Excess Reserves During the U.S. Financial Crisis: A Bank-level Perspective∗ Silvio Contessi Federal Reserve Bank of St. Louis
Johanna Francis Fordham University
February 2010
Abstract The level of aggregate excess reserves held by U.S. depository institutions increased by a factor of 15 at the peak of the financial crisis of 2008-09. Although the amount of aggregate reserves is almost entirely determined by the policy initiatives of the Central Bank, individual bank’s motivations differ. This paper undertakes a systematic analysis of this massive accumulation of excess reserves using bank-level data for more than 7,000 commercial banks and almost 1000 saving institutions during the U.S. financial crisis that intensified in 2008. We propose a testable stochastic model of reserves determination with interest on reserves. The model handily returns a log-linearized version that can be estimated using bank-level data and censored regressions method. We find evidence of a precautionary motive for reserves accumulation and discuss alternative explanations for the accumulation of reserves. We then combine propensity score matching and a difference-in-difference approach to determine whether the beneficiaries of the TARP-CPP program accumulated more reserves than non-beneficiaries. We find no significant differences between the two groups.
VERY PRELIMINARY - PLEASE DO NOT QUOTE
JEL Classification: E44, E51, G21 Keywords: Commercial Banks, Financial Crisis, Excess Reserves, TARP
∗ Hoda El-Ghazaly provided excellent assistance with the data management. Silvio Contessi: Federal Reserve Bank of St. Louis, Research Division, P.O. Box 442, St. Louis, MO 63166-0442,
[email protected]; telephone: 314-444-7410; fax: 314- 444-8731. Johanna Francis: Fordham University, Department of Economics, E-507 Dealy Hall, 441 East Fordham Road, Bronx, NY, 10458,
[email protected]; telephone: 718-817-4055. The views expressed are those of the authors and do not represent official positions of the Federal Reserve Bank of St. Louis, the Board of Governors, or the Federal Reserve System.
1
Introduction
The aggregate level of deposits held by held by U.S. depository institutions (DIs) at the Federal Reserve Banks (Fed) increased by a factor of 50 at the peak of the financial crisis between the end of August 2008 and the end of December 2008, in parallel with the unprecedented expansion of the Fed balance sheet (Figure 1).1 This huge change in deposits at the Fed, most of which are excess reserves, prompted some commentators to argue that precautionary hoarding of cash and reserves was impeding loan growth and potentially slowing the recovery from the 2007-2009 recession. Keister and McAndrews (2009) explain clearly that the level of aggregate reserves is determined by the policy initiatives of the Federal Reserves, and may only marginally affect lending. However, an individual bank can alter the composition of its balance sheet and lend to firms and consumers, while reducing its excess reserves. Therefore, studying the determinants of bank level reserves accumulation helps understand the heterogenous effects of the financial crisis - and to some extent of the Fed actions - in the commercial bank sector. This paper undertakes a systematic analysis of this massive accumulation of excess reserves using bank-level data for more than 7,000 commercial banks. We first discuss a stochastic model of reserves choice that incorporates interest on reserves (IOR). The model handily delivers a log-linearized version that can be estimated using banklevel data and censored regressions method. We empirically identify the determinants of reserves accumulations and find evidence of a precautionary motive. Specifically, we find that banks in weak financial health appear to accumulate relatively more excess reserve than sounder banks unlike similar to the Japanese case during the Lost Decade. We show that changes in the spread between yields on safe short-term assets and the IOR are negatively correlated with reserves accumulation. When alternative investment opportunities yield larger returns relative to the IOR, commercial banks then to decumulate reserves. We discuss alternative explanations for the accumulation of reserves, chiefly unwanted accumulation due to management of cash inventories. In the second part our analysis we study the effect of TARP on cash accumulation. To ascertain the effect of the capital injections one we would want an answer to the following question: What would have happened to cash accumulation of those bank 1 DIs (commercial banks, savings institutions, credit unions, and foreign banking entities) may hold their required reserves as either vault cash or deposits at their regional Federal Reserve Bank. Deposits at the Federal Reserve are the sum or reserve balances with Federal Reserve Banks and required clearing balances; on Aug 27, 2008 total deposits at Federal Reserve Banks were $20.394 billion; on December 29, 2010 they were $1020.937 billion.
2
that benefitted from the TARP-CPP, had they not received it? The question as stated is impossible to answer because we do not observe an exact counterfactual. However we can observe banks that were ex-ante similar to the CPP beneficiaries but did not received the capital injections. Operationally, we use propensity score matching to construct a control group of non-CPP that we then compare to the CPP beneficiaries. The term of comparison is then a estimate of the difference-in-difference between couple of indicators for the two groups of banks. This approach filters unobservable differences between the two groups of banks and constitutes the best synthetic counterfactual we can reconstruct. When we consider the possibility that the beneficiaries of the capital injection operated through the Capital Purchase Program (CPP) that was part the Troubled Asset Program accumulated more reserves than non-beneficiaries, we find no significant differences between the two groups. The paper is organized as follows. In Section 2, we present an overview of the existing literature and in Section 3 we discuss the specific case of the United States. In Section 4 we develop the testable model and Section 5 presents the data and the estimation . Section 6 studies the effects of the CPP. Finally, Section 7 concludes.
2
Related Literature
The literature on banks’ reserves during times of economic crisis is truly extensive and includes a number of key contributions whose discussion cannot possibly be exhausted in this section. Its premise is that banks use reserves to manage their liquidity risks, informed by a precautionary motive that suggests to keep a sufficient level of cash to meet their obligations, and affected by investment alternatives that present comparable (low) risks. Cash and reserves, however, imply an opportunity cost to the extent that they are remunerated less than higher interest-yield asset. In a theoretical perspective, a non-exhausting list of papers include Orr and Mellon (1961), Poole (1968), Frost (1971), and Hanes (2006). Frost (1971) models banks demand functions for excess reserves as an inventory problem, assuming that excess reserves allow them to reduce the cost of meeting requirements when they face random reserve flows and transaction curves. The model generates kinked demand curves that rationalize excess reserves accumulation during the 1930s. 3
Hanes (2006) develops a model in which an increase of reserves supply reduces longterm rates, holding fixed expectations of future overnight rates, and shows that the model is consistent with the relationship between reserve quantities and bond yields over 1934-39. Describe From an more empirical perspective, there is a large literature that studies banks reserves during the Great Depression and a small literature focused on Japan during the Lost Decade. The studies that tried to explain why banks accumulated so much reserves during and after the Great Depression either favor a “precautionary motive” for reserves accumulations or an alternative asset” view. Friedman and Schwartz (1963) argued that banks desired a higher level or reserves after the panic of the early 1930s, motivated by a precautionary motive against the risk of large unexpected withdrawals by depositor. Horwich (1963) suggested an explanation based on the lack of profitable alternative to holding cash. A similar conclusion, hinging on the role high risks and low return of alternatives, was reached in Bernanke (1983), and Bernanke and Gertler (1990). Calomiris and Wilson (2004) further argue that high levels of reserves amounted to signal liquidity to depositors and competitors. Van Horn (2009) looks at earlier years showing that Federal Reserve System non-members – that had no access to the lender of last resort – increased their ratio of excess reserves to assets after the first banking panic much more than non-member banks – that could access emergency lending. ?? use bank-level data to understand whether the doubling of reserves requirements imposed by the Federal Reserve in 1936-37 increased reserve demand and induced a credit contraction, contributing to the recession of 1937-38. They find that reserve requirements were not binding on bank reserve demand in 1936 and 1937 and therefore had little impact on credit availability, and argue that increases in reserve demand occurred between 1935 and 1937 reflected changes in the fundamental determinants in reserves demand, not changes in reserve requirements. The Japanese Lost Decade also offered an important case of steep reserves accumulation. Ogawa (2007) studies the determinants of bank-level reserves accumulations in a fashion similar to our approach in the next section. He determines that a strong precautionary motive induced banks with a large amount of bad loans to accumulate relatively more reserves. Uesugi (2002) studies the liquidity effect – the propositions 4
that monetary expansion lowers short-term nominal interest rates – in Japan.2 Finally, there is a growing literature that discusses excess reserves in the context of contemporary monetary policy when the monetary policy authority pays an IOR, as practiced in Canada, the U.S., New Zealand, the Euro-area, Hungary, and Poland, for example. Goodfriend (2002) argued in favor of paying IOR to add an additional policy instrument to the toolkit of modern central bankers. Keister and McAndrews (2009) explain how the payment of IOR can generate a floor-system that “divorces” money from monetary policy, as the supply of reserves is not necessarily tied to a target interest rate, allowing Central Banks to increase the supply of reserves without driving market rates below the target. Hornstein (2010) develops a stylized monetary model to explain the mechanics of monetary policy when the central bank pays an IOR and finds that although the responses of inflation and output are slightly different from models in which reserves yield zero-interest, such differences are small. With specific reference to the U.S. financial crisis, Bech and Klee (2010) studies a theoretical model in which bargaining power among trader explains why the federal funds rat remained below the target rate and the IOR during the U.S. financial crisis.
