1 Introduction

The 2008 U.S. financial crisis centered on the short-term debt markets. In response, the Federal Reserve (Fed) developed a number of programs for banks to address the crises in the different short-term debt markets. We examine one of those programs: the term auction facility (TAF) which was purported to act as a substitute for the discount window to provide liquidity to banks, to determine whether there are borrowing differences during the crisis between large banks, other publicly traded banks, and privately held banks.

The Federal Reserve operates the discount window as the lender of last resort for depository institutions with liquidity problems. This facility exists to address liquidity issues that can result from crises in short-term debt markets. Thus, even before the programs created during the 2008 financial crisis, an emergency management program existed for banks with immediate liquidity problems. Still, the Fed chose to create a new liquidity program for banks; the term auction facility as a substitute for the discount window because of concerns over a potential stigma associated with accessing the discount window (DW). Bernanke (2009), then Chairman of the Board of Governors of the Federal Reserve System, argued “(T)he perceived stigma of borrowing at the discount window threatened to prevent the Federal Reserve from getting much-needed liquidity into the system. To address this issue, in late 2007, the Federal Reserve established the term auction facility.” We hypothesize that, if TAF is not an effective reputation-free substitute for the discount window, then privately held banks should borrow relatively more than publicly traded banks since the publicly traded banks are subject to more market scrutiny.Footnote 1

Research on TAF is developing along two lines: (1) the market’s reaction to TAF borrowing and (2) TAF borrowings as a substitute for other sources of funds. Cyree et al. (2013) find that the market viewed TAF borrowing negatively for all banks, but particularly negative for investment banks, reinforcing the argument that publicly traded banks face scrutiny over borrowing in times of crisis. Blau et al. (2016) examine TAF borrowing for publicly traded banks in an event study context and show that the market responded quickly and negatively to TAF borrowings. Further, they found the underperformance was greatest for institutions with the largest loans or amount of loans outstanding. Hence, there exists prima facie evidence that a stigma is associated with TAF borrowings for publicly traded firms.

Berger et al. (2014) examine which banks used the funds, how the use of other funding sources changed, and whether the use of these funds affected bank lending. They report three main findings: small banks receiving funds were weak banks whereas large banks generally were not; the funds substituted to a limited degree for other funding sources; and banks receiving funds increased their lending relative to other banks. Our results differ from those of Berger et al. as we separate TAF borrowings from discount window borrowings to focus on the former, which may result in different samples and, we examine on the possible stigma of TAF by comparing public and privately held bank borrowings.

We focus on the existence of a TAF stigma as perceived by both publicly traded and privately held bank borrowings because, during the crisis, very large banks that are subject to the most market scrutiny could be concerned that borrowing from the Fed would be perceived as a sign of weakness. Cyree et al. (2013) contend that although one might expect access to additional short-term credit through Fed programs during a crisis would increase bank stock returns, accessing Fed credit facilities could be viewed negatively by investors. For example, the discount window exists to provide liquidity, however repeated visits to the discount window, while necessary and successful during crises, has been viewed as having a stigma (Furfine (2003)), such that banks generally avoid discount window borrowing. Armantier et al. (2013) also use TAF data to empirically show the existence of the borrowing stigma.We address the following questions:

  1. 1.

    Was the TAF program a stigma-free substitute for discount window borrowing for all banks?

  2. 2.

    Was there a difference in TAF borrowing behavior between publicly traded banks and privately held banks?

  3. 3.

    Does the nature of the collateral influence TAF borrowing, and are there differences across publicly traded and privately held banks?

Our hypothesis is that publicly traded banks are more apt to experience a stronger stigma effect than privately held banks since publicly traded bank borrowings are more likely to be known and evaluated by market participants.

Collateral can constrain borrowing as shown by Shleifer and Vishny (2010) who find that banks borrow on a short-term basis against their inventory of securities to maximize short-term profits. As discussed by Gorton and Metrick (2012), the financial crisis was, in large part, a reaction to the reduced collateral values for repo transactions which greatly restricted banks’ ability to borrow to meet short-term fluctuations in the volatile times of the crisis. Accordingly, we also contribute to the literature by examining the differences in collateral offered by both publicly traded and privately held banks in exchange for receiving TAF funds.

We find that while publicly traded banks are several orders of magnitude larger and borrowed more in dollar terms under the TAF program than privately held institutions, the latter borrowed considerably more on a relative basis (more than twice as much as a proportion of assets) even though they had a higher deposit base, a higher capitalization ratio, higher returns on assets, higher net interest margin, higher growth, greater collateral and fewer off balance-sheet liabilities. Only past-due loans are weaker for the privately held banks. That is, privately held banks exploited the opportunity to borrow at the fixed TAF rates to a much greater extent. Our findings are consistent with Boyson et al. (2014) who find weak banks borrow less during a crisis.

Further, we verify prior studies’ initial results, finding that TAF borrowing is a substitute for discount window borrowing. However, we extend the analysis to show that both investment banks (I-banks) and too-big-to-fail (TBTF) banks borrow even less (relatively) than other publicly traded banks consistent with a TAF stigma due to increased scrutiny received by these very large and closely-followed banks. Since privately held banks borrow more, ceteris paribus, and since these institutions are subject to less market discipline, managers of these institutions likely have less concern about any perceived stigma effects.

