Central Bank Credibility, Monetary Policy, and Bank Capital: what
role in affecting liquidity decisions of banks?
By Bas Butler Student ID: 6040624
July 2014
Presented to the Faculty of Economics and Business University of Amsterdam
In Partial Fulfilment of the Requirements for the Degree of
Master of Science in Economics – Specialization in Monetary Policy and Banking
Academic supervisor:
dr.
R.E. (Razvan) Vlahu
Faculty of Economics and Business Department of Macro & International Economics
Abstract
While the literature typically considers the bank risk-taking channel through the impact of changes in policy rates on the risk-taking behaviour of banks, this study examines the impact of the degree of central bank credibility on bank risk-taking in terms of liquidity risk. In addition, this study aims to shed light on how both policy rates relative to banks’ target rates of return and the level of bank capital relative to bank assets affect liquidity imbalances in the banking sector. This study uses quarterly data on the U.S. economy and estimates a simple distributed lag model with the period of analysis covering 1994Q1-2006Q4. The main findings of this study can be summarized as follows: (i) banks take on more exposure to liquidity risk against the background of a higher degree of central bank credibility; (ii) lowered policy rates relative to banks’ target rates of return induce banks to take riskier positions to liquidity risk, and the results indicate that is effect is more pronounced in the case of favourable financial conditions, and weaker otherwise; (iii) given higher levels of bank capital relative to bank assets, banks tend to hold less liquid assets against their liquid liabilities, and thereby this results is supportive of the “risk absorption theory”.
Table of Contents
1. Introduction ... 4
2. Literature Review ... 6
2.1 Review of central bank credibility and the conventional link between monetary stability, credibility, and financial stability ... 6
2.2 The bank risk-taking channel of monetary policy ... 7
2.3 Bank risk-taking and liquidity ... 10
2.4 Liquidity risk and liquidity risk management of banks ... 11
3. Methodology ... 12
3.1 The econometric model: baseline specification ... 12
3.2 Extensions ... 17 3.3 Estimation method ... 18 3.4 Descriptive statistics ... 19
4. Results ... 20
4.1 Baseline results ... 20 4.2 Robustness analysis ... 224.2.1 Extending the period of analysis ... 22
4.2.2 Alternative proxies ... 23
4.2.3 Controlling for non-stationary series ... 24
4.2.4 Adjustment costs ... 24
4.3 Extensions ... 24
4.3.1 Asset price dynamics and the “financial accelerator” mechanism ... 25
4.3.2 Bank Capital and liquidity imbalances: skin in the game vs. risk absorption ... 26
4.3.3 A state-dependent bank risk-taking channel ... 27
5. Conclusion ... 27
References ... 31
List of Figures
Figure 1 Liquidity Mismatch Indicator ... 34
Figure 2 Central Bank Credibility ... 34
Figure 3 Taylor Rule Gap ... 35
Figure 4 VIX ... 36
Figure 5 Slope Yield Curve ... 36
Figure 6 TED Spread ... 37
Figure 7 Bank Capital ... 37
Figure 8 House Market Returns ... 38
1. Introduction
According to Brunnermeier & Sannikov (2014a), “the vulnerability of the financial system depends on the degree of leverage in the economy as well as on its liquidity
mismatch1: the difference between the difficulty of selling assets in crisis, and the reliability
of access to funding in such times”. The recent financial crisis has highlighted the importance of the latter as this crisis is in essence a banking liquidity crisis (Banque de France, 2008). To get a more concrete sense of the severity of the recent crisis, Fleming (2012) stresses out that the Federal Reserve (Fed) provided around $1,599 billion of liquidity support to the financial system during the crisis.
Despite the apparent role of liquidity in the stability of the financial system, the dynamics of liquidity imbalances in the run up to the recent crisis have not yet been studied extensively elsewhere. Presumably, this lack of empirical studies on liquidity imbalances is related to measurement problems. However, in recent years two useful liquidity mismatch indicators have been proposed by Brunnermeier et al. (2011) and Berger & Bouwman (2009) that can be used to fill this gap in the literature.
Therefore, this study aims to contribute to the literature by examining how liquidity imbalances in the banking sector tend to build-up, with particular focus on the prevailing monetary environment. Thereby this study builds upon the strand of literature that has examined the bank risk-taking channel of monetary policy. Yet, whereas the literature typically considers the bank risk-taking channel through the impact of changes in policy rates on bank risk-taking, this study seeks to shed light on whether this channel operates, in addition, through the degree of central bank credibility.
Rationale for considering the effect of central bank credibility on bank risk-taking in terms of liquidity risk, is that the increased credibility over the recent decades may have reduced banks’ perceptions of risk regarding to liquidity risk owing to that a high degree of credibility signals macroeconomic stability through monetary stability (Borio & Lowe, 2002). More specifically, note that liquidity is a bank’s ability to meet both expected and unexpected cash flows and collateral needs without incurring unacceptable losses (BIS, 2008). Hence, banks determine their optimal level of exposure to liquidity risk against the assumed future outcome regarding to liquidity. Then, as higher credibility signals macroeconomic stability, it may follow that banks assume a future outcome where it needs to hold less liquid buffers against their liquid liabilities without incurring unacceptable losses. Therefore, higher credibility may induce banks take on more exposure to liquidity risk – a “credibility effect”. Riskier liquidity positions resulting from improved fundamentals is not per se bad. However,
1 This study uses the notions of liquidity mismatch and liquidity imbalances interchangeably. Both refer to a bank’s level of exposure to liquidity risk.
as Borio (2005) argues that the increased credibility also has made it less likely that signs of unsustainable economic expansion emerge first in rising inflation, higher credibility may give banks a false optimistic feeling about the future.
In addition, while the literature mainly examines Rajan’s (2005) arguments about the “search for yield effect” on bank risk-taking in terms of credit risk, this study explores whether this particular effect applies to liquidity risk as well. Specifically, it is examined whether a decrease of the policy rate relative to banks’ target rates of return, induces banks to increase their exposure to liquidity risk through a search for yield by banks. Finally, as there exist no consensus in the literature on the effect of bank capital on liquidity imbalances in the banking sector, this study aims to contribute to this debate by assessing the effect of increased bank capital on liquidity imbalances.
All together, the main hypothesis of this study is that banks take on more exposure to liquidity risk against the background of a higher degree of central bank credibility. In addition, it seeks to shed light on the dynamics of liquidity imbalances in the banking sector with respect to both monetary policy and bank capital. In order to examine these questions, a simple distributed lag model will be estimated at quarterly frequency for the U.S. economy with the period of analysis covering 1994Q1-2006Q4.
The obtained results from the econometric model statistically confirm the hypothesis of this study, meaning that banks do take riskier position to liquidity risk when the degree of central bank credibility has increased. Moreover, it is found that Rajan’s (2005) arguments about the “search for yield effect” apply as well to bank risk-taking in terms of liquidity risk. The robustness of both findings is tested along several dimensions and the generated results remain supportive of both findings. In addition, it is found that the “search for yield effect” is highly state-dependent, and thereby confirming the ideas of Borio & Zhu (2008). This result implies that the “search for yield effect” on liquidity imbalances is more pronounced in an environment of favourable financial conditions, and weaker otherwise. Finally, with respect to bank capital, the results are supportive of the “risk absorption theory”; that means, that by requiring banks to hold higher capital levels, banks tend to increase their exposure to liquidity risk.
