QE-
INDUCED DEPOSITS AND INTEREST RATE PASS
-THROUGH
Marinus Stefanus Johannes Heijmerikx* June, 2016
Supervisor: Dr. M.A. Lamers
ABSTRACT
This paper investigates the impact of three rounds of Quantitative Easing (QE) performed by the Federal Reserve on deposit funding of US commercial banks and its subsequent implications for their deposit rate setting behavior. The results indicate that QE induced an inflow of deposits at commercial banks. Additionally, I find suggestive evidence that this inflow resulted in lower deposit rates, indicating a supply shift of deposits which facilitates interest rate pass-through. Interest rate pass-through of asset purchases to deposit rates is found to be heterogeneous in bank rollover risk.
JEL classifications: E43; E52; G21.
Keywords: Unconventional monetary policy; Quantitative Easing; Interest rate pass-through;
Deposit funding.
1
1
Introduction
As a response to the Global Financial Crisis (GFC) many central banks initiated Large-Scale Asset Purchases (LSAPs) to combat inflation and increase economic activity as overnight market rates were around or at the zero lower bound (ZLB) (Labonte, 2014). With the commencement of unconventional monetary policy measures by central banks a new large academic debate about the effects and potential transmission channels of LSAPs came about. The debate mainly considers the impact of LSAPs on the yields of LSAP-targeted long-dated assets and other substitute financial assets. (see, e.g., Krishnamurthy and Vissing-Jorgensen, 2011; Gagnon et al., 2011; Engen, Laubach, and Reifschneider, 2015). This paper analyzes a different way through which LSAPs affect interest rates. In line with Joyce and Spaltro (2014) and Butt et al. (2014), asset purchases from non-banks induce a supply effect of new bank deposits. This would potentially force downward pressure on deposit rates and facilitate interest rate-pass-through. Transmission differs from the common interest rate pass-through literature in which a reduction in market rates, instead of asset purchases, put downward pressure on deposit rates. This paper starts the discussion how LSAPs affected deposit funding and subsequent interest rate setting behavior of commercial banks.
Gagnon et al. (2011) and Krishnamurthy and Vissing-Jorgensen (2011) find in the US case that LSAPs significantly reduce yields on Mortgage-Backed Securities (MBS), Treasuries and corporate bonds. Moreover, evidence is presented of a portfolio rebalance channel in which sellers of LSAP-targeted assets rebalance their portfolio towards other substitute assets, reducing the yields of non-LSAP-targeted assets. Two papers in the context of the UK consider transmission via asset purchases from non-banks which induce new bank deposits and its potential impact on bank lending (Joyce and Spaltro, 2014; Butt et al., 2014). They both find that banks only marginally increase lending in response to a LSAP-induced deposit inflow. This thesis considers the same transmission mechanism but with respect to deposit rates in the US context.
At the start of Quantitative Easing (QE)1 with the Fed Funds rate at the ZLB, the average interest rate on small time deposits was still around 2.5% and the average interest rate on small money market deposit accounts was around 1.2%, as presented in Fig. 1. This implies that interest rate pass-through for deposits was far from complete, which is compatible with the findings of Darracq Paries et al. (2014). They argue that interest rate
1
2
pass-through for deposits becomes less complete in a low interest rate environment. However, during three rounds of QE, average deposit rates declined significantly with the Fed Funds rate at the ZLB. Thereby suggesting that there could be a role for QE-induced deposits to facilitate interest rate pass-through. For central bankers this might be a tool to enhance completeness of interest rate pass-through and affect the interest rate sensitive saving and spending decisions of households and firms. Deposits remain a very important source of funding for the banking sector at large. The data show that US commercial bank deposit funding accounts on average for approximately 83% of total assets for the period between 2007 and 2013. Furthermore, deposits represent a very important safe and liquid asset for households and firms. The importance of deposits for banks as a main funding source and for non-banks as their primary safe and liquid asset, implies that thorough research with respect to QE and potential interest rate pass-through to deposit rates is needed. This paper attempts to fill part of this gap.
Fig. 1: Interest rate environment and monetary policy.
This graph shows the average interest rate of $10,000 12 month Certificates of Deposits (12MCD10K) and the average rate for $10,000 money market deposits accounts (MM10K) for US commercial banks in the dataset with the discontinued target Federal Funds rate and the total asset holdings of the Federal Reserve.
0 1 2 3 4 T o ta l A s s et s F ed er a l Res e rv e ( T ri llio ns ) 0 1 2 3 4 5 In te re st ra te (% )
2007Q4 QE1 2009Q4 QE2 2011Q4 QE3 2013Q4
Date
MM10K rate 12MCD10K rate
Fed Funds rate Total Assets Federal Reserve
Source : Ratewatch - US FRED - Federal Reserve Bank of New york USA, 2007-2013
3
This is the first study that formalizes and measures a direct impact of asset purchases on the quantity of bank deposits and deposit rates. To capture the impact of QE on deposits, I constructed a new microeconomic dataset containing deposit rates and extensive balance sheet information at bank level for the period between 2007 and 2013. I focus on two insured deposit products that are most widely offered in the US banking sector, represented by $10,000 12-month Certificates of Deposits (12MCD10K) and $10,000 Money Market deposit accounts (MM10K). To systematically analyze the transmission of asset purchases from non-banks, I start with measuring the impact of QE on total deposits and subsequently estimate the impact of QE-induced deposits on deposit rates. Finally, I also explore heterogeneity in deposit rate setting.
The contribution to the existing literature is twofold. First, this paper contributes to the literature of interest rate pass-through by formalizing and measuring the inflow of QE-induced deposits and its implications for deposit rates with market rates at the ZLB. Second, this study adds to the literature of interest rate setting behavior of commercial banks in which bank characteristics give rise to heterogeneous deposit rate setting in response to QE. The results show a significant inflow of deposits induced by QE and find suggestive evidence of a subsequent decline in deposit rates. The findings imply that the downward adjustment of deposit rates, as depicted in Fig. 1, can be partly attributed to a QE-induced supply effect of deposits. Thereby, facilitating interest rate pass-through. Heterogeneity in interest rate setting behavior exists with respect to the degree of rollover risk. Banks whose funding is more weighted towards shorter maturities of time deposits lower their rate less in response to QE.
4
2
Theoretical framework
This section sheds light on the potential impact of monetary policy on deposits. First, I introduce monetary policy and theoretical transmission. Subsequently, the implications of LSAPs for balance sheets, deposits and interest rates are discussed. There is no literature that addresses the direct impact of asset purchases on deposit rates. Therefore, I contribute a part of this section to formalize a proposed impact of QE-induced deposits on deposit rates. Finally, potential sources of heterogeneity in interest rate setting behavior with respect to bank characteristics are proposed.
Conventional monetary policy is often referred to as a central bank steering short-term interest rates and unconventional monetary policy as extraordinary measures in response to extreme macroeconomic conditions (Di Maggio, Kermani, and Palmer, 2015). The unconventional measure of interest is the broad initiation of LSAPs by central banks. Both conventional and unconventional measures affect the amount of liquid reserves, deposits and market rates in the economy by performing Open Market Operations (OMOs).2 The main difference between LSAPs and conventional monetary policy is the scale of asset purchases, the targeted maturities (LSAPs target long-dated assets) and the type of assets (LSAPs include not only Treasuries).
During the GFC many central banks initially took conventional measures by lowering their policy rate significantly to boost economic activity, effectively reaching the ZLB of nominal interest rates. Joyce et al. (2012) stipulate two issues that potentially reduced the effectiveness of conventional monetary policy during the crisis. First, the uncertainty about the financial viability of banks and borrowers caused a breakdown in the usually stable relation between the policy rate and market interest rates. Moreover, instead of lending to the private sector financial institutions held on to their funds. This harmed full interest rate pass-through in which a decrease in the policy rate does not lead to a one-for-one decrease in retail rates. When using the Taylor rule, the severe economic downturn would suggest negative interest rates. However, the ZLB constrains central banks to do this because economic agents would hold non-interest bearing cash. In that sense conventional monetary policy was rendered ineffective. Subsequently, central banks were urged to look for unconventional measures and broadly implemented LSAPs.