3
Reserves Accumulation: The U.S. Experience during the 2007-10 period
Institutional Details. In the U.S. reserve requirements are the amount of funds that a depository institution must hold in reserve against specified deposit liabilities, in the form of vault cash or deposits with Federal Reserve Banks. The Federal Reserve Board’s Regulation D specifies the dollar amount of a depository institution’s reserve requirement through a reserve ratio applied to reservable liabilities (see Table 1). Although reservable liabilities consist of net transaction accounts, non-personal time deposits, and eurocurrency liabilities, since December 27, 1990, only net transaction accounts carry a non-zero reserve requirement.3 The reserve ratio depends on the amount of net transactions accounts at the depository institution. After the the Garn-St Germain Act of 1982, the first $2 million of reservable liabilities from reserve requirements. The amount of net transaction ac2 The
latter study is related to work by Hamilton (1997) and Thornton (2001), among many others. Board of Governors has sole authority over changes in reserve requirements, within limits specified by law. See http://www.federalreserve.gov/monetarypolicy/reservereq.htm 3 The
5
counts subject to a reserve requirement ratio of 3 percent was set under the Monetary Control Act of 1980 at $25 million.4 Net transaction accounts over the low-reserve tranche are subject to a 10 percent reserve (see Table 1). In order to ensure that sound depository institutions can meet their funding needs, eligible depository institutions can borrow under the primary credit program of the discount window.5 The IOR after October 2008. Historically, depository institutions have preferred to minimize the amount of reserves they hold because neither vault cash nor reserves at the Federal Reserve used to yield interest income. However, on October 9, 2008, the Federal Reserve Banks started paying interest on required reserve balances and excess balances. Excess reserved jumped as a response to both this policy changed and likely to a precautionary effect as discussed in the next sections. The first three months of non-negative IOR witnessed a distinction between balances held to fulfill reserve requirements (so-called “required reserve balances”) and balances held in excess of required reserve balances and contractual clearing balances (“excess reserve balances”). The rate paid on required reserve balances was 10 basis points below the average target federal funds rate, while the rate paid on excess balances was 75 basis points below the lowest target. The reference window was the maintenance period FFR. The spreads were subsequently reduced twice before the end of 2008.6 These intraquarter changes do not affect our discussion as we will be studying quarterly data. How are do these institutional details related to our discussion? Whether banks have an incentive to lend reserves out or not depends on the relationship between the target interest rate, the Federal Funds Rate and the floor rate, i.e. the IOR, as explained in Keister and McAndrews (2009). When they are equal, banks never face an opportunity cost of holding reserves. However, during most of the financial crisis, and into 2010, these two rates were different: The target FFR was 4 The exemption amount is adjusted each year according to a formula specified by the act. The “low-reserve tranche” is also adjusted each year. 5 For example, if a DI experiences operational difficulties with its funds management systems, it is at risk of an overnight overdraft. Funding need at an individual institution can also arise from circumstances in which aggregate reserves in the banking system are significantly lower than what the Open Market Desk was anticipating in its management of the Federal Funds Rate target. 6 The first reduction brought the spread to zero for required balances and 35 basis points for excess balances, and zero for both by the maintenance period ending on November 19, 2008. However, after the December 2008 FOMC meeting the interest rates on required reserve balances and excess balances were both set at 25 basis points, upper bound of the newly established target range for the federal funds rate of 0 to 25 basis points. See Bech and Klee (2009) for an excellent discussion.
6
between 0 and 0.25%, while the IOR was kept at 0.25%. At the times, monetary policy authority were concerned with the strains in the interbank market, rather than weak bank lending to firms and households. Implications for Monetary Policy. What happens if the target interest rate and interest on reserves are not equal?7 If the FFR is higher than the interest on deposits, there are non-negative opportunity costs of holding reserves, and the IOR represents a floor. (TO BE COMPLETED)
4
A Simple Model of Excess Reserves Accumulation
In this section, we develop a simple of model of reserve determination that allows us to focus on the determinants of reserve accumulation that can be identified empirically using bank-level data. Consider a bank i that allocates a given level of deposits Di between an interestbearing asset and cash or reserves at the Federal Reserve. Consider the case in which the IOR (rIOR ) is non-negative. When rIOR is zero, any positive differential between the returns on the asset (rA , for example the rate on 1-month Treasury bills) and the zero-yield reserves, induces a profit-maximizing bank to keep reserves (Ri ) at the minimum required level (δ¯i Di , a share δ¯i of deposits). When rIOR is positive, banks may have an incentive to hold excess reserves depending on the relationship between interest rates. In both cases, deposits can be withdrawn at any time, so the bank also faces the risk of large unanticipated withdrawals and potentially of a bank run if the funds for such withdrawals are not available. The bank run does not need to be due to customers withdrawals but can potentially occur on the interbank market. If this happens, the bank can obtain funds paying a penalty rate of rp,i > rA . Thus, the bank problem is maximizing expected interest income, subject to a constraint due to the risk of reserves shortage beyond required reserves: 7 The ECB, as explained in Keister, Martin, and system that currently pays 0.25% on bank deposits whereas the main refinancing operations rate is 1% (and the marginal lending facility rate is 1.75%).
7
maxrA (Di − Ri ) + rIOR Ri − rp,i E [M ax (0, Li − Ri )]
(1)
s.t.δ¯i Di ≤ Ri
(2)
Ri
where δ¯i is the reserve requirement for bank i, and Li is a stochastic variable representing deposit withdrawal, or reserves losses. As the objective function is concave, by first order condition, the optimal amount of reserves is determined by this first order condition: rp,i Pr [Li ≥ Ri ] = (rA − rIOR ) − λ.
(3)
On the left-hand-side, the marginal benefit of accumulating reserves is determined by the interaction of the penalty rate and the probability of suffering larger-than-reserves withdrawals Pr [Li ≥ Ri ]: the larger the stock of reserves, the lower such probability and the lower the cost of emergency borrowing at the penalty rate. On the right-handside, the marginal cost of increasing reserves is determined by the forgone revenues of investing at larger-than-rIOR returns on alternative assets net of the benefit of relaxing the constraint (λ ≥ 0 is the Lagrange multiplier associated with the required reserves constraint). When the optimal reserve holding Ri∗ exceeds the required reserves, then λ = 0, and the constraint is not binding. The constrained solution in which λ > 0, instead, identifies situations in which the bank accumulates only the required reserves δ¯i Di . The demand for reserves increases when the ratio between the interest rate differential rA − rIOR and penalty rate rp,i rises. Ogawa (2007) use a zero-yield reserves version of this model, similar to Freixas and Rochet (1997) to interpret the increase of excess reserves holdings in the Japanese experience of the Lost Decade. In Japan, banks with more bad loans had larger reserves presumably to prevent the possibility of a bank run. In the model this is captured by by a decrease of P r [Li ≥ Ri ]. The model handily extends to a log-linearized version that facilitates reduce-form estimation using bank-level data. We can assume that banks perceive deposit withdrawals Li as draws from a Pareto distribution with density function is
8
f (Li ) =
θLθ0,i Lθ+1 i
; L0,i < Li < ∞
(4)
with location parameter L0,i > 0 and shape parameter θ < 0. Under this this specification the probability that withdrawals exceed reserves becomes Pr [Li ≥ Ri ] =
Ri L0,i
−θ (5)
that can be inserted in Equation 3, as follows rp,i
Ri L0,i
−θ = rA − rIOR − λ,
(6)
that becomes, in logarithmic terms, 1 log Ri = log L0,i − log θ
rA − rIOR − λ rp,i
.
(7)
In this specification, reserves depend negatively the ratio between interest rate and penalty rate, scaled by a parameter θ that governs the variance of deposit withdrawals (the larger θ, the higher the probability of relatively larger withdrawals). Larger L0,i translate into a right-shift of the withdrawal distribution:for a given level of the ratio of interest-to-penalty rates, the ith bank desires to keep larger reserves when L0,i is higher. We assume that this scale parameter depends on the strength of a bank’s precautionary motive, and is an inverse function of the financial health of a bank. Specifically, we assume the following L0,i = αDiη F IN HEALT Hi− ; α, η, > 0,
(8)
that can be plugged in 7 as follows: 1 log Ri = log α + η log Di − log F IN HEALT Hi − log θ
rA − rIOR − λ rp,i
(9)
Because under λ > 0, the reserves requirements are constraining(Ri = δ¯i Di ) we 9
have a system of equations as follows:
log Ri − log δ¯i Di = 0 log Ri − log δ¯i Di = log α − log δ¯i + (η − 1) log Di − log Di − 1 rA − rIOR − log F IN HEALT Hi − log θ rp,i
(10) (11) (12)
that can be estimated using censored regression methods.