Based on these empirical findings, this study draws the following policy implications. First, monetary authorities can eliminate the problem of the “credibility effect” by adopting Borio’s (2005) suggestions that monetary policy should respond more symmetrical over the financial cycle. Second, the results found in favour of the “search for yield effect” provides prima facia evidence that monetary policy can be used to curb liquidity imbalances in the banking sector. Third, the obtained evidence supportive of the “risk absorption theory” implies that prudential policy should take into account the interaction between capital and liquidity buffers of banks.
The remainder of this study is organized as follows. Section 2 reviews the existing literature on central bank credibility, the bank risk-taking channel, and liquidity risk management of banks. Section 3 outlines the methodological framework and describes the dataset. Section 4 reports the findings of the baseline specification, the robustness analysis and the extensions. Finally, section 5 concludes and draws implications for policy.
2. Literature Review
This section starts by reviewing the notion of central bank credibility, and the conventional links between monetary and financial stability. Subsequently, it surveys the bank risk-taking literature and relates it to central bank credibility. Finally, by exploring the strand of the literature that examined liquidity risk management of banks, it seeks to shed light on the relationship between central bank credibility and bank risk-taking in terms of liquidity risk.
2.1 Review of central bank credibility and the conventional link between
monetary stability, credibility, and financial stability
Over the recent two decades, the degree of central bank credibility in advanced economies, commonly defined as the absolute difference between the central bank’s plans and the public’s belief about those plans, has substantially increased given that monetary authorities delivered low and stable inflation in line with their mandate (Blinder, 2000; Cuikerman & Meltzer, 1986). Therefore, an environment of low and stable inflation, i.e. monetary stability, results from high degree of central bank credibility.
As the public’s beliefs about the central bank’s plans are getting more aligned with the actual central bank’s goals, it implies that inflation expectations of economic agents have become increasingly well anchored. A corollary of well-anchored inflation expectations is that movements in actual inflation tend to have dampened (Ball, 2006). In addition, along the lines of the Ball-Mankiw-Romer theory, lower trend inflation decreases the slope of the Phillips curve (Ball et al., 1988). As a result, inflation has become less sensitive to output gaps. In this sense, increased central bank credibility might have made it less likely that unsustainable economic expansion appear first in rising inflation (Borio, 2005).
After all, this is exactly what central banks perused since a higher degree of central bank credibility (i) makes disinflation less costly in terms of output losses; (ii) helps to keep inflation low once it is low; (iii) gives a central bank greater tactical flexibility; (iv) helps to gather public support for central-bank independence; and (v) makes it easier to defend its currency (Blinder, 2000). All together, the gain in central bank credibility is a powerful and
essential tool in the conduct of monetary policy, and leads to greater efficiency in terms of resource allocation.
Besides these potential gains, the conventional wisdom regarding the link between financial and monetary stability is that monetary stability tends to promote financial stability (Bordo et al., 2000). Hardy & Pazarbasioglu (1999) find empirical evidence supporting this view by stressing out that a high degree of volatility in inflation significantly increases the likelihood of a financial crisis. In addition, Demirguc-Kunt & Detregiache (1997) find that the probability of a financial crisis is a positive function of the absolute level of inflation in a country. In that sense, increased central bank credibility contributes to the stability of the financial system. Yet, the experience of the recent financial crisis made it clear that financial imbalances and financial instability can arise in an environment of monetary stability. This idea is not new. In the run up to the recent financial crisis Borio & Lowe (2002) and Borio (2005) already warned that changes in the monetary and financial regimes over the recent decades have made the current environment more vulnerable to the build-up of financial imbalances. In order to explore how financial imbalances can develop in an environment of monetary stability, the next subsection examines the bank risk-taking channel of monetary policy. Rationale for only considering the role of monetary policy is that this study is primarily concerned about the channels through which monetary policy can trigger financial imbalances in an environment of monetary stability.
2.2 The bank risk-taking channel of monetary policy
As mentioned above, increased central bank credibility has made it less likely that signs of unsustainable economic expansion appear first in rising inflation. On the other hand, financial liberalization has made it more likely that these signs emerge first as excessive increases in both asset prices and credit. Thus, the control of inflation reduces the need to tighten the monetary environment against the background of financial imbalances; that is, it has become more likely that financial imbalances develop in an environment of monetary stability. This is what Borio (2005) refers to as the “paradox of credibility”.
Given the “paradox of credibility”, the channel through which monetary policy may contribute to the build-up of financial imbalance is the bank risk-taking channel of monetary policy, which gained great popularity after the financial debacle of 2008. According to Borio & Zhu (2008), the bank risk-taking channel mainly operates through “the impact of changes in policy rates on either the risk perceptions or risk tolerance of banks”. For example, a reduction in the policy rate may increase collateral values and profits, which tends to reduce banks’ perceptions of risk and induce banks to increase risk-taking.
Turning to the empirical evidence, the impact of changes in policy rates on different proxies for bank risk-taking has been studied quite extensively. For instance, Dell’Ariccia et
al. (2013) find evidence in favour of a risk-taking channel of monetary policy for the U.S. banking system; that is, that policy rates that are “too low for too long” can induce the banking system to increase risk-taking. The authors’ identification of risk-taking is based on the risk rating of the bank’s loan portfolio and they find a negative relationship between the policy rate and risk rating of the bank’s loan portfolio. Similarly, Jimenez et al. (2007) provide additional evidence along the same lines using ex ante loans characteristics and ex-post loan performance. Nevertheless, the literature is relatively modest on the relationship between policy rates and bank risk-taking in terms of liquidity risk. For instance, Adrian & Shin (2010) find that a decrease in policy rates is generally associated with a higher growth rate of short-term liabilities. But they do not derive the implications for banks’ liquidity mismatch, as they do not take the dynamics of banks assets into account.
In addition, a second set of effects operates through the relationship between market rates and target rates of return (Rajan, 2005). A change in the policy rate relative to target rates of return may influence both the risk incentives and risk aversion of banks. Target rates of returns can be sticky given quasi-fixed costs of banks (e.g., labour costs) and compensation structures that were prevalent prior to the financial crisis. Hence, for example, a reduction in the policy rate relative to target rates may induce a “search for yield effect” and this is likely to increase the risk-taking of banks since a reduction in the policy rate reduces the returns especially on low risk investments. Empirically, Altunbus et al. (2010) find evidence supportive of the “search for yield effect” by approximating this effect through the difference between the actual policy rate and a benchmark interest rate on the average probability for a bank to go into default. However, the strand of the literature that examined the “search for yield effect” has not yet examined the implications for this effect on bank risk-taking in terms of liquidity risk. More on this particular relationship follows below.