2
5
2.1 Monetary transmission
Before I move on to explain the implications of LSAPs, I discuss the theoretical transmission of monetary policy. There are two views that dominate the literature of monetary transmission: the money view and the credit view. The money view assumes that financial institutions are mere intermediaries and do not affect real economic activity. In line with this view, markets are often assumed to be complete and informational asymmetries do not exist. The interest rate channel falls under the umbrella of the money view (Mishkin, 1996). The key assumption of this channel is price stickiness. In the presence of price stickiness, a decline in nominal interest rates due to an increase in the money supply also leads to reduction in real interest rates. Interest rate sensitive spending would increase, thereby affecting real economic output.
The credit view is represented by the existence of asymmetric information with imperfect markets for debt and equity which give rise to an amplification of monetary transmission. (see, e.g., Bernanke and Gertler, 1995; Peek and Rosengren, 2010). The existence of information asymmetries between lenders and borrowers causes an external finance premium. Relative to a world with perfect information agency costs are incorporated in the price of external finance. The premium is inversely related to the borrower’s riskiness, reflected by its balance sheet composition. The bank lending channel is directly related to the credit view. This channel can be described by a monetary tightening that causes bank deposits to decline as the opportunity costs of holding deposits is increased. Assuming no excess reserves, this leads to a decline in the supply of loans if the banks do not fully insulate the supply of loans by rearranging their portfolio of assets and liabilities (Oliner and Rudebusch, 1995). Amplification of this channel is considered to be present for smaller, low-liquid and low-capitalized banks (Gambacorta, 2008; Kashyap and Stein, 2000). These banks are less capable to liquidate assets, raise equity or raise market funding, because they have to pay a lemon’s premium. Subsequently, they have to raise their interest rates on deposit funding more than larger, more liquid and more capitalized banks to adequately contain the drain in deposits induced by a monetary tightening.
2.2 Quantitative Easing
6
already at the ZLB. This would further flatten the yield curve and, in theory, should stimulate economic activity in line with the monetary transmission views discussed above.
There are two cases that can be considered with respect to the direct implications of QE for balance sheets, presented in Fig. 2. In case of asset purchases from banks there is no balance sheet expansion for these banks (Panel 2.c) and liquid reserves are swapped for long-dated assets. Ceteris paribus, the increase in reserves on the bank’s balance sheet increases its liquidity cushion. When the stock of liquid assets increases, banks are better able to settle their financial obligations with immediacy. The inflow of liquidity would also reduce the profitability of banks as the return on holding liquid reserves is low. Assuming that banks target stable interest rate margins, banks would be very responsive to lower their deposit rates in response to the liquidity inflow induced by QE. When assets are bought from non-banks, balance sheets of banks expand as liquid reserves directly end up as deposits at the bank (Panel 1.c). In this case banks act as an intermediary in asset purchases by the central bank. Hence, the implications for bank balance sheets differ between entities involved in the transaction of asset purchases. Asset purchases from non-banks induce a supply effect of deposits and would, ceteris paribus, lead to downward pressure on deposit rates. This thesis will therefore mainly focus on asset purchases from non-banks that might directly affect deposit rates.
Benford et al. (2009) postulate that the impact of QE can be parsed in three components.3 These are asset prices and portfolio effects, bank lending and quantity effects, and expectations. I use this framework to provide a structured view of the potential transmission channels of QE to appropriately identify the potential impact of QE on deposit funding and deposit rates.
2.2.1 Bank lending and quantity effects
QE leads to an increase in bank reserves and deposits when assets are bought from non-banks. When the inflow of liquidity exceeds the demand for liquidity, banks increase their willingness to issue loans. Increased lending to companies and households can boost
3
7
consumption and investment. Joyce et al. (2012) formalize this idea under the bank lending channel.
Fig. 2: Implications of asset purchases by central banks for balance sheets
8
There are two studies that try to identify a bank lending channel through an increase in QE-induced deposits. Joyce and Spaltro (2014) find in the UK context that in the period between 2009 and 2010 there may have been a small increase in bank lending due to an inflow of deposits caused by QE, albeit statistically insignificant. Butt et al. (2014) make a distinction between “sticky” and “flighty” deposits induced by QE. Their paper imposes an important question on the stability of the deposits induced by QE. When non-banks rebalance their portfolios, deposits move from the bank balance sheet. Hence, the potential increase in the variance of deposits could lead to a reduction in the willingness to supply credit and reduced downward pressure on deposit rates.
2.2.2 Asset prices and portfolio effects
Joyce et al. (2012) argue that QE could potentially lead to a portfolio rebalancing effect of assets. When this channel is operative, asset purchases should also lower yields of other non-QE-targeted assets. The sellers of QE-targeted assets regard cash and long-dated assets as imperfect substitutes and subsequently rebalance their portfolio towards other long-dated assets with similar characteristics. This would lead to an increase in prices and a decline in yields of other comparable non-QE-targeted assets. This channel is likely to affect deposits. When non-banks swap QE-induced deposits for other long-dated assets, deposits move from a given bank balance sheet. This would give rise to “flighty” deposits, reducing downward pressure on deposit rates.
The empirical literature with respect to the impact of QE on interest rates mainly discusses the impact on government bonds, MBS and corporate bonds using an event study methodology. Gagnon et al. (2011) find that QE had a large effect on the longer-term yields of Agency debt and MBS through improved market liquidity and the reduction in assets with high prepayment risk. They also document a decline in the swap rate and the yield on corporate bonds capturing the spillover effect of QE due to a portfolio rebalance channel. This is in line with Krishnamurthy and Vissing-Jorgensen (2011) who also find evidence for a portfolio rebalance channel. Their findings suggest that investors have preferences for particular maturities within the safe asset class in rebalancing their portfolios, in line with preferred habitats.4
4
9
2.2.3 Expectations
QE has implications for the expectations of households and firms. Changed expectations about the economic outlook can have a direct effect on asset prices, inflation and investment and spending decisions of households and firms (Benford et al., 2009). If for example, a QE announcement improves the perceived economic outlook, consumption and investment might rise. This potentially affects deposits as households and firms increase their deposit holdings when income rises. The effect is more present when there is a transitory shock in income as unexpected income is generally deposited (Gambacorta, 2008).
2.3 Deposit rates
I have described the channels through which QE might affect deposits. The question remains what the effect of QE would be on deposit rates as there is no empirical literature describing this effect. I discuss the existing literature considering deposit rates below.
Deposit rates are considered to be slow-moving variables and banks infrequently update their rates in response to changing economic conditions. A plausible explanation is that there are menu costs involved which give rise to this behavior of infrequent updating. Inter alia, Hannan and Berger (1991) and Neumark and Sharpe (1992) show that deposit rates often adjust months after changes in money market rates. Moreover, deposit rates tend to adjust asymmetrically and are upwards sticky and downwards flexible. The empirical literature about interest rate pass-through is in consensus over the stickiness of retail rates and the incompleteness of interest rate through in the short run. Short run interest rate pass-through is heterogeneous with respect to several factors such as the size of the bank, their liquidity and capitalization (Gambacorta, 2008). In the long run interest rate pass-through is more complete and sometimes even full (see, e.g., Borio and Fritz, 1995; De Bondt, 2002).
Darracq Paries et al. (2014) find that there is a slowdown in interest pass-through when interest rates are low. In the phase when policy rates5 were lowered from 1.5% to 0.5%, pass-through became less complete and banks showed more resistance to reduce their deposit rate. A possible explanation could be that through the reduction of the opportunity cost of holding money, people reduce their deposits at the bank, effectively causing banks to prevent a further decrease in deposit rates. Illes and Lombardi (2013) find that since the global
5
10
financial crisis there has been a structural change in the relation between lending rates and the policy rate in the United States and find that spreads between these rates significantly widened. These findings could imply that interest rate pass-through is also far from complete with respect to deposit rates of US commercial banks.