Our results are also consistent with the conventional wisdom that publicly traded banks will avoid participating in government lending programs if they believe it can be perceived as a whole or partial bailout. The results are consistent with both Cyree et al. (2013) and Blau et al. (2016) since those publicly traded institutions that did borrow had negative wealth effects. From a policy standpoint, future crises programs should aim to reduce or eliminate the stigma of borrowing to banks to increase their effectiveness.

2 Background on Federal Reserve crisis facilities

The two principal Fed financial crisis facilities under which banks could access short-term credit are the discount window (DW) and the term auction facility (TAF). However, we control for the other major Fed crisis liquidity facilities available to banks during our sample period namely: the asset-backed commercial paper money market mutual fund liquidity facility (AMLF), the primary dealer credit facility (PD) and, the commercial paper funding facility (CPFF). Borrowing in these facilities could potentially offset the need for TAF or discount window access for some banks, thus we add them as control variables. A brief discussion of each of these facilities follows.Footnote 2

The discount window (DW) facility has been in operation for decades as a means of implementing the Fed’s lender-of-last-resort function. All discount window borrowing is immediately available but must be supported by collateral and, during the financial crisis, the Fed broadened the classes of acceptable collateral to include any asset of sound financial quality. In general, the Federal Reserve seeks to value all pledged collateral at an internal fair market value estimate. Our data are consistent with this guidance from the Fed.

Adrian et al. (Adrian et al. 2009, Chart. 4) show that discount window (primary credit) usage increased from approximately zero to roughly $10 billion in April 2008. They also report that discount window primary credit increased sharply to about $100 billion following the Lehman bankruptcy and remained above $40 billion through the end of 2009.

The term auction facility provided short-term (28-day or 84-day) funds for depository institutions. The facility requires depository institutions to bid in a single-price auction for collateralized term funds as an alternative to accessing the discount window. The use of the bidding process was thought to avoid the stigma of discount window borrowing (Cecchetti 2009; Armantier et al. 2013). These loans were required to be over-collateralized by at least a factor of two (i.e., a haircut of at least 50 %) although the Fed allowed a wide variety of illiquid collateral. Thus, while there are no differences in the rates paid by banks at the time of TAF access, there may be considerable differences in the nature and liquidity of the collateral provided. We examine these differences in Sect. 4.3.

The first TAF loan was made in December 2007 and the last loan was executed in March 2010. The program extended a total of $3.8 trillion of loans with $2.6 trillion at maturities of 28 days or less and, $1.2 trillion with maturities between 35 days and 85 days.Footnote 3 The maximum in TAF loans outstanding was about $266 billion in February and March 2009. Since TAF was designed to infuse liquidity into banks, borrowing under this program could be seen by the market as admission of serious structural problems.Footnote 4,Footnote 5

Gorton and Metrick (2012) use the term “securitized banking” to describe the process of packaging and selling loans and note that this process is funded using repurchase agreements (repos).Footnote 6 From the second quarter of 2007 to the first quarter of 2009, net repo financing provided to U.S. banks and broker-dealers fell by about $1.3 trillion—more than half of its pre-crisis total. The primary dealer credit facility (PD) was created to provide primary dealers collateral for participation in the repo market. Specifically, PD was an overnight loan facility for primary dealers secured by collateral with appropriate haircuts which was created to reduce the strain on the overnight repo market.Footnote 7 Acharya et al. (2012) state that this program played the role of a dealer “discount window.” PD began in March 2008 with loans reaching $40 billion in the April following the Bear Stearns collapse. The facility peaked at approximately $200 billion in loans outstanding following the bankruptcy of Lehman Brothers. Loans under PD decline to zero in the second quarter of 2009 and the facility was closed in February 2010.

The securitization process allows for the creation of securities on mortgage pools which were funded by asset-backed commercial paper (ABCP). Griffiths et al. (2011) show that ABCP declined from $1.2 trillion at the beginning of the crisis to $600 billion by the end of the crisis. Two programs were created to address this decline: the Asset-backed commercial paper money market mutual fund liquidity facility (AMLF) and the commercial paper funding facility (CPFF).

AMLF provided loans to depository institutions and bank holding companies to purchase high-quality asset-backed commercial paper (ABCP) from money market mutual funds. AMLF began in September 2008 and was closed in February 2010. Akay et al. (2013) examine the AMLF program and document a total of $217 billion in loans to seven depository institutions resulting in 3249 purchases of ABCP from MMFs.

CPFF was the commercial paper funding facility and was somewhat unique in that the Fed created and funded a limited liability special purpose vehicle (SPV) with the Federal Reserve Bank of New York as the sole beneficiary of the new company. The vehicle purchased 3-month unsecured and asset-backed commercial paper from eligible issuers. The CPFF began in October of 2008 and was closed in February of 2010. Data from the Federal Reserve report a total of $738 billion in loans across the entire program. Griffiths et al. (2011) show that CPFF contributed to stabilizing rates in the ABCP market.

