and normalize

the bank’s total liabilities Dt + Wt to 1.

In effect, this bank has Wt of floating-rate wholesale debt, β(Rt )Dt of floating-rate deposits, and [1 −

β(Rt )]Dt of zero-coupon, zero-interest debt. The duration of the bank’s deposits and its total liabilities

depends on the exogenous decay rate α as well as the relationship between β and Rt .

For simplicity, we will work in continuous time and calculate duration by considering perturbations around

a steady state where Rt is constant (both on a spot basis and in expectation with no term premium; i.e., a

flat yield curve). Each dollar of non-interest-bearing deposits costs the bank less than one dollar because, in

expectation, the bank will repay this dollar in the future and can discount that future repayment at rate R.

This savings represents the value of the bank’s deposit franchise. The wholesale deposits and interest-bearing

deposits, on the other hand, cost the bank exactly as much as their par value. The present value of the

bank’s liabilities when it has deposits D0 , treating Rt = R as constant and adopting the convention that

liabilities are negative, is thus

 Z ∞ 
−(R+α)t α + β(R)R
L̃R = −1 + [1 − β(R)]Re dt D0 = − D0 − W0 . (1)
0 R+α

Taking the derivative with respect to R to get duration:

dL̃R 1 α R
∆R = = [1 − β(R)] 2
− β ′ (R) . (2)
dR D0 (R + α) (R + α)

Equation (2) captures key pieces of intuition for both the standard model of constant-beta deposits and

our model of dynamic betas.

7
 Consider first the special case where β(R) equals a constant β̄. Then (2) simplifies to

α
∆R,β̄ = (1 − β̄) . (3)
(R + α)2

The bank’s liabilities have positive duration, i.e., the present value of the bank’s payments to its creditors

falls as interest rates rise. This feature allows the bank to hedge the interest rate risk on long-duration assets.

The duration of the bank’s liabilities, and thus the size of this hedge, is increasing in the share of liabilities

that are deposits D0 and decreasing in the share of deposits that pay market rates β̄. For R < α, which is

the case for typical ranges of market interest rates and typical deposit longevity of 5–7 years, duration is

also decreasing in the deposit runoff rate α.

With dynamic betas, equation (2) shows that deposit duration is reduced relative to what it would be

with a constant beta. Intuitively, deposits have less duration when rising rates can cause them to reprice

faster.4

Next, we consider the convexity of the bank’s liabilities:

d2 L̃R 1 2α 2α R
ΓR = = −[1 − β(R)] − β′ − β ′′ . (4)
dR2 D0 (R + α)3 (R + α)2 R+α

For constant β, convexity simplifies to:

2α(1 − β̄)
ΓR,β̄ = − , (5)
(R + α)3

the standard result that a bond has negative convexity (from the issuer’s perspective).

When β depends on R, convexity has additional terms related to the first and second derivatives of β, as

shown in (4). The second term in the numerator of the fraction in (4) is non-negative given our assumption

that β ′ ≥ 0. However, the sign of the first term in the numerator is ambiguous. In particular, because β(R)

is bounded above by 1, β ′′ must become negative for sufficiently large R, and if β ′′ is sufficiently negative,

convexity becomes positive. Where that transition occurs depends on the shape of β(R). To the extent that

these higher-order terms dominate, the complexity of managing dynamic interest rate risk is only increased.

Finally, Luck et al. (2023) and Drechsler et al. (2021, 2023) point out that deposits’ capacity to fund long-

duration assets relies on stable or, at least, predictable deposit balances. While aggregate banking deposits

have been broadly stable in most interest rate regimes in the United States, empirical studies suggest that

runoff (parameterized as α in our model) can rise with interest rates (Duffie and Krishnamurthy, 2016; Kang-

Landsberg and Plosser, 2022). Our model can accommodate such a relationship by making α an increasing
4 Sufficiently β ′ (R) α 1
large local variation in β with R can even change the sign of ∆R . Specifically, ∆R < 0 if 1−β(R) > R R+α
.
This condition rarely holds in practice: Banks tend to vary deposit rates gradually in light of both the empirical behavior of
their deposits and practical constraints on their ability to manage dynamic interest rate risk.