5 5.1
The Determinants of Reserves Accumulation Data and Descriptive Statistics
We create a panel of commercial banks, using multiple datasets. The main sources for financial information the quarterly Reports of Condition and Income database (commonly called the Call Reports). The CRs contain regulatory information for all banks regulated by the Federal Reserve System, the Federal Deposit Insurance Corporation (FDIC), and the Comptroller of the Currency. In these datasets, banks report their individual-entity activities on a consolidated basis for the entire group of banks owned by the reporting entity at the end of each quarter. We use data for the quarters between 2007:Q1 and 2010:Q2. During this period the number of entities in the CRs fell from 8,209 to 7,424 banks as a result of failures, mergers, and acquisitions. Entities in each quarter’s CR typically belong to Bank Holding Companies (BHC). The most frequent proprietary structure is an individual BHC controlling an individual bank. In many instances however, an individual BHC may control many banks or a combination of banks and thrifts. We call this banking groups “multi-unit” BHC. Several adjustments are made to deal with troublesome entities. We first regroup “new” commercial banks – financial entities that were not typically be regulated as banks (and hence did not file CR) – but acquired charters in 2008-09 because either they applied for a charter or were acquired by regulated commercial banks.8 These “banks” 8 Namely, Goldman Sachs, Morgan Stanley, Merrill Lynch, American Express, CIT Group Inc., Hartford Financial Services, Discover Financial Services, GMAC Financial Services, IB Finance Holding Company, and Protective Life Corporation.
10
are likely to show peculiar reserves accumulation patterns (significantly different from other commercial banks) due to the nature of the intermediation function they carry out. Finally, We drop foreign owned banks as they are likely to manage their cash as part of an international network. Figure ?? is the plot of the cross-section distribution of our measure of excess-torequired reserves between 2008Q2 and 2010Q2. Each histogram represents one quarter and is left-censored at zero (as no bank holds less than required reserves) and rightcensored at 75 (as our data contain a number of banks that have ratios well in excess of 75).9 Over the 2008:Q32010:Q2 period, the distribution of the excess-to-required reserves ratio has become more dispersed (less peaked and with a fatter tail), indicating that more banks have accumulated larger amounts of excess reserves, in parallel with the expansion of the Federal Reserve Budget. The reasons why the distributions are so dispersed, however, are an open research question that we discuss in the next Section. 5.2
Results
Consistent with the simple theoretical model we developes in Section 4We use two regression techniques to investigate what has caused the build-up of excess reserves during the 2008-2010 period. We first estimate a tobit model in which we consider two different dependent variables: the level of reserves and the ratio of excess reserves to total assets (we only present the level of reserves results in this version of the paper). We include three main groups of variables, one to measure the effective interest rate on other assets, one to measure the penalty rate for holding insufficient reserves, and several variables to measure bank health. We also include the log of deposits as a scale variable and two measures of uncertainty, the conditional variance of industrial production (over the same time frame as the reserves data), and the BAA corporate to 3 month Treasury bill spread. These measures of uncertainty vary over time but not across banks. Using a tobit model initially, with the log of excess reserves as the dependent variable, we find that the coefficient on deposits is positive and significant, the difference between the 3 month Treasury Bill rate and the interest rate paid on reserves has a negative and significant effect on excess reserves, while the penalty rate has a positive 9 For each of the 9 quarters, we counted the following number of banks exceeding a excess-to-required reserve ratio of 75: 38, 45, 77, 107, 114, 143, 126, 3, and 4.
11
and significant effect. These results are as expected: the more deposits a bank retains, the larger its reserves (this co-variate is a scale variable). The coefficient on the difference between the 3 month t-bill and interest rate paid on reserves measures the responsiveness of excess reserves to changes in the opportunity cost of holding reserves, thus excess reserves respond negatively to increases in this ratio as banks should reduce reserve holdings when more lucrative economic opportunities arise, ceteris paribus. On the other hand, when banks face uncertainty about withdrawal rates, they should increase reserves. Thus we find excess reserves respond strongly and positively to increases in the penalty rate, which measures the penalty banks face for maintaining inadequate reserves. The strength of this response may also be picking up the fact that during the financial crisis banks did not want provide negative market signals regarding their liquidity by borrowing in the overnight market. Maintaining excess reserves insures that they won’t need to use the overnight market. We use several measures of bank health, two of which we report here. We use the Tier 1 capital ratio which is equal to the ratio of tier 1 capital to risk adjusted assets and net loan charge-offs.10 If either of these measures of bank health increase, it implies that banks either are taking on more risk (their Tier 1 capital ratio rises) or the quality of their loan portfolio is declining. In either case, increases in these measures (which implies poorer health) results in larger excess reserves. We also consider two measures of uncertainty in production and the financial market: the conditional variance of industrial production (estimated from an AR process with GARCH error structure) and the ratio of BAA rated corporates to the 3 month Treasury bill rate. (need to do write-up of ip-cov variable) We find a large negative coefficient on the conditional variance of industrial production, which is puzzling, but a positive coefficient, as anticipated, on the BAA corporate to 3 month Treasuries. One problem with using the conditional variance of industrial production is that banks already began responding the financial crisis before the volatility of industrial production spiked in 2008Q3. Changes in industrial production are going to mainly effect the behavior of commercial loans, of which banks hold relatively little. The behavior of the junk to risk free bond ratios reflects market liquidity that banks were likely more responsive too. 10 We also tried banks’ loan loss provision and non-performing loans, but these measures of bank health gave conflicting results across regressions.
12
As a robustness check, we also run the same set of regressions using Powell (1984) censored least absolute deviations estimator (CLAD) model for data that admits a corner solution (i.e., zero excess reserves is an optimal choice). The advantage of the CLAD estimator over the tobit is that it is robust to heteroscedasticity and is consistent and asymptotically normal for a variety of error distributions, unlike the tobit. This specification is consistent with the tobit results reported in the previous tables are confirmed under this specification. Banks respond to increases in deposits and the penalty rate by increasing reserves, while increases in the differential between rates on 3 month treasuries and IOR strongly reduces reserve holdings. Further, banks respond to a deterioration in their loan portfolio by increasing reserves as expected. The regressions in Tables ?? through ?? all use the tobit model with a left censored value (0); standard errors are in brackets. The three types of banks we control for are: banks with excess reserve ratios lrager than 50 (PB); credit card companies with bank charters (CC); foreign owned banks (FR). We look at interactions of each of these three dummy variables with either log-deposits or tb3m-IOR. For the log-levels regressions we used four measures of bank health: x17 (tier 1 capital ratio), riad4230, rcfd1407 and riad4365. x17 is used in all regressions, the other three are used individually. For the ratios regressions we consider eight measures of bank health: x17, x2, x3, x20, x27, x25 and (y11, y12), the latter of which are used together. For measures of general economic uncertainty, we use a combination of ip-cov and baa-tb3m. IP-Cov is the conditional variance from an AR(1,1) with GARCH(2,1) error terms of industrial production over the time frame of the sample 2007:Q3-2010Q2. If we use them separately, we get a large negative sign on ip-cov and a negative sign on baa-tb3m. The CLAD model was run for 1000 replications of the bootstrap standard errors Standard errors are in brackets Level 1 and level 2 differ by the interaction effect for each of the 3 dummy variables The three types of banks we control for are: banks with excessive excess reserve ratios (PB); credit card companies with bank charters (CC); foreign owned banks (FR) We look at interactions of each of these three dummy variables with either log-deposits or tb3m-IOR There are 4 measures of bank health: x17 (tier 1 capital ratio), riad4230, rcfd1407 and riad4365 x17 is used in all regressions, the other three are used individually (they are highly co-linear) The CLAD model was run for 1000 replications of the bootstrap standard errors. Standard errors are in brackets 13
6
Did TARP Beneficiaries Accumulate Larger Excess Reserves?
In this section we discuss whether the CPP program under the TARP umbrella induced beneficiaries banks to over-accumulate cash and reserves. 6.1
Data and Methodology
We match the Treasury data on the CPP disbursements to the unbalanced panel created from the CRs. With few exceptions, most capital injections were granted to BHC, not to bank. In the case of single-bank BHC we attribute the capital injection to the bank that maintains its CR identifier. In the case of multi-unit BHC there is no possibility to detect the ultimate beneficiary of the CPP so we maintain the BHC, instead of each subsidiary bank, in our dataset. We sum the relevant CR variables for all subsidiaries that belong to the BHC group that received the CPP, and use the BHC identifier. We analyze case by case and include the banks in the panel only if the substantial majority of the banking group activity (measured by deposits) is carried out by commercial banks in the group.11 After creating appropriate banking groups for the case of multi-unit banks, we matched the CPP information collected from the Transaction Reports using either the CR identifiers or the BHC identifiers. Our TARP/CPP information includes the amount of CPP, the date of the CPP announcement, dummy variables and dates for double payments, repayments, the number of banks and thrifts in the multi-unit BHCs. We match individual banks identifiers to BHC identifiers. Every information available at the BHC level is assumed to carry over to individual banks in the group. For example, (ii) we collect information on whether BHC are publicly traded from a publicly available dataset from the Federal Reserve Bank of New York – that we updated until the end of 2009 – and construct a dummy variable equal to 1 for each bank in the publicly traded BHC. (ii) We use BHC identifier to match each bank with our proxy from management quality.12 To evaluate the impact of CPP, ideally we would like to have information on the set of CPP applicants and compare the performance of a bank that receives a capital injection to the performance of the bank’s most similar 11 The largest imbalance we found was a three-units BHC in which a thrift held about 5% of total group deposit. In all other case, the share of the thrifts was substantially smaller. While there is a chance that all CPP injection was channelled in the thrift, we think this is an unlikely event. 12 The “CRSP-FRB Link” dataset is available at this url: http://www.newyorkfed.org/research/banking_research/ datasets.html.
