«Research Division Federal Reserve Bank of St. Louis Working Paper Series Understanding the Accumulation of Bank and Thrift Reserves during the U.S. ...»
small banks. Diﬀerences in the regulatory framework for thrifts and banks may explain these observed diﬀerences. For instance, on average, thrifts have a larger portion of their loan portfolios invested in residential loans and small bank lending. Thrifts also have greater restrictions on other assets and securities they can hold. Their loan portfolios are also subject to much stricter controls, in terms of capital adequacy and the percentage of assets they can leverage, than are banks’ portfolios. This may make thrifts especially sensitive to the penalty rate, but their loan portfolios may be better managed.
We also performed a number of robustness checks. We considered whether a censored regression technique was appropriate, whether there were selection concerns, and we tested diﬀerent speciﬁcations for the dependent variable, the penalty rate, the opportunity cost of holding cash and ER, and the measure of capital adequacy. We found that our results were qualitatively and quantitatively robust under these alternative speciﬁcations. As a last robustness check, we considered whether we ought to consider reserves held at the bank holding company level rather than the individual bank level.
We found our results to be quantitatively similar using this higher level of aggregation. A more detailed discussion of these robustness checks is available in an online appendix.29
4.3 Responsiveness of cash and reserves accumulation
Table 7 analyzes the responsiveness of ER and cash accumulation to changes in the penalty rate, opportunity cost, and distressed loans. We ﬁrst look at the response of ER and cash to unit variations in these covariates for all banks, large banks (top 2 percent by assets), small banks (bottom 95 percent), and thrifts.
We ﬁnd that ER and cash do not respond very strongly (less than unit elasticity) to the opportunity cost (yield on 1-year Treasury bills minus IOR) or the average interest rate on Treasury bill repurchase agreements, but there are some diﬀerences between banks by size. The elasticity of the opportunity cost and repo rate for large banks is signiﬁcantly higher than for small banks (though still below unit elasticity).
The elasticity of distressed loans is signiﬁcantly higher for large banks than for small banks—in fact, the elasticity on the broadest category of distressed loans is above unity and positive, meaning that a 1-unit increase in distressed loans generates more than a 1 percent increase in ER and cash holdings. This suggests that cash and ER 29 Detailed discussion of our robustness tests is available here in an online appendix available here
accumulation by large banks was strongly inﬂuenced by precautionary motives, much more so than small bank accumulation. The elasticity of penalty rates and opportunity costs for thrifts is similar to that for large banks, but the elasticity of distressed loans is more similar to that of small banks: very small and not signiﬁcant.
4.4 Response of reserve and cash accumulation to uncertainty
We consider two measures of macroeconomic uncertainty. One is based on the CBOE VIX, which is an estimate of market expectations of short-term volatility based on movements in the S&P 500 stock index option prices (see Goldberg and Grisse, 2013), for a discussion of the relationship between the VIX and asset prices). The other is based on a measure of macroeconomic uncertainty derived from movements in industrial production.
Our second measure of uncertainty is calculated based on a technique outlined in Baum, Caglayan, and Ozkan (2009) of ﬁtting a generalized autoregressive conditionally heteroskedastic (GARCH) model to the monthly industrial production series and using the conditional variance derived from the model as a measure of uncertainty (they also ﬁt a GARCH model to the consumer inﬂation series, but we focus only on industrial production). They ﬁnd that macroeconomic uncertainty generates a misallocation of banks’ loanable funds. Assuming that a similar mechanism might aﬀect ER and cash holdings (which is part of the portfolio allocation problem), we believed this would be a good measure of uncertainty relevant for bank behavior.
We ﬁnd no eﬀects of these measures when we combine all banks, but separating them into large and small banks, we realize the reason for this. Table 8 includes the results for the eﬀect of the VIX level, changes in the VIX, and the conditional variance of industrial production for large and small banks. We report the coeﬃcients on only our three measures of uncertainty, but the Tobit regressions include the covariates from our previous regressions, focusing on the bad loans 1 case (the signs and signiﬁcance of the other coeﬃcients remain the same as previously reported). The dependent variable is the log of the ratio of ER (and cash) ratio to deposits. For large banks, a 1-unit increase in the VIX results in a 0.1 percent decrease in ER and cash holdings; a similar eﬀect occurs for the growth in the VIX. For small banks, there is no eﬀect. This is not necessarily the result we expected: We assumed increased macroeconomic uncertainty would cause all banks to increase their cash and reserve holdings.
