The changing transmission mechanism of US monetary policy

The balance sheet channel operates through the net worth of business firms and arises from the problems of adverse selection and moral hazard (Mishkin 1995). The bank-lending channel emphasizes the role of banks in determining the supply of loans in an environment where information is not symmetric.

Most papers that study monetary transmission channels, specifically the bank-lending channel, consider a single sample period (such as Ramey 1993; Bernanke and Gertler 1995; Kashyap and Stein 1995, 2000; Kishan and Opiela 2000; Den Haan et al. 2007, among many others). Bernanke and Gertler (1995) did question the validity of some of the assumptions needed for the operation of the bank-lending channel since the early 1980s, although they did not empirically assess the changes. Dave et al. (2009), the working paper version of Dave et al. (2013), did, as a robustness check, analyze the post-1984 sample period and find that the strength of the bank-lending channel weakens for this period.

Contraction of bank loans reduces spending of firms and households that depend on bank loans. Capital market imperfections imply that some, perhaps most, agents cannot directly issue securities in imperfect capital markets. These agents depend on intermediated credit for external finance. See Fazzari et al. (1988). Also, bank loans are usually a precondition for bond issuance by firms (Gorton 2009), so without a bank loan firms may be shut out of capital markets altogether.

Bernanke and Gertler (1995) discuss the justification for why banks cannot easily replace the lost deposits with other source of funds. In contrast, Kashyap and Stein (1994) show that it is sufficient to argue that banks do not face a perfectly elastic demand for their open-market liabilities and, hence, central bank operations that shrink their core deposit base will force them to rely more on managed liabilities and also increases their cost of funds. The latter will shift the supply of loans inward and in turn will negatively affect bank-dependent borrowers and raise the external finance premium.

For a critical survey of SVAR analysis based on sign restrictions, see Fry and Pagan (2011).

Faust (1998) provides anecdotal and quantitative examples of the danger in restricting contemporaneous interactions among variables.

Canova and De Nicoló (2002) present a model based on an economy with limited participation to derive the signs of cross-correlation functions to use as sign restrictions. For example, a monetary disturbance generates a positive contemporaneous comovement between output and the price level, between the price level and real money balances, and between real money balances and output. A technology disturbance, on the other hand, would generate a negative contemporaneous comovement between output and the price level, and between the price level and real money balances, but a positive contemporaneous movement between real money balances and output. The various sign restrictions are sufficient to distinguish between monetary, fiscal, and technology shocks.

As a caveat, we assume the structural impact matrix remains the same when shutting down a channel. We do so to ensure the interpretation of shocks does not change in a fundamental way when comparing to the benchmark. This is somewhat analogous to Fry and Pagan (2011) arguing that it is important for interpretation to fix the structural impact matrix when considering sign restrictions in order to avoid mixing different structural models. However, it means that we are really only comparing the differences in dynamic responses, not impact responses, when shutting down a channel. We note that this approach is in contrast with some other studies that consider shutting down transmission channels, such as Ludvigson et al. (2002), in which the structural impact matrix is recalculated when treating a given variable as exogenous.

Such reduced-form counterfactual analysis is possibly susceptible to the Lucas Critique in the sense that a change in structural parameters related to policy might change all of the reduced-form VAR parameters. However, we assume that the changes in the reduced-form VAR parameters would be relatively small, an approach implicitly and sometimes explicitly taken in other studies that employ reduced-form counterfactuals. As empirical support for our argument, we note the results in Liu and Morley (2014) that show reduced-form parameters for the “private-sector” equations in a three-variable VAR of the US economy do not appear to change significantly at the same time as parameters for the policy equation. But we acknowledge that a counterfactual based on setting certain SVAR parameters to zero while assuming the remaining parameters are unchanged may not be as relevant or informative about the importance of a given channel as a counterfactual based on a fully specified structural model in which the channel can be shut down by changing deep structural parameters that determine its importance for the macroeconomy.  

Instead of using total bank loans, we opt for the “mix” variable, which is constructed as the ratio of total bank loans to the sum of bank loans and commercial paper issuance. We refer readers to Kashyap et al. (1993) for the full argument as to why the “mix” variable is better than total bank loans in identifying the bank-lending channel.

Many studies in the monetary policy shock literature ignore this issue and proceed to estimate models in levels. See, for example, Bernanke and Blinder (1992), Eichenbaum and Evans (1995), and Leeper et al. (1996).