Drechsler, Savov, and Schnabl (2016) use banking competition to show the existence of a deposits channel of monetary policy. The authors find a causal relation between the Fed Funds rate and the deposit spread.6 An increase in the Fed Funds rate leads to an increase in market power of banks as a dominant provider of liquid deposits, because the opportunity costs of holding money has increased. Subsequently, increased market power causes banks to increase their deposit spreads. This is even more pronounced in relatively concentrated markets. In response to the higher price of deposits, households withdraw their deposits and substitute them with higher yielding but lower liquidity assets. They identify this effect as a shift in the deposit supply7 of banks represented by an increase in the price (deposit spread) and a decrease in quantity (total amount of deposits).
2.4 Hypotheses
In case of asset purchases by central banks from non-banks, every exchange between long-dated assets and liquid reserves immediately show up on their deposit accounts at the bank as the transaction is performed (Panel 1.c of Fig. 2). At bank level there is an expansion of the balance sheet in which it holds more liquid reserves and deposits. This indicates a supply effect of QE-induced deposits which, all else equal, leads to lower deposit rates. Considering the potential inflow of deposits induced by QE, I formulate the following hypothesis:
H1: Asset purchases performed by the Federal Reserve have a negative effect on
deposit rates via a QE-induced inflow of deposits.
2.4.1 Heterogeneity in interest rate setting behavior
Bank size, liquidity and capitalization are considered to be important determinants for heterogeneity in interest rate pass-through (Gambacorta, 2008). Assuming an imperfect
6
The spread is represented by the difference between the Fed Funds rate and the deposit rate. 7
11
market for bank debt and equity, small, low-liquid and low-capitalized banks potentially reduce their deposit rates less. Because of asymmetric information about the value of a bank’s assets, these banks pay a lemon’s premium on their uninsured funding as they are perceived to be more risky. For larger, more liquid and more capitalized banks, this is less the case as they can more easily raise market funding at lower costs. In that sense they are less dependent on insured deposit funding such that they are better able to lower their deposit rate in response to asset purchases by the Fed. This leads to the following hypothesis:
H2: Banks that are large, highly liquid and highly capitalized lower their deposit rates
more in response to asset purchases by the Federal Reserve than banks that are relatively smaller, low-capitalized and low-liquid.
Gambacorta (2008) finds that banks which are heavily weighted towards market funding increase their deposit rate more and quicker than banks that are more deposit funded in response to an increase in market rates. Gambacorta (2008) gives a clear explanation of this result: “The intuition of this result is that, other things being equal, it is more likely that a bank adjust her terms for passive deposits if the conditions of her own alternative form of refinancing change.” (p. 9). QE puts downward pressure on market rates and thus reduces the cost of market funding. It is more likely that market funded banks are better able to reduce their deposit rates. Accordingly, I formulate the following hypothesis:
H3: Banks that are heavily weighted towards market funding lower their deposit rates
more in response to asset purchases by the Federal Reserve than banks that are less market funded.
12
Federal Reserve as they have more pressure on their funding structure than OTH banks. This leads to the following hypothesis:
H4: OTD banks lower their deposit rates less in response to asset purchases by the
Federal Reserve than banks that are more weighted towards an OTH business model.
Acharya and Mora (2015) find that banks that were heavily weighted towards a shorter maturity profile of time deposits offered higher rates on Certificates of Deposits (CDs) during the first year of the crisis due to higher rollover risk. These banks have more urgent funding needs and are not able to reduce their rates as much compared to banks that are more weighted towards time deposits with a longer maturity. In line with this argument, rollover risk seems to be positively related to deposit rates, which leads to the following hypothesis:
H5: Banks that have a short maturity profile of time deposits lower their deposit rates
less in response to asset purchases by the Federal Reserve than banks that have a relatively longer maturity profile of time deposits.
3
Background
The GFC induced a blow to the capital positions of US banks. The interbank provision of liquidity evaporated as banks became uncertain about the financial viability of their counterparties. Consequently, several institutions provided a policy response during the GFC to relieve the pressures on the banking system. I briefly summarize policy responses that have direct implications for the balance sheets of US commercial banks below.
On October 2008, the US treasury department issued a Troubled Asset Relief Program (TARP) and invested approximately $245 billion in the US banking system to improve solvency and stabilize the banking system. In the same month, the FDIC increased depositor insurance from $100,000 up to an amount of $250,000. This was initially a temporary measure to prevent bank runs and facilitate financial stability, and on July 2010 it was made permanent in the Dodd-Frank act. Besides increasing the depositor insurance to $250,000, the FDIC introduced an interest rate cap on deposit products for less than well capitalized financial institutions in May 2009.8 These less than well capitalized banks were also prohibited to accept brokered deposits. The interest rate cap was set as the average national
8
13
interest rate on insured deposit products plus 75 basis points. The intention of an interest rate cap and the prohibition of accepting brokered deposits is to prevent low-capitalized banks from competing for funds at very high interest rates and prevent unsustainable pressure on their financial performance. The interest rate cap inevitably affects interest rate setting behavior of less than well capitalized banks, I address this issue in the limitations and discussion section below.
The Federal Reserve provided a policy response to the GFC and ZLB by purchasing a variety of long-dated assets by performing Open Market Operations (OMOs) and holding them in its System Open Market Account (SOMA).9 QE is an expansion of the central bank’s balance sheet by mainly buying longer maturity assets in the open market at pre-specified quantities. Asset purchases should lower long-term interest rates to broadly improve financial conditions. Improved financial conditions should subsequently lead to boosted economic activity and reduce deflationary pressures through higher aggregate demand by increased interest rate sensitive spending and investment. (Engen, Laubach, and Reifschneider, 2015). The Federal Reserve bought not only US Treasuries but also Agency debt and MBS. As depicted in Fig. 3, the Federal Reserve mainly expanded its holdings of MBS and Agencies10 from September 2008 until the announcement of the second round of QE. During the second round of QE the Federal Reserve expanded its Treasury holdings and in the third round both Treasuries and MBS. This led to a significant expansion of the balance sheet of the Federal Reserve from approximately $0.9 trillion in September 2008 to $4 trillion in total assets in 2014. Bank balance sheets are inherently affected by the asset expansion by the Fed, either acting as an intermediary in case of asset purchases from non-banks or directly participating in the transaction of long-dated assets from their balance sheets (see Fig. 2).
9
The discrete actions taken by the Federal Reserve regarding QE are listed in Appendix E. The announcements of QE are presented in Appendix F and other policy measures are described in appendix G.
10
14
Fig. 3: Total Assets Federal Reserve
This figure shows the balance sheet expansion and asset composition of the Federal Reserve over time for three rounds of QE (Board of Governors of The Federal Reserve System, 2016).
4
Data & Methodology
In this section I discuss the data and empirical specifications used in this thesis. First, I describe the specific deposit products that are analyzed and explain the structure and characteristics of the dataset. Subsequently, I consider trends in the composition of balance sheets of commercial banks during QE. Finally, I discuss the identification strategy and model specifications employed in this thesis.
I consider interest rates of two deposit products, represented by $10,000 12-month certificates of deposits (12MCD10K) and $10,000 money market deposit accounts (MM10K). Time deposits can be seen as an illiquid deposit and therefore carry a liquidity premium. Terms for money market deposit accounts are different.11 The number of withdrawals from these accounts is often capped. However, they can be seen as a relatively liquid savings account. Money market deposit accounts invest in low risk, short-term, fixed income securities like Treasuries and therefore they have the potential to offer a higher return than
11
15
savings accounts, but returns are lower than time deposits. Both MM10K and 12MCD10K accounts are FDIC insured.
4.1 Data
Deposit rate data of the deposit products mentioned above are provided by Ratewatch. This dataset is proprietary, contains deposit rates for separate bank accounts and is documented at bank-branch level for different deposit products. Extensive balance sheet and income statement data are collected from the Call Reports at a quarterly frequency for all US commercial banks. Total asset holdings of the Federal Reserve and Fed Funds rate data are respectively retrieved from the Federal Reserve Bank of New York and the Federal Reserve Bank of St. Louis.