8
function of R, in which case duration becomes

′
ˆ R = ∆R − [1 − β(R)] Rα (R) ,
∆ (6)
[R + α(R)]2
| {z }
>0

where ∆R shown in (2) is the duration assuming α is constant. Equation (6) shows that if rising rates

accelerate deposit runoff, the duration of deposits is reduced. This effect is similar to the dynamic beta

effect described above, but in this case value is destroyed by depositors withdrawing their funds rather than

demanding higher rates. Additionally, the value of the bank’s deposit franchise would be reduced by the

cost to acquire replacement depositors.

The remaining sections use the model to examine implications of deposit convexity and dynamic betas for

the bank lending channel of monetary policy, the transmission of central bank asset runoff, and the fragility

of banks.

5 The bank lending channel

The bank lending channel of monetary policy refers to the idea that changes in monetary policy may influence

not only the interest rate on loans but also the quantity of loans that banks are willing to make. In its original

formulation (Bernanke and Blinder, 1988), the bank lending channel of monetary policy operated through

reserve requirements. However, bank reserve requirements have not significantly constrained banks in the

United States for many decades (Bennett and Peristiani, 2002), are generally not expected to bind in the

ample reserves operating regime that the Federal Reserve has employed since the Global Financial Crisis

(Ihrig et al., 2020), and were set to zero by the Federal Reserve Board in 2020 (Board of Governors of the

Federal Reserve System, 2020). More recently, Drechsler et al. (2017) describe another mechanism for the

bank lending channel: the behavior of deposits. In that framework, deposits have a constant beta to market

rates; the spread between market rates and deposit rates therefore widens when the central bank raises the

policy rate, giving depositors an incentive to move balances out of the banking system, which constrains

bank lending. We extend Drechsler et al. (2017) by incorporating dynamic betas.

Many bank assets are relatively illiquid, so banks typically do not continuously re-optimize their entire

asset books, but rather make marginal decisions about whether to add particular assets. Suppose a bank has

the opportunity to make a floating-rate, T -year loan with book value A that earns a risk-adjusted interest

rate Rt + s. For example, this could be a commercial loan where the bank is able to earn a risk-adjusted

spread s over market rates due to comparative advantage in sourcing or evaluating loans in a particular

industry or geography.

9
 If the bank funds the loan with deposits that must be replaced with wholesale funding as they run off at

rate α, the present value of the bank’s marginal net income from the loan over the loan’s life is

Z T
e−Rt t Rt + s − e−αt β(Rt )Rt − (1 − e−αt )Rt dt
 
N =
0
Z T Z T Z T
(7)
−Rt t −(Rt +α)t −(Rt +α)t
=s e dt + R̃T (1 − β̃T ) e dt + e (Rt − R̃T )[1 − β(Rt ) − (1 − β̃T )]dt
0 0 0

where R T −(R +α)t RT

e t
Rt dt e−(Rt +α)t β(Rt )dt
R̃T = 0R T , β̃T = 0
RT . (8)
e −(R t +α)t dt e −(Rt +α)t dt
0 0

The first term in (7) is the present value of the bank’s spread s on the loan and does not depend on funding

costs. The second term is the benefit to the bank of funding the loan with deposits rather than wholesale

funds, if market rates remain constant over the life of the loan. The final term is the present value of the

covariance between market rates and the bank’s funding advantage 1 − β(Rt ). This covariance is strictly

negative for dynamic betas (β ′ ≥ 0). Thus, dynamic betas reduce banks’ marginal net income from lending

and motivate banks to constrain even floating-rate lending more than they would if deposit betas were

constant.

For fixed-rate lending, dynamic betas have an additional effect. Suppose that, as is typical, the bank has

a risk limit on its consolidated net duration exposure. Acquiring a fixed-rate asset increases the bank’s net

duration exposure, while issuing deposits decreases the bank’s net duration exposure. Equation (2) shows

that dynamic betas reduce net deposit duration, which means the marginal increase in total duration exposure

from acquiring a fixed-rate asset and funding it with deposits is larger. Thus, as rates rise, dynamic betas

reduce banks’ ability to take on additional net duration risk. Even if the marginal loan would be profitable

on a funding cost basis, duration risk considerations could cause banks to curtail the supply of term credit

more than they would with static betas.