14
institution. While the information on entire list of CPP applicants was not released by the U.S. Treasury in part to avoid potential stigma associated to the applying bank. However, the U.S. Treasury revealed that thousands of applications were submitted. Therefore, we use propensity score matching techniques to identify a control group of non-CPP that closely match the CPP beneficiaries. The basic idea is to placed a bank into the control group if it is sufficiently similar to a CPP beneficiary on the basis of the key determinants of the capital injection. In other words, our goal is to find a set of control firms that are a priori equally likely to received capital injection as those banks which ultimately receive the capital injection. This method is allows matching without an ex-ante specifying a particular model. The first part of our strategy to address the endogeneity of TARP status is to focus on banks that were granted the capital injection. This approach implies a substantial reduction of the number of banks considered. However, our data set is large enough that we are still left with a sufficient number of observations to generalize our results with confidence. The advantage of focusing on banks observed before and after the capital injection is that differencing over time allows us to eliminate the influence of all observable and unobservable non-random elements of the TARP decision that are either constant or strongly persistent over time. Using a difference-in-differences technique allows us to compare TARP banks with banks that did not receive TARP.13 To address the selection issue, we combine this approach with propensity score matching. The matching procedure controls for the selection bias by restricting the comparison to differences within carefully selected pairs of banks with similar observable characteristics. Each pair consists of unit that received the capital injection and a bank with similar observable characteristics in the quarter preceding the concession of the TARP. Let CP Pi,t = 0, 1, be a dummy variable indicating whether a bank i received CPP support during quarter. The aim of the analysis is to estimate the causal effect of TARP on an outcome Yi,t – cash as a share of total assets – during quarter t, or the average treatment (ATT). Ideally we would want to measure the following quantity E Yi,t1 − Yi,t0 |CP P = 1 = E Yi,t1 |CP P = 1 − E Yi,t0 |CP P = 0 13 One alternative would be to use the Heckman selection model to control for the selection of TARP recipients using bank-specific information from the periods before the capital injection. This approach, however, is still vulnerable to problems of non-random sample selection.
15
which is the difference between the performance paths of banks that received TARP (first term) and the analogous outcome of the same banks, had they not been granted TARP (second term). The complication here is that the outcome of any one bank is observed under either the TARP or non-TARP status, but never both, thus making the latter outcome an unobserved counterfactual. Because each bank is only observed as either receiving the treatment (Yi,t1 ) or not receiving the treatment (Yi,t0 ). The average effect of CPP support contains an additional selection component that is zero only if bank i is randomly assigned to the treatment or the control group: Given the fact that we conduct our evaluation in a non-experimental setting, the selection bias is likely to be non-null, as banks have observable characteristics that make them more or less likely to receive the treatment. Therefore, we use a matching techniques to approximate the (unavailable) counterfactual by drawing comparisons conditional on the vector X of observable banks characteristics. As stated in Dehejia and Wahba (2002), “When the relevant differences between any two units are captured in the observable (pretreatment) covariates, which occurs when outcomes are independent of assignment to treatment conditional on pretreatment covariates, matching methods can yield an unbiased estimate of the treatment impact.” The necessary assumption for the validity of the procedure is that conditional on the observable characteristics that are relevant for the acquisition decision, potential outcomes for the treated (TARP recipients) and non-treated banks (non-TARP recipients) are orthogonal to treatment status, given the observable covariates: E Yi,t0 |X, CP P = 1 = E Yi,t0 |X, CP P = 0 = E Yi,t0 |X under the assumption that all banks have the same performance if non-treated, i.e. the conditional independence assumption that holds if X contains all variables that affect both selection into treatment and performance indicators. The selection bias represents the difference between the outcome of TARP banks, under the hypothetical circumstances that they had not been granted TARP, and those banks that never received TARP, in the same (actual) situation of no TARP injection. If the selection bias represented by the second term is zero for given realizations of the vector X, then this approach returns the causal effect. One can then argue that the performance difference of performances between TARP and a carefully selected group 16
of non-TARP banks in the control group is an unbiased estimate of the causal effect under the matching assumption. The high dimensionality of the observable characteristics , i.e. a number of variables that have significant predictive power for determining whether a given bank will become receive TARP makes propensity score matching techniques particularly attractive. Although conditioning on a vector of variables requires a choice along which dimensions to match units or which weighting scheme to apply, Rosenbaum and Rubin (1983) and Dehejia and Wahba (2002) demonstrate that the propensity score provides a natural weighting scheme that yields unbiased estimates of the treatment impact. Since conditioning on the propensity score is equivalent to conditioning on all variables in the treatment model, the dimensionality problem can be solved by conditioning on the propensity score rather than a full vector of variables. The propensity score is the predicted probability of treatment, which in our case is the probability of a bank receiving TARP. pi,t = P rob(CP Pi,t = CP P (Xi,t−1 )) Making use of this result, we compare the performance of banks within the couples of observations matched on the basis of the propensity score. The combination of matching and a difference-in-differences approach means that we are trying to determine differences in the paths of performance between the TARP banks and the the matched control banks that had similar characteristics in the pre-TARP quarter. The performance analysis begins in the TARP period and focuses on the change in performance over the following quarters, either cumulatively or quarterly. ? emphasize the benefits of combining matching and a difference-in-differences approach for controlling for observable and unobservable but constant differences between treatment and control units. While matching accounts for differences in observable characteristics, its combination with difference-in-differences analysis provides scope for an unobserved determinant of participation as long as it can be represented by separable individualand/or time-specific components of the error term. Examples of such determinants include a particular banks being chosen for TARP because of a political connection or a of a bank manager not applying for TARP because of strong anti-Government intervention beliefs (CNNMoney.com, 2010). Duchin and Sosyura (2010) for example, argue that political connections – as measured by to contribution House members on 17
finance committees and representation at the Federal Reserve via board members – have significant positive marginal effects on the probability of banks being granted TARP. In our environment, such effect is likely to be time-invariant over the period of TARP and wiped out by the difference-in-difference approach. Therefore the final measures we are interested in are the following definitions of ATT: n
AT T
1
AT T
2
n
1X 1 1X 1 0 = Yi,t − Yi,t0 − Yi,t−1 − Yi,t−1 n 1 n 1 n
(13)
n
1X 1 1X 1 0 = Yi,t+1 − Yi,t+1 − Yi,t − Yi,t0 n 1 n 1
(14)
.. AT T 5 =
(15)
n 1X
n
1 0 Yi,t+4 − Yi,t+4 −
1
n 1X
n
1 0 Yi,t+1 − Yi,t+1
(16)
1
when we consider quarter-on-quarter growth rates; we consider the pre-TARP quarter t − 1 = 2008 : Q3. We also consider a definition that considers cumulative growth rates between the pre-TARP quarter and the last quarters for which the data are comparable 2009:Q4. n
AT T
1
AT T
2
n
1X 1 1X 1 0 = Yi,t − Yi,t0 − Yi,t−1 − Yi,t−1 n 1 n 1 n
(17)
n
1X 1 1X 1 0 0 = Yi,t+1 − Yi,t+1 − Yi,t−1 − Yi,t−1 n 1 n 1 ..
AT T 5 =
(19)
n 1X
n
(18)
1 0 Yi,t+4 − Yi,t+4 −
1
18
n 1X
n
1
1 0 Yi,t−1 − Yi,t−1
(20)
n
AT T
1
n
1X 1 1X 1 0 = Yi,t − Yi,t0 − Yi,t−1 − Yi,t−1 n 1 n 1 n
AT T 2 =
AT T 5 =
6.2
(21)
n
1X 1 1X 1 0 0 Yi,t+1 − Yi,t+1 − Yi,t−1 − Yi,t−1 n 1 n 1
(22)
..