Turning to the conditional variance of industrial production, we ﬁnd that for large banks, a 1-unit increase in this measure results in a 0.8 percent increase in the ER and cash-to-deposit ratio. For small banks, a 1-unit increase is consistent with a 0.03 percent decrease in the ratio of ER and cash-to-deposits ratio. These results are closer to our prior assumptions. We also considered the St. Louis Fed Financial Stress Index and the impact on reserve and cash accumulation of movements in the spreads between high-yield and risk-free bonds and found no eﬀect.
Although these results are suggestive of the mechanism through which heightened macroeconomic uncertainty could impact banks’ precautionary behavior, they are to some extent inconsistent with the model we developed. Thus we are cautious about their interpretation. The only role for uncertainty in our theoretical model is through an increase in payment volatility, captured by the θ parameter. We expect that an increase in the VIX or its growth rate (or other measures of uncertainty) would aﬀect reserves and cash accumulation only through its eﬀect (if any) on the volatility of payments (or withdrawals), not as an independent additive eﬀect. The primary eﬀect of macroeconomic uncertainty for banks is likely to be associated with new lending or the allocation of liquidity across diﬀerent assets. If a bank has a good portfolio of loans and is adequately capitalized, it is unclear how an increase in macroeconomic uncertainty would independently aﬀect reserves and cash accumulation, setting aside the risks associated with securities holdings. Our results also point to the fact that banks were primarily concerned about their own balance sheets and managing their own liquidity risks and not as concerned about counterparty risk, a conclusion that Acharya and Merrouche (2013) also reach.
One aspect of uncertainty we have not explored is the eﬀect of regulatory uncertainty. The ﬁnancial crisis generated many proposals for new regulations on both banking activity and sources of bank funding, such as overdraft fees. In addition, DIs may also have been concerned about future litigation. This type of heightened regulatory and litigation uncertainty could generate excessive cash and reserves accumulation.
This is an avenue we intend to explore in future work.
5. Did TARP beneﬁciaries accumulate more ER?
In this section, we discuss whether the CPP program under the TARP umbrella induced banks that were beneﬁciaries of the program to overaccumulate cash and reserves. We ﬁrst describe our data and empirical methodology and then the results of our analysis.
To allow for this comparison, we attempt to identify systematic diﬀerences between these two groups. We ﬁrst describe the notable features of CPP beneﬁciaries using non-CPP DIs for comparison. We group institutions using information on the TARP funds distribution from the TARP Transaction Reports that were updated weekly by the U.S. Treasury after the program’s inception in October 2008.30 Fig. 10 plots the patterns of monthly disbursements and repayments derived from these data using the TARP Transaction Reports releases.31 The ﬁgure shows the total number of beneﬁciaries by month (vertical bars), the total disbursement (open circles), and the monthly disbursement net of repayments (solid circles). Over its ﬁrst 15 months of life, the CPP allowed the injection of almost $205 billion of capital into approximately 730 ﬁnancial entities (Department of the U.S. Treasury, 2009).32 As of December 31, 2009, 71 institutions had redeemed their preferred stocks and about $83 billion remained invested in the remaining beneﬁciaries. It should be noted that the observational units in the Transaction Reports are ﬁnancial holdings (as detailed in footnote 30) and not individual banks or thrifts per se. The institutions that received funding under the program could allocate the funds to any of the institutions (e.g., banks and thrifts) they control. Therefore, in the remainder of the analysis we reaggregate individual DIs that have a charter (and an entity number in the CRs and TFRs) into a consolidated entity. In our dataset, 28 CPP “multi-unit” beneﬁciaries control 110 banks and thrifts.
