The Effects of Unconventional Monetary Policies on Bank Soundness
Author:
Frederic Lambert
Search for other papers by Frederic Lambert in
Current site
Google Scholar
Close
and
Mr. Kenichi Ueda
Search for other papers by Mr. Kenichi Ueda in
Current site
Google Scholar
Close

Contributor Notes

Authors’ E-Mail Addresses: flambert@imf.org; kueda@imf.org

Unconventional monetary policy is often assumed to benefit banks. However, we find little supporting evidence. Rather, we find some evidence for heightened medium-term risks. First, in an event study using a novel instrument for monetary policy surprises, we do not detect clear effects of monetary easing on bank stock valuation but find a deterioration of medium-term bank credit risk in the United States, the euro area, and the United Kingdom. Second, in panel regressions using U.S. banks’ balance sheet information, we show that bank profitability and risk taking are ambiguously affected, while balance sheet repair is delayed.

Abstract

Unconventional monetary policy is often assumed to benefit banks. However, we find little supporting evidence. Rather, we find some evidence for heightened medium-term risks. First, in an event study using a novel instrument for monetary policy surprises, we do not detect clear effects of monetary easing on bank stock valuation but find a deterioration of medium-term bank credit risk in the United States, the euro area, and the United Kingdom. Second, in panel regressions using U.S. banks’ balance sheet information, we show that bank profitability and risk taking are ambiguously affected, while balance sheet repair is delayed.

I. Introduction

To combat the financial turmoil and subsequent “Great Recession”, major advanced countries have adopted unconventional monetary policies: keeping the policy rate near zero, attempting to manage expectations actively (forward guidance), expanding central banks’ balance sheets by purchasing long-term government bonds and risky assets, and introducing schemes to facilitate bank lending.2 In particular, when the financial market faced acute dysfunction, central banks’ actions prevented banks, and the economy at large, from falling into a “bad” equilibrium or debt-deflation spiral (for a review, see for example, IMF, 2013a).

In theory, even after the acute phase of the crisis ended, unconventional monetary policies may benefit banks. In the short run, banks engaging in maturity transformation should gain from low short-term rates as long as the long-term rates remain relatively stable. Similarly, banks can gain from borrowing at low cost and investing in assets delivering higher returns provided that policies do not depress the returns on those assets as well. Moreover, banks may take advantage of any reduction in term premia to replace short-term debt with long-term debt and reduce the risk of maturity mismatches in their balance sheets (Stein, 2012).

However, in the medium term, too easy monetary policies may hurt banks. The boost in spread income wanes as unconventional policies flatten the yield curve and reduce risk premia. Consequently, banks may rationally take extra leverage and risk (Borio and Zhu, 2008). This happens, for example, with an extraordinary relaxation of collateral rules that makes funding available at low cost to all banks almost regardless of the strength of their balance sheets. Also, with low interest rates, banks may prefer to roll over loans to non-viable firms rather than declaring them non-performing and registering a loss in their income statement. Previous studies have found evidence of such “evergreening” policies in Japan in the 1990s and 2000s (Peek and Rosengren, 2003; Caballero, Hoshi and Kashyap, 2008).

The overall effect of unconventional monetary policies on banks’ profitability and risk is thus theoretically unclear. Both the benefits and costs, however, should be reflected in the changes in banks’ stock prices and their bond risk premia at the time of announcement of new monetary policy measures. We analyze the sign and magnitude of such changes by regressing daily changes in bank asset prices on the surprise changes in monetary policy on all Federal Open Market Committee’s announcement days from January 2000 until October 2012. In addition to this event study analysis, we examine the effects of low interest rates—both their level and the duration of the period of low rates—and central banks’ asset purchases on banks’ profitability, risk taking, and balance sheet repair using bank-level data for the United States over the period 2007–2012. While the event-study regressions uncover changes in medium-term credit risks and profitability of banks, the panel analysis looks for any symptoms of risk already present in banks’ balance sheets.

The event-study regressions show that bank credit risk increases with monetary easing over the medium term without clear effects on bank stock valuation. To accurately gauge the effects, we use the surprise component of policy announcements. This is because the expected element should not affect market prices at the time of announcement as it should be already priced in. In particular, in the benchmark regressions, we use the change in the one-year-ahead three-month futures rates as the surprise measure, so as to capture both the contemporaneous part of a monetary policy announcement (reflected in the target policy rate change) and any expected developments for near-term future rates (focus of the forward guidance and quantitative easing). However, this measure may also reflect the expected effect on one-year-ahead rates of today’s policy. Besides, downward changes are potentially limited once the policy rate hits the zero lower bound. As an alternative, we thus construct a novel instrument for the surprise based on the number of news articles before and after each policy announcement. This measure is not constrained by the zero lower bound. The instrumental variable estimates confirm the negative effect of monetary easing on bank credit risk.

The results of bank balance sheet panel regressions point to contrasting effects on banks’ profitability and risk. While unconventional monetary policies do not appreciably affect the profitability of banks, they seem to reinforce banks’ soundness by allowing them to reduce leverage and extend the maturity of their debt. However, a prolonged period of low interest rates is also associated with an increase in the ratio of risk-weighted assets to total assets. In addition, we find that increases in central banks’ balance sheets could delay loss provisioning for existing loans, thereby potentially increasing the overall credit risk as also found in the event study results.

To the best of our knowledge, our paper is the first to provide a comprehensive assessment of the effects of unconventional monetary policies on the soundness of the banking sector. While the empirical literature on unconventional monetary policy has been growing, most studies focus either on the transmission question, i.e., the effects of unconventional policies on long-term government bond yields and risky asset prices, or on the macroeconomic question, i.e., the effects on inflation and GDP growth rate (see review papers, for example, Woodford, 2012, and IMF, 2013b).

A few papers have examined the relationship between monetary policy and bank risk-taking but primarily over the pre-crisis period. Altunbas, Gambacorta, and Marqués-Ibáñez (2010) found evidence that low interest rates over an extended period of time contributed to an increase in banks’ risk, measured by their expected default frequency, over the pre-crisis period 1999–2008. Using data on U.S. banks’ corporate loan ratings over a longer period that includes the first years of the crisis (1997–2011), Dell’Ariccia, Laeven, and Suarez (2013) also found a negative relationship between risk-taking by banks and increases in real policy rates. The strength of that relationship depends on banks’ capitalization and the effect of interest rates on risk-taking is smaller for poorly capitalized banks. Similar results were obtained by Jiménez, Ongena, Peydrò and Saurina (2009) with data on loans granted by Spanish banks over the period 1984–2006, and by Ioannidou, Ongena and Peydrò (2009) with Bolivian data.