(23)
1 n
n X 1
1 0 1 − − Yi,t+4 Yi,t+4 n
n X
0 1 − Yi,t−1 Yi,t−1
(24)
1
Timing
As the program was announced in October 2008, a number of applications were submitted to the U.S. Treasury. At the time of writing and despite various lawsuits under the Freedom of Information Act of 1966 the Treasury has not disclosed the list and the timing of applications, so we need to rely on informal evidence on the application timing and the pool applicants. To begin with, treasury officials revealed that “thousands of applications” were received. United States Department of Financial Stability (2010) also reports that the rate at which applications were submitted declined rapidly in early 2009.The report cites three chief reasons for this decline: (i) In February 2009, the U.S. Congress adopted more restrictive executive compensation requirements on all TARP recipients. (ii) Many banks felt there was a stigma associated with participation in the program. (iii) The impact of the crisis on DIs started to appear less dramatic. Based on this information we treat the entire population of banks (more than 7,000 institutions) as the pool of applicants, excluding foreign banks that are excluded from the program since its inception. Moreover we conjecture that the vast majority of applications were submitted in the Fall of 2008. Based on this assumption, we proceed to estimate the probability of receiving TARP based on observables measured at the end of the third quarter of 2008. 6.3
An Empirical Model of the Capital Purchase Program
TARP beneficiaries differ from non-TARP banks along many dimensions (Contessi and Francis, 2011). In particular, our data reveal substantial dissimilarities in terms of capital ratios, size and loan composition. However, these difference are not informative 19
about the direction of causality. Using an approach developed in the literature on welfare programs evaluation and labor economics, we address the causality concern by combining a matching estimator, as described in Section 6.1. The first part of our approaches relies on a reduced-form empirical model of CPP participation. We observe banks becoming TARP beneficiaries (a 1-0 outcome), along with a matrix of observable indicators. A natural approach to model these events is a probit model of binary outcome of a bank receiving CPP that uses observable characteristic as explanatory variables. We assume that local economic conditions affect the probability of applying for and being granted TARP, along with key bank-level characteristics. The explanatory variables are measured at the end of the third quarter of 2008, as the application process opened in the fourth quarter of 2008 and according to U.S. Treasury documents, most applications were received by the beginning of 2009.14 As foreign banks were excluded from participating in the program we drop them from the pool of potential applicants. We also drop “new” commercial banks (credit card companies and investment banks) for seasons explained in Section 5. We report our results for the probit estimates in Tables 7. We estimate the probit using three groups of variables, a set of standard financial indicators for banks, geographic variables meant to capture changes in demand, and other variables likely to affect selection in the program. We use the following variables in the preferred specification for the probit model we estimate • Capital Adequacy: Our first measure of capital adequacy is computed as the ratio of total equity divided by total assets. We also consider Tier 1 Capital Ratio in levels and squared • Asset size and composition: We consider the logarithm of total assets, commercial and industrial loans as a share of total assets, cash and Reserves as share of total assets, all “other securities” (quarterly average) as a share of total assets. • For bad loans and reserves as a measure of asset quality, we use are loan loss reserves, the provision of possible loan losses as a share of earning assets, loan losses as a share of equity, net loans charge-offs as a share of total loans.15 14 An alternative route is to estimate the probit based on observables measured at the end of the quarter in which the CPP was granted; however, anecdotal evidence suggests that a large number of applications were submitted in the first few months of the program. 15 The sum of net loans charge-offs and the loan loss provision is defined as gross charge-offs
20
• As for liabilities composition, we use deposits, as a share of total assets, borrowed funds with maturity longer than one year as a share of total assets. • Other variables include a measure of leverage (the ratio between total loans and deposits), and a dummy variable with value one if the bank holding company is publicly traded. • We construct a proxy of management quality using the number of corrective actions taken against bank management by their banking regulator in the 2006-09 period.16 • Earnings are considered as the ratio of pre-tax net income and total earning assets (the sum of total loans and total securities) and the ratio of net income to operating income; In estimating the probit model, we have no clear prior about the sign of the coefficient of the capital adequacy variable (Tier 1 capital ratio). A positive coefficient would suggests that the decision to grant TARP aimed at reinforcing the capital position of healthy banks. A negative coefficient on the contrary would suggest that the support targeted mostly banks that were in relatively weaker position. As the relationship may be non-linear we also introduce a quadratic term. The estimated coefficient for this variable suggest a negative concave relationship that could be interpreted as supportive of the spirit of the TARP legislation. Less capitalized banks, likely above a certain TIER 1 capital level that separates healthy from non-healthy (likely to fail) institutions, are more likely to apply and be granted the capital injection 6.4
Average Treatment on the Treated
The outcome variable that we consider (yi t) is the ratio of cash-to-assets in various quarters between the end of September 2008 (2008Q3) and the end of December 2009 (2008Q10) . We report the results for the ATT estimation in Table 8. We consider two variants of the ATT. Average Treatment Effect on the Treated - 1 is the difference between an outcome variable in a quarter and the previous quarter. The Average Treatment Effect on the Treated - 2 is the difference between an outcome variable in a 16 As in Duchin and Sosyura (2010) - who generously provided the raw data - we have a total of 1,681 orders issued to 961 commercial banks. Enforcement actions include prohibitions from further participation in banking activities, orders to cease and desist, and orders to pay civil money penalties.
21
quarter and the last TARP quarter before the banks was granted TARP. Most banks received tarp in 2008Q4 and and 2009Q4 (see Figure and also (Contessi and Francis, 2011)
7
Conclusion
This paper undertakes a systematic analysis of this massive accumulation of excess reserves using bank-level data for more than 7,000 commercial banks during the U.S. financial crisis that intensified in 2008. We propose a simple stochastic model of reserves determination with interest on reserves. The model handily returns a log-linearized version that can be estimated using bank-level data and censored regressions method. We find no evidence of a precautionary motive for reserves accumulation and discuss alternative explanations for the accumulation of reserves. We then combine propensity score matching and a difference-in-difference approach to determine whether the beneficiaries of the TARP-CPP program accumulated more reserves than non-beneficiaries. We find no significant differences between the two groups.
22
References Bech, Morten, and Elizabeth Klee (2010): “The Mechanics of a Graceful Exit: Interest on Reserves and Segmentation in the Federal Funds Market,” Federal Reserve Board, Finance and Economics Discussion Series, No. 7. Bech, Morten L., and Elizabeth Klee (2009): “The Mechanics of a Graceful Exit: Interest on Reserves and Segmentation in the Federal Funds Market,” Federal Reserve Bank of New York, Staff Report No. 426. Bernanke, Ben (1983): “Nonmonetary Effects of the Financial Crisis in the Propagation of the Great Depression,” American Economic Review, (73). Bernanke, Ben, and Mark Gertler (1990): “Financial Fragility and Economic Performance,” Quarterly Journal of Economics, (105), 87–114. Calomiris, Charles, and Berry Wilson (2004): “Bank Capital and Portfolio Management: The 1930’s Capital Crunch and the Scramble to Shed Risk,” Journal of Business, (105). CNNMoney.com (2010): “The Buzz: The Bankers Who Said ’Hell No’ to Bailouts,” available at . Contessi, Silvio, and Johanna Francis (2011): “TARP Beneficiaries and Their Lending Patterns during the Financial Crisis,” Federal Reserve Bank of St. Louis, March/April,. Dehejia, R. H., and S. Wahba (2002): “Propensity Score Matching Methods for NonExperimental Casual Studies,” Review of Economics and Statistics, 84(1), 151–191. Duchin, Ran, and Denis Sosyura (2010): “TARP Investments: Financials and Politics,” University of Michigan, Ross School of Business manuscript. Freixas, Xavier, and Jean-Charles Rochet (1997): Microeconomics of Banking. The MIT Press. Friedman, Milton., and Anna J. Schwartz (1963): A Monetary History of the United States, 1867. 1960. Princeton, NJ: Princeton University Press.
23
Frost, Peter (1971): “Banks’s Demand for Excess Reserves,” Journal of Political Economy, (79(4)), 805–25. Goodfriend, Marvin (2002): “Interest on Reserves and Monetary Policy,” Federal Reserve Bank of New York Economic Policy Review, (8), 77–84. Hamilton, James D. (1997): “Measuring the Liquidity Effect,” American Economic Review, (87), 80–97. Hanes, Christopher (2006): “The Liquidity Trap and U.S. Interest Rates in the 1930s,” Journal of Money, Credit and Banking, 38(1), 163–194. Hornstein, Andrea (2010): “Monetary Policy with Interest on Reserves,” Federal Reserve Bank of Richmond Economic Quarterly, (96), 153–177. Horwich, George (1963): Effective Reserves, Credit, and Causality in the Banking System of the Thirties. Homewood, IL: Richard D. Irwin. Keister, Todd, and James McAndrews (2009): “Why Are Banks Holding So Many Excess Reserves,” Federal Reserve Bank of New York, Staff Report No. 380. Ogawa, Kazuo (2007): “Why Commercial Banks Held Excess Reserves: The Japanese Experience of the Late 1990s,” Journal of Money, Credit, and Banking, (39), 241– 257. Orr, Daniel, and W. G. Mellon (1961): “Stochastic Reserve Losses and Expansion of Bank Credit,” American Economic Review, (51), 614–23. Poole, William (1968): “Commercial Bank Reserve Management in a Stochastic Model: Implications for Monetary Policy,” Journal of Finance, (23), 769–91. Powell, James L. (1984): “Least Absolute Deviations Estimation for the Censored Regression Model,” Journal of Econometrics, (25), 303–25. Rosenbaum, P., and D. Rubin (1983): “The Central Role of Propensity Score in Observational Studies for Casual Effects,” Biometrika, 70, 41–55. Thornton, Daniel (2001): “The Federal Reserve’s Operating Procedures, Nonborrowed Reserves, Borrowed Reserves and the Liquidity Effect,” Journal of Banking and Finance, (25), 1717–39. 24
Uesugi, Iichiro (2002): “Measuring the Liquidity Effect: The Case of Japan,” Journal of the Japanese and International Economies, (16(2)), 289–316. United States Department of Financial Stability (2010): “Troubled Asset Relief Program: Two Year Retrospective,” . Van Horn, Patrick (2009): “Excess Reserves during the Great Contraction: Evidence from the Central Money Market of New York City, 1929 to 1932,” University of Michigan at Dearborn, manuscript.