The magnitude of these effects can be substantial. As of December 2021, domestically chartered commer-

cial banks had approximately $16.5 trillion of deposit balances (Board of Governors of the Federal Reserve

System, 2021). We estimate the duration of these deposits by considering separate betas for retail deposits,

which tend to be sticky; wholesale operational deposits, which are associated with institutional payment

activity and are less sticky; and wholesale non-operational deposits, which are wholesale customers’ funds in

excess of operational needs and have interest rates that move most closely with market rates.

On average, respondents to the May 2023 SFOS reported realized retail, operational and non-operational

betas of 24%, 41% and 50%, respectively, over the March 2022–April 2023 period, when the federal funds

10
rate averaged 2.75%.5 Respondents projected cumulative retail, operational and non-operational betas of

35%, 50% and 59% for March 2022–November 2023, a period when interest rates were expected to average

3.4% based on swap rates. Public disclosures by banks designated as Global Systemically Important Banks

(GSIBs), representing about half of aggregate deposits, indicate that approximately 55% of deposits are

retail deposits, 25% are wholesale operational deposits, and 20% are wholesale non-operational deposits.

That distribution of deposit types implies an aggregate deposit beta of 33% for March 2022 to April 2023

and 44% for March 2022 to November 2023, and (comparing the two as well as the interest rates across

the periods) an aggregate beta of roughly 60% for the May–November 2023 period when swaps forecast an

interest rate of 4.94%. Assuming betas are never modeled below 25%, these aggregate betas and interest

rates can be fitted to a quadratic relationship of β(R) ≈ 1.79R2 − 1.90R + 25.4 (where β(R) and R are both

expressed in percentage points).

We make three simplifying assumptions to estimate the aggregate duration of deposits. First, we assume

deposits have a half-life of 5 years.6 Second, in practice, banks manage to a medium-term rate outlook.

Therefore, we use 5-year Treasury rates instead of EFFR as the R in our calculations. In addition to

providing a somewhat conservative estimate of the change in interest rates and resulting duration effect, we

believe this is a more accurate reflection of internal modeling choices made by the banks themselves in their

asset-liability management decisions.

Third, we calculate duration as if beta is static but updated by bank management at different rate levels

(i.e., β(R) is stepwise). This corresponds to typical bank practice, as discussed above, and allows us not to

estimate β ′ (R). Including the terms in β ′ (R) would increase our estimates.

At a beta of 25% with 5-year Treasury yields at 1.27%, banks’ deposits as of December 2021 corresponded

to negative duration risk of approximately $7.5 trillion of 10-year Treasury securities (often referred to as

10-year equivalents, or TYE). By the end of July 2023, higher rates and betas reduced the magnitude of

that duration risk to approximately negative $3.0 trillion TYE. In other words, the change in interest rates

effectively delivered the same amount of duration risk to commercial banks as if those institutions had bought

nearly $4.5 trillion of 10-year Treasury securities. Of that, roughly $1.8 trillion (40% of the total duration

delivery) was due to dynamic betas. And, as noted above, that figure represents a lower limit under our

framework.

The total quantity of duration delivery to banks via deposit dynamics is comparable to the cumulative

duration removed from the private sector by all of the Federal Reserve’s large-scale asset purchase programs
5 The survey response options were categorical. We compute averages by assigning each response to the midpoint of the

chosen category.
6 This assumption is roughly consistent with modeling assumptions reported by national banks in a spring 2023 survey

by the Office of the Comptroller of the Currency (OCC) (https://www.occ.gov/publications-and-resources/publications/

interest-rate-risk-statistics-reports/files/pub-interest-rate-risk-statistics-spring-2023.pdf).

11
(LSAPs) since the Global Financial Crisis (which total roughly $5.5 trillion TYE; Gulati and Smith, 2022).

Focusing specifically on the role of dynamic betas, the $1.4 trillion TYE impact is larger than any of the

pre-Covid-19 asset purchase programs and comparable to market functioning purchases in spring 2020 as

well as the subsequent LSAPs from approximately May 2020 to March 2022.

These estimates assume a constant deposit runoff rate α. In practice, banks may revise runoff rates

upward when interest rates reprice more than expected.7 The March 2023 banking stresses may also have

motivated some banks to assume faster runoff. All else equal, we estimate a 1-year reduction in the half-life

of deposits corresponds to an additional $600 billion TYE of duration delivery to banks.