Yet, while the literature typically identify the bank risk-taking channel through the impact of changes in policy rates on either risk-tolerance or risk perceptions of banks, it might be that this channel also operates through the degree of central bank credibility, and thereby monetary stability. The idea is that central bank credibility signals macroeconomic stability owing to that an environment of monetary stability results from high central bank credibility. As a result of the signalled macroeconomic stability, it may reduce the uncertainty that banks feel about the future (Borio & Lowe, 2002), and hence it may reduce banks’ perceptions of risk. Then, given banks’ perceptions of risk decreased, it can induce banks to take riskier positions – a “credibility effect”.
On reflection, riskier position resulting from higher credibility is not per se bad as it signals stability, meaning that fundamentals have improved. However, given the “paradox of credibility”, low and subdued inflation may not necessarily imply stability because signs of unsustainable economic activity are less likely to emerge first in rising inflation. Higher
credibility therefore may give banks a false optimistic feeling about the future. Moreover, due to the financial evolution in the 1980s and 1990s, the financial system transformed to a more complex and opaque system, and as a result it has become harder to measure risk exposures. For instance, as intermediation chains became longer, since it involves more players and markets, the informational content decreased (Shin, 2010).
The idea that protracted periods of stability endogenously lead to greater risk tolerance is closely related to Minsky’s (1992) financial instability hypothesis. In particular, Minsky states that endured prosperity, stability and growth lead to greater risk tolerance, which in turn may increase complexities, interdependencies and thereby increasing the fragility of the financial system. In Minsky’s view the endogenous acceptance of greater risk tolerance is accelerated by inflation. But, as mentioned above, higher credibility tends to dampen inflation movements. Yet, in the current landscape this process can be accelerated by the self-reinforcing link between liquidity and the bank risk-taking channel, as will be discussed below in greater detail.
The hypothesis concerning the “credibility effect” is somewhat similar to what Borio & Zhu (2008) call the “insurance effect”. More specifically, it is argued that a higher degree of monetary policy predictability, which results from credibility, may induce bank risk-taking owing to that banks may expect monetary policy to ease the monetary environment and act as a “lender of last report” in the event of a bad economic outcome. As a result, monetary policy generates insurance for risk-taking – an “insurance effect”. This effect is the results of what White (2009) calls the highly asymmetrical response of the Fed over the financial cycle. That is, “the combinations of the Fed’s refusal to lean against the build-up of financial imbalances and its eagerness to aggressively clean in the case of a financial downturn”. Hence, this effect underlines “classical” moral hazard issues resulting from higher credibility.
The empirical literature on the “credibility effect” is relatively scarce. Most recently, however, Montes & Peixoto (2014) find that in the case of the Brazilian economy, an environment of higher credibility reduces banks’ perceptions of risk, which they proxy by the amount provisioned by banks in relation to the expected losses in lending. In addition, they find that increased credibility leads to greater risk-taking in terms of a reduced credit spread. In contrast, Granville & Mallick (2009) find by using a sign-restriction-based VAR approach that there exists a pro-cyclical relationship between monetary and financial stability; therefore, they conclude that the idea that monetary stability may result in financial instability does not seem to hold. However, the authors simulate monetary stability and higher credibility using a negative inflation shock, which is not per se the result of higher credibility. More specifically, higher credibility implies that inflation expectations are better anchored, which tends to dampen inflation movements (Ball, 2006), rather than implying a necessary decline in inflation.
All together, it seems that higher central bank credibility may induce banks to increase risk-taking. Yet, the question remains how to proxy risk-taking of banks in the context of central bank credibility? As Montes & Peixoto (2014) aim to verify the functioning of the credit channel in Brazil, they propose to measure bank risk-taking in terms of the credit spread. Yet, the credit spread as measure for bank risk-taking can be criticized on the following ground. A decline in the credit spread may also result from an improvement in the quality of the borrower, and thus the implications for risk-taking are ambiguous
(
Dell’Ariccia et al., 2013).Alternatively, this study argues that the “credibility effect” is most apparent in terms of liquidity imbalances in the banking sector. Thereby, this proxy is somewhat similar to Montes & Peixoto’s (2014) proxy of the amount provisioned by banks because both say something about how banks feel about the future. Besides conceptual linkages, this study aims to fill the gap of a lack of empirical evidence on the build-up of liquidity imbalances in the banking system. This gap is likely to be related to measurement problems, however, in recent years some useful liquidity imbalance indicators have been proposed.
2.3 Bank risk-taking and liquidity
Before proceeding to the relationship between central bank credibility and bank taking in terms of liquidity risk, it is useful to briefly explore the link between bank risk-taking and the notion of liquidity.
According to Borio & Zhu (2008), the bank risk-taking channel is tightly linked to the concept of liquidity. They think of liquidity as the relative ease at which perceptions of value can be converted into purchasing power. The advantage of this definition is that it captures both concepts of liquidity: funding and market liquidity. The former refers to the risk of being unable to satisfy claims without impairment of financial capital, whereas the latter refers to the risk that an economic agent cannot offset a position at the market price (BIS, 2008).2 In that sense, funding liquidity and market liquidity can be thought of as an “external
funding constraint” and a “saleability constraint”, respectively (Kiyotaki & Moore, 2001). While the literature mainly takes liquidity as exogenous, Borio & Zhu (2008) argue that liquidity should be partly regarded as endogenous. Meaning that changes in both constraints, and thus the impact on liquidity, are not solely driven by changes in both profits and collateral values but depends as well on perceptions of risk. Therefore, lowered perceptions of risk may result in both higher bank risk-taking and increased liquidity through its effect on both constraints.
2 The advantage of emphasizing both concepts of liquidity risk is that the interaction of both can explain how liquidity risk suddenly materializes (Brunnermeier, 2009). For instance, funding liquidity risk materializes when assets can only be sold against a discount; that is, when market liquidity risk materializes.
2.4 Liquidity risk and liquidity risk management of banks
This subsection briefly explains both the concept of liquidity risk and how banks manage these risks with the aim of exploring the link between central bank credibility and the bank risk-taking in terms of liquidity risk. In addition, as this study also aims to examine the relationship between liquidity imbalances in the banking sector and both the monetary policy and bank capital, it provides very brief notes on both relationships.
To start, banks are inherently exposed to funding liquidity risk because of the maturity transformation role of banks in the financial system. That is, as banks traditionally borrow short and lend long, they are exposed to a maturity mismatch. Moreover, even if a bank has a duration-matched balance sheet, it remains exposed to liquidity risk because of informational asymmetries about the asset quality.For instance, regarding to funding liquidity risk, there exist an informational gap between a bank and its financiers about a bank’s solvency position. Moreover, given that banks are highly interconnected through the payment system, balance sheet linkages and cross-holdings of liabilities across banks3, funding
liquidity risk in a single bank can be propagated to more than one bank through these interlinkages in the presence of asymmetric information (Nikolaou, 2009). Therefore Nikolaou (2009) explains that the root of liquidity risk lies in information asymmetries. Because of the information asymmetries, liquidity risk tends to materialize in an environment of uncertainty owing to that increased uncertainty makes it harder for lenders to screen good from bad credit risk (Mishkin, 2009).