25
Figure 1: Federal Reserve Bank Balance Sheet, 2007-2010 Billions $ 2,500
Short-Term Lending to Financial Firms and Markets 2,000
Rescue Operations
1,500
Operations Focused on Longer-Term Credit Conditions Traditional Portfolio
1,000
500
ASSETS 0
-500
LIABILITIES
-1,000 Fed Reserve Notes -1,500
Deposits: Depository Institutions Treasury Financing Account
-2,000
Traditional Liabilities and Capital Account
-2,500 2007
2008
2009
Source: Federal Reserve Board.
26
2010
Figure 2: Cross-Section Distribution of Excess Reserves Ratio for Commercial Banks whose Ratio is lower than 70, 2008Q2-2010Q2 2008Q3 700 600 500 400 300 200 100 0
0 15 30 45 60 >=75 Excess Reserves/Required Reserves
Frequency
Frequency
Frequency
0 15 30 45 60 >=75 Excess Reserves/Required Reserves
2010Q1 700 600 500 400 300 200 100 0
2010Q2 Frequency
Frequency
Frequency
2009Q3 700 600 500 400 300 200 100 0
0 15 30 45 60 >=75 Excess Reserves/Required Reserves
2009Q4
0 15 30 45 60 >=75 Excess Reserves/Required Reserves
0 15 30 45 60 >=75 Excess Reserves/Required Reserves
2009Q2 700 600 500 400 300 200 100 0
0 15 30 45 60 >=75 Excess Reserves/Required Reserves
700 600 500 400 300 200 100 0
700 600 500 400 300 200 100 0
0 15 30 45 60 >=75 Excess Reserves/Required Reserves
2009Q1 700 600 500 400 300 200 100 0
2008Q4 Frequency
Frequency
Frequency
2008Q2 700 600 500 400 300 200 100 0
0 15 30 45 60 >=75 Excess Reserves/Required Reserves
700 600 500 400 300 200 100 0 0 15 30 45 60 >=75 Excess Reserves/Required Reserves
Source: Authors’ calculations based on CR.
27
Figure 3: Ratio of Excess-to-Required Reserves in Japan and the U.S., 1998:M1-2010:M12
20 18 16
United States
14 12
Japan
10 8 6 4 2 1998 1998 1999 1999 2000 2000 2001 2001 2002 2002 2003 2003 2004 2004 2005 2005 2006 2006 2007 2007 2008 2008 2009 2009 2010 2010 2011
0
Source: Bank of Japan, Federal Reserve Board
28
Figure 4: Call Rate and Excess Reserves Ratio in Japan, 1998:M1-2010:M12
ERR
Call rate (average)
Series3 1
7
0.9
6
0.8 5
0.7 0.6
4
0.5 3
0.4
2
0.3 03 0.2
1
0.1 0 1998 1998 1998 1999 1999 2000 2000 2000 2001 2001 2002 2002 2003 2003 2003 2004 2004 2005 2005 2005 2006 2006 2007 2007 2008 2008
0
Source: Bank of Japan.
29
Figure 5: Federal Funds Rates, Interest on reserves, Primary Credit Rate, and adjusted Differential between 1 Year Treasury Bonds and Interest on Reserves, 2008:M7-2010:M12
Penalty Rate
IOR
eFFR
(1y Treas ‐ IOR)/Penalty rate
2.50 2.00 1.50 1.00 0.50 0 00 0.00
Source: Fred II.
30
2010‐09‐10
2010‐08‐10
2010‐07‐10
2010‐06‐10
2010‐05‐10
2010‐04‐10
2010‐03‐10
2010‐02‐10
2010‐01‐10
2009‐12‐10
2009‐11‐10
2009‐10‐10
2009‐09‐10
2009‐08‐10
2009‐07‐10
2009‐06‐10
2009‐05‐10
2009‐04‐10
2009‐03‐10
2009‐02‐10
2009‐01‐10
2008‐12‐10
2008‐11‐10
2008‐10‐10
2008‐09‐10
‐0.50
Figure 6: Cross-Section Distribution of Excess Reserves Ratio for CPP beneficiaries and nonbeneficiaries for Commercial Banks, 2008Q2 through 2010Q2)
30
45
60
>=75
.2 .15
Non-CPP
0
0 15
CPP
.1
Density
Non-CPP
.05
.15 .1
Density
CPP
.05
.15 .1 .05
Non-CPP
0
0
15
30
45
60
>=75
0
15
30
45
60
>=75
2009 Q1
2009 Q2
2009 Q3
0
15
30
45
60
>=75
.15
CPP
.1
Density
Non-CPP
Non-CPP
0
.05
.15 .1
Density
CPP
0
0
.05
Non-CPP
.05
.15 .1
CPP
.2
Excess Reserves/Required Reserves
.2
Excess Reserves/Required Reserves
.2
Excess Reserves/Required Reserves
0
15
30
45
60
>=75
0
15
30
45
60
>=75
Excess Reserves/Required Reserves
2009 Q4
2010 Q1
2010 Q2
0
15
30
45
60
>=75
Excess Reserves/Required Reserves
.15
CPP
.1
Density
Non-CPP
Non-CPP
0
.05
.2 .15
CPP
0
0
Non-CPP
.1
Density
Density .05 .1 .15
CPP
.2
Excess Reserves/Required Reserves
.2
Excess Reserves/Required Reserves
.05
Density
CPP
0
Density
2008 Q4
.2
2008 Q3
.2
2008 Q2
0
15
30
45
60
>=75
Excess Reserves/Required Reserves
0
15
30
45
60
>=75
Excess Reserves/Required Reserves
Source: Authors’ calculations based on CR.
31
Figure 7: CPP disbursements and repayments, October 2008-December 2010
Source: Authors calculations based on Treasury data.
32
Figure 8: Cross-Section Distribution of Excess Reserves Ratio for CPP beneficiaries and nonbeneficiaries for large and small commercial Banks, 2008Q2 through 2010Q2)
30
45
60
>=75
.15
.2 0
15
30
45
60
>=75
0
15
30
45
60
>=75
2009 Q1
2009 Q2
2009 Q3
15
30
45
60
>=75
.15
Large
.1
Density
Small
Small
0
.05
.15 .1
Density
Large
0
0
.05
Small
.05
.15 .1
Large
.2
Excess Reserves/Required Reserves
.2
Excess Reserves/Required Reserves
.2
Excess Reserves/Required Reserves
0
0
15
30
45
60
>=75
0
15
30
45
60
>=75
Excess Reserves/Required Reserves
Excess Reserves/Required Reserves
2009 Q4
2010 Q1
2010 Q2
Small
15
30
45
60
>=75
Excess Reserves/Required Reserves
.15
Small
0
0 0
Large .1
Density
.15
Large
.05
.05
Small
.05
.1
Large
.1
Density
.15
.2
.2
.2
Excess Reserves/Required Reserves
0
Density
Small
0
0 15
Large
.1
Density
Small
.05
.15 .1
Density
Large
.05
.15 .1 .05
Small
0
Density
Large
0
Density
2008 Q4
.2
2008 Q3
.2
2008 Q2
0
15
30
45
60
>=75
Excess Reserves/Required Reserves
0
15
30
45
60
>=75
Excess Reserves/Required Reserves
Source: Authors’ calculations based on CR.
33
Figure 9: CPP disbursements and repayments, October 2008-December 2010
Source: Authors calculations based on Treasury data.
34
Table 1: U.S. Reserves Requirements Liability Type Net Transaction Accounts
Percent of Liabilities
Effective Date
$0 to $9.3 million $0 to $10.3 million $0 to $10.7 million
0 0
12/20/2007 1/1/2009 12/31/2009
$9.3 million to $43.9 million $10.3 million to $44.4 million $10.7 million to $55.2 million
3 3 3
12/20/2007 1/1/2009 12/31/2009
10
12/20/2007 1/1/2009 12/31/2009
Non-personal time deposits
0
12/27/1990
Eurocurrency liabilities
0
12/27/1990
More than $43.9 million More than $44.4 million More than $55.2 million
Source: Board of Governors of the Federal Reserves.