16
In line with Gatev and Strahan (2006), I control for mergers and acquisitions by dropping all observations that have an increase in total assets of more than 10%. Thereby, I control for potential confounding effects of mergers and acquisitions on deposits. Additionally, all dependent and independent variables used in the analysis, which are discussed in detail in the methodology section, are truncated at the 1st and 99th percentile to control for outliers. Level variables are corrected for inflation by using a GDP implicit price deflator indexed at 2007Q1 and is provided by the Federal Reserve Bank of St. Louis. This yields the summary statistics of the dependent variables, the bank characteristics of interest and the control variables used in the analysis, presented in Table 1. For bank characteristics a correlation matrix is presented in Table 7 in appendix B. There is no collinearity detected between any of the bank characteristics. Therefore, there seems no problem of collinearity in estimating the relations proposed in the next section.
The average quarterly inflow of deposits is approximately 0.4% with a standard deviation of 4.2%. Acharya and Mora (2015) report a quarterly deposit growth of 1.1% with a standard deviation of 3.1% for the period between 1990 and 2009. The average quarterly growth of deposits in our sample is relatively small. During the period under consideration, there could be a significant deposit inflow induced by QE as proposed above. On the other hand, due to the subprime mortgage crisis banks experienced a significant blow to the value of their assets. This would mean that banks were induced to deleverage and shrink the size of their balance sheets. Also households and firms tend to deleverage and shrink the size of their balance sheet (Labonte, 2014). These factors, among others, could explain the moderate average quarterly growth in deposits.
17
Table 1: Summary statistics
(1) (2) (3) (4) (5)
VARIABLES N mean sd min max
Dependent variables
∆ Ln Total Deposits 164,201 0.004 0.042 -0.134 0.119
∆ 12MCD10K Rate 130,598 -0.083 0.101 -0.400 0.216
∆ MM10K Rate 124,790 -0.075 0.128 -0.568 0.333
Main explanatory variable
∆ Ln Total Assets Fed 167,656 0.053 0.155 -0.314 0.591
Bank characteristics
Capitalization 171,698 0.108 0.031 0.040 0.265
Liquid Asset Ratio 171,697 0.318 0.152 0.051 0.780
Ln Total Assets 171,698 11.931 1.151 9.517 16.308
Maturity Profile Time Deposits 171,185 0.759 0.144 0.345 1.000
Market Funding 171,698 0.243 0.106 0.040 0.615
Off Balance sheet items/Total Assets 170,779 0.100 0.067 0.000 0.375
Control variables
Implicit Interest Rate 171,471 0.004 0.003 0.000 0.013
Real Estate Loans/Total Loans 171,698 0.694 0.173 0.117 0.973
Non-performing Loans/Total Loans 173,449 0.014 0.017 0.000 0.107
Cost/Income 171,469 0.738 0.221 0.336 2.194
ROA 171,475 0.002 0.003 -0.020 0.008
∆ Fed Funds Rate 175,202 -0.065 0.203 -1.000 0.000
∆ Ln Wholesale Funding 163,805 -0.003 0.110 -0.435 0.422
∆ Ln Total Loans 164,205 0.002 0.042 -0.138 0.155
∆ Ln Small Time Deposits 130,483 -0.016 0.051 -0.231 0.236
∆ Ln Money Market Deposits 123,699 0.014 0.126 -0.456 0.521
Target MBS/Total Securities 169,738 0.062 0.115 0.000 0.678
Target Treasuries & Agencies/Total Securities
170,865 0.613 0.302 0.000 1.000
This table shows summary statistics for the variables used in the regression analyses. The dependent variable ∆ Ln Total Deposits is the quarterly percentage change in total deposits. ∆ 12MCD10K Rate and ∆ MM10K Rate represent the quarterly change in the deposit rate of the specific deposit product. The main explanatory variable
is ∆ Ln Total Assets Fed representing the quarterly asset expansion of the Fed. A set of bank characteristics
18
Since the initiation of QE a more or less proportional relation between loans and deposits broke down, as presented in Fig. 4. Loans grew much slower than deposits such that the gap between loans and deposits widened. This could imply that banks started hoarding liquidity or buy other financial assets in response to a QE induced inflow of liquid reserves. Labonte (2014) confirms this and finds that banks primarily held excess reserve and did not expand the credit supply. When banks hold more liquid reserves, all else equal, the return on their asset portfolio would decline putting pressure on their interest rate margins and implying that they would be very responsive to lower their deposit rates.
Banks decreased their wholesale funding and increased their core deposit funding over time, as depicted in panel A of Fig. 5. This suggests that commercial banks prefer core deposit funding over wholesale funding since the initiation of QE, indicating a substitution effect. Ceteris paribus, this would lead to upward pressure on deposit rates. Panel B of Fig. 5 shows that the amount of small and large time deposits declined over time while the total amount of money market deposit accounts increased significantly. This seems to indicate a substitution effect of depositors in which they prefer to hold liquid deposits rather than illiquid time deposits. Drechsler, Savov, and Schnabl (2016) find that when market rates increase, depositors prefer to hold illiquid deposits over liquid deposits. They argue that the liquidity premium increases in a high interest rate environment leading depositors to substitute liquid deposits for illiquid deposits. Conversely, the liquidity premium decreases in a low interest rate environment leading depositors to substitute illiquid deposits for liquid deposits as is observed in the data.
19
non-banks in which QE induces a deposit inflow. This would put significant downward pressure on deposit rates as new bank deposits flow into the banking system.
Fig. 4: Commercial bank balance sheet and QE
This graph shows the average amount of deposits, loans and assets of commercial banks with the total asset holdings of the Federal Reserve over the sample period. Bank deposits, loans and assets are quarterly averages calculated by summing up the respective quantities for the entire sample and subsequently dividing it by the total number of banks in the dataset at that
point in time. This corrects for the decline in total banks in the dataset over time.12
Fig. 5: The composition of commercial bank funding and deposit funding
Panel A: The light shaded grey area represents the
average amount of wholesale funding in the commercial banking system and the darker shaded grey represents the average amount of core deposits.
Panel B: The two lower shaded grey areas represent
respectively the average amount of small time deposits and the average amount of large time deposits in the commercial banking system. The upper area represents the amount of money market deposit accounts.
12
All graphs that incorporate balance sheet averages are calculated in the same way as described here. 0 1.0 2.0 3.0 4.0 T o ta l A s s et s F ed er a l Res e rv e ( T ri llio ns ) 0.5 0.75 1.0 1.25 1.5 1.75 2.0 2.25 A v er a ge Q ua nt it y ( M illi on s )
2007Q4 QE1 2009Q4 QE2 2011Q4 QE3 2013Q4
Date
Total Deposits Total Assets
Total Loans Total Assets Federal Reserve
Source : Call Reports - Federal Reserve Bank of New York
USA, 2007-2013
Deposits and Loans Commercial Banks
0 0.5 1.0 1.5 2.0 A v er a ge Q ua nt it y ( M illi on s )
2007Q4 QE1 2009Q4 QE2 2011Q4 QE3 2013Q4 Date
Source : Call Reports
USA, 2007-2013
Funding Composition Commercial banks
0 0.25 0.5 0.75 1.0 A v er a ge Q ua nt it y ( M illi on s )
2007Q4 QE1 2009Q4 QE2 2011Q4 QE3 2013Q4 Date
Source : Call Reports
USA, 2007-2013
20
4.2 Methodology
This paper analyzes whether QE induced an inflow of deposits that consequently led to a decline in deposit rates. I use a Two-Stage Least squares (2SLS) model specification to identify deposit inflows that are induced by QE in a first stage regression. The estimated inflow of QE-induced deposits is subsequently used in a second stage regression on deposit rates, potentially capturing the proposed downward pressure on deposit rates. The Federal Reserve does not provide transaction data of asset purchases. Therefore, I must stress that this identification method is imperfect in the sense that it cannot directly identify which banks received QE-induced deposits.