A recent paper by English, Van den Heuvel, and Zakrajšek (2012) is most similar to our study. Using pre-crisis period data, they study the effects of changes in the level of policy rates and the slope of the yield curve on bank stock valuations and profitability. They show a drop in bank stock prices following an unexpected increase in the level of policy rates and a steepening of the yield curve. Although higher short-term rates and a steeper yield curve increase banks’ return on assets, the positive effect on near-term profitability is offset by a slowdown in asset growth and an outflow of core deposits, which represent an inexpensive source of funding compared to market alternatives. Yet a policy rate cut is typically associated with a steepening of the yield curve. This is consistent with the assumption that monetary easing is effective at boosting economic activity, which should increase inflation and growth expectations.

The rest of the paper is organized as follows. Section II reports the results of the event-study regressions. It explains our measures of monetary policy surprises and then examines the forward-looking effects of those surprises on bank stock returns and credit risk measures. Section III analyzes bank-level panel data and sheds light on the channels through which monetary policy easing affects banks. Section IV concludes.

II. Event Study

A. Surprise Component of Monetary Policy Announcements

We compare the average effects on banks of unconventional monetary policies with those of conventional monetary policies. We do so by regressing daily bank stock returns and daily changes in credit risks on monetary policy surprises on (almost) all FOMC announcement days since 2000. We define the conventional monetary policy period as the period before August 2007 while the so-called unconventional policy period starts after the collapse of Lehman Brothers on September 16, 2008 (see more detailed discussion below). The effects of unconventional policies could differ from those of conventional policies both quantitatively and qualitatively.

A difficulty in conducting an event study on monetary policy is that the expected actions are already priced into asset prices at the time of the policy announcement. In other words, only the surprise component of an announcement can affect asset prices. Yet, in the conventional policy period, when the interest rate is the only policy tool, once the impact of one unit (e.g., one basis point) of surprise is estimated, the overall impact of the total policy rate change (e.g., 25 basis point cut) can be calculated simply by multiplying the coefficient by the total policy rate change (e.g., 25 times the estimated coefficient). Unfortunately, a similar estimate for the overall effect by an unconventional policy is difficult to obtain, as unconventional policy measures do not take the form of a change in the policy rate. Instead, we investigate whether the effects of unconventional policies differ in terms of signs and magnitude for one unit of surprise from those of conventional policies.

To compare the results, it is essential to gauge the surprise component for both conventional and unconventional monetary policy actions in a similar way. For conventional policies, Bernanke and Kuttner (2005) and Kuttner (2001) used daily changes in the 30-day federal funds futures contract rate as a measure of the surprise element of a monetary policy action for the United States. However, mirroring the behavior of the federal funds rate, the federal funds futures rate has been near zero and moving very little since late-2009. Moreover, the zero lower bound on short-term interest rates has caused central banks to target interest rates on longer-term securities as well as riskier assets.

As our benchmark surprise measure we use one-year-ahead futures of the three-month Eurodollar rate. This is because the one-year-ahead futures rate is less affected by the zero lower bound than the current-month futures contract rate and also because it relates more to the unconventional monetary policy’s intention to influence longer-term interest rates. Moreover, even in conventional times, changes in the one-year-ahead futures contract rate reflect not only the target policy rate changes but also implicit forward guidance on future economic conditions and future policy rates implied by the FOMC statements, which is an important part of monetary policy (Gürkanyak, Sack, and Swanson, 2005, and Campbell, Evans, Fisher and Justiniano, 2012).

We acknowledge that the change in the one-year-ahead three-month Eurodollar futures rate is not a perfect measure of monetary policy surprises. First, rate changes are also constrained by the zero lower bound in later years, even though unconventional policy measures such as quantitative easing may not be constrained. In this case, the measured surprise can be smaller than the true surprise, and this would create an upward bias in the estimated magnitude of the effects of surprise. Figure 1 appears to show a reduction in the variance of the changes in the one-year-ahead futures rate in the unconventional policy period. Yet, changes are larger than those in the current-month federal funds futures rates, which stay at essentially zero after late-2009.

Figure 1.
Figure 1.

Surprise-Change in One-Year Ahead Three-Month Eurodollar Futures

Citation: IMF Working Papers 2014, 152; 10.5089/9781498363563.001.A001

Sources: Bloomberg L.P. and authors’ calculations.

Second, a bold monetary easing action by the Fed, especially in the acute crisis phase, may be also seen as signaling a pessimistic view about the economy, and the market may accordingly revise its own assessment of economic growth downward. In this case, the one-year-ahead futures rate would decline more than in the absence of such unintentional signals.3 However, the direction of the overall bias is unclear because the opposite case is also possible (i.e., a less-than-expected action may signal that the Fed holds an optimistic view).

In addition to the measurement issues, endogeneity problems may also arise in the regression when using the one-year-ahead futures rate. The one-year-ahead futures rate reflects expectations of economic conditions a year later, which are influenced by current monetary policy. Put differently, changes in the one-year-ahead futures rate may capture not only the monetary surprise itself but also the expected outcome one year from now of the announced monetary actions. If, for example, today’s policy rate cut is effective at increasing growth and inflation a year later, changes in the one-year-ahead futures rate may be reduced or even reversed. Moreover, the underlying interest rate of the Eurodollar futures contract is the London interbank rate (LIBOR), which picks up the credit risk of banks in addition to the policy rate.4 Thus, any change in the expectations of future credit risks due to current monetary policy is also reflected in the change in the one-year-ahead Eurodollar futures rate. In this case, if the current policy rate cut is effective at lowering credit risks, the movements in the one-year-ahead futures rate may be even larger.