Unit root test results are available upon request. We keep interest rate variables in levels because the evidence for unit roots is borderline, and it is standard to treat them as stationary in the SVAR literature.

We also considered a shortened second subsample (1984Q1–2008Q3) to avoid the recent zero-lower-bound period. The results are qualitatively similar; hence, we only report results for the longer subsample below.

Canova and Paustian (2010) believe that being too agnostic in the identification process may have important costs for inference. They advocate imposing enough sign restrictions to make the results of monetary SVAR analysis meaningful since monetary shocks are typically considered a minor source of contemporaneous output growth and inflation fluctuations. Disturbances with small relative variability and with an insufficient number of restrictions may lead to mismeasurement in transmission properties.

The nominal exchange rate here is an index, where an increase in the exchange rate is an appreciation of the US dollar, while a decrease is a depreciation of the US dollar.

There is probably less consensus in the literature regarding this particular cross-correlation sign restriction since earlier studies of the bank-lending channel often have a hard time finding convincing empirical evidence that support the idea that a decline in aggregate bank lending should follow a contractionary monetary policy shock, as predicted by theory (see discussion in Gertler and Gilchrist 1993). However, Dave et al. (2013), using a factor augmented VAR (FAVAR) approach, show that aggregate loans do decrease in response to a contractionary monetary policy shock. Here we take the view that presupposes the existence of the bank-lending channel in aggregate by imposing this particular restriction in our benchmark model. But we have also considered models where we remove this restriction. Even in the absence of the restriction, we still find candidate decompositions that produce very similar results to what we present in this section.

The bootstrapped bands were constructed conditional on the selected candidate decomposition using the MT approach.

This delayed overshooting feature of the nominal exchange rate is not uncommon in the empirical literature. See, for example, Eichenbaum and Evans (1995).

Den Haan et al. (2007) argue that different types of loans in a bank’s loan portfolio react differently in response to monetary policy shocks. Real estate and consumer loans are most sensitive, and both decrease in response to a contractionary policy shock, while commercial and industrial loans tend to increase instead. As robustness, we make use of the Call Report data provided by the authors to see whether different types of loans behave differently within our empirical estimation framework. We find qualitatively similar results to our benchmark model regardless of the type of loan considered, and our general conclusion regarding the strength of the bank-lending channel across the subsamples remains consistent.

Boivin et al. (2010) estimate their model over the sample periods of 1962:1–1979:9 and 1984:1–2008:12.

Even though there are few studies that have shown an increase in the strength of the interest rate channel for the USA, Angeloni et al. (2003) find that the interest rate channel is important and is the dominant transmission channel of monetary policy in most Euro countries using data from the 1980s to the 1990s.

Chen (2004) finds that financial constraints among firms have become less severe. Brady (2011) reports the statistical and economic significance of the consumer loan supply effect has weakened over time.

It should be noted that a decline in the bank-lending channel of the monetary transmission mechanism does not necessarily imply a decline in the credit channel of monetary policy transmission. The credit channel includes various other avenues through which monetary policy shocks could influence aggregate output, such as the balance sheet channel or bank capital channel. Indeed, after the Great Recession, there has been a resurgence of interest in the financial accelerator framework of Bernanke et al. (1999). Brady (2011) also finds suggestive evidence of a strengthening of the balance sheet channel. However, our results still make it clear that the bank-lending channel, as the traditional center of the credit view, does not seem to operate as it did prior to the early 1980s.

There is much debate in the literature regarding the causes of the Great Moderation (shock vs. propagation or good luck vs. good policy), particularly for output growth. Giannone et al. (2008) provide an excellent summary of the debate. Our counterfactual analysis approach using SVAR is similar to that employed by Stock and Watson (2002), and Ahmed et al. (2004), among many others, and we obtain similar results that suggest shocks are the main cause for the reduction in output growth volatility. Using a theoretical approach through the estimation of dynamic stochastic general equilibrium (DSGE) models, many authors (such as Lubik and Schorfheide 2004; Boivin and Giannoni 2006) find more support for the good policy hypothesis. Recently, Benati and Surico (2009) have also suggested that SVAR methods for studying the Great Moderation may lead to misinterpretation of the good policy explanation as being due to good luck.