In order to analyze the impact of QE-induced on deposit rates the following equations are estimated: ∆ ln 𝑇𝑇𝑇𝑇𝑇𝑇𝑇𝑇𝑇𝑇 𝐷𝐷𝐷𝐷𝐷𝐷𝑖𝑖,𝑡𝑡 = 𝛼𝛼𝑖𝑖 + 𝛿𝛿𝑡𝑡+ 𝛽𝛽1∗ ∆ 𝑇𝑇𝑙𝑙 𝑇𝑇𝑇𝑇 𝐹𝐹𝐷𝐷𝐹𝐹𝑡𝑡+ 𝛽𝛽2 ∗ 𝐼𝐼𝐼𝐼𝐷𝐷𝑇𝑇𝐼𝐼𝐼𝐼𝐼𝐼𝑇𝑇 𝐼𝐼𝑙𝑙𝑇𝑇𝐷𝐷𝐼𝐼𝐷𝐷𝐼𝐼𝑇𝑇 𝑅𝑅𝑇𝑇𝑇𝑇𝐷𝐷 𝑖𝑖,𝑡𝑡−1 + 𝛽𝛽3∗ 𝐶𝐶𝑖𝑖,𝑡𝑡−1+ 𝜖𝜖𝑖𝑖,𝑡𝑡 (1) ∆𝑅𝑅𝑇𝑇𝑇𝑇𝐷𝐷𝑖𝑖,𝑡𝑡 = 𝜌𝜌𝑖𝑖 + 𝜂𝜂𝑡𝑡+ 𝜆𝜆1∗ 𝑄𝑄𝑄𝑄 𝐼𝐼𝑙𝑙𝐹𝐹𝐼𝐼𝐼𝐼𝐷𝐷𝐹𝐹 𝐷𝐷𝐷𝐷𝐷𝐷 𝑡𝑡+ 𝜆𝜆2∗ ∆ ln 𝑄𝑄𝐼𝐼𝑇𝑇𝑙𝑙𝑇𝑇𝐼𝐼𝑇𝑇𝑄𝑄 𝑆𝑆𝐷𝐷𝐷𝐷𝐼𝐼𝐼𝐼𝑆𝑆𝐼𝐼𝐼𝐼 𝐷𝐷𝐷𝐷𝐷𝐷𝑖𝑖,𝑡𝑡−1 + 𝜆𝜆3∗ 𝐶𝐶𝑖𝑖,𝑡𝑡−1+ 𝜃𝜃𝑖𝑖,𝑡𝑡 (2)
Eq. (1) estimates the inflow of deposits induced by QE and Eq. (2) captures its subsequent impact on deposit rates using 2SLS. As dependent variables the log change in total deposits, represented by ∆ ln 𝑇𝑇𝑇𝑇𝑇𝑇𝑇𝑇𝑇𝑇 𝐷𝐷𝐷𝐷𝐷𝐷𝑖𝑖,𝑡𝑡, and the change in deposit rate, represented by ∆𝑅𝑅𝑇𝑇𝑇𝑇𝐷𝐷𝑖𝑖,𝑡𝑡, are used. They represent the dependent variables for bank 𝐼𝐼 at time 𝑇𝑇. The change in
21
impact of monetary policy on deposits and deposit rates is at time 𝑇𝑇. The amount of long-dated assets bought by the Federal Reserve were pre-specified. This implies that the amount of assets bought by the Federal Reserve could affect deposit rates, but deposit rates do not affect the amount of asset bought and thereby do not affect the amount of QE-induced deposits. Therefore, I assume that there is no simultaneity between both variables.
The implicit interest rate, represented by 𝐼𝐼𝐼𝐼𝐷𝐷𝑇𝑇𝐼𝐼𝐼𝐼𝐼𝐼𝑇𝑇 𝐼𝐼𝑙𝑙𝑇𝑇𝐷𝐷𝐼𝐼𝐷𝐷𝐼𝐼𝑇𝑇 𝑅𝑅𝑇𝑇𝑇𝑇𝐷𝐷 𝑖𝑖,𝑡𝑡−1, is included in Eq. (1) to control for deposit movements that can be attributed to the price level of deposits.13 In Eq. (2), the quarterly change in quantity of the deposit product under consideration is incorporated to control for changes in the interest rate induced by deposit product specific in- or outflows, represented by ∆ ln 𝑄𝑄𝐼𝐼𝑇𝑇𝑙𝑙𝑇𝑇𝐼𝐼𝑇𝑇𝑄𝑄 𝑆𝑆𝐷𝐷𝐷𝐷𝐼𝐼𝐼𝐼𝑆𝑆𝐼𝐼𝐼𝐼 𝐷𝐷𝐷𝐷𝐷𝐷𝑖𝑖,𝑡𝑡−1. 𝐶𝐶𝑖𝑖,𝑡𝑡−1 is a vector of control variables which is used in both equations. Bank fundamentals14 are incorporated in the vector of control variables. These are commonly used in the existing literature to control for bank riskiness that may affect the price of bank funding (see, e.g., Ben-David, Palvia, and Spatt, 2015; Demirgüç-Kunt and Huizinga, 1999). In line with Acharya and Mora (2015), I control for the maturity profile of time deposits that could have an effect on deposit funding, thereby controlling for rollover risk. Bank specific weighting towards QE-targeted MBS, Treasuries and Agencies should control for asset purchases directly from banks (Panel 2.c of Fig. 2).15 The change in the Fed Funds rate affects deposits according to Drechsler, Savov, and Schnabl (2016) and is therefore included. The change in wholesale funding is incorporated to control for the collapse of the wholesale funding market during the GFC and its implication for deposit funding. Off balance sheet items over total assets will serve as a proxy for the OTD business model. Off-balance sheet activities is assumed to be positively related to securitization activity, representing the OTD model. The change in total loans is included to control for money creation in which “loans create deposits”16, in line with Ben-David, Palvia, and Spatt (2015). This has probably a small effect during periods of QE as it seems that there is a break in the proportional relation between deposits and loans (Fig. 4). The change in total
13
I did not include the weighted average of actual interest rates of commercial banks, because extensive quantity data of all deposit products covered in Ratewatch are not available in the Call Reports. Therefore, it is impossible to calculate a weighted average deposit rate.
14
Bank fundamentals consist in this case of bank size, liquidity, capitalization, cost over income, return on assets, real estate loans over total loans and non-performing loans over total loans.
15
In line with Gagnon et al. (2011) I include only long-dated asset holdings of commercial banks that are compatible with the QE-targeted assets. The quantities and maturities of assets bought by the Federal Reserve are listed in Appendix E.
16
22
loans could also capture bank specific lending opportunities that might affect bank funding needs and willingness to raise deposits.
The vector of control variables, the implicit interest rate and the change in quantity of the specific deposit product are lagged to avoid simultaneity problems, which should remove potential correlation between the explanatory variables and the error term. This is common procedure in the literature (see, e.g., Acharya and Mora, 2015; Kashyap and Stein, 2000). With respect to the deposit products specific quantity used in Eq. (2), I implicitly assume that the asset purchases by the Fed only induces an inflow of deposits on transaction accounts and not on money market deposit accounts or time deposits. Therefore, I argue that there is no endogeneity involved. There are potential endogeneity problems in Eq. (1), as the asset purchases by the Federal Reserve affect both total deposits as the implicit interest rate. I come back to this issue in the discussion and limitations section. The quarterly change in the Fed Funds rate is the only non-lagged control variable. As discussed above, the timing impact of monetary policy on deposits is assumed to be at time 𝑇𝑇. In Eq. (1) and (2) bank fixed effects and time fixed effects are incorporated to control for unobserved heterogeneity. Both equations represent the full and preferred specifications used in the analysis.17 I incorporate robust standard errors clustered at bank level to correct standard errors for potential heteroskedasticity and autocorrelation in the data. In performing 2SLS, standard errors are corrected for the original regressors used in the first stage regression, otherwise standard errors are wrong. Standard error correction is only performed in case of no robust standard errors as the transformation of robust standard errors when using 2SLS lies outside the scope of this paper.18
I also analyze whether banks responded heterogeneously to asset purchases by the Federal Reserve in setting deposit rates by estimating the following equation presented below: ∆𝑅𝑅𝑇𝑇𝑇𝑇𝐷𝐷𝑖𝑖,𝑡𝑡 = 𝜉𝜉𝑖𝑖 + 𝜏𝜏𝑡𝑡+ 𝜑𝜑1∗ ∆𝑇𝑇𝑙𝑙 𝑇𝑇𝑇𝑇 𝐹𝐹𝐷𝐷𝐹𝐹𝑡𝑡+ 𝜑𝜑2∗ ∆𝑇𝑇𝑙𝑙 𝑇𝑇𝑇𝑇 𝐹𝐹𝐷𝐷𝐹𝐹𝑡𝑡∗ 𝐵𝐵𝑇𝑇𝑙𝑙𝐵𝐵 𝐶𝐶ℎ𝑇𝑇𝐼𝐼𝑖𝑖,𝑡𝑡−1
+ 𝜑𝜑3∗ 𝐶𝐶𝑖𝑖,𝑡𝑡−1+ 𝜓𝜓𝑖𝑖,𝑡𝑡
(3)
This specification captures heterogeneity in deposit rate setting in response to asset purchases with respect to bank size, liquidity and capitalization, degree of market funding, maturity profile of time deposits and off balance sheet ratio. These variables are incorporated
17
A Hausman test is performed which indicates that fixed effects estimation is preferred over random effects estimation, because there is significant correlation between the unobservables and independent variables.