So far researchers have not come up with a clean surprise measure of unconventional monetary policy. Only a few attempts have been made: Rosa (2012) provides a measure of the unanticipated component of asset purchases programs by the Fed based on the Financial Times’ coverage, and Joyce, Lasaosa, Stevens and Tong (2011) relies on market participants’ expectations of asset purchases by the Bank of England. However, Rosa’s measure can only take three values (-1, 0, and 1) depending on whether the announcement is deemed more restrictive, similar, or more expansionary than expected and his classification relies on a single news source. The data used by Joyce et al. are unfortunately unavailable for the United States. In any case, asset purchases are only one type of unconventional monetary policy measure.

We expanded Rosa’s approach and counted the number of news articles on monetary policy three days before and after each policy announcement (total news coverage) using Factiva, which is a news-article search service provided by the Dow Jones company. We also collected the numbers of “positive” and “negative” news references in terms of monetary easing.5 These three variables, specifically, the before-after changes in total, positive, and negative news coverage, are used as instruments to avoid the endogeneity issues discussed above. Unlike Rosa’s measure, our surprise measure does not involve a judgment call by the researcher.

Figure 2 shows the plot of the fitted surprise measure constructed from the first stage regression of a two stage least squares estimation.6 Note that the movements of the fitted surprise are no longer constrained by the zero-lower bound in the later period unlike the changes in the one-year-ahead three-month Eurodollar futures rates (shown in Figure 1). In this sense, our two stage least squares strategy addresses not only the endogeneity problem but also corrects a possible bias due to the zero lower bound.

There is still a caveat. The number of news articles used as an instrument for the monetary policy surprise may increase just because unconventional monetary policy is newsworthy and not because of a surprise related to monetary policy easing or tightening. However, because many unconventional policies were already discussed and announced (for example, in speeches) before actual FOMC meetings and we only use FOMC episodes (with just two exceptions), the potential bias stemming from the novelty of the policy tools should be small. Indeed, the mean and the variance of our fitted surprise measure in Figure 2 are more or less the same during the conventional policy period and the unconventional policy period.

Figure 2.
Figure 2.

Surprise-Fitted Values in the First State of TSLS By News Based Instruments

Citation: IMF Working Papers 2014, 152; 10.5089/9781498363563.001.A001

Sources: Bloomberg L.P. and authors’ calculations.

B. Benchmark Regressions

We run simple regressions for key bank-related asset prices based on our sample of monetary policy decision dates in the spirit of Bernanke and Kuttner (2005). Specifically, we regress daily bank stock returns and daily changes in spreads between bank corporate bond yields and the Treasury bond yields on the surprise component of the monetary policy announcements.7 By including a dummy variable for unconventional policy announcements, we allow for possible differences in the level effect (that is, the constant term) and the marginal effect (that is, the coefficient on surprise) between conventional and unconventional monetary policy announcements:

y z = β 0 + β 1 U M P d u m m y z + β 2 S u r p r i s e z + β 3 ( S u r p r i s e z * U M P d u m m y z ) + ε z , ( 1 )

where the subscript z corresponds to each FOMC announcement and UMPdummy takes value one for announcements during the unconventional policy period and zero otherwise. Unscheduled meetings are included in our sample of events but speeches (outside FOMC meetings) are in general excluded to ensure consistency as well as to contain any biases that might arise from pooling different types of events.

We are interested in the overall average effects of monetary policy announcements and thus use as many policy announcement events as possible. This is in contrast to most previous studies on unconventional monetary policy which focus only on a few events and look at asset prices movements for each event separately. In particular, we include FOMC days in which there was no policy change. This is because the surprise component can be negative or positive depending on market expectations before the FOMC meeting, even if no change in monetary policy was announced. Note that in these events without any policy changes, the expectations for the use of some specific policy tool (like asset purchases or forward guidance) cannot be identified but the size of the expectations for monetary easing can still be gauged by our measure of surprise.

We define the conventional policy period as the period up to July 31, 2007, because the FOMC held its first ad hoc meeting (conference call) on August 10, 2007, in response to the beginning of the financial turmoil.8 Thus, in the regression, the conventional policy period (i.e., UMPdummy = 0) covers the period from January 1, 2000, to July 31, 2007. The unconventional policy period (i.e., UMPdummy = 1) covers the period following the collapse of Lehman Brothers, that is, September 16, 2008 to October 31, 2012. Essentially all FOMC meeting dates are included except for September 12, 2001, corresponding to the FOMC meeting held one day after the September 11 attack on New York and Washington, D.C. 9

The first column of Table 2 shows the effect of conventional monetary policy surprises on bank stock daily returns. It is insignificant. Similarly, the simple average effect of monetary policy surprises in both conventional and unconventional policy periods (second column) is not significant. The third column shows the results for the full specification, allowing different constant terms and coefficients for both the conventional and unconventional monetary policy periods. Again, no significant result arises. This finding is consistent with previous studies. Bernanke and Kuttner (2005) report positive effects of surprise policy rate cuts on the market-wide stock index. However, as already discussed, a policy rate cut typically steepens the yield curve. Considering the positive effect of the policy rate cut and the negative effect from a steepened yield curve found in English, et al. (2012), the overall effect is ambiguous and we interpret our results in this light.

However, we found negative, significant effects of monetary easing on bank credit risk measured by the daily changes in the spread between the bank corporate bond yields and the Treasury bond yields for the similar maturities (Table 2, columns 4 to 12). For all maturities, an increase of about 0.08 basis point (bp) in spreads is observed for any unit of monetary easing surprise.10 Roughly speaking, 1 bp of monetary easing increases the credit spread by about 0.1 bp. Over the three years since the beginning of the easing cycle after the onset of the financial crisis, the policy rate came down from about 5 percent. Assuming that the cumulative easing from interest cuts, QE, and forward guidance is “equivalent” to 6 percent in interest rate terms, the impact on the credit spread would correspond to a 50 to 60 bp increase, which is substantial. From a study on bank funding cost and credit ratings (Ueda and Weder di Mauro, 2013), we know that a 60 bp funding cost increase is equivalent to a downgrade of almost 3 notches in the credit rating scale used by most credit rating agencies.

Since unconventional monetary policies aim at easing monetary conditions, they may be better compared only to monetary easing episodes in the conventional monetary policy period. Bernanke and Kuttner (2005) indeed find slightly different coefficient estimates in regressions using all monetary policy events and in those using only tightening episodes. This is consistent with anecdotal evidence that monetary tightening is often gradual, while easing is often executed more rapidly. Also, even if tightening and easing proceed at the same pace, their impact might be asymmetric.