6 The central bank balance sheet channel

As described by Li and Wei (2013), changes in central bank asset holdings are typically modeled as affecting

term premia by changing the amount of duration held by the private sector. The negative duration of

deposits creates the potential for an additional channel through which central bank asset holdings can affect

term premia.

When a central bank acquires assets, it pays for the purchase by creating reserves. Typically, the seller

of the assets is some entity other than a commercial bank, such as a securities dealer or an investment fund;

the seller’s bank is credited with reserves and in turn credits the seller’s deposit account. Thus, central bank

asset purchases typically create bank deposits (see, e.g., Leonard et al., 2017; Acharya and Rajan, 2022;

Acharya et al., 2023).8 And when a central bank reduces its asset holdings, whether by selling assets or by

not rolling over maturing assets and leaving the government to issue replacement debt to the public, bank

deposits typically fall. (If banks respond to these dynamics by acquiring or selling assets, there can also be

a multiplier effect on the initial change in deposits.)

Because deposits are both a liability of commercial banks and an asset of non-banks, changes in the

aggregate quantity of deposits in the banking system do not change the aggregate amount of duration risk

on private-sector balance sheets. However, changes in aggregate deposits do change the distribution of

duration risk: The creation of bank deposits reduces the aggregate amount of duration held by banks and

increases the amount held by non-banks, while the destruction of deposits has the opposite effect.

To the extent that banks and non-banks have different risk appetites or preferences for hedging the

duration risks that they face, these distributional changes may influence term premia. For example, while

banks may employ sophisticated hedges for duration risk, it seems less likely that a typical household would
7 For example, large banking institutions reported a shorter median weighted average deposit life in the OCC’s

spring 2023 survey than in the fall 2022 survey (https://www.occ.gov/publications-and-resources/publications/

interest-rate-risk-statistics-reports/files/pub-interest-rate-risk-statistics-fall-2022.pdf).
8 In the unusual case where the seller is a depository institution, no deposits are created.

12
hedge the duration risk of its checking account in the same way. Thus, while banks typically seek to hedge

the duration of deposits by acquiring longer-dated assets, households may be less likely to respond directly

to an decrease in their deposit holdings by increasing their holdings of longer-dated assets. If so, banks may

be the marginal buyers or sellers of duration risk, and banks’ duration hedging could disproportionately

influence term premia. The creation of bank deposits as a result of central bank balance sheet expansion

would then increase banks’ demand for duration and reduce term premia, above and beyond the effect of the

change in assets held by the central bank versus the private sector. Conversely, when the central bank allows

its asset holdings to mature, any resulting decrease in bank deposits would increase term premia, above and

beyond the effect of the change in asset holdings.

Additionally, with dynamic betas, the effect of creating or reducing bank deposits depends on the interest

rate environment. Typically, central banks buy assets to stimulate the economy at the same time as they are

lowering interest rates or after reducing interest rates to the lower bound (e.g., Hamilton and Wu, 2012), and

reduce asset holdings in conjunction with removing accommodation by raising interest rates. With dynamic

betas, that correlation amplifies the effect of asset holdings on term premia through the deposits channel in

both directions —removing duration from bank balance sheets more quickly when the central bank is adding

accommodation, and putting duration back on bank balance sheets more quickly when the central bank is

removing accommodation.

However, from June 2022, when the Federal Reserve began reducing its asset holdings, through late-

September 2023, reserve balances actually increased by $50 billion from a total of more than $3 trillion,

while non-reserve liabilities fell by $960 billion. The central bank balance sheet channel’s effect on bank

deposits has likely therefore been modest to date.

7 Run risk with dynamic betas

Drechsler et al. (2023) argue that interest rate risk can drive the transition from the stable to unstable

equilibrium of Diamond and Dybvig (1983) by shifting value between the asset and liability sides of a

bank’s balance sheet. When the bank’s equity is greater than the value of its deposit franchise, depositors

are motivated to retain their balances; when the deposit franchise value is greater than the bank’s equity,

depositors’ incentives shift in favor of withdrawing funds. We find that dynamic betas can amplify this

mechanism.