We match the Treasury data on the CPP disbursements with the unbalanced panel created from the CRs. With few exceptions, most capital injections were granted to BHCs, not to banks. In the case of a single-bank BHC, we attribute the capital injection to the bank that maintains its CR identiﬁer. In the case of a multi-unit BHC, 30 The allocation of CPP funds to BHCs, instead of individual banks and thrifts, has raised some criticism (Coates and Scharfstein, 2009) in terms of whether it promotes more lending at the bank level. It also creates various issues in our dataset because, unlike the TFRs and the CRs, which provide us with ﬁnancial information, the TARP Transaction Reports list the BHCs. Therefore, we organized the data as follows. First, we determined the entity identiﬁcation numbers for all DIs listed to make the TARP information compatible with our CR and TFR information. By using the Competitive Analysis and Structure Source Instrument for Depository Institutions (CASSIDI) database managed by the Federal Reserve Bank of St. Louis and the Federal Financial Institutions Examination Council’s institutional history database we determined the set of institutions each BHC controls, BHC by BHC. We organized our data into four categories. (i) If the BHC controls only a single bank or thrift, we match the TARP Transactions Report information with the single bank or thrift’s Federal Reserve entity identiﬁcation number. (ii) When the BHC controls several diﬀerent banks or a mix of banks and thrifts, all of the loans (and other ﬁnancial information) at the individual bank and thrift level are totaled and the group is given the BHC’s entity identiﬁcation number. (iii) Additionally, we diﬀerentiated between the funds distributed to large lenders and other beneﬁciaries that are either non-ﬁnancial institutions (namely, General Motors and Chrysler) or (iv) new commercial banks and thrifts.
31 See the relevant ﬁles on the Financial Stability website. The Congressional Oversight Panel (2009) reported some diﬃculties in conﬁrming the exact value of the Treasury disbursements using these ﬁgures.
32 The latest available TARP Transaction Report was accessed on January 31, 2010, and contains information for the period ending January 13, 2010. See http://www.ﬁnancialstability.gov/latest/reportsanddocs.html for details.
it is impossible to determine the ultimate beneﬁciary of the CPP, so we retain the BHC identiﬁer. We sum the relevant CR variables for all subsidiaries that belong to the BHC group that received the CPP funds 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.33 After creating appropriate banking groups for multi-unit banks, we matched the CPP information collected from the TARP Transaction Reports using either the CR identiﬁers or the BHC identiﬁers. Our TARP/CPP information includes the amount of the CPP, the date of the CPP announcement, dummy variables and dates for double payments, repayments, and the number of banks and thrifts in the multi-unit BHCs.
Tables 9 (all DIs) and 10 (banks and thrifts separately) compare relevant variables and ratios across institutions that received CPP funding (ﬁrst column) and those that did not (second column), as well as the entire population of DIs (third column). Summary statistics are calculated before the regrouping of multi-unit DIs, which leaves 614 banks and 54 thrifts for a total of 668 CPP beneﬁciaries. The number of observations is reported in the tables. All variables for banks and thrifts are comparable except for cash.
The comparison between CPP and non-CPP DIs shows that CPP beneﬁciaries are larger than non-beneﬁciaries in terms of total loans and total assets (on average about 20 times larger, but this is skewed by the fact that the largest DIs, e.g., Citibank, JP Morgan Chase, and Bank of America, received CPP support). CPP DIs extend a slightly larger share of real estate and commercial and industrial loans and have slightly larger leverage and lower deposits-to-assets ratios. These diﬀerences characterize both the thrifts and the banks that received CPP funds.
5.2 Estimation strategy
To evaluate the impact of the CPP, ideally we would like to compare the performance of a BHC that receives a capital injection with its performance had it not received support. Although this counterfactual is not available, performance comparisons between the beneﬁciaries and the non-beneﬁciaries can be made provided we can minimize the econometric problems that arise from such a comparison. The main econometric concern is the sample selection problem—namely, the BHCs receiving CPP funds are not 33 The largest imbalance found was a three-unit BHC in which a thrift held about 5 percent of the total group deposit.