35
Table 2: OLS Regressions for Excess Reserves (log), All Banks and Thrifts All Banks log Dep r1yr IOR penalty log Tier1KR log badloans 1
(1)
(2)
(3)
(4)
(5)
(6)
0.815*** (0.00997) -108.9*** (3.662) 56.42*** (2.675) 0.250*** (0.0152) 0.0446*** (0.00545)
0.801*** (0.0101) -106.3*** (3.582) 54.41*** (2.618) 0.272*** (0.0152)
0.796*** (0.0115) -109.4*** (3.499) 55.82*** (2.564) 0.295*** (0.0148)
0.792*** (0.0112) -114.3*** (3.767) 59.79*** (2.759)
0.792*** (0.0104) -110.3*** (3.583) 56.88*** (2.622)
0.787*** (0.0119) -112.7*** (3.512) 57.86*** (2.576)
log badloans 2
0.0308*** (0.00526) 0.0439*** (0.00567)
log badloans 3
0.0263*** (0.00548)
-0.827*** (0.0947)
-0.596*** (0.0936)
0.0438*** (0.00747) -0.516*** (0.0923)
Observations R-squared
50,662 0.522
53,730 0.514
56,122 0.509
50,753 0.498
53,790 0.510
56,182 0.503
All Thrifts
(1)
(2)
(3)
(4)
(5)
(6)
0.753*** (0.0214) -86.78*** (12.81) 42.41*** (9.294) 0.430*** (0.0366) 0.0426*** (0.0144)
0.767*** (0.0185) -93.86*** (12.54) 47.66*** (9.120) 0.366*** (0.0341)
0.796*** (0.0197) -92.24*** (12.19) 45.91*** (8.878) 0.337*** (0.0335)
0.747*** (0.0225) -93.05*** (13.02) 46.12*** (9.447)
0.762*** (0.0194) -98.44*** (12.67) 50.24*** (9.216)
0.795*** (0.0206) -94.83*** (12.31) 47.23*** (8.963)
Constant
log Dep r1yr IOR penalty log Tier1KR log badloans 1 log badloans 2
Observations R-squared
-0.908*** (0.0911)
0.0115 (0.0140) 0.0185 (0.0132)
log badloans 3 Constant
-0.971*** (0.105)
0.0211*** (0.00728) -0.817*** (0.0907)
-0.0128 (0.0127)
0.485** (0.198)
0.351** (0.174)
-0.0139 (0.0171) 0.228 (0.163)
5,227 0.507
5,700 0.490
5,903 0.493
0.0246 (0.202)
0.0125 (0.179)
-0.0527*** (0.0161) -0.0248 (0.167)
5,242 0.492
5,717 0.480
5,920 0.484
Notes: Standard errors in parentheses. *** p<0.01, ** p<0.05, * p<0.1 .
36
Table 3: OLS Regressions for Excess Reserves (log), Banks by Size Large Banks log Dep r1yr IOR penalty log Tier1KR log badloans 1
(1)
(2)
(3)
(4)
(5)
(6)
0.313*** (0.0501) -173.3*** (47.71) 110.0*** (35.64) 0.439* (0.255) 0.280*** (0.0654)
0.290*** (0.0457) -159.1*** (46.26) 106.3*** (34.51) 0.411* (0.247)
0.269*** (0.0448) -176.1*** (45.58) 113.8*** (34.22) 0.381 (0.245)
0.293*** (0.0500) -196.4*** (45.09) 123.9*** (34.25)
0.275*** (0.0451) -180.5*** (44.31) 119.1*** (33.41)
0.254*** (0.0436) -195.6*** (43.77) 125.5*** (33.23)
log badloans 2
0.272*** (0.0654) 0.431*** (0.0696)
log badloans 3 Constant
Observations R-squared
0.425*** (0.0696)
4.793*** (0.863)
3.153*** (0.856)
0.538*** (0.0820) 1.858** (0.944)
4.189*** (0.784)
2.511*** (0.732)
0.534*** (0.0818) 1.262 (0.824)
1,021 0.123
1,054 0.167
1,055 0.187
1,022 0.120
1,055 0.164
1,056 0.185
0.837*** (0.00590) -109.8*** (3.558) 56.64*** (2.600) 0.221*** (0.0143) 0.0199*** (0.00319)
0.824*** (0.00658) -106.9*** (3.493) 54.36*** (2.551) 0.242*** (0.0146)
0.826*** (0.00708) -108.7*** (3.430) 55.03*** (2.507) 0.260*** (0.0143)
0.829*** (0.00595) -113.3*** (3.558) 58.85*** (2.602)
0.812*** (0.00686) -110.3*** (3.496) 56.49*** (2.556)
0.814*** (0.00742) -111.6*** (3.441) 56.81*** (2.517)
Small Banks log Dep r1yr IOR penalty log Tier1KR log badloans 1 log badloans 2
0.00602* (0.00310) 0.0146*** (0.00364)
log badloans 3 Constant
Observations R-squared
-0.00143 (0.00346)
-0.979*** (0.0613)
-0.724*** (0.0676)
0.00443 (0.00460) -0.634*** (0.0649)
48,888 0.471
51,899 0.464
54,128 0.459
-1.226*** (0.0593)
-0.956*** (0.0656)
-0.0161*** (0.00440) -0.851*** (0.0638)
48,946 0.468
51,957 0.459
54,186 0.453
Notes: Standard errors in parentheses. *** p<0.01, ** p<0.05, * p<0.1 .
37
Table 4: Tobit Regressions for Excess Reserves (log), All Banks and Thrifts All Banks log Dep r1yr IOR penalty log Tier1KR log badloans 1
(1)
(2)
(3)
(4)
(5)
(6)
0.815*** (0.00461) -109.0*** (3.600) 56.45*** (2.614) 0.250*** (0.0119) 0.0446*** (0.00275)
0.801*** (0.00431) -106.3*** (3.527) 54.45*** (2.568) 0.272*** (0.0114)
0.796*** (0.00418) -109.4*** (3.469) 55.85*** (2.525) 0.295*** (0.0110)
0.793*** (0.00432) -114.4*** (3.546) 59.85*** (2.583)
0.792*** (0.00433) -110.3*** (3.539) 56.91*** (2.578)
0.787*** (0.00420) -112.7*** (3.487) 57.89*** (2.540)
log badloans 2
0.0304*** (0.00264) 0.0438*** (0.00264)
log badloans 3 Constant
Observations Sigma
0.0262*** (0.00255)
-0.828*** (0.0487)
-0.597*** (0.0454)
0.0438*** (0.00276) -0.517*** (0.0423)
-0.978*** (0.0427)
-0.909*** (0.0433)
0.0211*** (0.00268) -0.817*** (0.0409)
50,662
53,730
56,122
50,753
53,790
56,182
1.011*** (0.00140)
1.022*** (0.00140)
1.026*** (0.00140)
1.049*** (0.00149)
1.027*** (0.00141)
1.032*** (0.00141)
0.753*** (0.0153) -86.81*** (12.46) 42.40*** (9.157) 0.430*** (0.0316) 0.0427*** (0.0109)
0.767*** (0.0154) -93.95*** (12.23) 47.69*** (8.971) 0.366*** (0.0305)
0.796*** (0.0161) -92.33*** (11.97) 45.93*** (8.787) 0.337*** (0.0296)
0.796*** (0.0161) -92.33*** (11.97) 45.93*** (8.787) 0.337*** (0.0296)
0.762*** (0.0163) -98.52*** (12.71) 50.26*** (9.292)
0.795*** (0.0163) -94.92*** (12.10) 47.26*** (8.885)
All Thrifts log Dep r1yr IOR penalty log Tier1KR log badloans 1 log badloans 2
0.0185 (0.0113)
log badloans 3 Constant
Observations
Sigma
-0.0128 (0.0113)
0.486*** (0.148)
0.352** (0.142)
-0.0141 (0.0136) 0.228* (0.133)
-0.0141 (0.0136) 0.228* (0.133)
0.0132 (0.148)
-0.0529*** (0.0131) -0.0245 (0.132)
5,227
5,700
5,903
5,903
5,717
5,920
1.131*** (0.00708)
1.154*** (0.00722)
1.148*** (0.00709)
1.148*** (0.00709)
1.171*** (0.0110)
1.163*** (0.00729)
Notes: Standard errors in parentheses. *** p<0.01, ** p<0.05, * p<0.1 .