18
23
in a vector of bank characteristics, represented by 𝐵𝐵𝑇𝑇𝑙𝑙𝐵𝐵 𝐶𝐶ℎ𝑇𝑇𝐼𝐼𝑖𝑖,𝑡𝑡−1. Bank characteristics are lagged for the same endogeneity reasons mentioned above. The same vector of control variable is used as in Eq. (1) and Eq. (2) but now ∆ ln 𝑄𝑄𝐼𝐼𝑇𝑇𝑙𝑙𝑇𝑇𝐼𝐼𝑇𝑇𝑄𝑄 𝑆𝑆𝐷𝐷𝐷𝐷𝐼𝐼𝐼𝐼𝑆𝑆𝐼𝐼𝐼𝐼 𝐷𝐷𝐷𝐷𝐷𝐷𝑖𝑖,𝑡𝑡−1 is incorporated in this vector. Also bank and time fixed effects are included. 𝐵𝐵𝑇𝑇𝑙𝑙𝐵𝐵 𝐶𝐶ℎ𝑇𝑇𝐼𝐼𝑖𝑖,𝑡𝑡−1 is interacted with ∆𝑇𝑇𝑙𝑙 𝑇𝑇𝑇𝑇 𝐹𝐹𝐷𝐷𝐹𝐹𝑡𝑡 such that the marginal effect of asset purchases on deposit rates ∆𝑅𝑅𝑇𝑇𝑇𝑇𝐷𝐷𝑖𝑖,𝑡𝑡 becomes:
𝜕𝜕∆𝑅𝑅𝑇𝑇𝑇𝑇𝐷𝐷
𝜕𝜕∆ 𝑇𝑇𝑙𝑙 𝑇𝑇𝑇𝑇 𝐹𝐹𝐷𝐷𝐹𝐹 = 𝜑𝜑1+ 𝜑𝜑2∗ 𝐵𝐵𝑇𝑇𝑙𝑙𝐵𝐵 𝐶𝐶ℎ𝑇𝑇𝐼𝐼𝑖𝑖,𝑡𝑡−1
(4)
The estimate of 𝜑𝜑2 captures the potential heterogeneous bank response to asset purchases in setting deposit rates with respect to the bank characteristics mentioned above. To allow for an economic interpretation of the estimated heterogeneity, the estimates are used to generate average marginal effects of asset purchases on deposit rates with the specific bank characteristic of interest at the 75th percentile and the 25th percentile.19 Thereby, only varying the bank characteristic of interest, keeping all the other variables at the mean.
5
Results
The results of the analysis are presented in this section. First, I test whether asset purchases induced an inflow of deposits. In a second stage regression the estimated inflow of QE-induced deposits is regressed on deposit rates to test the proposed shift in deposit supply. Finally, I analyze whether US commercial banks differ in their interest rate setting behavior in response to asset purchases by the Federal Reserve.
5.1 First stage regression
In Table 2 the results of Eq. (1) are presented. The results show that the quarterly change in total asset holdings of the Federal Reserve is positively associated with the quarterly change in total deposits for all model specifications. For the full and preferred specification of deposits including bank fixed effects, the marginal effect of asset purchases is positive and statistically significant at the 1% level. Therefore, the null hypothesis of no significant inflow of deposits in response to asset purchases by the Fed can be rejected. The marginal impact of an average asset expansion (5.3%) by the Fed corresponds to an average inflow of deposits of 3.6% on the balance sheets of US commercial banks.
19
24
As transmission via asset purchases directly from commercial banks is controlled for, the marginal impact of asset purchases on deposits can be interpreted as an inflow of deposits largely attributed to asset purchases from non-banks. The results suggest that the Federal Reserve bought a significant amount of long-dated assets from non-banks. However, the estimate might be confounded by other transmission channels causing deposit in- or outflows at time 𝑇𝑇 coinciding with asset purchases at time 𝑇𝑇. I discuss this issue below in the discussion and limitations section.
The coefficients of the dummy variables indicate that during each respective round of QE the inflow of deposits is significantly larger than the average inflow of deposits during the non-QE period. During three rounds of QE the inflow of deposits is on average respectively 3.4%, 4.6% and 5.7% larger compared to the non-QE period. The estimated inflow of deposits is economically very significant and suggests that asset purchases could lead to significant downward pressure on deposit rates via asset purchases from non-banks with commercial banks acting as an intermediary in asset purchases during QE.
Table 2: Least squares estimates of the impact of asset purchases on total deposits.
(1) (2) (3)
VARIABLES ∆ Ln Depositsi,t ∆ Ln Depositsi,t ∆ Ln Depositsi,t
∆ Ln Total Assets Fedt 0.120*** 0.671***
(0.014) (0.024) QE1 dummyt 0.034*** (0.002) QE2 dummyt 0.046*** (0.002) QE3 dummyt 0.057*** (0.002) Observations 127,128 127,128 127,128 R-squared 0.107 0.141 0.141 Number of cert 6,897 6,897 6,897
Bank FE NO YES YES
Quarter FE YES YES YES
Bank controls YES YES YES
25
5.2 Second stage regression
Before estimating Eq. (2), the fitted values for QE-induced deposits are generated: 𝑄𝑄𝑄𝑄 𝐼𝐼𝑙𝑙𝐹𝐹𝐼𝐼𝐼𝐼𝐷𝐷𝐹𝐹 𝐷𝐷𝐷𝐷𝐷𝐷𝑡𝑡 = 𝛽𝛽̂1∗ ∆ ln 𝑇𝑇𝑇𝑇 𝐹𝐹𝐷𝐷𝐹𝐹𝑡𝑡. As discussed in the methodology section, this
variable should be a good proxy for QE-induced deposits through asset purchases from non-banks. The summary statistics of QE-induced deposits are presented in Table 3. The estimated average quarterly inflow of QE-induced deposits is 3.7%. Therefore, it can be expected that deposit rates decline concurrently with the estimated QE-induced inflow of deposits.
Table 3: Summary statistics of fitted values for QE-induced deposits
(1) (2) (3) (4) (5)
VARIABLES N mean sd min max
QE-induced Deposits 133,172 0.037 0.105 -0.219 0.411
Table 4: Least squares estimates of the impact of QE-induced deposits on 12MCD10K and MM10K rates.
