What is the role of bank liquidity in determining commercial banks’ responses to monetary policy shocks? And second, what is the role of the information effect of monetary policy announcements in this context? These questions are motivated by the fact that, at the peak of the Great Financial Crisis (GFC henceforth), the Fed implemented (for that time) unconventional policies that relied on increasing transparency and public communication, such as forward guidance. Additionally, the Fed provided liquidity injections via large-scale asset purchases (LSAPs henceforth)1 that significantly expanded the amount of excess liquidity in the financial system. Commercial banks are at the core of the financial system and play a key role in the transmission of monetary policy to the real economy through several channels.

Among the various transmission mechanisms that have been analysed in the literature the bank-lending channel and the information channel of monetary policy are the most closely related to our research questions. First, the broad interpretation of the bank-lending channel2 predicts that, in response to a monetary tightening, liquidity constrained banks will contract their credit supply while leaving their securities portfolio unchanged. Arguably, the assumption of banks being at their liquidity constraint was largely valid until the early 1990’s. However, making this assumption when looking at more recent data turns into a more difficult task. As shown in figure 1, in the decades preceding the GFC the share of banks with high excess liquidity was always very low, and thus, largely in line with the assumption of the bank-lending channel. This share was below 20% at the beginning of the 1990’s and was also decreasing over time. However, the implementation of LSAPs to mitigate the effects of the GFC reversed this trend, and after 2008Q3 the share of cash-liquid banks reached a peak of almost 75% in 2011 and stabilized above 60% by the end of our sample.3

Second, the information channel of monetary policy, as proposed by Nakamura & Steinsson (2018), highlights that Fed announcements do not only affect beliefs about (the future path of) monetary policy “but also about other economic fundamentals.” That is, economic agents

are also updating their beliefs about the economic outlook in reaction to Fed announcements. Disregarding of the information content of Fed announcements leads to biased inference, as has been manifested in long-standing puzzles in the literature on monetary policy shocks. For example, Miranda-Agrippino & Ricco (2021) show that in reaction to a monetary contraction, established instruments (which do not account for the information effect) produce an expansion in industrial production and a decline in the unemployment rate.

The persistent changes in the financial system’s liquidity levels in the aftermath of the GFC, together with an increasing reliance on central bank communication policies, raises the question of whether banks’ reactions to monetary shocks are still in line with pre-GFC estimates. Specifically, we first ask whether banks’ reactions to monetary policy shocks depend on their cash-liquidity levels and, second, whether they also depend on the information effect from policy announcement.

In order to answer these questions, we employ a state-dependent, lag-augmented (panel) local projection instrumental variables (LP-IV) setup, which allows us to distinguish between liquid and illiquid banks, and also to control for large sets of bank-level characteristics. We further use an informationally-robust monetary policy instrument (MPI henceforth) following Miranda-Agrippino & Ricco (2021). This instrument has two significant advantages: First, it is orthogonal to both the central bank’s projections as well as past market surprises. And, second, we can differentiate between total, conventional and unconventional monetary policy shocks.4

Our results provide two key insights. First, monetary policy transmission crucially depends on banks’ individual liquidity levels, as well as the aggregate liquidity composition of the banking sector, i.e., our findings depend on the inclusion of post-2007 data. In the pre-GFC sample, we find supporting evidence for an active bank-lending channel independently of a bank’s liquidity state, i.e., we identify homogeneous bank responses. In response to a 100bp monetary policy tightening, banks reduce lending by three percent after three years (cumulative). Additionally, we show that expansionary and contractionary shocks affect bank lending in a qualitatively symmetric fashion. However, the magnitude of the responses after a contractionary shock are two to three times larger than those after an equally large expansionary shock.

Our full sample results, however, highlight that banks do react systematically differently to monetary policy shocks depending on their liquidity state. In reaction to a 100bp contractionary monetary policy shock, highly liquid banks show a cumulative increase in their loan supply by 9% after three years. In contrast, illiquid banks show a largely muted response to the same shock. In aggregate, the responses of liquid banks dominate those from illiquid banks, as our analysis suggests that the average bank increases total lending after a contractionary monetary policy shock. This implies that the bank-lending channel is no longer active in the U.S. A more detailed look, however, reveals that illiquid banks still behave in line with the broad interpretation of the bank-lending channel and that the counter-intuitive results are driven by contractionary shocks only. In particular, we find that banks now increase their lending both after an expansionary and contractionary shock, i.e., they react asymmetrically to policy shocks.