3 Data and sample size

Our sample includes both publicly traded and privately held banks that participated in the TAF program. When we examine the different crisis programs together, we use only publicly traded banks because privately held banks did not participate in some of the crisis programs. When we focus directly on TAF, we include privately held banks along with publicly traded banks in the analysis. As we discuss each set of our results, we will clearly identify whether or not privately held banks are in the analysis.

3.1 Data description

The data for Federal Reserve lending under the crisis management programs come from Bloomberg L.P. and comprise each bank’s dollar amount by day for each program. In addition, we have Fed lending data from 8/1/2007 through 4/30/2010 and, as in Cyree et al. (2013), we restrict our sample to those banks which accessed one or more of the Fed programs during the crisis period. The end date in the Fed lending data is not a data collection restriction, but instead represents the end of Fed lending to banks under the various crisis programs.

We estimate the model separately over the following sub-periods to be consist with CGW (2013): (1) the BNP Paribus (8/9/07) announcement of the inability to value certain mortgage backed derivative structures through the Bear Stearns (3/14/08) failure, (2) the Bear Stearns failure (3/14/08) through the Lehman Brothers (9/12/08) bankruptcy, (3) the Lehman Brothers bankruptcy (9/12/08) through the end of 2008, and (4) from 1/3/09 through 12/31/10 which covers the period through the termination of all the Fed crisis facilities under analysis. These are well accepted breaks in the crisis timeline.Footnote 8

For comparison purposes, we separate the sample into investment banks, banks named too-big-to-fail and traditional banks. Cyree et al. (2013) label Bank of America, Bank of New York-Mellon, Goldman Sachs, JP Morgan-Chase, and Morgan Stanley as investment banks based on the self-descriptions of their business model and we continue this labeling for consistency. Banks classified as too-big-to-fail banks are based on whether or not the Federal Reserve publicly announced their requirement to be “stress tested” in April 2009, and were not classified as investment banks. The too-big-to-fail banks are BB&T, Capital One, Citigroup, Fifth Third, Keycorp, PNC Financial, Regions Bank, State Street Bank, SunTrust, US Bancorp, and Wells Fargo.Footnote 9 Banks that are neither I-banks or TBTF banks are labeled as traditional banks. We hypothesize that the larger the publicly traded bank, the more widely followed and therefore the most likely to have a stronger stigma effect due to increased scrutiny.

Table 1 contains descriptive statistics for these three groups of banks on all Fed crisis programs available to all banks. Since privately held banks did not participate in PD or AMLF, the data reported for traditional banks in Table 1 includes only publicly traded banks. The financial data are reported quarterly, but the program amounts outstanding and market data are daily figures. Because the banks are of substantially different sizes, we standardize the amount borrowed by the amount of total assets reported. Further, to provide additional insight into the institutions involved, we provide various bank-specific ratios including: DEP2ASSET—quarterly deposits scaled by assets, ROA—the return on assets, NIM—net interest margin calculated as annualized quarterly net interest income, divided by earning assets, PD902TA—the level of past-due loans 90 days or more, scaled by total assets, GAP12—a measure of interest rate risk and calculated as the value of assets that will re-price in the next 12 months, scaled by assets, OBSL—off balance-sheet items including derivatives and loan commitments, scaled by total assets and GROWTH—the quarterly growth rate in total assets.

Table 1 Descriptive statistics for investment banks, too-big-to-fail banks, and traditional (publicly traded) bank daily borrowings under the four major Fed lending programs and related financial variables for the period 8/1/07 through 4/30/10
Full size table

In addition, we provide several market value control variables including: FINPREM—the difference between 30-day financial and non-financial commercial paper rates in percentages, VIX_CLOSE—the closing value of the Chicago Board Options Exchange Market Volatility Index measuring the implied volatility of S&P 500 index options, and ZSCORE—the log of (ROA + CAPITAL) divided by the standard deviation of ROA. The FINPREM variable accounts for the direct premium for financial firm risk compared to non-financial firms. VIXCLOSE is a measure of overall market volatility and therefore overall market risk without taking into account industry effects. The Z-score provides a measure of insolvency with a higher value representing lower probability of default. Thus, if troubled banks borrow more there would be a negative relation with Z-score and borrowings.

Table 1 reports that traditional publicly traded banks borrow significantly more (as a percent of total assets) than either I-banks or TBTF banks under both DW and TAF. Since both I-banks and TBTF banks have access to virtually the same alternate funding sources, the fact that TBTF banks borrow approximately five times as much as I-banks but only eight percent less than other publicly traded banks is consistent with TAF representing a stigma to avoid, if possible. This is also consistent with DW performing its traditional role of providing liquidity and with TAF acting as its substitute. I-banks are the major players under PD, which is designed to provide securities to dealers for repos. I-banks and TBTF banks both use AMLF more than traditional banks. According to Akay et al. (2013), more than 90 % of all AMLF transactions were executed by JP Morgan and State Street. Generally, the changes in the amounts borrowed are small and not statistically significant, nor are the market variables that are measured each day.