A bank’s economic value of equity (EVE) is the net value of its tangible and intangible assets, including

the deposit franchise, minus its liabilities. Importantly, EVE ignores the accounting and regulatory treatment

of various assets and liabilities, instead capturing the financial values that influence depositors’ incentives.

13
 Consider a simple bank that funds assets A with deposits D0 and capital CA, where C ∈ (0, 1) is the

bank’s required capital ratio. In our simple model, the present value of the bank’s deposit liability is given

by (1) and is smaller than D0 . Thus, EVE is

 
α + β(R)R
EVE = A 1 − (1 − C) (9)
R+α

where we have used the relationship A = D0 + CA.

Equation (9) shows that EVE depends on the interest rate R, but in practice, banks do not leave

themselves entirely unhedged against this risk. The hedge may come from the duration of the assets A

or from derivatives with no initial cash value. In either case, as a practical matter, the hedge must be

established over some finite time horizon and typically with instruments that are (nearly) linear in interest

rates. For simplicity, we consider hedges based on one risk-free, zero-coupon asset, but our qualitative results

would generalize to a variety of other hedges. We also assume for simplicity that the asset can be liquidated

at fair value with no transaction cost and real-time settlement, and thus that the bank does not need to hold

cash to manage liquidity risk.9

Suppose the zero-coupon asset matures at date T . If its present value is A0 at t = 0 when interest rates

are R0 , the asset must pay A0 (1 + R0 )T at maturity. The value of this asset as a function of R is then

A(R) = A0 [(1 + R0 )/(1 + R)]T .

Suppose the bank acquires an asset whose maturity T is such that the asset perfectly hedges EVE as of

time 0. That is, the bank chooses T = T (R) such that dEVE/dR = 0. This requires

T (R) = Tβ̄ (R) − Y (R)β ′ (R) (10)

where
(1 + R)(1 − C)R
Y (R) = (11)
R + α − [α + β(R)R](1 − C)

and Tβ̄ (R) is the duration that would provide a perfect hedge if deposits had a constant beta β(R) = β̄.

There is no underhedging when R = 0. Since R ≥ 0, C ∈ (0, 1), β ∈ (0, 1),α ∈ (0, 1), we have

T (R) < Tβ̄ (R) whenever β ′ > 0 and R is not at the ELB. In consequence, a bank that uses a constant-beta

model to construct an exact hedge will be underhedged against dynamic betas.

The gap between T (R) and Tβ̄ (R), or the amount of underhedging, is increasing in R so long as

β ′′ (R) Y ′ (R)
> − . (12)
β ′ (R) Y (R)
9 Accounting for liquidity requirements would result in a simple adjustment to our calculation of asset duration.

14
A bit of arithmetic shows that Y ′ (R)/Y (R) ≥ 0 for the range of α, β, C, and R above. We then define Rm
 
as the level of interest rates at which this inequality is an equality and T (R) − Tβ̄ (R) is maximized—in

other words, the level of rates at which the adjustment for incorporating dynamic betas is largest.

Some further assumptions about β(R) can help in characterizing the gap. As noted earlier, our assumption

that 0 ≤ β ≤ 1 for R > 0 and β ′ (R) > 0 argues for a shift in the sign of β ′′ (R) from positive when rates are

at the ELB to negative at higher rates. Let Rs be the transition point where the sign changes: β ′′ (Rs ) = 0.

Based on the sign requirements of equation 12 and assumptions around the range of input parameters above,

we can conclude Rm > Rs > 0. To give a concrete example, using a sigmoid fit to the SFOS-based β(R)

above, we find Rs ≈ 4% and Rm ≈ 4.5%, compared with EFFR around 5% and 5-year Treasury rates around

3.5% at the time of the survey. This means the gap grows with R at low rates but shrinks with R at higher

rates.

This exercise is a formal way of recovering an intuitive point. To the extent that d∆R /dβ < 0 and β is an

increasing function of interest rates, then hedges ignoring those dynamics will be oversized (EVE will have

net long duration) at low rates and undersized (EVE will have net short duration) at higher rates. That has

important consequences for bank portfolio risk management and, when combined with the Drechsler et al.

(2023) framework for run risk, implications for financial fragility emanating from the banking system.