On the basis of the monetary easing events only, the effect of monetary surprises is found to be slightly lower than the one found in the benchmark regression (results omitted). Note that all episodes in the unconventional policy period are defined as monetary easing episodes. As for the conventional policy period, an FOMC announcement is classified as an easing episode if the expected change in the one-year-ahead futures rate is negative. The expected change is defined as the actual change in policy rates minus the surprise change.

Table 3 shows the results using our news-based surprise measures as instruments in two stage least squares (TSLS) regressions. Specifically, in the first stage, logarithms of the before-after ratios of the number of total news, positive news, and negative news references are used as three regressors, substituting for the one-year-ahead three-month Eurodollar futures rate. 11 Then, the fitted value is used as our Surprise variable in the key regression (1). The reported standard errors account for the two stage least squares estimation. To avoid any bias which may stem from the ambiguity in the search terms used to identify “positive” or “negative” surprises, we only consider monetary easing episodes in the conventional policy period and the whole sample of events in the unconventional period.

The results are broadly consistent with those of the benchmark regressions. Monetary policy easing surprises are associated with a deterioration of bank credit risk at the 1–3 year and 3–5 year maturities, but statistical significance is lost for the longer, 5–7 year, maturity. Again, we do not find any significant effect on bank stock returns.

C. Robustness Checks

The results are robust to changes in the sample of events (taking out observations when the surprise measure exceeds its sample mean by more than two-standard deviations) and to alternative definitions of the conventional and unconventional monetary policy periods (results omitted).12

As an additional robustness check, it is, however interesting to investigate whether the effects of unconventional policies have changed over time after the collapse of Lehman Brothers. Unconventional policies were explicitly or implicitly employed to calm down the acute financial turmoil, especially in 2009 and 2010, but the stress subsided as time went by. Again, rather than describing what happened in each event, we use regression analysis but compare each year after 2008 to the conventional policy period as previously defined (January 2000, to July 2007). The periods we consider are (i) September 16, 2008 to September 30, 2009; (ii) October 1, 2009 to September 30, 2010; (iii) October 1, 2010 to September 31, 2011; and (iv) October 1, 2011 to October 31, 2012.13

Table 4 shows the year-by-year regression results. During the first year after the Lehman collapse, the effects of monetary policy announcements are the same as in the benchmark regression. In other words, unconventional policies do not seem to have qualitatively different effects compared to conventional policies. Perhaps the crisis was so acute that the unconventional policies did not have much effect. Or, on the contrary, if non-linear dynamics are at play, unconventional policies may have prevented the economy from falling into a “bad equilibrium” that cannot be observed. In this case, the hypothesis of zero effect of unconventional monetary policy as a counterfactual is not correct. The linear effect that is not significantly different from zero may in fact conceal a larger impact. However, a formal technique to study such non-linear counterfactuals has not yet been developed.

Unconventional policies had qualitatively different effects in the second year after the Lehman collapse (October 2009 to September 2010). They had a significant negative effect on stock returns but improved medium-term (3–5 years) credit risk significantly. This suggests that bold policies may have lowered the default probability of banks but also involved weaker expectations for profitability, and is consistent with the view that some monetary policy measures worked as merely a life support system for distressed banks at that time. In the third year after the Lehman collapse, unconventional monetary policy, though it became less active, appears to increase both bank stock prices and credit risk. Lastly, in the fourth year, unconventional monetary policies seem to have lost any distinct effect on banks’ profitability and soundness and the results are similar to those of the benchmark regressions.

In summary, unconventional policies do not have robust specific effects, as those effects differ depending on the year (and surely depending on each event). However, the common effects of both conventional and unconventional monetary policies on bank credit risk remain significant over all sample years.

D. Euro Area and the United Kingdom

A natural question is whether the effects of monetary policy are different in other countries. To answer this question, we estimated similar regressions for the euro area (Table 5a) and the United Kingdom (Table 5b). As regards bank credit risk, the regression results show similar results to the ones obtained for the United States in that monetary easing surprises appear to cause a deterioration of bank credit risk. The magnitude of the effect for the United Kingdom is similar to that for the United States, while the magnitude for the euro area is about twice as large. In addition, for the euro area and the United Kingdom, bank stock prices fall with monetary easing, an effect not observed in the United States. Again, most of these effects are common to both conventional and unconventional policies.

The results from the two stage least squares estimation are a bit weaker. Like in the United States, we do not find a significant effect of monetary policy on bank stock returns. Credit risk in the euro area still deteriorates, while the effect is not significant anymore in the United Kingdom (results omitted).

E. U.S. Bank-level Regressions

This section examines whether the effects of monetary policy on bank stock returns and credit risk vary with individual bank characteristics such as asset size and capitalization. We use quarterly balance sheet data (i.e., total assets and the equity/assets ratio) for a balanced sample of 88 U.S. banks from the SNL Financial database.14 Total assets are divided by GDP to ensure stationarity. The variables are lagged by one quarter and then used as the additional controls in the regression. We also include interaction terms with the surprise measure and the unconventional monetary policy dummy. Note that bank balance sheet data are quarterly while monetary policy announcements typically occur once a month. To control for differences in the information content of balance sheet variables in different months within a quarter, we include a dummy variable for the first month of each quarter (Mon1 dummy) and another dummy variable for the second month of each quarter (Mon2 dummy). 15 The following equation is estimated with bank fixed effects.

y i z = γ 0 i + γ 1 M o n 1 d u m m y + γ 2 M o n 2 d u m m y + γ 3 U M P d u m m y 2 + γ 4 S u r p r i s e z + γ 5 ( S u r p r i s e z * U M P d u m m y z ) + γ 6 L a g ( A s s e t / G D P ) i z + γ 7 L a g ( E q u i t y / A s s e t ) i z + γ 8 ( L a g ( A s s e t / G D P ) i z * U M P d u m m y z ) + γ 9 ( L a g ( E q u i t y / A s s e t ) i z * U M P d u m m y z ) + γ 10 ( Surprise z * L a g ( A s s e t / G D P ) i z ) + γ 11 ( Surprise z * L a g ( E q u i t y / A s s e t ) z * U M P d u m m y i z ) + γ 12 ( Surprise z * L a g ( A s s e t / G D P ) i z * U M P d u m m y z ) γ 13 ( Surprise z * L a g ( E q u i t y / A s s e t ) z * U M P d u m m y z ) + v i z , ( 2 )

where subscript z denotes a FOMC announcement and i refers to a bank.