38
Table 5: Tobit Regressions for Excess Reserves (log), Banks by Size Large Banks log Dep r1yr IOR penalty log Tier1KR log badloans 1
(1)
(2)
(3)
(4)
(5)
(6)
0.312*** (0.0428) -176.2*** (59.98) 112.1*** (43.38) 0.436** (0.209) 0.280*** (0.0499)
0.288*** (0.0398) -161.7*** (59.48) 108.3** (43.05) 0.406* (0.208)
0.266*** (0.0397) -178.8*** (59.23) 115.7*** (42.95) 0.375* (0.207)
0.293*** (0.0418) -199.2*** (58.98) 125.9*** (42.89)
0.273*** (0.0391) -182.9*** (58.52) 121.0*** (42.58)
0.252*** (0.0390) -198.0*** (58.29) 127.3*** (42.49)
log badloans 2
0.271*** (0.0497) 0.433*** (0.0516)
log badloans 3 Constant
Observations Sigma
0.428*** (0.0515)
4.783*** (0.820)
3.120*** (0.808)
0.544*** (0.0538) 1.793** (0.843)
4.183*** (0.770)
2.486*** (0.741)
0.540*** (0.0537) 1.205 (0.779)
1,021
1,054
1,055
1,022
1,055
1,056
2.248*** (0.0508)
2.273*** (0.0506)
2.274*** (0.0506)
2.252*** (0.0508)
2.276*** (0.0506)
2.277*** (0.0506)
0.838*** (0.00535) -109.8*** (3.521) 56.66*** (2.563) 0.221*** (0.0119) 0.0199*** (0.00294)
0.824*** (0.00530) -106.9*** (3.458) 54.38*** (2.521) 0.242*** (0.0114)
0.826*** (0.00559) -108.7*** (3.406) 55.04*** (2.485) 0.260*** (0.0110)
0.829*** (0.00534) -113.3*** (3.527) 58.86*** (2.570)
0.812*** (0.00529) -110.3*** (3.468) 56.51*** (2.529)
0.814*** (0.00559) -111.6*** (3.420) 56.83*** (2.496)
Small Banks log Dep r1yr IOR penalty log Tier1KR log badloans 1 log badloans 2
0.00599** (0.00285) 0.0145*** (0.00302)
log badloans 3 Constant
Observations Sigma
-0.00151 (0.00294)
-0.980*** (0.0546)
-0.726*** (0.0522)
0.00433 (0.00363) -0.636*** (0.0503)
-1.227*** (0.0531)
-0.958*** (0.0512)
-0.0162*** (0.00353) -0.853*** (0.0496)
48,888
51,899
54,128
48,946
51,957
54,186
0.957*** (0.00166)
0.967*** (0.00180)
0.972*** (0.00181)
0.961*** (0.00171)
0.971*** (0.00182)
0.977*** (0.00184)
Notes: Standard errors in parentheses. *** p<0.01, ** p<0.05, * p<0.1 .
39
Table 6: CLAD Regressions for Excess Reserves (log), All Banks and Thrifts All Banks log Dep r1yr IOR penalty log Tier1KR log badloans 1
(1)
(2)
(3)
(4)
(5)
(6)
0.838*** (0.00658) -142.8*** (4.661) 77.47*** (3.413) 0.272*** (0.0155) 0.0336*** (0.00395)
0.838*** (0.00551) -140.5*** (3.930) 76.52*** (2.876) 0.280*** (0.0129)
0.838*** (0.00613) -136.7*** (4.096) 72.66*** (3.002) 0.324*** (0.0132)
0.837*** (0.00647) -137.4*** (4.660) 73.19*** (3.414)
0.836*** (0.00603) -139.4*** (4.254) 74.49*** (3.117)
0.843*** (0.00601) -141.1*** (4.042) 75.47*** (2.959)
log badloans 2
0.0146*** (0.00380) 0.0208*** (0.00341)
log badloans 3 Constant
Observations
-0.000188 (0.00359)
-1.022*** (0.0666)
-0.899*** (0.0544)
0.0173*** (0.00424) -0.792*** (0.0549)
-1.396*** (0.0630)
-1.287*** (0.0573)
-0.0116*** (0.00404) -1.272*** (0.0525)
50,833
53,859
56,361
50,749
53,719
56,171
0.793*** (0.0179) -110.5*** (14.21) 56.82*** (10.40) 0.486*** (0.0347) 0.0200 (0.0125)
0.754*** (0.0164) -100.1*** (12.81) 48.55*** (9.359) 0.483*** (0.0310)
0.747*** (0.0224) -123.5*** (16.30) 68.62*** (11.92) 0.403*** (0.0404)
0.721*** (0.0175) -99.17*** (14.61) 46.03*** (10.70)
0.785*** (0.0210) -113.4*** (15.92) 55.60*** (11.62)
0.790*** (0.0156) -80.90*** (11.38) 36.16*** (8.305)
All Thrifts log Dep r1yr IOR penalty log Tier1KR log badloans 1 log badloans 2
0.0170 (0.0121) 0.0548*** (0.0116)
log badloans 3 Constant
Observations
-0.0199 (0.0141)
0.328* (0.172)
0.555*** (0.153)
0.0356* (0.0188) 0.520*** (0.183)
5,220
5,704
5,796
0.396** (0.165)
-0.0960 (0.192)
-0.0530*** (0.0123) 0.197 (0.125)
5,252
5,802
5,989
Notes: Standard errors in parentheses. *** p<0.01, ** p<0.05, * p<0.1 .
40
Table 7: Probit Model for CPP injection Specification
Estimates
Std.
2.7411 -6.1895 2.2497 24.0359 -14.8823 0.2105 -9.7692 1.9762 0.6200 -2.0678 -49.8456 -47.4067 34.8196 -0.0854 -12.1189 -1.8522 0.6227
1.5022 1.6079 0.3844 8.1263 2.7643 0.0264 1.3054 0.3706 0.1501 0.9176 17.9284 22.8579 16.9477 0.0357 4.9688 0.6696 0.8486
1.0051 0.0106
0.0805 0.0072
Business Bankruptcy Filings y-y Conventional Mortgage Home Price Index y-y Delinquencies of 1-to-4 Unit Residential Mortgage Loans y-y
0.0023 -0.0464 0.0084
0.0009 0.0165 0.0040
ints55y5 Constant
0.0946 -3.4877
0.0265 0.4306
Total Equity/TA Total Noncurrent Loans/TL Commercial Loans/TA Loan Loss Reserves/TL Pretax Net Income/TA Log of TA Tier 1 Capital Ratio Tier 1 Capital Ratio Squared Loans/Deposits Loan Losses /Equity Provision for Possible Loan Loss/Earning Assets Net Loan Charge-offs /TL Gross Charge-offs /TL Net Income/Operating Income–totals Net Interest Income/Earning Assets All other securities (Q average)/TA Cash and reserves/TA Publicly Traded Management Penalties
PsedoR2 Log-Likelihood
0.3122 -1421.3443
Table 8: Difference-in-Difference Results: Cash-to-Assets Pre-TARP
TARP
TARP+1
TARP+2
TARP+3
TARP+4
4.68 3.29
4.45 6.59
7.04 7.68
8.87 7.5
7.95 9.79
9.05 8.37
0.75 0.75
-1.74 -1
-1.91 -2.91
3.76 0.85
-1.87 -1.02
618
597
567
482
170
TARP group Non-TARP group Average Treatment Effect on the Treated - 1 Average Treatment Effect on the Treated - 2 No of matched pairs
627
The Average Treatment Effect on the Treated - 1 is the difference between an outcome variable in a quarter and the previous quarter; the Average Treatment Effect on the Treated - 2 is the difference between an outcome variable in a quarter and the last TARP quarter before the banks was granted TARP.
41
43
svgl2170 Total assets (SC60) Sum of svgl3885, svgl 3909, svgla650, and svgla674 - Total
rcfd2170 Total assets
riad4635 Charge-Offs on Allowance for Loan and Lease Losses
svgl0484 Provision for loan and lease losses (SO321) svgl3936 Past Due 30-89 Days and still accruing, total ( )
svgl3942 Past Due 90 Days or more and still accruing, total ( ) svgl3948 Non-accrual, Total ()
Sum of svgl2339, svgl 2728, and svgl2071 - Deposits (Sum of SC710, SC715, and SC712) svcc5279 Tier 1 Capital (CCR20) svcc2375 Net Risk-Weighted Assets (CCR78) svcc7205 Risk-based Capital Ratio
riad4230 Loan Loss Provision
rcfd1406 Total Loans and Lease financing receivables: Past Due 30-89 Days and Still Accruing
rcfd1407 Total Loans and Lease financing receivables: Past Due 90 Days and Still Accruing
scfd1403 Total Loans and Leases Finance Receivables: Nonaccrual
rcfd2200 Total deposits
rcfd8274 Tier 1 capital
rcfd a223 Risk-adjusted assets (Net Risk-weighted Assets)
rcfd 7205 Risk-based Capital Ratio
Loans Charge-offs (Sum of VA46, VA56, VA48, and VA58)]
Thrift Financial Reports
Call Reports
Table 9: Variables List and Correspondence from the CRs and TFRs