The financial variables indicate the difference in business models with traditional banks having higher deposits-to-assets, but lower ROA, NIM, and interest rate risk as measured by GAP. Investment banks have the highest ROA and NIM, likely due to their non-interest income advantages. I-banks also have the highest ZSCORE. Too-big-to-fail banks have the highest capital ratios, past-due loans, off-balance-sheet liabilities, and quarterly growth in assets. In general, all of the financial variables have significantly different means and medians.

Table 2 provides descriptive statistics comparing privately held banks that borrowed in the TAF program to all publicly traded banks that borrowed in the TAF program. The privately held bank sample contains 184 banks with 1190 observations. The publicly traded bank sample contains 61 banks with 946 observations. There are only two I-banks (JP Morgan and Bank of America) and eight TBTF banks in the subsample of publicly traded TAF borrowers, with the remaining 51 banks being publicly traded traditional banks. If our hypothesis that publicly traded banks experience greater stigma effects, and the largest TBTF or I-banks even more so, then these banks should borrow relatively less than privately held banks.

Table 2 Descriptive statistics for publicly traded versus privately held banks on the TAF loan date for those banks that participated in the TAF program from 8/01/2007 through 4/30/2010
Full size table

While publicly traded banks are several orders of magnitude larger and borrowed more in dollar terms under the TAF program than privately held institutions, privately held banks borrowed considerably more on a relative basis (more than twice per level of assets—see: TAF loans/assets) even though they had a higher deposit base, a higher capitalization ratio, higher returns on assets, higher net interest margin, higher growth, greater collateral and fewer off balance-sheet liabilities. Only past-due loans are weaker for the privately held banks.Footnote 10 As shown by the mean of the indicator variables in Table 2, only about 4 % of the TAF borrowings for publicly traded banks were made to I-banks, and about 16 % of the TAF borrowings for publicly traded banks were to TBTF banks.

The mean of the financial premium is larger when publicly traded banks borrow in the TAF program, and volatility as measured by the closing VIX level is also higher, both indicating increased borrowing in times of greater uncertainty. We conclude that the nature of the average privately held bank borrowing in the TAF program is substantially different from the average publicly traded bank. While this may be somewhat attributable due to the availability of different funding sources, the differences between the three categories of publicly traded banks reinforce the argument for a difference in the perception of and willingness to participate in the TAF program.

Figure 1 is a plot of the TAF borrowings by quarter, as a percent of non-deposit borrowings for the banks accessing the TAF facility. Each bar represents the mean proportion borrowed. As shown, the proportion of TAF borrowing is much higher for privately held banks as a percentage of non-deposit borrowed funds. In other words, the TAF program was relatively a much more important source of funds for privately held banks compared to publicly traded banks. Further, the steady increase in TAF as a percent of non-deposit borrowings for privately held banks indicates that TAF was recognized as an important and ongoing source of funds as the financial crisis and the subsequent recession progressed.

Fig. 1
figure 1

Privately held bank versus publicly traded bank TAF borrowings

Full size image

3.2 Empirical model

To study whether or not TAF is a substitute for the discount window, we model the change in the percentage of TAF borrowings scaled by assets as:

$$\begin{aligned} dTAFA_{i,t} & = \alpha + \omega_{1} dDWA + \omega_{2} dPDA\alpha_{2} + \omega_{3} dAMLFA + \alpha_{1} TBTF + \alpha_{2} IBANK \\ & \quad + \lambda_{1} FINPREM + \lambda_{2} VIX\_CLOSE + \delta_{1} DEP2ASSET + \delta_{2} CAPRATIO + \delta_{3} ROA \\ & \quad + \delta_{4} NIM + \delta_{5} PD902TA + \delta_{6} GAP12 + \delta_{7} OBSL + \delta_{8} GROWTH + \theta_{1} QTREND \\ & \quad + \vartheta_{2} YEAREND + \varepsilon_{i,t} \\ \end{aligned}$$
(1)

The variables are defined as followsFootnote 11:

dDWA :

is the daily change in the amount of discount window borrowings scaled by assets, in percentage terms

dPDA :

is the daily change in the amount of primary dealer borrowings scaled by assets, in percentage terms

dAMLFA :

is the daily change in the amount of AMLF borrowings scaled by assets, in percentage terms

TBTF :

is an indicator variable for the too-big-to-fail banks that were stress tested by federal regulators

IBANK :

is an indicator variable for investment banks, and equals one for Goldman Sachs, Bank of America, Bank of New York/Mellon, JP Morgan/Chase, and Morgan Stanley, and zero otherwise

FINPREM :

is the difference between 30-day financial and non-financial commercial paper rates in percentages

VIX_CLOSE :

is the closing value of the Chicago Board Options Exchange Market Volatility Index measuring the implied volatility of S&P 500 index options

DEP2ASSET :

is quarterly deposits scaled by assets

ROA :

is return on assets, measured as annualized quarterly net income divided by quarterly total assets

NIM :

is net interest margin calculated as annualized quarterly net interest income, divided by earning assets

PD902TA :

is the level of past-due loans 90 days or more past due, scaled by total assets

GAP12 :

is a measure of interest rate risk and is calculated as the value of assets that will re-price in the next 12 months, scaled by assets