Table 6 shows the results.16 We are in particular interested in the coefficients on the Surprise measure and its interactions. Results from the fixed effect estimation are shown in columns 1 to 5 and those from two-stage-least-squares estimation where we use our news variable are shown in columns 6 to 10.

The results generally support the benchmark regression results—little effect of monetary easing on bank stocks. Conventional monetary policy surprises have no effect on bank stock returns (column 1) though unconventional monetary easing has a positive effect (γ5, around 0.08). Moreover, this effect is larger for larger banks (γ12, around 0.001). These effects come out strongly only in the third year after the collapse of Lehman Brothers (column 4). However, TSLS regressions indicate that these effects disappear once the endogeneity problems and the measurement problems introduced by the zero-lower-bound are accounted for by use of the news-based surprise variable.

Similar regressions are run on bank credit default swap (CDS) spreads. Bank CDS spreads are in principle a better measure of bank credit risk compared to bank corporate bond spreads from Treasury bond yields. This is because monetary easing may primarily alter Treasury bond yields without affecting bank bond yields. In the presence of market frictions that prevent arbitrage between the two markets, the observed increase in spreads may not reflect an increase in the bank risk premium. Such arbitrage failures may have happened in the few months immediately following the collapse of Lehman Brothers.

The CDS market for U.S. banks is relatively new (the earliest available data are from 2005) and notoriously thin in earlier years. We therefore focus on the period after the collapse of Lehman Brothers and on four banks only: Bank of America, JP Morgan, U.S. Bancorp, and Wells Fargo. The regressions are run separately for each bank using simple OLS.

Tables 7a (OLS estimation) and 7b (TSLS estimation) show robust evidence of a negative effect of monetary policy surprises only for U.S. Bankcorp. This does not disprove the results of the benchmark regressions. The sampled four banks are quite large compared to the average bank whose bond is included in the index of bank corporate bond spreads provided by Bank of America Merrill Lynch. In particular, perhaps with the exception of U.S. Bancorp, these banks are most likely protected by “too-big-to-fail” considerations. This would limit the downward movements of CDS spreads for these banks. In our view, finding a consistent and significant result for at least one of the largest banks corroborates the benchmark results for the average bank.

III. Panel Analysis on Balance Sheets and Income Statements of Banks

In this section, we use bank balance sheet data to shed light on the channels through which unconventional monetary policies affect bank credit risks. We focus on three channels: bank profitability, risk-taking, and efforts toward balance sheet repair. We consider the following dynamic panel regressions:

x i , t = α 1 x i , t 1 + α 2 M o n e t a r y P o l i c y t + α 3 B a n k C h a r a c t e r i s t i c s i , t + α 3 ( M o n e t a r y P o l i c y t * B a n k C h a r a c t e r i s t i c s i , t ) + α 4 C o n t r o l s t + ξ i , t , ( 3 )

where xi,t denote variables of bank i at time t, which are indicators for bank profitability, risk taking, and balance sheet repair (see below for details).

MonetaryPolicyt measures conventional aspects as well as unconventional monetary policy measures. Three specifications are considered. The first one includes “Taylor rule” residuals (the Taylor gap17) as a measure of the monetary policy stance. When the Taylor rule indicates that the interest rate should be below zero, the central bank may choose to employ unconventional measures (such as QE). In the regression, such measures are summarized by the change in the ratio of central bank assets to GDP. In addition, the regression includes a measure of the length of time during which the policy rate stayed below the Taylor rule rate over the previous 5 years, to represent prolonged periods of exceptionally low interest rates (in itself an unconventional measure). Because the Taylor gap by construction combines several variables that may affect banks in different ways, the second specification separately includes the effective federal funds rate, the inflation rate, and the output gap instead (the last two variables as Controlst). The slope of the yield curve computed as the difference between 10-year and three-month Treasury yields is added in a third specification to better capture the effect of forward guidance. All monetary policy variables in the three specifications, except the one measuring the length of time during which the Taylor gap is negative, are lagged by one period to address endogeneity issues that may result from monetary policy reacting to the banking sector issues.

BankCharacteristicsi,t corresponds to individual bank characteristics: equity-to-assets ratio, log asset size and an indicator variable for banks that are on the Financial Stability Board’s list of global systemically important banks. Both the equity ratio and the asset size variables are lagged by one period. We include the interaction terms of these bank characteristics with the monetary policy indicators to investigate possible heterogeneous effects across banks.

Controlst comprises the real growth rate, to control for the business cycle; the ratio of the cyclically adjusted government balance to GDP to control for fiscal policy; and the VIX to control for the stress in the financial system. We also include time dummies.

The dataset consists of quarterly balance sheet data for U.S. commercial listed banks from the SNL database and of U.S. macroeconomic data over the period 2007Q3-2012Q3. The full sample includes data for 614 banks. Because not all variables are available for all banks in every period, the sample composition varies depending on the variable of interest. We exclude observations that are three standard deviations away from the sample mean. For each regression, the panel is balanced by keeping only banks for which data are available for every quarter over the estimation period (see descriptive statistics provided in Table 8).

We employ the system GMM estimation method (Arellano and Bover, 1995; Blundell and Bond, 1998) to alleviate the endogeneity issues. The analysis uses the monetary policy variables as independent variables, assuming they “cause” the changes in the bank soundness indicators. However, the central bank actions since 2007 have been partly in response to problems in banks, so that they may not be truly independent. This system GMM estimator is consistent provided the instruments are valid, which is tested with a Sargan test. Lags of all variables (except the interaction terms) are used as instruments for the differenced equation. The number of lags used (and hence the number of instruments) varies according to the dependent variable and the sample size. The first, second, or third lag of the difference of each variable (except the interaction terms) is used as an instrument for the level equation. The estimator is computed in two steps (using the inverse of the covariance matrix of the moment vector from the first-step estimation as the weighting matrix in the second step). Standard errors are computed using the Windmeijer bias-corrected estimator (Windmeijer, 2005).18

The results still need to be interpreted with caution. Besides the influence of monetary policy, bank balance sheets have been affected by fiscal, financial, and other factors over the period. To limit a possible “omitted variable bias,” the regressions therefore also include a number of variables controlling for output growth, fiscal policies, and stress in the financial system, along with time effects.