To see these implications, we need to generalize our framework to consider uncertainty in R. To do so,

we assume, consistent with empirical research (e.g., Rogoff et al., 2022, and references therein) that interest

rates are mean reverting over long periods of time with a long-term average of R̄ > 0. When R is at the

ELB, mean reversion implies that the expected value of R is above the ELB. If the bank were to hedge with

Tβ̄ when R is at the ELB, as shown above it would be net long duration in expectation. That, combined

with mean reversion in R, implies that EVE is expected to fall, putting our model bank at greater risk of

the run equilibrium of Drechsler et al. (2023).

As many have noted, deposit insurance can mitigate run risk, but Call Report data show that, as of the

second quarter of 2023, uninsured deposits and wholesale funding made up more than 50% of total commercial

bank liabilities in the U.S.10 Importantly, as SFOS responses indicate, banks often model uninsured deposits

with β < 1, which means they contribute significantly to the value of the deposit franchise.11 In addition,

individual institutions can have much higher exposures to these more runnable sources of funding, and the

events of March 2023 argue for the importance of a more granular view of run risk for assessing systemic

vulnerabilities in banking.
10 This calculation considers reported total liabilities (excluding minority interest), uninsured deposit balances, domestic

deposits and foreign deposits.

11 While the SFOS does not ask explicitly about insured versus uninsured deposits, the vast majority of operational and

non-operational wholesale deposits are likely uninsured.

15
 Our analysis shows that, unless properly accounted for in bank deposit models, dynamic betas will

increase run risk when interest rates are low. Importantly, to avoid overhedging duration risk, banks would

need to consider these dynamics ex ante. That is, while periodic revision of static beta models can result

in ex post dynamic betas, the impact of those dynamics will not affect the size of hedges until the bank

makes the revisions. Thus, banks using static models, even ones that are revised on an ongoing basis, will be

more likely to overhedge their duration in low-rate environments and therefore may be at increased risk of

runs compared with banks that model dynamic betas directly.12 As discussed in Section 3, industry practice

appears more consistent with static rather than dynamic models. Thus, banks’ interest rate risk modeling

practices are a potential source of financial fragility.

Practitioners may point out the opposite dynamic at higher rates. Were a bank to hedge its structural

interest rate risk with a static beta model at interest rates above Rm , it would find itself underhedged should

rates return to the ELB. Paired with the same assumptions as above, including mean reversion, this leads to

the same directional conclusion that hedging with static betas raises the risk that EVE will fall when rates

are high. However, the erosion of EVE in this scenario is driven by declining deposit franchise value, which

reduces run risk.

The behavior of deposit betas thus has an important asymmetric impact on the stability of a banking

franchise. Failing to model and hedge the dynamics of deposit betas increases financial fragility, but only

in the transition from low interest rates (e.g., rates at the ELB) to higher rates. The same cannot be said

for a scenario where rates fall from previously high levels; in fact, when dynamic betas are not adequately

hedged, declining rates enhance the stability of banks as asset values increase relative to the value of the

deposit franchise. This asymmetry represents a novel and concrete channel through which interest rates at

the ELB can pose vulnerabilities to financial stability.

8 Conclusion

Deposits are the primary funding source of the banking system. That makes deposit interest rates a key

driver of credit supply to the economy. The behavior of deposit interest rates is typically summarized by

the beta of deposit rates to market overnight rates. Low betas not only support bank net interest margins,

they also are the primary reason why banks can make long-term, fixed-rate loans, which make up a sizeable

fraction of credit demand.

In this paper, we show that allowing for dynamic rather than static betas more accurately reflects
12 If the assumptions in static models need to be revised more frequently, these models may also introduce problematic

incentives for risk managers to make ill-advised changes in the models. We note the “poorly supported” decision by Silicon
Valley Bank’s management to revise deposit betas lower to reduce the apparent duration of equity (Board of Governors of the
Federal Reserve System, 2023).

16
depositor behavior and significantly alters the interest rate risk profile of bank balance sheets. First, we

summarize the empirical and qualitative evidence for dynamic betas. Second, we use a simple analytic

model to consider the consequences of those dynamics for the interest rate risk on bank balance sheets.

Third, we show that dynamic betas amplify the transmission of monetary policy through both the bank

lending channel and the effect of the central bank’s balance sheet on the amount of duration held by the

private sector. Finally, we show that dynamic betas can affect the role of interest rates in banking fragility.