In order to manage liquidity risk banks engage in liquidity risk management. More specifically, banks evaluate their future need for funds to meet obligations and ensure sufficient liquid buffers to satisfy those needs in the future without incurring unacceptable losses (BIS, 2008). Put differently, given the assumed future outcome with regarding to banks’ liquidity needs, banks determine their optimal level of exposure to liquidity risk in the context of profit maximization. Then, as higher credibility signals macroeconomic stability, through monetary stability, it may follow that higher credibility lowers banks’ perceptions of risk regarding to liquidity risk. Accordingly, this may induce banks to increase their exposure to liquidity risk. Meaning that banks hold less liquid buffers against the funding liquidity of their liabilities. Given Borio & Zhu’s (2008) claim about the endogenous feature of liquidity, the decreased perceptions of risk in response to higher credibility, indeed increases liquidity to banks. As a result, banks may expose themselves to even more liquidity risk.
In addition, as the bank risk-taking literature finds evidence for the “search for yield effect” in terms of credit risk, this effect may apply as well to liquidity risk. Specifically, a decline in the policy rate relative to banks’ target rates of return represents a loss on holding
3
liquid assets and thus it may induce banks to substitute liquid assets away for more illiquid assets that generate a higher return.
Finally, the strand of the literature that examined the relationship between bank liquidity risk management and bank capital is relatively extensive. Nevertheless, there exist no consensus in this literature on this particular relationship. One the one hand, Diamond & Rajan (2001) argue that higher bank capital may lead banks to reduce their exposure to liquidity risk owing to that higher bank capital makes it harder for banks to commit to monitoring. This idea is akin to the “skin in the game effect”, which states that banks have more to lose when bank capital is relatively high and thus banks have fewer incentives to engage in higher risk-taking (Mishkin, 2009). On the other hand, Repullo (2004) argues that higher bank capital allows banks to absorb greater risk. For instance, higher bank capital reduces asymmetrical informational problems between a bank and its financiers, and thereby allowing a bank to take on more exposure to liquidity risk. Berger & Bouwman (2009) refer to this notion as the risk absorption theory”.
3. Methodology
In what follows, this section outlines the methodological framework used to examine the main questions of this study and describes the features of the main variables used in this study.
3.1 The econometric model: baseline specification
Given the main hypothesis of this study, that a higher degree of central bank credibility induces banks to increase their exposure to liquidity risk, the following baseline specification4 (1) is estimated.
𝐿𝑀𝐼! = 𝛽!+ 𝛽!𝐶𝐵𝐶!!!+ 𝛽!𝑇𝑔𝑎𝑝1!!!+ 𝛽!𝑂𝑔𝑎𝑝!!!+ 𝛽!𝑉𝐼𝑋!!!+ 𝛽!𝑆𝑙𝑜𝑝𝑒!!!+ 𝛽!𝑇𝐸𝐷!!!+ 𝜀! ,
(1)
Liquidity Mismatch Index (LMI) – This series represents the aggregated level of liquidity risk exposure in the banking system as proposed by Brunnermeier et al. (2011) and Bai et al. (2013), hereafter respectively BGK (2011) and BKW (2013).5 They construct the
Liquidity Mismatch Index (LMI) as the difference between the liquidity of bank assets minus
4 The final choice for the lag length is based on the Bayesian information criterion (BIC). See Table 1 in the Appendix. Moreover, Table 8 shows that all series are stationary, except the proxy for central bank credibility. If a regressor is non-stationary, this may pose a problem to the assumption of normal distributions of the t-statistic (Stock & Watson, 2011). Therefore, the robustness of the baseline results is examined by re-estimating the baseline specification in first difference rather than in levels. Furthermore, it is notable to say that the null hypothesis of no cointegration cannot be rejected using the Johansen tests for cointegration.
5 BGK (2011) provides the theoretical foundations for the LMI, whereas BKW (2013) deals with practical challenges to implement it.
the liquidity promised through bank liabilities, where each item on the bank balance sheet is multiplied by a specific liquidity weight. With respect to asset-side liquidity weights, the weights are a positive function of “the ease with which one can raise cash by selling the asset in times of distress”, while with respect to the liability-side liquidity weights, the weights are a negative function of “the reliability of the funding source in such times” (BGK, 2011).6
Therefore, one clear advantage of this approach is that it takes both market liquidity and funding liquidity into account, which is a desirable feature as liquidity risk tends to materialize due to the interaction of both notions of liquidity (Brunnermeier, 2009). In addition, whereas studies that consider both types of liquidity risk typically measure it as the ratio of very liquid assets over very liquid liabilities (see for example Altunbus et al., 2010 and Lichtenberger & Sørensen, 2007), the LMI takes more information into account by considering the total bank balance sheet, except for bank capital and fixed assets. More importantly, as Berger & Bouwman (2009) find that about half of the liquidity creation at commercial banks occurs via off-balance sheet commitments, i.e. a large part the liquidity mismatch of banks occur through off-balance sheet activities, it is essential to take a bank’s off-balance sheet items into account. This condition is satisfied by the LMI. Finally, an important property of the LMI is that it can be aggregated across banks to measure the liquidity imbalances in the banking sector, while liquidity measures such as Basel’s liquidity coverage ratio, do not possess this aggregation feature (BKW, 2013).
This approach of this study in constructing the LMI slightly deviates from that of BGK (2011) and BKW (2013) as the attached liquidity weights are not time varying. Besides that it would be time-consuming to construct time-varying weights that depend on haircut data from repo transactions and liquidity premiums, the time series for the LMI with and without time-varying weights follow each other closely in the run up to recent financial crisis (see BKW, 2013). Thus, little information will be lost by constructing the LMI with time-invariant liquidity weights, especially not as this study is primarily concerned about the build-up of liquidity imbalances. Moreover, this study uses both a funding liquidity premium and a market liquidity proxy as explanatory variables rather than including both in the dependent variable. As a result, the modified LMI is constructed as follows:
𝐿𝑀𝐼! = !𝜆!"𝑥!,!"− !!𝜆!!!𝑥!,!!!, (2)
where 𝐿𝑀𝐼! is the liquidity mismatch in the banking system; 𝜆 stands for the time-invariant specific liquidity weights for each balance sheet item7; and 𝑥 represent the corresponding
balance sheet item.
For robustness purposes, this study also measures liquidity imbalances in the banking sector by considering two alternative proxies proposed by Berger & Bouwman (2009), hereafter BB. The two proxies differ from another with respect to off-balance sheet activities. The BB’s (2009) LMI should be interpreted as the amount of liquidity created by banks rather than the amount of liquidity imbalances. Yet, while there are some modelling differences between the two approaches, such as that the BB’s (2009) LMI is less discriminating across the liquidity weights8, both do tell the same story as the BKG’s (2011) LMI can be interpreted
as the equal counterpart of the BB’s (2009) LMI.