A. Effects on Profitability

Unconventional monetary policies can have both positive and negative effects on banks’ profitability. On the positive side, banks can benefit from near-zero policy rates that reduce their funding costs. Quantitative easing policies supporting asset prices will also have positive valuation effects. On the negative side, a prolonged period of low rates accompanied by a flattening of the yield curve reduces revenues from long-term loans with variable rates, new loans and newly issued fixed income securities, compressing the interest margin of banks engaged in maturity transformation. This negative price effect on the interest margin may be partially offset by changes in the volume of loans or an increase in non-interest income (for instance, fees).

We attempt to disentangle those various effects of unconventional monetary policies by looking separately at banks’ net interest margin, the ratio of non-interest income over total assets, and the return on average assets. Figure 3 plots the three variables, averaged across all banks in the sample, over the sample period. After a sharp drop in 2007–2008, the net interest margin of U.S. banks has partially recovered but still falls short of its 2007Q3 level. The pattern of the return on assets is similar, with a decline at the beginning of the sample period and a recovery since 2009Q4.

Figure 3.
Figure 3.

U.S. Bank Balance Sheet Indicators

Citation: IMF Working Papers 2014, 152; 10.5089/9781498363563.001.A001

Sources: SNL Financial and authors’ calculations.

The regression results are reported in Table 9. An increase in central bank assets negatively affect banks’ net interest margin (column 1). The Taylor gap does not have a significant effect, possibly because of the zero lower bound constraint. However, if we allow for heterogeneous effects depending on bank characteristics (column 2), the Taylor gap comes out as significant for highly capitalized banks. Note that a negative Taylor gap implies monetary easing.

In the third column, we replace the Taylor gap by its components, that is, the effective federal funds rate, the inflation rate, and the output gap. In the fourth column, we further add the yield curve slope as a regressor. In both specifications, a reduction in the federal funds rate also has a negative and significant effect on the net interest margin (columns 3 and 4) especially for larger banks. Yet the effect is very small when compared to the sample mean of the net interest margin (Table 1) and so economically insignificant. This can be interpreted as the result of the two opposing forces described above: low rates reduce funding costs; but over time, revenues from new loans and debt securities also decline, offsetting the decline in funding costs.

The effects on the ratio of non-interest income to total assets are reported in columns 5 to 8. Both a negative Taylor gap and a low level of interest rates are associated with a higher ratio of non-interest income to total assets. The same is also true for a steep yield curve, which may be counterintuitive, as a steeper yield curve should imply higher margins and thus higher income from banks’ maturity transformation business. Yet, the length of time during which interest rates are below the Taylor rule has negative effects on non-interest income. Also, an increase in central bank assets reduces non-interest income. Overall, we cannot find clear effects of unconventional monetary policies on banks’ non-interest income, consistent with previous findings (e.g., English et al., 2012).

The overall effects of monetary policy on bank profits are reflected in the return on assets (columns 9 to 12). The return on assets increases with the decrease in interest rates, consistent with the effects found on non-interest income. However, this initial positive effect seems to be offset by the negative effect of every quarter during which the policy rate remains below the Taylor rate. Thus, the overall effect of unconventional monetary policies on bank profitability is negative or at best ambiguous. We do not find any evidence of heterogeneous effects as a function of banks’ size or capitalization.

B. Bank Risk-taking

The effect of MP-plus on bank risk-taking is theoretically less ambiguous than the effect on profitability. The relationship between short-term interest rates and bank risk-taking involves two reinforcing forces. Low interest rates along with lower asset price volatility encourage banks to reduce their demand for low-risk low-yield assets and increase their purchases of riskier assets offering higher returns (portfolio reallocation). This negative relationship between short-term interest rates and risk-taking is strengthened in a model of financial intermediation where banks, operating under limited liability and asymmetric information, can engage in costly monitoring to reduce the credit risk in their loan portfolios, and endogenously modify their capital structure in response to a monetary policy change (Allen and Gale, 2004, and Dell’Ariccia, Laeven, and Marquez, 2010). Under limited liability, a policy change that decreases banks’ profits, as unconventional monetary policies do to a limited extent, reduces the franchise value of banks and hence the incentive for monitoring the borrowers and investing prudently. A policy rate cut also reduces the incentive for banks to finance themselves with equity as a commitment device to prevent excessive risk-taking and decreases the cost of debt and deposits, so that leverage increases.

We examine bank risk-taking by looking at three different measures. First, we consider the ratio of risk-weighted assets to total assets. Risk-weighted assets are a weighted sum of a bank’s assets with weights determined by the riskiness of each asset according to banking regulations and the bank’s internal models. This measure is also used in De Nicolò, Dell’Ariccia, Laeven, and Valencia (2010). Yet, the ability of banks to manipulate this ratio by adjusting the risk-weights may bias the comparison across banks. Second, we look at the ratio of equity to total assets, which is inversely related to banks’ leverage. Third, we use the z-score, which is the ratio of the return on total assets plus the ratio of equity over total assets, divided by the standard deviation of returns over 12 quarters. It measures the number of standard deviations a return realization has to fall in order to deplete equity and so is inversely related to a bank’s probability of insolvency. A higher z-score is then interpreted as lower bank risk (De Nicolò, 2000).

The risk-weighted assets to total assets ratio increases with the period of low interest rates and a flatter yield curve (Table 10, columns 1 to 4). This is consistent with the portfolio reallocation and risk-shifting theories and the results from our event study. It is also consistent with previous empirical studies on the pre-crisis period (De Nicolò et al., 2010; Altunbas, Gambacorta, and Marqués-Ibáñez, 2010, and Dell’Ariccia, Laeven, and Suarez, 2013). Yet, the signs and very small magnitude of the coefficients on the interaction terms do not provide much support to the hypothesis that those effects are stronger for weak or poorly capitalized banks.