They do so asymmetrically, introducing additional stability risks in the transition from the ELB to policy

rates above the lower bound. These results highlight the importance of a proper understanding of depositor

behavior and the resulting dynamics in structural interest rate risk on balance balance sheets in formulating

monetary policy and maintaining financial stability.

17
References

Viral Acharya and Raghuram Rajan. Liquidity, liquidity everywhere, not a drop to use – why flooding banks

with central bank reserves may not expand liquidity. NBER Working Papers 29680, National Bureau of

Economic Research, Inc, 2022.

Viral V. Acharya, Rahul S. Chauhan, Raghuram Rajan, and Sascha Steffen. Liquidity dependence and the

waxing and waning of central bank balance sheets. NBER Working Papers 31050, National Bureau of

Economic Research, Inc, 2023.

Bank for International Settlements. Basel III: The Liquidity Coverage Ratio and liquidity risk monitoring

tools, 2013. https://www.bis.org/publ/bcbs238.htm.

Paul Bennett and Stavros Peristiani. Are U.S. Reserve Requirements Still Binding? Federal Reserve Bank

of New York Economic Policy Review, 8(1), May 2002.

Ben S Bernanke and Alan S Blinder. Credit, Money, and Aggregate Demand. American Economic Review,

78(2):435–439, May 1988.

Board of Governors of the Federal Reserve System. Federal Reserve Actions to Support the Flow of Credit

to Households and Businesses, 2020. https://www.federalreserve.gov/newsevents/pressreleases/

monetary20200315b.htm.

Board of Governors of the Federal Reserve System. Assets and liabilities of commercial banks in the united

states - h.8, December 30 2021.

Board of Governors of the Federal Reserve System. Review of the Federal Reserve’s Supervision and Regu-

lation of Silicon Valley Bank. Technical report, Board of Governors of the Federal Reserve System, April

2023.

Douglas Diamond and Philip Dybvig. Bank runs, deposit insurance, and liquidity. Journal of Political

Economy, 91(3):401–19, 1983.

Douglas W. Diamond. Banks and liquidity creation : a simple exposition of the Diamond-Dybvig model.

Economic Quarterly, 93(Spr):189–200, 2007.

Itamar Drechsler, Alexi Savov, and Philipp Schnabl. The Deposits Channel of Monetary Policy*. The

Quarterly Journal of Economics, 132(4):1819–1876, 05 2017. ISSN 0033-5533. doi: 10.1093/qje/qjx019.

18
Itamar Drechsler, Alexi Savov, and Philipp Schnabl. Banking on deposits: Maturity transformation without

interest rate risk. The Journal of Finance, 76(3):1091–1143, 2021. doi: 10.1111/jofi.13013.

Itamar Drechsler, Alexi Savov, Philipp Schnabl, and Olivier Wang. Banking on uninsured deposits. NBER

Working Papers 31138, National Bureau of Economic Research, Inc, 2023.

Darrell Duffie and Arvind Krishnamurthy. Passthrough efficiency in the fed’s new monetary policy. Designing

Resilient Monetary Policy Frameworks for the Future. Federal Reserve Bank of Kansas City, Jackson Hole

Symposium, page 1815–1847, 2016.

Federal Deposit Insurance Corporation. Consolidated Reports of Condition and Income for a Bank with Do-

mestic and Foreign Offices—FFIEC 031, 2023a. https://www.ffiec.gov/pdf/FFIEC_forms/FFIEC031_

202303_f.pdf.

Federal Deposit Insurance Corporation. Consolidated Reports of Condition and Income for a Bank with

Domestic Offices Only—FFIEC 041, 2023b. https://www.ffiec.gov/pdf/FFIEC_forms/FFIEC041_

202303_f.pdf.

Federal Deposit Insurance Corporation. Consolidated Reports of Condition and Income for a Bank with

Domestic Offices Only and Total Assets Less than $5 Billion—FFIEC 051, 2023c. https://www.ffiec.

gov/pdf/FFIEC_forms/FFIEC051_202303_f.pdf.

Emily Greenwald and Doug Gray. Essentials of effective interest rate risk measurement. Community Banking

Connections, pages 2–3 and 12–15, third quarter 2013. https://www.communitybankingconnections.

org/articles/2013/q3/essentials-of-effective-interest-rate-risk-measurement.