The two considered LMI’s proposed by BB (2009) are constructed as follows:
𝐿𝑀𝐼! =1 2∗ 𝑖𝑙𝑙𝑖𝑞𝑢𝑖𝑑 𝑎𝑠𝑠𝑒𝑡𝑠!+ 𝑙𝑖𝑞𝑢𝑖𝑑 𝑙𝑖𝑎𝑏𝑖𝑙𝑖𝑡𝑖𝑒𝑠!+ 𝑖𝑙𝑙𝑖𝑞𝑢𝑖𝑑 𝑔𝑢𝑎𝑟𝑎𝑛𝑡𝑒𝑒𝑠! − 1 2∗ (𝑙𝑖𝑞𝑢𝑖𝑑 𝑎𝑠𝑠𝑒𝑡𝑠!+ 𝑖𝑙𝑙𝑖𝑞𝑢𝑖𝑑 𝑙𝑖𝑎𝑏𝑖𝑙𝑖𝑡𝑖𝑒𝑠! + 𝑙𝑖𝑞𝑢𝑖𝑑 𝑔𝑢𝑎𝑟𝑎𝑛𝑡𝑒𝑒𝑠!+ 𝑒𝑞𝑢𝑖𝑡𝑦!+ 𝑙𝑖𝑞𝑢𝑖𝑑 𝑑𝑒𝑟𝑖𝑣𝑎𝑡𝑖𝑣𝑒𝑠!) (3) 𝐿𝑀𝐼! = 1 2∗ 𝑖𝑙𝑙𝑖𝑞𝑢𝑖𝑑 𝑎𝑠𝑠𝑒𝑡𝑠!+ 𝑙𝑖𝑞𝑢𝑖𝑑 𝑙𝑖𝑎𝑏𝑖𝑙𝑖𝑡𝑖𝑒𝑠! − 1 2∗ (𝑙𝑖𝑞𝑢𝑖𝑑 𝑎𝑠𝑠𝑒𝑡𝑠! + 𝑖𝑙𝑙𝑖𝑞𝑢𝑖𝑑 𝑙𝑖𝑎𝑏𝑖𝑙𝑖𝑡𝑖𝑒𝑠!+ 𝑒𝑞𝑢𝑖𝑡𝑦!) (4)
Central bank credibility9 (CBC) – This variable is a proxy for the degree of central
bank credibility. According to Cukierman & Meltzer (1986), central bank credibility is defined as “the absolute value of the difference between the policymaker’s plans and the public’s beliefs about those plans.” Based upon these suggestions, Cecchetti & Krause (2002) construct an index of central bank credibility that takes into account the deviations of expected inflation from the central bank’s target level. The index is normalized between 0 and 1 in the following way:
𝐶𝐵𝐶! = 1, 𝑖𝑓 𝐸 𝜋! = 𝜋∗ 1 − 𝐸 𝜋! − 𝜋 ∗ 𝜋 − 𝜋∗ , 𝑖𝑓𝜋 < 𝐸 𝜋!) < 𝜋 0, 𝑖𝑓 𝜋 ≤ 𝐸(𝜋!) 𝑜𝑟 𝐸(𝜋!) ≥ 𝜋 , (5)
7 The asset and liability specific time-invariant weights are somewhat ad hoc. Yet, see BKW (2013) as they tested the robustness of the liquidity weights and find that their results remain similar under alternative weights.
8 For a complete overview of the used liquidity weights to construct the Berger & Bergman LMI see Table 3 in the Appendix. The series are provided by the personal website of Christa Bouwman under the topic data.
where 𝐸 𝜋! represent the expected annual inflation rate; 𝜋 refers to the upper limit and 𝜋 to the lower limit; and 𝜋∗ stands for the central bank’s inflation target rate. Cecchetti & Krause (2002) propose to set the upper limit at 20%. That means, that they believe that if inflation expectations exceed this upper ceiling, the central bank would lose control over inflation (de Mendonça, 2007). Yet, as Mendonça (2007) argues that a limit of 20% is too high, this study uses a more stringent upper limit at 4%. In addition, the lower limit is set at 0%; that means, that the tolerance interval is set equal to ±2%. Regarding the inflation target rate, Mishkin (2007) argues that the Fed is committed to keep inflation close to 2 per cent and that the public has come to understand this fact. To give an idea about the credibility index, note that the index equals 1 if inflation expectations are fully anchored, and decreases in a linear way if expectations differ from the announced target. Finally, the index assumes zero if inflation expectations are outside the tolerance interval.
Given the main hypothesis of study it is expected that central bank credibility is negatively related to the proxy for liquidity imbalances in the banking system (LMI).
Taylor rule gap10 (Tgap) – This series represents the difference between the actual
policy rate from a benchmark rate and is used to evaluate the relative stance of monetary policy. That is, the Taylor rule gap evaluates the monetary policy stance relative to a bank’s target rates of return. This variable is introduced in order to examine whether Rajan’s (2005) arguments about the “search for yield effect” apply as well in the context of liquidity risk management of banks. Rationale for considering the relative stance of monetary policy, rather than simply using the policy rate, is that in order to identify substitution effects one needs to consider relative prices. Too see this, assume a decrease in the policy rate. Then, ceteris paribus, the impact of this changed policy rate on liquidity decisions of bank is ambiguous, as it is unknown how this change is related to banks’ target rates of returns.
Altunbus et al. (2010) propose the following two alternative ways to proxy the relative stance of monetary policy: (i) the difference between the actual policy rate and that generated by a Taylor rule with interest rate smoothing (Tgap1); (ii) the difference between the policy rate and that generated by a Taylor rule without interest rate smoothing (Tgap2). Hence, note that the Taylor rule serves as a proxy for banks’ target rates of return.
𝑇𝑔𝑎𝑝! = 𝑖!− 𝑖!!" = 𝑖!− 1 − 𝛾 𝛼 + 𝛽! 𝜋!− 𝜋∗ + 𝛽!𝑥! + 𝛾𝑖!!!!" , (6) where parameter 𝛾 represents the degree of interest rate smoothing. Following Altunbas et al. (2010), this parameter is set at 0.85 under the first alternative and 0 under the second
10 The Federal Reserve provides the series for the effective federal funds rate; the Federal Reserve Bank of Cleveland provides the inflation series; and the OECB provides the series for the ouput gap.
alternative; α represent the steady-state equilibrium real interest rate, and it is assumed to equal 2 following Taylor (1993) and Dale (2012); and finally, the weights on inflation and output are set at 1.5 and 0.5, respectively, under both alternatives.
The baseline specification (1) uses the difference between the actual policy rate and a Taylor rule with interest rate smoothing (Tgap1) to proxy the relative stance of monetary policy. Along the lines of Rajan’s (2005) claims, it is expected that banks take riskier positions to liquidity risk when the policy rate decreases relative to the Taylor rule produced interest rate.
Output gap11 (Ogap) – This series controls for economic conditions. The output gap
is measured using real-time data and is defined as the difference between actual and potential output expressed as a per cent of potential output. The idea of including this variable is to examine the effect of economic conditions on the behaviour of liquidity risk management of banks. As periods of economic expansion may lower expectations of future financial distress, it is expected that a positive output gap induce banks to take on more liquidity risk.