(1) (2) (3) (4)
VARIABLES ∆12MCD10Ki,t ∆12MCD10Ki,t ∆MM10Ki,t ∆MM10Ki,t
QE Induced Dept -0.140*** -0.010 (0.040) (0.060) QE1 dummyt -0.107*** -0.074*** (0.002) (0.003) QE2 dummyt -0.061*** -0.046*** (0.002) (0.003) QE3 dummyt -0.008*** -0.001 (0.002) (0.003) Observations 97,948 97,948 92,346 92,346 R-squared 0.307 0.307 0.118 0.118 Number of cert 5,515 5,515 5,286 5,286
Bank FE YES YES YES YES
Quarter FE YES YES YES YES
Bank controls YES YES YES YES
This table shows the results of estimating Eq. (2). This regression includes the estimate of QE-induced deposits from Eq. (1). The QE-induced inflow of deposits at time t is regressed on the quarterly change in the 12MCD10K and MM10K rate of bank i at time t. In the regression specification standard errors are corrected for the original regressors in the first stage regression. In columns (2) and (4) QE-induced deposits are replaced by three dummy variables indicating the three respective rounds of QE. Standard errors in parentheses. Statistical significance: *** p<0.01, ** p<0.05, * p<0.1
26
rates and MM10K rates. For 12MCD10K rates this relation is significant. The marginal impact of an average inflow of QE-induced deposits (3.7%) corresponds to a 0.5% decline in 12MCD10K rates. For MM10K rates the impact of QE-induced deposits is not statistically significant but all coefficients do show the expected negative sign.
A possible explanation for the relatively small downward rate adjustment is that banks could change their rate at time 𝑇𝑇 + 1 or even later in response to a QE-induced deposit inflow at time 𝑇𝑇. The data show that commercial bank do not change their rates in 63,070 observations of the total 133,565 observations for MM10K rates. For 12MCD10K rates 33,480 observations of the total 139,701 observations do not alter their rates. This implies considerable inertia in changing interest rates. Another explanation could be that the marginal impact of QE-induced deposits on deposit rates is declining over time as deposit rates become closer to the ZLB. When rates are closer to the ZLB, deposits face more competition from holding cash as a substitute, because the opportunity costs of holding cash is lower. The average MM10K rate adjustment in response to asset purchases by the Federal Reserve is estimated to be smaller than the downward adjustment of 12MCD10K rates. This might be attributed to the fact that the potential for downward adjustment is much smaller for MM10K rates as the average MM10K rate is relatively closer to the ZLB, which is illustrated in Fig. 1. Driscoll and Judson (2013) find that rate adjustment varies significantly with respect to the specific deposit type. They show that CD rates are more flexible than interest rates of money market deposit accounts which show considerable inertia changing on average every 20 weeks. This is also observed in this dataset, which could explain the relatively smaller estimated downward adjustment for MM10K rates.
27
The null hypothesis that QE-induced deposits put no downward pressure on deposit rates can only be rejected for small time deposits. The null hypothesis cannot be rejected for MM10K rates considering the transmission via QE-induced deposits. The results provide suggestive evidence that there was a QE-induced supply shift of deposits in which the increase in the quantity of deposits induced by QE at time 𝑇𝑇 is associated with a decrease in deposit rates in the same period.20 This finding is in line with the proposed transmission in this thesis, formalized in Hypothesis 1. Synthesizing the results presented in Table 4, it seems that there is a role for asset purchases by the Fed to facilitate interest rate pass-through with respect to deposit rates. With market rates at the ZLB, the total impact of QE seems to put significant downward pressure on deposit rates.
5.3 Heterogeneity in interest rate setting
Using Eq. (3) I test whether interest rate pass-through is heterogeneous with respect to a set bank of characteristics. In order to scrutinize heterogeneity in interest rate setting in response to asset purchases by the Fed, I focus on the coefficient estimates of 𝜑𝜑2 and the difference in marginal impacts considering the respective bank characteristics of
interest.
The heterogeneous impact of asset purchases on deposit rates is documented in Table 5. Banks that are more weighted towards a short maturity profile of time deposits reduced their MM10K and 12MCD10K rate significantly less as the estimated coefficient is positive and statistically significant at the 1% level. The average marginal effect on 12MCD10K rates for the bank at the 75th percentile and the bank at the 25th percentile is respectively -0.47% and -0.60%. This implies a difference in marginal effects over the interquartile range of 0.13%. Results for MM10K rates are similar and the difference in marginal effects is 0.09%. Bank size seems to give rise to heterogeneity in 12MCD10K rate setting. The estimated coefficient for bank size is negative and statistically significant at the 1% level. The average marginal effects at the 75th percentile and the 25th percentile with respect to bank size are respectively -0.58% and -0.49% with a difference of 0.09% over the interquartile range. This is not the case for MM10K rates and no significant relation is found. Regarding the degree of liquidity, a negative and significant coefficient estimate is documented for MM10K rates. Comparing the difference in average marginal effects suggests that a bank on the 75th
20
28
percentile with respect to liquidity lowered its MM10K rate with 0.10% more than a bank on the 25th percentile. Considering 12MCD10K rates, banks seem to respond homogeneously to asset purchases with respect to the liquid asset ratio as the coefficient estimate is statistically insignificant. For both deposit products there seem to be no significant heterogeneous response to asset purchases regarding the degree of capitalization and market funding. With respect to off balance sheet items, the estimated coefficient is positive and significant for MM10K rates. A bank on the 75th percentile lowered its rate with 0.14% less than a bank on the 25th percentile. The difference in marginal effects between the 75th and 25th percentile for 12MCD10K rates is very small and the estimated coefficient is insignificant.
Table 5: Heterogeneous marginal impact of asset purchases on 12MCD10K and MM10K rates.
(1) ∆12MCD10Ki,t (2) ∆MM10Ki,t VARIABLES 𝜑𝜑2 p75 p25 (p75-p25) 𝜑𝜑2 p75 p25 (p75-p25) Market Funding -0.037 -0.555 -0.527 -0.028 0.060 -0.040 -0.084 0.044 (0.023) (0.039)
Off Balance Sheet Items -0.006 -0.543 -0.541 -0.002 0.307*** -0.003 -0.138 0.135
(0.035) (0.062)
Maturity Time Deposits 0.116*** -0.472 -0.604 0.132 0.083*** -0.010 -0.104 0.094
(0.018) (0.029)
Capitalization 0.084 -0.537 -0.550 0.013 0.149 -0.051 -0.075 0.024
(0.078) (0.138)
Liquid Asset Ratio -0.006 -0.545 -0.538 -0.007 -0.097*** -0.107 -0.004 -0.103
(0.017) (0.029) Ln(Total Assets) -0.012*** -0.583 -0.493 -0.090 0.005 -0.044 -0.081 0.037 (0.002) (0.004) Observations 97,948 0.308 5,515 YES YES YES 92,346 0.120 5,286 YES YES YES R-squared Number of cert Bank FE Quarter FE Bank controls
Column (1) and (2) present the estimates of Eq. (3) under 𝜑𝜑2 for each bank characteristic. The average marginal
effects of asset purchases with the specific bank characteristic of interest at the 75th and 25th percentile are
presented under p75 and p25. They are respectively calculated using Eq. (7) and Eq. (8) in Appendix C. The difference between both are presented under (p75-p25) and calculated using Eq. (9) in Appendix C. A complete
list of estimates is presented in Table 9 in Appendix D including the estimate for 𝜑𝜑1 which is used in the
29
The results in Table 5 are in two cases in line with Hypothesis 2. Larger banks lowered their 12MCD10K rate relatively more than smaller banks. The other case is that more liquid banks reduced their MM10K rate more than less liquid banks. This is consistent with the argument that larger or more liquid banks are able to lower their rate more because they do not have to pay a lemon’s premium on their funding. There is no such evidence found with respect to capitalization. Banks that are heavily weighted towards market funding did not significantly decrease their deposit rate more in response to QE than banks that are less market funded. This is not in line with Hypothesis 3. The results regarding the proxy for an OTD business model (off balance sheet items) is in line with Hypothesis 4, only for MM10K rates. OTD banks lowered their MM10K rate less than banks with an OTH business model. An explanation of this result would be that OTD banks experienced more pressure on their funding structure during the collapse of the secondary loan market. In this situation they are less able to repackage and sell their assets in the secondary market. This would induce OTD banks to raise deposit funding which would lead them to reduce their rate less. The results with respect to the maturity profile of time deposits are in line with Hypothesis 5. I find a positive and significant relation between the maturity profile of time deposits and the change in deposit rates in response to QE for all specifications and deposit products under consideration. Banks that have a shorter maturity profile of time deposits have more rollover risk. This suggests a positive relation between rollover risk of time deposits and a change in deposit rates in response to asset purchases.