OBSL :

is off balance-sheet items including derivatives and loan commitments, scaled by total assets

GROWTH :

is the quarterly growth rate in total assets

QTREND :

is an indicator variable for the last 2 days of the quarter except for the fourth quarter

YEAREND :

is an indicator for the last two trading days of the year end

Under normal conditions, when modeling a market-based process such as returns, the financial data would be lagged since it is not public until reported with some delay. Here, the decision to borrow or not from the Fed and under which program is internal, and thus, decision makers would be aware of their then current financial status. Accordingly, we do not lag our financial data, nor do we use changes in the financial data. The market data, such as FINPREM and VIX_CLOSE change daily, and thus are better proxies for concurrent changes in the environment than the financial data. However, in the case of the change in borrowings (through TAF or the discount window), it is important to scale by the level of assets since such market value introduces excessive volatility during this period.Footnote 12 In addition, we cannot compare privately held banks to publicly traded banks if we use only market values. The use of a scaling variable is necessary as the banks in our sample vary widely in size.

For examining the second hypothesis of differences in TAF borrowing for publicly traded versus privately held banks, we estimate TAF borrowings scaled by assets in the following model:

$$\begin{aligned} TAFA_{i,t} & = \alpha + \rho PUBLIC + \alpha_{1} TBTF + \alpha_{2} IBANK + \lambda_{1} COLLATER + \lambda_{2} FINPREM \\ & \quad + \lambda_{3} ZSCORE + \lambda_{4} VIX\_CLOSE + \delta_{1} DEP2ASSET + \delta_{2} CAPRATIO + \delta_{3} ROA + \delta_{4} NIM \\ & \quad + \delta_{5} PD902TA + \delta_{6} OBSL + \delta_{7} GROWTH + \varepsilon_{i,t} \\ \end{aligned}$$
(2)

where PUBLIC is an indicator variable equal to one if the bank is publicly traded and zero otherwise. A significant coefficient for PUBLIC is consistent with less TAF borrowing due to a stigma. The other variables are as defined in Eq. 1 with the addition of COLLATER which is collateral pledged as a percent of the loan.

4 Results

We first analyze whether TAF functioned as a liquidity substitute and whether it provided necessary liquidity or simply bailed out poorly performing banks. This analysis will also verify whether our sample is consistent with other studies that find TAF is a stigmatized substitute for the discount window.

4.1 Is TAF a substitute?

Table 3 presents the results for the OLS model (1) that estimates the change in TAF (relative to total assets) and whether these changes represent substitutions from other crisis programs. The model is estimated only for publicly traded banks because privately held banks did not have access to or participate in all the crisis programs. The results are presented for the four previously defined sub-periods of the crisis: Period 1 is labeled BNP as this period starts when BNP halted redemptions on some money market funds, Period 2 is labeled Bear as this period starts with the failure of Bear Stearns, Period 3 is labeled Lehman as this period starts with the bankruptcy of Lehman Brothers, and Period 4 is labeled post-crisis and starts at January 2009. The variable dDWA estimates whether borrowing at the discount window substitutes for TAF borrowing. The coefficient for this variable is negative and highly significant across the crisis indicating that decreases in discount window borrowing relate to increases in TAF borrowings and is consistent with banks using the TAF program as a substitute for discount window borrowing. That is, it appears that banks used the TAF program instead of the discount window during the crisis. It is not knowable without surveying the decision-makers directly whether or not they were exactly substituting TAF for DW borrowings or, if they were borrowing TAF funds and would not have borrowed at the discount window in the absence of TAF.Footnote 13 The results for the variables PD and AMLF are insignificant. These two programs were focused on dealers and money market funds and thus, a priori, should not substitute for TAF borrowings, but serve as control variables to mitigate the effects of borrowing by some banks in these programs.

Table 3 OLS regression for the daily percentage change of TAF borrowings scaled by assets for banks that borrowed TAF funds, by period
Full size table

Importantly, both the too-big-to-fail and I-bank indicator variables are statistically insignificant and thus, are not different from other publicly traded banks throughout the sample period. We will explore whether publicly traded banks are different from privately held banks in a subsequent section, but this result indicates that those banks we believe to be under the most scrutiny—the largest publicly traded banks—do not significantly borrow more or less TAF funds generally during the crisis than other publicly traded banks. (We provide a more complete analysis of this point in the following section.) That is, the largest banks do not appear to have additional stigma associated with borrowing in the TAF program when compared only to other publicly traded banks.

Few of the control variables are significant and none are significant across all periods. This is not surprising as Discount Window borrowing and TAF crisis borrowing are for emergency liquidity and the control variables measure normal bank performance. The one control variable that is proxy for liquidity is DEP2ASSETS and it is significant and negative both during the first period (p value = 0.041) and negative with a p value of 0.1068 during the Lehman period. This suggests that banks with more deposits to total assets have less need of emergency liquidity. During the first period of the crisis, PD902TA is significant and positive. During this period, the crisis was considered to be a liquidity issue and TAF is a substitute for DW liquidity. Banks borrowed less when they had more deposits and borrowed more when they had larger past-due loans. VIX_CLOSE is significant and positive in the last two sub-periods which follows the Lehman bankruptcy. The VIX results indicate more TAF borrowings occurred when market uncertainty increased.