However, the equity ratio increases (i.e., the leverage decreases) with low interest rates, especially when kept so for a long period of time (columns 5 to 8). Likely because of this positive effect on bank capitalization, the overall effects on of bank distress measured by z-score turns out beneficial (columns 9 to 12). Note that a special caution may be warranted. While we included many control variables and time effects in the regressions, the monetary policy variables might still capture part of the effect of the ongoing financial reforms which directly affect banks’ capital requirements and hence banks’ leverage. We could not find good variables to control for the effects of those reforms.

C. Balance Sheet Repair

Lastly, we consider the effects of unconventional monetary policies on banks’ efforts to repair their balance sheets. On the asset side, balance sheet repair implies removing toxic assets and writing off loans whose beneficiaries are insolvent and incurring losses. When interest rates are very low, banks can however rollover existing loans or even extend new loans to nonviable firms at nearly zero cost. Therefore, banks tend to avoid repairing their balance sheets. Several empirical studies (e.g., Peek and Rosengren, 2003, and Caballero, Hoshi, and Kashyap, 2008) show the prevalence of “evergreening” practices among large Japanese banks in the 1990s. Moreover, the ongoing financial reforms and higher capital requirements may add to the perceived cost of recognizing any losses. On the liability side, banks can take advantage of lower term premium to issue longer-term debt to replace short-term debt, thereby extending the overall maturity of their liabilities. This reduces the risk of maturity mismatches (Stein, 2012).

We proxy banks’ efforts towards balance sheet repair by three measures. The first measure is the ratio of the provisions for possible losses on loans and leases (excluding provisions for possible losses on real estate owned) to total (gross) loans. The second one is the ratio of non-performing loans to total loans. The third one is the share of short-term debt in banks’ total borrowing.

Banks’ loan loss provision ratio declines with the expansion of central bank’s assets (Table 11, columns 1 to 4), although this relationship is weaker for larger banks. This suggests a risk of evergreening. An alternative view is that with unconventional monetary policies supporting economic activity, existing loans become more viable and hence need fewer provisions. We indeed find some evidence that non-performing loans are reduced with low interest rates and a positive output gap (columns 5 to 8). On the liability side, we find a decrease in the short-term debt ratio when policy rates decrease (columns 9 to 12, taking significant interaction terms into account). This suggests that banks, especially large ones, do take advantage of lower rates to extend the maturity profile of their debt.

IV. Conclusion

We examined the effects of unconventional monetary policy on banking sector soundness. We could not find clear supporting evidence for the common perception that unconventional monetary policy helped banks. Rather, we find some evidence for heightened medium-term risks, which is likely due to delayed balance sheet repair by banks. These findings are the result of two sets of analyses: an event study on bank stock valuation and credit risk, and panel regressions on bank level measures of profitability, risk taking, and balance sheet repair.

In the event study, using a novel instrument for monetary policy surprises, we find robust evidence that unexpected monetary policy easing tends to increase bank medium-term credit risk in the United States, the euro area, and the United Kingdom.

Using quarterly U.S. bank-level data after the crisis started, we find that unconventional monetary policy measures have ambiguous effects on banks’ profitability. They are also associated with a reduction in some aspects of bank risks, namely leverage and short-term debt ratios. However, they lead to a rise in the ratio of risk-weighted assets to total assets and may potentially delay balance sheet repair by smaller banks, for example, by allowing for an evergreening of non-performing loans.

A caveat is that the crisis-period data may require an analysis based on non-linear counterfactuals. We implicitly examine whether the effects are different from zero, but the economy could have taken much more negative paths without the support of unconventional monetary policy. Further technical developments are warranted to address this issue when analyzing crisis-period data.

Table 1a.

Summary Statistics for the Variables Used in the Event Study

Surprise is the daily change in one-year ahead three-month futures rates: Eurodollar futures for the United States, Euribor futures for the euro area, and Sterling futures for the United Kingdom. UMP dummy takes the value one for the unconventional monetary policy period, which is defined as after August 1, 2007. Total news ratio is the ratio of the number of English-language news articles reported in Factiva in three days after the monetary policy announcement to the number of news articles in three days before the announcement. Positive news ratio is calculated in the same way as the Total news ratio but contains only expressions indicating positive surprise for monetary easing. Negative news ratio is calculated in the same way as the Total news ratio with negative expressions only. Due to the smaller coverage of English language news for the ECB, in Positive and Negative news ratios, several outliers are observed and winsorized at 3 for the euro area.

article image
Table 1b.

Correlation Matrix for the Variables Used in the Event Study

article image
Table 1c.

Summary Statistics and Correlation Matrix for the Variables Used in the U.S. Bank-Level Event Study

article image
Table 2.

Benchmark Regression Using Surprise Measure Computed from Changes in 1-year Ahead 3-month Futures Rates

The dependent variable is either the daily bank stock return (%) or the daily change in yield spread (bp) between bank corporate bonds and Treasury bonds with the similar maturity. The events are FOMC announcements between January 2000 and October 2012, except for September 12, 2001. Also, two Jackson Hole speech dates are added for 2010 and 2012. The regressors are: Surprise, which denotes for the daily change in one-year ahead three-month Eurodollar futures rate; UMP dummy, which takes the value one for the unconventional policy period (defined in the row just above the column numbers); and the interaction term, Surprise * UMP dummy. T-statistics are presented in parenthesis based on robust standard errors: * denotes significance at the 10 percent threshold, ** at 5 percent, and *** at 1 percent.

article image
Table 3.

Two Stage Least Square Using News-based Instruments, Easing Episodes Only

The dependent variable is either the daily bank stock return (%) or the daily change in yield spread (bp) between bank corporate bonds and Treasury bonds with the similar maturity. The events are FOMC announcements between January 2000 and October 2012, except for September 12, 2001. Also, two Jackson Hole speech dates are added for 2010 and 2012. In the conventional period, only monetary easing episodes are used. The regressors are: Surprise, which is fitted value based on the first-stage regression of the daily change in one-year ahead three-month Eurodollar futures rate on the logarithms of after-before ratio of number of total news, positive news, and negative news (see text); UMP dummy, which takes the value one for the unconventional policy period (defined in the row just above the column numbers); and the interaction term, Surprise * UMP dummy. T-statistics are presented in parenthesis based on robust standard errors: * denotes significance at the 10 percent threshold, ** at 5 percent, and *** at 1 percent.

article image
Table 4.