Financial market volatility12 (VIX) – This variable reflects financial market volatility
and is approximated by the VIX, which measures market expectations of near-term volatility conveyed by S&P 500 stock index option prices. Rationale for including this variable is to control for liquidity conditions in the asset market. More specifically, as both haircuts are known to positively vary with measures of asset price volatility and haircuts are negatively related to the degree of asset liquidity (BKW, 2013), the VIX is an important factor in the conduct of liquidity risk management of banks. Therefore, it is expected that banks reduce their exposure to liquidity risk in the light of heightened financial market volatility.
Slope of the yield curve13 (Slope) – This variable reflects the profitability of the
typical maturity transformation function performed by banks measured by the slope of the yield curve (Albertazzi & Gambacorta, 2009; Altunbus et al., 2010). This variable is measured as the difference between the interest rate on a 10-year Treasury bond and the interest rate on a 3-month Treasury bill. As an increase in the slope reflects an opportunity for a bank to increase profits by mismatching the maturity of its balance sheet, it is expected that an increase in the slope induce banks to take on more exposure to liquidity risk.
The three-month LIBOR-Treasury Bill spread14 (TED) – This series controls for
funding liquidity conditions of banks. The literature proposes several measures to proxy funding liquidity conditions such as the TED spread and Libor-Overnight Indexed Swap
11The OECD provides the series for the output gap. One robustness exercise performed in this study is to extend the period of analysis up to 2014Q1, yet output gap data from the OECD is only available up to 2007Q4. Therefore, the output gap in this particular robustness check is measured using own calculations. The Federal Reserve Bank of St. Louis provides the series for real potential output and actual output.
12 The Chicago Board Options Exchange provides this series.
13 The series for the interest rates on both the U.S. 10-year Treasury bond and the 3-month Treasury bill are provided by Federal Reserve.
spread. Yet, both measures are imperfect as they are likely to be contaminated by default risk (Thornton, 2009). Alternatively, BKW (2013) propose to proxy funding liquidity conditions by the three-month Treasury-Overnight Indexed Swap spread, as it likely to be minimally contaminated by default risk and more closely aligned with the funding conditions of banks. Due to data availability problems, however, the TED spread is used as the main variable to proxy funding liquidity condition. It is expected that an increase in the TED spread results in lower liquidity imbalances in the banking sector.
3.2 Extensions
To test whether changes in asset prices affect liquidity decisions of banks, the baseline regression (1) is enriched by taking the evolution of asset price dynamics into account. The idea is that changes in asset prices may influence the quality of a bank’s assets through its effect on both borrowers’ collateral values and overall credit risk, and thereby affecting liquidity risk management of banks. This notion is consistent with the standard ‘financial accelerator’ mechanism of Bernanke & Gertler (1989).
𝐿𝑀𝐼! = 𝛽!+ 𝛽!𝐶𝐵𝐶!!!+ 𝛽!𝑇𝑔𝑎𝑝1!!!+ 𝛽!𝑂𝑔𝑎𝑝!!!+ 𝛽!𝑉𝐼𝑋!!!+ 𝛽!𝑆𝑙𝑜𝑝𝑒!!!+ 𝛽!𝑇𝐸𝐷!!!+ 𝛽!𝑃𝐻!!!+ 𝛽!𝑃𝑆!!!+ 𝜀! ,
(7)
Housing market returns15 (PH) – This variable controls for house price dynamics and
is measured by the real home price index. Along the lines of the standard ‘financial accelerator’ mechanism of Bernanke & Gertler (1989), increased house prices may induce banks to increase their exposure to liquidity risk as overall credit risk tends to decline in response to increased collateral values.
Stock market returns16 (PS) – This variable controls for stock price dynamics and is measured by the real stock price index. Both the rationale for including this variable and the expected relationship are equivalent to the series for housing market returns.
Subsequently, in order to complement the strand of the literature that has examined the role of bank capital on the liquidity decisions of banks, the dynamics of bank capital are taking into account by enriching the baseline specification (1) with bank capital.
𝐿𝑀𝐼! = 𝛽!+ 𝛽!𝐶𝐵𝐶!!!+ 𝛽!𝑇𝑔𝑎𝑝1!!!+ 𝛽!𝑂𝑔𝑎𝑝!!!+ 𝛽!𝑉𝐼𝑋!!! + 𝛽!𝑆𝑙𝑜𝑝𝑒!!!+ 𝛽!𝑇𝐸𝐷!!!+ 𝛽!𝐵𝐶!!!+ 𝜀! ,
(8)
15 The series are provided by the personal webpage of Robert Shiller; http://www.econ.yale.edu/~shiller/data.htm. 16 The series are provided by the personal webpage of Robert Shiller; http://www.econ.yale.edu/~shiller/data.htm.
Bank capital17 (BC) – This variable controls for the bank capital dynamics of the banking sector. Following Berger & Bouwman (2009), bank capital is measured as the ratio of total equity over total assets.
Finally, to test whether the effect of the relative stance of monetary policy (Tgap) on liquidity imbalances depends on financial conditions, i.e. it is state-dependent, the baseline specification (1) is enriched by including an interaction term between the Taylor rule gap and a time dummy that controls for financial conditions. The idea is that the impact of changes in policy rate relative to banks’ target rates of return is more pronounced in an environment of favourable financial conditions, and weaker otherwise (Borio & Zhu, 2008).
𝐿𝑀𝐼! = 𝛽!+ 𝛽!𝐶𝐵𝐶!!!+ 𝛽!𝑇𝑔𝑎𝑝1!!!+ 𝛽!𝑂𝑔𝑎𝑝!!!+ 𝛽!𝑉𝐼𝑋!!! + 𝛽!𝑆𝑙𝑜𝑝𝑒!!!+ 𝛽!𝑇𝐸𝐷!!!+ 𝛽!𝐷 + 𝛽!𝐷_𝑇𝑔𝑎𝑝1!!!+ 𝜀!,
(9)
Time dummy (D) – this variable represents a time dummy that controls for financial market conditions. The time dummy assumes value 1 in periods that features increased financial market volatility, defined as the periods when the VIX exceeds the threshold value of 20, and zero otherwise. This threshold value is chosen because it represents the sample average.
Time dummy*Taylor rule gap (𝐷_𝑇𝑔𝑎𝑝1!!!) – this seies represent the interaction term between the variable that controls for the relative stance of monetary policy and a time dummy variable that captures periods featured by financial market distress. Consistent to Borio & Zhu’s (2008) claims about the state-dependent nature of the impact of changes in policy rates relative to banks’ target rates of return, it is expected that the estimated coefficient on the interaction term is negative.