Finally, we note the low adjusted R-square for the BNP and Bear periods. BNP halted redemptions in August of 2007 and TAF started lending in December of 2007. Bear was rescued in March of 2008. Accordingly, TAF existed for only the second half of our first period hence, the model is unlikely to explain much about TAF borrowing. The second period follows the failure and assisted sale (i.e. bailout) of a large investment bank (Bear Stearns) which reduced the need for systematic emergency borrowing. Lehman signaled a full blown financial crisis and the need for emergency lending. More than half of all the TAF transaction occur after January 1, 2009 during our post-crisis period.

4.2 Regression results for TAF borrowings for privately held and publicly traded banks

To this point, our results suggest that TAF is a substitute for the discount window (Table 3) and that the largest TAF borrowers (on a relative basis) tend to be privately held banks which are not particularly distressed (Table 2). These results suggest TAF worked as designed, and that publicly traded banks borrow at the same rate as very large systemically important banks. We now turn to whether or not the stigma for TAF borrowing is different for publicly traded versus privately held banks.

Table 4 contains the regression results for all banks in the sample, both public and private, and the percentage of assets for their TAF borrowings. The key variables designed to answer our second research question are PUBLIC, and to a lesser extent, TBTF and IBANK. If any of these variables are significant and positive, it would suggest not only a lack of stigma, but perhaps a bailout of the largest banks vis-à-vis more traditional, non-publicly traded banks. If any of these variables have significantly negative coefficients, it suggests a stigma exists for publicly traded banks, and that the TAF program was more useful for smaller, traditional banks. Results show that publicly traded banks borrow significantly fewer TAF funds as a percent of assets than privately held banks. Further, TBTF and I-banks borrow even less than other publicly traded banks during Period 2 between the Bear Stearns and Lehman failures, a period when the crisis was rapidly worsening. That is, when liquidity was likely needed, the largest and most susceptible publicly traded banks borrowed less than other banks.Footnote 14

Table 4 OLS regression for the percentage of TAF borrowings scaled by assets on the loan date for banks that borrowed TAF funds, by period
Full size table

Several control variables also provide interesting insights. First, as the crisis progresses banks with more deposits borrow less under TAF. Second, banks with more capital tend to borrow more under TAF. This is suggestive of prudent management toward available opportunities in the middle of a crisis (Ivashina and Scharfstein 2010 and Berger et al. 2014). Consistent with prudent management following the Lehman Brothers bankruptcy, banks that are growing, as measured by increasing assets, borrow more from TAF.

4.3 Collateral

As shown in Table 2, privately held banks provide significantly more collateral for TAF borrowing than publicly traded banks. Lending against distressed collateral would be a method for bailing out banks that previously made poor business decisions. In this section, we examine the collateral available for TAF borrowing.

Each TAF transaction is accompanied by a list of collateral available to support that loan. The collateral is referred to as unencumbered collateral and defined as: “lendable value of TAF borrower’s discount window collateral, after the application of appropriate margins (haircuts) and minus the value of any outstanding loans, in millions of dollars.”Footnote 15 That is, our variable represents net collateral not total collateral, thus a TAF loan can be up to 100 % of the collateral listed with each transaction.

In Table 4, the collateral variable is significant and positive in each regression, even though it is used primarily in that regression as a control variable. As expected, the size of collateralized loans should be positively related to the amount of collateral available.

To directly examine collateral pledged, we regress our control variables on the amount of collateral expressed as a percentage of total assets and estimate the regression for each of the four sub-periods. Table 5 contains these results. None of the control variables is significant with the same sign across all four sub-periods. The control variables measure bank performance, so the lack of results on performance measures in Table 5 suggest that weaker banks are not pledging more collateral to access TAF. The most consistent results are: (1) publicly traded banks tend to pledge less collateral and, (2) better capitalized banks pledge less collateral.

Table 5 OLS regression results for collateral pledged on the loan date for TAF borrowings as a percent of assets, by period
Full size table

The negative and significant coefficient for the PUBLIC variable in the collateral regression indicates publicly traded banks pledge less collateral, all else equal. This decision creates more flexibility for publicly traded banks in the future since they do not have to pledge as much collateral. In other words, publicly traded banks have more incentive to borrow in the TAF program since they do not have to pledge as much collateral and therefore their opportunity cost is lower compared to privately held banks. This result, coupled with the findings that publicly traded banks borrow fewer TAF funds than privately held banks (Table 4) reinforces the argument of a TAF stigma.

At this point, there is nothing in our results that suggests that weaker banks use more collateral to access more TAF loans and thus receive a bailout. We now extend our analysis to examine the types of collateral.