Benchmark Regression, Year by Year

The dependent variable is either the daily bank stock return (%) or the daily change in yield spread (bp) between bank corporate bonds and Treasury bonds with the similar maturity. The events are FOMC announcements between January 2000 and October 2012, except for September 12, 2001. Also, two Jackson Hole speech dates are added for 2010 and 2012. The unconventional policy period is divided into four subperiods: the first year after the collapse of Lehman Brothers (columns 1, 5, 9, 13); the second year (columns 2, 6, 10, 14); the third year (columns 3, 7, 11, and 15); and the fourth year (columns 4, 8, 12, and 16). The regressors are: Surprise, which denotes for the daily change in one-year ahead three-month Eurodollar futures rate; UMP dummy, which takes the value one for the unconventional policy period (defined in the row just above the column numbers); and the interaction term, Surprise * UMP dummy. T-statistics are presented in parenthesis based on robust standard errors: * denotes significance at the 10 percent threshold, ** at 5 percent, and *** at 1 percent.

article image
Table 5a.

Events Study Results—Euro Area

The dependent variable is either the daily bank stock return (%) or the daily change in yield spread (bp) between bank corporate bonds and Treasury bonds with the similar maturity. The events are monetary policy official announcements between January 2000 and October 2012. The regressors are: Surprise, which denotes for the daily change in one-year ahead three-month Euribor futures rate; UMP dummy, which takes value one for the unconventional policy period (defined in the row just above the column numbers); and the interaction term, Surprise * UMP dummy. T-statistics are presented in parenthesis based on robust standard errors: * denotes significance at the 10 percent threshold, ** at 5 percent, and *** at 1 percent.

article image
Table 5b.

Events Study Results—United Kingdom

The dependent variable is either the daily bank stock return (%) or the daily change in yield spread (bp) between bank corporate bonds and Treasury bonds with the similar maturity. The events are monetary policy official announcements between January 2000 and October 2012. The regressors are: Surprise, which denotes for the daily change in one-year ahead three-month Sterling futures rate; UMP dummy, which takes value one for the unconventional policy period (defined in the row just above the column numbers); and the interaction term, Surprise * UMP dummy. T-statistics are presented in parenthesis based on robust standard errors: * denotes significance at the 10 percent threshold, ** at 5 percent, and *** at 1 percent.

article image
Table 6.

The U.S. Bank-Level Panel Regressions of Stock Returns

The dependent variable is the daily bank stock return (%). Events are FOMC announcements between January 2000 and October 2012, except for September 12, 2001. Also, two Jackson Hole speech dates are added for 2010 and 2012. The aggregate-level regressors are: Surprise, which denotes for the daily change in one-year ahead three-month Eurodollar futures rate in fixed effect estimation (columns 1-5); or, for the two stage least square specifications (columns 6-10), Surprise is the fitted value based on the first-stage regression of the daily change in one-year ahead three-month Eurodollar futures rate on the logarithms of after-before ratio of number of total news, positive news, and negative news (see text); UMP dummy, which takes the value one for the unconventional policy period (defined in the row just above the column numbers); and the interaction term, Surprise * UMP dummy. Coefficients for the level controls are not reported: bank fixed effects, constant, dummy for the first month in each quarter, and the dummy for the second month in each quarter. The bank-level regressors are: Lagged Asset/GDP ratio, Lagged Equity/Asset Ratio, and interaction terms between these bank-level variables and Surprise as well as UMP dummy. Triple interaction terms are also included. T-statistics are presented in parenthesis based on robust standard errors clustered at bank level: * denotes significance at the 10 percent threshold, ** at 5 percent, and *** at 1 percent.

article image
Table 7a.

The U.S. Bank-level Regressions of Credit Default Swap Spreads for Major Banks

The dependent variable is the daily change in CDS spread (bp) for each bank. Events are FOMC announcements after September 16, 2008, until October 2012. Two Jackson Hole speech dates are added for 2010 and 2012. The aggregate regressor is Surprise, which denotes for the daily change in one-year ahead three-month Eurodollar futures rate. The coefficients for the level controls are not reported: constant, dummy for the first month in each quarter, and the dummy for the second month in each quarter. The bank-level regressors are: Lagged Asset/GDP ratio, Lagged Equity/Asset Ratio, and interaction terms between these bank-level variables and Surprise. T-statistics are presented in parenthesis based on robust standard errors: * denotes significance at the 10 percent threshold, ** at 5 percent, and *** at 1 percent.

article image
Table 7b:

The U.S. Bank-Level Regressions of Credit Default Swap Spreads for Major Banks

The dependent variable is the daily change in CDS spread (bp) for each bank. Events are FOMC announcements after September 16, 2008, until October 2012. Two Jackson Hole speech dates are added for 2010 and 2012. The aggregate regressor is Surprise, which is the fitted value based on the first-stage regression of the daily change in one-year ahead three-month Eurodollar futures rate on the logarithms of after-before ratio of number of total news, positive news, and negative news (see text). The coefficients for the level controls are not reported: constant, dummy for the first month in each quarter, and the dummy for the second month in each quarter. The bank-level regressors are: Lagged Asset/GDP ratio, Lagged Equity/Asset Ratio, and interaction terms between these bank-level variables and Surprise. T-statistics are presented in parenthesis based on robust standard errors: * denotes significance at the 10 percent threshold, ** at 5 percent, and *** at 1 percent.

article image
Table 8.

Summary Statistics for the Variables Used in the Panel Regressions

article image
Sources: SNL; Haver Analytics; and IMF staff estimates.
Table 9.

Panel Regressions on Measures of Banks’ Profitability

The dependent variables are the net interest margin, the ratio of non-interest income to total assets, and the return on assets. The Taylor gap is the difference between the policy rate and the rate given by a standard Taylor (1993) rule. Different estimates of the Taylor gap produce different results (magnitude, sign and significance). To reduce bias that may result from using any specific estimate of the Taylor gap, we use an average of four possible measures of the Taylor rate. Cyclically adjusted government balances are annual series from the Fiscal Monitor. The coefficients on the time dummies are not reported. System GMM estimation (Arellano and Bover (1995) and Blundell and Bond (1998)) is used: * denotes significance at the 10 percent threshold, ** at 5 percent, and *** at 1 percent.

article image