Chaitri Gulati and Andrew Smith. The evolving role of the fed’s balance sheet: Effects and challenges.

Economic Review, 107(no.4), 2022. doi: 10.18651/ER/v107n4GulatiSmith.

James Hamilton and Jing Cynthia Wu. The effectiveness of alternative monetary policy tools in a zero lower

bound environment. Journal of Money, Credit and Banking, 44:3–46, 02 2012. doi: 10.1111/j.1538-4616.

2011.00477.x.

Jane Ihrig, Zeynep Senyuz, and Gretchen Weinbach. The fed’s ‘ample-reserves’ approach to implementing

monetary policy. Finance and Economics Discussion Series 2020-022, Board of Governors of the Federal

Reserve System, 2020.

Alena Kang-Landsberg and Matthew Plosser. How Do Deposit Rates Respond to Monetary Policy? Liberty

street economics, Federal Reserve Bank of New York, November 2022.

19
Deborah Leonard, Antoine Martin, and Simon M. Potter. How the Fed Changes the Size of Its Balance

Sheet. Liberty street economics, Federal Reserve Bank of New York, July 2017.

Canlin Li and Min Wei. Term structure modeling with supply factors and the federal reserve’s large-scale asset

purchase progarms. International Journal of Central Banking, 9(1):3–39, 2013. doi: 10.2139/ssrn.2191189.

Stephan Luck, Matthew Plosser, and Josh Younger. How Do Interest Rates (and Depositors) Impact Mea-

sures of Bank Value? . Liberty street economics, Federal Reserve Bank of New York, April 2023.

Office of the Comptroller of the Currency, Board of Governors of the Federal Reserve System, and Federal

Deposit Insurance Corporation. Joint agency policy statement: Supervisory policy statement concerning

a supervisory framework for measuring and assessing banks’ interest rate risk exposure. Federal Register,

60(148):39495–39572, August 1995.

Kenneth Rogoff, Barbara Rossi, and Paul Schmelzing. Long-run trends in long-maturity real rates 1311-2021.

NBER Working Papers 30475, National Bureau of Economic Research, Inc, 2022.

20
Figure 1: Percentage of domestic bank respondents reporting various levels of actual betas from March 2022
to October 2022 as of the November 2022 survey, actual betas from March 2022 to April 2023 as of the May
2023 survey, and cumulative expected betas through November 2023 as of the May 2023 survey.
Source: November 2022 and May 2023 Senior Financial Officer Survey.

(a) Interest-bearing domestic deposit rates paid (b) Total cost of funds

Figure 2: Bank funding costs over three recent tightening cycles

Quarterly average data; aggregate commercial bank interest expense on domestic deposits as a percentage
of total interest-bearing domestic deposits and aggregate commercial bank interest expense as a percentage
of book value of assets, including only commercial banks with a domestic parent.
Source: Call Reports, FRED

(a) Change in large time deposits as a % of total (b) Annualized interest expense from time deposits

Figure 3: The role of time deposits in overall deposit betas

Left panel: Annual average data, domestically chartered commercial banks (H.8); Right panel: Quarterly
average data, commercial banks with a domestic parent (Call Reports).
Source: Call Reports, FRB H.8, FRED

21
 Monetary Policy Cycle
Q2 2004–Q2 2006 Q4 2015–Q2 2019 Q1 2022–Q2 2023
RD RF RD RF RD RF
EFFR -0.03 0.03 0.06 0.05 0.11 0.06
(0.12) (0.14) (0.024) (0.019) (0.03) (0.055)
EFFR2 8.58 6.96 11.73 9.66 6.10 6.13
(1.78) (2.08) (0.95) (0.73) (0.61) (1.10)
Intercept 0.014 0.013 0.003 0.003 0.002 0.02
(0.002) (0.002) (0.0001) (0.0001) (0.0002) (0.00005)

Table 1: Linear regressions of interest-bearing deposit expense RD and total cost of funds RF on EFFR
and EFFR2 , where EFFR is the quarterly average of the effective federal funds rate. Standard errors in
parentheses. Includes all commercial banks in the United States with a domestic parent.
Source: Call Reports, FRED, authors’ calculations.

22