4.3.1 Collateral mix

If a bank has a liquidity problem, then it borrows from the discount window (or TAF during the crisis) and should pledge the least liquid collateral that the discount window will accept. In general, bank loans are less liquid than securities and, during the crisis, mortgage backed securities were considered low quality. The discount window would accept both of these types of collateral. Accordingly, the mix of collateral pledged for discount window (and thus TAF) borrowing provides insights into whether TAF borrowing is for liquidity or, to provide a bail out of poorly performing banks. In addition to being distressed, another plausible explanation for pledging collateral is that a bank will pledge the least liquid collateral acceptable to maintain financial flexibility in case additional borrowing is required in the future.

Table 6 provides summary statistics on the collateral mix for each type of bank. Panel A reports the median percent of total collateral of each type and Panel B reports the number of TAF loans supported by each type of collateral.

Table 6 TAF collateral
Full size table

I-banks pledge loans as collateral and use both commercial and consumer loans. This is consistent with pledging illiquid collateral and/or maintaining flexibility. TBTF banks also lean heavily toward loans as collateral, but do use some securities as collateral. For example, over half (88 of 161) of the TBTF transactions are supported by other MBS securities, but these securities represent only 8 % of the collateral available to support a loan. Publicly traded banks continue the pattern of submitting loans as collateral. Publicly traded banks do use securities, but securities of any type are used in less than 25 % of their TAF transactions. Privately held banks use less loans and more securities than I-banks, TBTF banks and traditional publicly traded banks. Only slightly more than half of the privately held bank TAF transactions are supported by loans and, when securities are used, they tend to be a higher percentage of the total collateral. Privately held banks pledge both a substantial dollar amount and numbers of high quality securities (Treasury, Agency, and MBS/agency guaranteed).

There are two important insights from the summary statistics on collateral. First, the collateral tends toward loans over securities consistent with the banks pledging illiquid collateral instead of distressed securities as well as maintaining flexibility. Second, privately held banks used securities as collateral more than publicly traded banks, but these securities are of high quality. These results support the argument that, in a crisis, one limiting factor for loans is the amount of highly liquid securities held, and especially for privately held banks.

Finally, there is a large group of TAF transactions that is not included in our sample. Our sample is limited to domestic banks to allow us access to bank financial data. Foreign banks account for 1485 TAF transactions. Hence, foreign banks are the largest group in this program.

We provide the collateral data for the foreign banks in the last row of each panel of Table 6 and, there are two important findings. First, foreign banks tend to prefer securities as collateral over loans implying that foreign banks were less concerned with flexibility or liquidity than US banks. Second, the securities provided are from the classes of securities that tended to show distress during the crisis (corporate, MBS/other and asset-backed).

4.4 Robustness of results

For robustness, we estimate the model across all periods and use indicator variables for the different periods omitting the post-crisis indicator. We do not present the specific results in the interest of brevity. The variable PUBLIC representing publicly traded firms is insignificant, largely because of the insignificance in the latter part of the crisis as shown in the period-specific tables. IBANK and TBTF continue to indicate that, during the whole sample period, both I-banks and too-big-to-fail banks borrowed less in the TAF program as a proportion of assets. Deposits are negatively related to TAF borrowings for the whole period supporting the argument of the importance of a stable depositor base. The negative sign on the capital ratio indicates that lower proportions of equity translate into higher TAF borrowings, ceteris paribus.Footnote 16 Banks with higher net interest margins and growth also borrow more in the TAF program.

We then perform panel regressions for one-way (cross-sectional) fixed effects. This model assumes a mean for each bank and subtracts this mean from the dependent variable. The results (available upon request) indicate that there are no significant fixed-effects. The dDWA variable remains significant in all cases with only minor changes in the control variables.Footnote 17

We also perform a random effects model which assumes there is some underlying structure that is not necessarily the same for each bank and it is therefore “random” but correlated with the underlying structure. Excluding I-banks and TBTF banks results in the failure to reject the null hypothesis that a random-effects model appropriately measures the correlation between the effects variables and the dependent variables.Footnote 18 Collectively, the robustness tests indicate the OLS model is effective in capturing changes and levels of TAF borrowings.

4.4.1 3SLS system of equation results

For additional robustness and, because it is likely that financing decisions are jointly determined and therefore endogenous, we also estimate a three-stage least squares model to form a system of equations for the type of borrowing:

$$\begin{aligned} dY_{i,t} & = \alpha + \alpha_{1} TBTF + \alpha_{2} IBANK + \lambda_{1} FINPREM + \lambda_{2} VIX\_CLOSE + \delta_{1} DEP2ASSET \\ & \quad + \delta_{2} CAPRATIO + \delta_{3} ROA + \delta_{4} NIM + \delta_{5} PD902TA + \delta_{6} GAP12 + \delta_{6} OBSL + \delta_{5} GROWTH \\ & \quad + \lambda_{3} QTREND + \lambda_{4} YEAREND + \varepsilon_{i,t} \\ \end{aligned}$$
(3)

where the dependent variable (dY) in the 3SLS model is alternatively: dTAFA, dDWA, dPDA, and dAMLFA as defined earlier.