3.3 Estimation method
In the above-considered regressions, causation is presumably running from both sides. For instance, while a positive output gap is likely to increases liquidity imbalances, liquidity imbalances tend to reinforce the business cycle as well (Cornett et al., 2011). Moreover, even though that there exist no consensus in the literature on the relationship between bank capital and liquidity imbalances, it is very likely that higher liquidity imbalances tend to increase bank capital. This is attributable to that banks with a higher liquidity mismatch are likely to earn a higher return through their maturity transformation function (BKW, 2013). Finally, while risk premiums may affect the build-up of liquidity imbalances, risk premiums itself may be influenced by liquidity imbalances; for instance, greater liquidity imbalances effectively implies more liquidity creation by banks, which in
turn may lower risk premiums. In other words, risk premia and liquidity imbalances may reinforce another. This idea is consistent with the Brunnermeier & Sannikov’s (2014b) notion of the “paradox of volatility”.
Even though the above-considered specifications by coincidence indirectly control for endogeneity as the regressors are in lags, this study uses the generalized method of moments (GMM) in order to directly control for simultaneous causality. Regarding to the choice of the instrumental variables, it is required that the instruments do not directly affect the dependent variable to generate consistent estimators. To satisfy this restriction, the chosen instruments are lagged levels of the dependent variable and the explanatory variables. A standard Hansen J-test is performed to test the exogeneity requirement for the chosen instruments.
3.4 Descriptive statistics
18This subsection describes the features of the main variables used in this study in the case of the U.S. economy.
From Figure 1, it can be seen that at the beginning of the sample period, 1994Q1, the aggregated liquidity mismatch indicator (LMI) was negative around 0.2 trillion dollars. Recall that, following BKW (2013), a negative value indicates that the banking sector is more vulnerable to liquidity stress. Thereafter, liquidity imbalances continued to develop moderately up to the Dot-com crash. Then, after a brief period of calmness in the aftermath of Dot-com crash, there is a pronounced increase liquidity imbalances and it peaks during the recent financial crisis. In the recovery phase of the crisis, liquidity imbalances reversed, and this improvement coincides with the Fed’s liquidity injections.
Regarding to central bank credibility, Figure 2 shows that the degree of central bank credibility (CBC) started relatively low in the beginning of the sample period compared to the period before the financial crisis when inflation expectations were nearly fully anchored. During that period, the degree of central bank credibility increased rather steeply but similar to liquidity imbalances there was a fall in the degree of credibility during the Dot-com crash. At last, it is noteworthy to observe that while the repercussions of the financial crisis on the real economy were enormous and are still ongoing, inflation expectations remained relatively well anchored.
Figure 3 displays that the relative stance (Tgap) of the Fed was rather contractionary in the mid 1990s. Then, after LTCM crash in 1998, it can be seen that the policy rate started to decrease relative to the benchmark Taylor rule. Subsequently, in the aftermath of the Dot com crash, the relative stance became expansionary. This is likely to be related to that the Fed at that time was concerned about a possible unwelcome disinflation (Bernanke, 2010). As a
result, the Fed substantially lowered its policy rate. But as the Fed’s inflation forecasts were too low as inflation increased rather than decreased in 2002-2005 (Taylor, 2009), the relative stance became expansionary. By comparing the Figures 1 and 3, it can be see that the expansionary stance of the Fed during that period is followed by a pronounced increase in liquidity imbalances in the banking sector with some lags. With respect to the post-crisis period, it can be seen that the liquidity imbalances reversed against the background of a contractionary relative stance of monetary policy. Yet, it should be noted that the current contractionary stance is likely to be the result of the Fed’s policy rate being constrained by the zero lower bound.
Finally, from figure 7 it can be seen that banks significantly increased bank capital relative to bank assets in the run up to the recent financial crisis. More specifically, the bank capital ratio was above 10% in two years preceding the crisis. Then, given the Basel capital requirements, this means that banks were well capitalized. Moreover, as this study measures the bank capital ratio using total assets rather than risk-weighted assets, banks were even better capitalized than that Figure 7 shows in the years before the crisis, given the Basel capital requirements.
4. Results
This section analyses the empirical findings obtained from the econometric model. The parameters of each regression are estimated using GMM at quarterly frequency, the period of analysis runs from 1994Q1 up 2006Q419, and the explanatory variables are
instrumented by their lagged levels and lags of the dependent variable.20
4.1 Baseline results
The main results of the baseline specification (1) are reported in column (1) of Table 4 in the Appendix.
The estimated coefficient for the influence of the degree of central bank credibility (CBC) on the liquidity imbalances in the banking sector has the expected negative sign and is highly significant. Thereby, this result is supportive of the main hypothesis of this study that banks take on more exposure to liquidity risk against the background of higher central bank credibility. Yet, as the “credibility effect” is akin to the “insurance effect”, this result may also be attributable to “classical” moral hazard issues rather than that higher credibility induces bank risk-taking through its signalling of macroeconomic stability. Nevertheless, the
19 With exception of the regression in column (2) of Table 4.
essential point to understand is that while credibility leads to greater efficiency in terms of resource allocation, it also tends to favour the creation of expectations that may stimulate the build-up of liquidity imbalances in the banking system either through the “credibility effect”, or the “insurance effect”, or both.
Regarding to the impact of policy rates relative to banks’ target rates of return on bank risk-taking in terms of liquidity risk, the estimated coefficient on the relative stance of monetary policy (Tgap1) is positive and highly significant. That means that, ceteris paribus, if the actual policy rate is below what a standard Taylor rule with interest rate smoothing would call for, then banks do take on more exposure to liquidity risk. Quantitatively this result implies, for example, that if the policy rate is 100 basis points below the benchmark Taylor rule, then on average the liquidity imbalances in the U.S. banking system tend to increase by $0.145 trillion (i.e., LMI turns negative). Thereby, this result indicates that Rajan’s (2005) arguments about the “search for yield effect” also apply in the context of bank risk-taking in terms of liquidity risk.
Concerning the influence of economic activity on liquidity imbalances, the estimated coefficient on the output gap (Ogap) has the expected negative sign and is significant. Thus, periods of above normal output may lower expectations of future financial distress, which in turn may induce banks to increase their exposure to liquidity risk.
The estimated coefficient for the variable that controls for liquidity conditions in the asset market (VIX) has the expected positive sign and is significant. Thus, an improvement of liquidity conditions in the asset market can induce banks to increase their exposure to liquidity risk. Thereby, it follows that this result is consistent with the arguments made by BKW (2013).
The estimated coefficient on the variable that controls for the profitability of the typical maturity transformations function performed by banks (Slope) does not have the expected sign while it is highly significant. However, this result is not necessarily at odds with the theory. Too see this, consider the impact of an increase in the slope of the yield curve on a bank’s balance sheet. As banks fund themselves with short-term debt and lend out long-term, Altunbas et al. (2010) argue that a steeper yield curve determines an increase in bank profits. As a result, both the bank’s assets and its net worth increase in the next period. In terms of a bank’s LMI, the increased value of bank assets unambiguously raises the market liquidity of bank’s assets, whereas the bank’s funding liquidity promised through its liabilities remains unchanged because bank capital is not part of the latter in the BKW’s (2013) LMI. That means, that banks’ liquidity imbalances decline in response to an increase in the slope of yield curve one period earlier.