5.4 Robustness checks
I perform two robustness checks. The first check uses a different empirical strategy with dummy variables to capture the possible heterogeneous response to QE in setting deposit rates. Additionally, I control for the FDIC interest rate cap that might confound the results.
The findings of the dummy variable regression21, presented in Table 10 in Appendix D, are broadly in line with the results presented in Table 5. However, the documented differences in estimated coefficients for the 75th and 25th percentile are larger than in Table 5. This is because QE dummies capture the total impact of QE and therefore the magnitude of the coefficients are larger. The interaction terms with the liquid asset ratio and bank size show
21
The following equation is estimated:
∆𝑅𝑅𝑇𝑇𝑇𝑇𝐷𝐷𝑖𝑖,𝑡𝑡= 𝛼𝛼𝑖𝑖+ 𝛿𝛿𝑡𝑡+ ∑3𝑘𝑘=1𝛽𝛽𝑘𝑘∗ 𝑄𝑄𝑄𝑄 𝐹𝐹𝐼𝐼𝐼𝐼𝐼𝐼𝑄𝑄𝑘𝑘+ ∑3𝑗𝑗=1𝛽𝛽𝑗𝑗∗ 𝑄𝑄𝑄𝑄 𝐹𝐹𝐼𝐼𝐼𝐼𝐼𝐼𝑄𝑄𝑗𝑗∗ 𝐵𝐵𝑇𝑇𝑙𝑙𝐵𝐵 𝐶𝐶ℎ𝑇𝑇𝐼𝐼𝑖𝑖,𝑡𝑡−1+ 𝛽𝛽4∗ 𝐶𝐶𝑖𝑖,𝑡𝑡−1+ 𝜖𝜖𝑖𝑖𝑡𝑡
QE dummies are created for three rounds of QE. Such that 𝑄𝑄𝑄𝑄 𝐹𝐹𝐼𝐼𝐼𝐼𝐼𝐼𝑄𝑄𝑗𝑗 is one for round 𝑗𝑗 of asset purchases and
zero otherwise. The coefficient 𝛽𝛽𝑘𝑘 captures the total impact of QE on deposit rates during round 𝐵𝐵 via all
30
similar signs as in the former analysis. Size is again negative and significant for 12MCD10K rates and the interaction with liquid asset ratio is also negative and statistically significant for MM10K rates. Remarkably the coefficient for capitalization has become significant indicating that more capitalized banks increase their rate more during QE than less capitalized banks. This finding is roughly in line with the finding of Ben-David, Palvia, and Spatt (2015). They find that banks that are more capitalized also offer higher 12MCD10K rates. However, their model is estimated in levels such that the interpretation is different. Regarding the degree of market funding, the interaction term for 12MCD10K rates is again negative but now significant. This is not the case for MM10K rates and coefficients remain insignificant. For all specifications the coefficients with respect to the maturity profile of time deposits are positive and statistically significant, implying that rollover risk is a strong determinant for interest rate setting behavior of US commercial during QE for both deposit products. The interaction with off balance sheet items is again only positive and significant for MM10K rates, which suggests that the bank’s exposure to off balance sheet activities is only relevant in MM10K rate setting behavior. As results remain largely unchanged, findings can be considered to be robust regarding the employed model specification.
The FDIC introduced an interest rate cap in 2009 in which less than well capitalized banks were not allowed to take brokered deposits and not to offer a deposit rate that is 75 basis points above the national average rate. The introduction of the interest rate cap should affect interest rate setting behavior of less than well capitalized banks. These banks with a total risk-based capital ratio of less than 10% or a Tier 1 risk-based capital ratio of less than 6% or a leverage ratio of less than 5% are therefore dropped from the dataset. Consequently, 3,944 observations and 3,831 observations are respectively omitted from the 12MCD10K and MM10K dataset. The results are presented in Table 11 and 12 in Appendix D. Comparing the results in Table 11 with Table 4 and comparing the results in Table 12 with Table 9, implies that standard errors and estimated coefficients remain largely unchanged. This is as expected because the sample remains largely the same. Hence, findings are broadly considered to be robust when controlling for the FDIC interest rate cap.
5.5 Discussion and limitations
31
solely attributed to deposit creation via asset purchases from non-banks. When non-banks rebalance their portfolio and substitute their QE-induced deposits received at time 𝑇𝑇 for other long-dated assets bought at time 𝑇𝑇 the estimate will be biased. When banks are setting interest rates they would not lower their deposit rate as much compared to a situation in which deposits would remain in their books and are perfectly stable. However, no empirical strategy can overcome the issue of missing transaction data of asset purchases and therefore the results should be interpreted with caution. The implicit interest rate and the asset purchases by the Federal Reserve might be endogenously related. It is not an option to omit the implicit interest rate from Eq. (1) as this would result in omitted variable bias. Three-Stage Least Squares (3SLS) or Generalized Method of Moments (GMM) estimation might overcome this issue. Another issue is the possible dynamic nature of the interest rate adjustment process in response to QE. Panel error correction models are better suited to broadly capture the rate adjustment process rather than the static model. The static model suffers from bias as second order and other long run effects cannot be included. Moreover, the static model assumes implicitly that banks change their rate in the same period that the Federal Reserve buys assets. The data show that deposit rates adjust slowly and therefore banks could also adjust their rates at time 𝑇𝑇 + 1 or even later, in response to asset purchases at time 𝑇𝑇. Panel error correction models, GMM and 3SLS estimation lie outside the scope of this thesis and I leave this for further research. The generalizability of the results is limited as I only investigate two deposit products because data coverage for other deposit products is only broadly available from 2011 onwards. However, the deposit products under consideration are the most widely offered deposits in the US commercial banking system which implies substantial representativeness.
The theoretical argument proposed in this thesis in which QE induces a supply effect of deposits is however strong. Asset purchases from non-banks lead theoretically to a one-for-one increase in deposits.22 Additionally, the finding that banks even increase their holdings of QE-targeted MBS, Treasuries and Agencies over the QE period seems to imply that the Federal Reserve bought a significant amount of long-dated assets from non-banks. This suggests that banks were mere intermediaries in asset purchases by the Fed, implying significant inflows of QE-induced deposits. In theory, this should reduce deposit rates and facilitate interest rate pass-through with market rates at the ZLB.
22
32
6
Conclusion
Central banks collectively adopted unconventional monetary policy measures as policy rates were at the ZLB while economies were still in deep recession. LSAPs were broadly implemented to reduce long-term interest rates and stimulate economic activity. This study proposes that LSAPs could also lower short-term deposit rates, inter alia, by creating new bank deposits. To test this proposition, $10,000 12-month maturity time deposits and $10,000 money market deposit accounts of US commercial banks were analyzed in the period between 2007Q1 and 2013Q4. Additionally, heterogeneity in interest rate setting behavior in response to LSAPs was scrutinized for both deposit types.
This thesis presents evidence of a significant deposit inflow induced by the asset expansion of the Federal Reserve during three rounds of QE. Additionally, suggestive evidence of a downward rate adjustment in response to a QE-induced deposit inflow is documented. The overall results imply that the total impact of QE, without differentiating between separate transmission channels, put significant downward pressure on insured deposit rates. Thereby suggesting that QE led to a supply shift in deposits. Additionally, this paper finds heterogeneity in deposit rate setting behavior regarding rollover risk. Banks with a shorter maturity profile of time deposits experience more pressure on their deposit funding. Hence, they were less able to lower their deposit rate in response to QE.
QE could provide an opening for central bankers to affect deposit rates when the Fed Funds rate is at the ZLB. By buying large amounts of assets, interest rate-pass-through to deposit rates could be facilitated, potentially affecting the saving and spending decision of households and firms. However, bank rollover risk could be a constraining factor as it seems to reduce downward pressure of QE on deposit rates. For effective monetary policy and completeness of interest rate pass-through, microprudential supervision with respect to the maturity structure of time deposits could be intensified to reduce this potentially constraining factor.
33