The three-stage least squares model in Eq. 3 accounts for the possible correlation across the available borrowing programs and possible endogeneity.Footnote 19 Again, we examine all the programs, so only publicly traded banks are included in the sample for this estimation since privately held banks did not have the opportunity to borrow in all programs. As specified, the model lends insight into how the different factors affect each individual program’s borrowings, as well as providing an overall correlation matrix to examine any relation among the programs. These correlations measure more directly whether any of the programs are substitutes (negative correlation) or complements (positive correlation).

In the interest of brevity, the results from the 3SLS are not reported. Instead, we focus on the correlation matrix, which is presented in Table 7. The correlation matrix indicates that only TAF and the discount window are significantly related and supports the argument that they are substitutes consistent with the stated goal of the Federal Reserve (Bernanke, 2009):

The introduction of this facility seems largely to have solved the stigma problem, partly because the sizable number of borrowers provides anonymity, and possibly also because the 3-day period between the auction and auction settlement suggests that the facility’s users are not relying on it for acute funding needs on a particular day.

Despite chairman Bernanke’s assertion of solving “the stigma problem,” Blau et al. (2016) show that the market responded quickly and negatively to TAF borrowings and, their results suggest that market was aware of which banks participated in TAF, so it is unlikely that anonymity plays a major role in a bank’s decision to participate in TAF.

Table 7 Cross model correlation from the 3SLS system of equations
Full size table

5 Summary and conclusions

In this study, we examine discount window and term auction facility (TAF) borrowings. Specifically, we address the following questions:

  1. 1.

    Was the TAF program a stigma-free substitute for discount window borrowing for all banks?

  2. 2.

    Was there a difference in TAF borrowing behavior between publicly traded banks and privately held banks?

  3. 3.

    Does the nature and extent of collateral influence TAF borrowing, and are there differences across publicly traded and privately held banks?

In period-specific OLS regression models and a 3SLS model, there is strong support that TAF borrowings were used as a substitute for discount window borrowings. Our regression results indicate that after banks borrowed under the TAF program, they borrowed less at the discount window. Our findings also indicate our sample is representative and supports findings by Berger et al. (2014) and Armantier et al. (2013).

The principal focus of our paper is in answering question 2: whether or not there is a difference in TAF borrowings between publicly traded and privately held banks. We hypothesize that publicly traded banks are more subject to attention and market scrutiny, and especially so if they are investment banks or large banks declared too-big-to-fail. If publicly traded banks borrowed more than privately held banks, the hypothesis that large banks benefitted at the expense of traditional banks is supported. If publicly traded banks borrowed less, it suggests that a stigma exists for TAF or similar borrowings that could be perceived as a bailout, thereby reducing the efficacy of the TAF program.

In general, we find that while publicly traded banks are several orders of magnitude larger and borrowed more in dollar terms under the TAF program than privately held institutions, the latter borrowed considerably more on a relative basis (more than twice as much as a proportion of assets) even though they had a higher deposit base, a higher capitalization ratio, higher returns on assets, higher net interest margin, higher growth, greater collateral and fewer off balance-sheet liabilities. Only past-due loans are weaker for the privately held banks.

An alternate explanation for the increased borrowings by privately held banks lies in the assertion made by Ivashina and Scharfstein (2010) that after the Lehman Brothers failure in September 2008, banks found it difficult to roll over their short-term debt while many borrowers drew down their credit lines. In this case, privately held banks may have attempted to acquire as much liquidity as possible in light of the ongoing crisis. Similarly, since Berger et al. (2014) find that banks that borrowed more subsequently engaged in more lending, privately held banks may have viewed TAF borrowings as reliable and inexpensive liquidity to be used for growth purposes.

Both I-banks and TBTF banks borrow even less (relatively) than other publicly traded banks consistent with the increased scrutiny received by these banks, as well as these large banks having other sources of borrowing that smaller banks do not have. However, since privately held banks borrow more, ceteris paribus, and since these institutions are subject to less market discipline, managers of these institutions likely would have fewer concerns about perceived stigma effects. Privately held banks not only borrow more as a percentage of assets, but also as a percentage of non-deposit borrowed funds in every time period. Our findings on higher proportions of borrowings for privately held banks indicate that policymakers need to carefully consider how to reduce or remove the stigma imposed upon large, publicly traded banks borrowing in these programs to help mitigate future crises.

Lastly, the incentive for banks to maintain flexibility through adequate collateral to pledge in times of crises is shown since those banks with higher proportions of collateral borrow more in the TAF program. As shown by Berger et al. (2014), banks that borrowed funds during the 2008 crisis were able to lend more, however the increased lending does not translate into higher stock returns as shown by Cyree et al. (2013). Consistent with our expectations however, we find that publicly traded banks pledge the least liquid collateral acceptable to maintain financial flexibility in case additional borrowing is required in the future.

Our finding that publicly traded banks in general borrowed relatively less than privately held banks, and TBTF and I-banks even less in the TAF program suggests that success in TAF was constrained for the most vulnerable and systemically important banks at the height of the crisis, likely due to the stigma of a perceived bailout. Publicly traded banks pledge less collateral than privately held banks, and therefore their reluctance to borrow should be based less on collateral constraints. TAF borrowing is substantial for foreign banks as shown by the number pledging collateral, and this raises questions about the participation of foreign banks in TAF, but we leave this policy question to others.