As will be discussed below, it is found that higher bank capital results in increased liquidity exposures of banks. Thus, as bank capital has increased in response to the steeper
yield curve, banks may increase liquidity imbalances after the first period. Therefore, the baseline specification (1) is enriched with two additional lags of the slope of the yield curve to examine this idea. Column (1) of Table 9 shows that banks indeed take on more exposure to liquidity risk after the first period. Yet, the cumulative effect of a steeper yield curve on the liquidity imbalances in the banking system remains positive; that is, liquidity imbalances decrease in response to a steeper yield curve.
Finally, the coefficient on the variable that controls for funding liquidity conditions of banks (TED) has the expected positive sign, yet this result is not statistically significant. In part, this may be related to that the TED spread is an imperfect measure to proxy funding liquidity conditions of banks. Though, this result is still suggestive of that banks reduce their exposure to liquidity risk against the background of deteriorated funding liquidity conditions.
Moreover, the result of the Hanson J-test suggests that the model is correctly specified and R2 indicates that the regressors are good at explaining how liquidity imbalances develop in the sample of data on hand.
4.2 Robustness analysis
In this subsection, the robustness of both the “credibility effect” and the “search for yield effect” are tested by extending the period of analysis; using a different proxy for liquidity risk exposures in the banking system; using a different specification of a Taylor rule; controlling for potential non-stationary problems by re-estimating the baseline specification in first difference rather than in levels; and finally, by controlling for potential adjustment costs that banks may face in altering their liquidity positions.
4.2.1 Extending the period of analysis
To examine the robustness of the baseline results, the period of analysis is extended up to 2014Q1. The obtained regression results are presented in column (2) of Table 4.
The estimated coefficient on the variable that evaluates the relative stance of monetary policy (Tgap1) remains positive in line with theory, but is no longer statistically significant. In retrospect, this result should be interpreted with care because the fact that liquidity imbalances improved during that period is likely to be related to the liquidity injections of the Fed. Moreover, the reason that the Taylor rule gap is positive in the post-crisis period is attributable to the fact that the actual policy rate is constrained by the zero lower bound. In addition, the ”search for yield effect” is likely to be less pronounced for banks in the post-crisis period as quasi-fixed costs decreased (e.g., lower labour costs related to retrenchment in the financial sector after the financial crisis).
The estimate for the influence of central bank credibility (CBC) remains to have the expected sign and is significant, but it should be noted that the strength of the “credibility
effect” decreased. In part, this is likely to be related to Ball & Mazumder’s (2011) finding of that inflation expectation remained well anchored during the recent financial crisis.
Furthermore, the results on the other coefficients are mixed; for example, while the estimated coefficient on the output gap (Ogap) has the expected sign and is highly significant, the estimated coefficient on the variable that controls for asset liquidity conditions (VIX) does not have the expected positive sign and is insignificant. In addition, compared to the baseline results, the R2 decreased significantly from 0.83 to 0.34. And the Hanson J-test indicates that the model is incorrectly specified as the null hypothesis that all the instruments are exogenous is rejected.
4.2.2 Alternative proxies
21As the literature proposes several different Taylor rules, the reliability of the baseline result for the “search for yield effect” is tested using an alternative proxy for banks’ target rates of return. The considered alternative is the Taylor rule without interest rate smoothing but with the same weights on the inflation gap and the output gap as under the baseline specification (1). As shown in column (3) of Table 4, the results are quantitatively similar with respect to the ones generated by the baseline specification (1). Hence, the results remain supportive of the “search for yield effect” in terms of liquidity risk.
The robustness of the baseline results is also examined by considering two alternative LMI measures proposed by BB (2009). While the first proxy includes off-balance sheet activities, the second proxy excludes those items (see literature review). This difference is exploited in order to examine the importance of off-balance sheet activities in mismatching liquidity. Before proceeding to the results, it is important to recall that the LMI proposed by BB (2009) measures the liquidity creation of banks rather than liquidity imbalances. However, both tell the same story since the equal counterpart of the BGK’s LMI can be interpreted as the supply of liquidity provided by banks (BKW, 2013).
Turning to the results, columns (1) and (2) of Table 7 show that the results are very similar to ones generated with the baseline specification (1), except for both controls on economic activity (Ogap) and financial market volatility (VIX). In addition, the estimated coefficients have the opposite sign compared to the baseline results, as was expected since the LMI of BB (2009) is more or less the equal counterpart of the BGK’s (2011) LMI. Hence, the baseline results are robust to different proxies for the liquidity imbalances in the banking sector. Moreover, by comparing columns (1) and (2) of Table 7, it can be seen that the magnitude of the estimated coefficients generated by taking off-balance sheet activities into
21 As an alternative to the TED spread this study also considered the three-month Treasury-Overnight Indexed Swap (TIOS) spread. However, due to data availability problems, the generated results were difficult to interpret. Consequently, the results are not reported in Appendix.
account are higher than by excluding those items. This provides prima facie evidence for the importance of taking off-balance sheet items into account.
4.2.3 Controlling for non-stationary series
Table 8 shows that all series are stationary using the DF-GLS test, except the series that serve as a proxy for central bank credibility. This may pose a problem to the assumption of normal distributions of the t-statistics. In other words, if a regressor has stochastic trend, then the non-normal distribution implies that hypothesis tests cannot be conducted as usual (Stock & Watson, 2011). Therefore, re-estimating the baseline specification with the series in first difference rather than in levels assesses the robustness of the baseline results.
The results of this robustness exercise are presented in column (1) of Table 5 and it can readily be seen that the estimated coefficient on credibility (CBC) has the expected negative sign and is slightly insignificant at the conventional levels. Hence, while somewhat less significant, this result is consistent with the baseline result for the “credibility effect”. Likewise, the estimates for the other regressors are in line with those generated by the baseline specification (1). Finally, it should be noted that the significance of all the other estimated coefficients substantially decreased. Yet, this was expected because with the dependent variable in first difference, information could be lost as well as explanatory power (Montes & Peixoto, 2014).
4.2.4 Adjustment costs
Finally, it may be that banks face adjustment costs in changing their liquidity positions. Suppose, for example, that a bank wishes to reduce its reliance on wholesale market funding by increasing its the share of deposits. Then, it may be that the bank has to offer higher interest rates to attract new deposits.
In order to control for potential adjustment costs, an autoregressive distributed lag model is estimated with one lag of the dependent variable (LMI). The obtained results are shown in column (2) of Table 9. It can be seen that all estimated coefficients are similar to the baseline results, except for the variable that controls for funding liquidity condition of banks (TED). In that sense, the baseline results are robust to the inclusion of potential adjustment costs. Finally, the estimated coefficient on the lagged dependent variable (LMI) is positive and highly significant.
4.3 Extensions
In this subsection, the results are extended in three dimensions. First, it is examined whether the “financial accelerator” mechanism plays a role in affecting liquidity imbalances in the banking sector. Second, the effect of bank capital on liquidity imbalances is assessed.