Second, as emphasized in various recent contributions on the information effect of monetary policy, monetary policy instruments neglecting this effect ultimately deliver biased impulse response functions when used as external instruments in (S)VAR or local projections (LPs) setups. In this regard we show that in our full sample analysis instrumenting the changes in the policy rate with the high-frequency market-based surprises coming from Swanson (2021)

(which do not control for the information effect of monetary policy) lead to opposite reactions compared to what we find when employing our own informationally-robust MPI in the spirit of Miranda-Agrippino & Ricco (2021) or the informationally-robust monetary shock series from Bu, Rogers & Wu (2021). This suggests that similar to the previously mentioned long-standing puzzles in the monetary literature, informational robustness of MPIs leads to qualitative differences in the dynamic consequences of monetary policy shocks. Specifically, as we show below, our results are more closely aligned with economic theory than those that are based on instruments not taking into account the information effect of monetary policy.

These results are robust to a number of factors: We show that they are robust to different lag specifications and to alternative liquidity thresholds. In fact, there is a clear positive relationship showing a growing responsiveness with rising liquidity thresholds. The qualitative responses of bank variables do not depend on the specific type of monetary policy regime or tools. That is, when constructing our MPI only with the forward guidance and LSAP factors from Swanson (2021) we still find the same qualitative differences across liquid and illiquid banks. Our results show that the different policy dimensions do not yield different impulse responses other than expected quantitative differences. Finally, our main results are also robust to the use of alternative policy indicators. Namely, replacing the Effective Federal Funds rate with the one-year Treasury rate or the short-term shadow rate from Wu & Xia (2016) leaves our results unchanged.

Let us relate our findings to the literature. First, we contribute to the empirical literature on the bank-lending channel by showing that banks’ credit supply decisions crucially depend on their cash-liquidity levels when including post-2007 data in our analysis. This strand of the literature can be traced back to the seminal contribution by Bernanke & Blinder (1992), who developed a simple VAR model to asses the relevance of the bank-lending channel using U.S. data. Other studies have built on Bernanke & Blinder’s approach by looking at different categories of bank loans (see, e.g., den Haan, Sumner & Yamashiro (2007, 2009), Dave, Dressler & Zhang (2013) and Greenwald, Krainer & Paul (2020)), or by focusing on the housing market (see Iacoviello & Minetti (2008)). Other related studies analysing bank-level data (see, e.g., Kashyap & Stein (1994, 2000), Cetorelli & Goldberg (2012), Cao & Dinger (2018) and Eggertsson, Juelsrud, Summers & Wold (2019)) or a mix between aggregate and micro-level data (see, e.g., Carpenter & Demiralp (2012) or Dave et al. (2013)) have found mixed evidence on the presence of an active lending channel. To the best of our knowledge, our contribution is the first to shed light on the systematically different responses based on bank level cash-liquidity after both conventional and unconventional monetary shocks.

Second, our paper further contributes to the literature on monetary policy shocks and the relevance of informational effects conveyed around monetary policy announcements. Our contribution lies in contrasting high-frequency market-based surprises with informationally robust shocks. We show the importance of information effects in determining the impact of monetary policy shocks on key bank variables using informationally-robust monetary policy instruments a` la Miranda-Agrippino & Ricco (2021).

Related to the high-frequency shock identifying literature, early path-setting contributions are Kuttner (2001) and Gu ̈rkaynak, Sack & Swanson (2005). The key element of these methods is to examine changes in a benchmark rate in 30-minute windows around FOMC announcements. For example, in Jarocin ́ski & Karadi (2020) this is done for the three-months Federal Funds futures rate, whereas in Barakchian & Crowe (2013) a factor model using different futures contracts is being employed. In Nakamura & Steinsson (2018) as well as Swanson (2021) the market surprises are extracted through principal components analyses using a number of Fed Funds futures with different maturities, allowing the latter to identify separate effects of the Fed Funds rate, forward guidance and LSAPs. High-frequency identified shocks are able to bridge periods of conventional and unconventional monetary policy, a key characteristic that has made them more appealing since the onset of the GFC compared to, e.g., a narrative approach as in Romer & Romer (2004). However, recently, a number of studies have highlighted the relevance of the “information channel” of monetary policy. Among others, this includes Hoesch, Rossi & Sekhposyan (2020), Jarocin ́ski & Karadi (2020), Cieslak & Schrimpf (2019) and Nakamura & Steinsson (2018). The major insight of this strand of the literature is that monetary policy instruments need to account for the information effect of monetary policy adjustments to be unbiased. Popular instruments such as the Gu ̈rkaynak et al. (2005) or Swanson (2021) high-frequency market surprises do not control for these information effects. Bu et al. (2021) also show that Jarocin ́ski & Karadi (2020)’s series still contains an information effect. In contrast, Miranda-Agrippino & Ricco (2021), as well as Bu et al. (2021), develop an approach to control for these information effects allowing them to find a solution for long-standing puzzles in the monetary policy literature, as further documented inEttmeier & Kriwoluzky (2019). In this spirit, our contribution is to show that failing to control for the information channel of monetary policy leads to significant qualitative differences when employing such policy instruments in a macro-finance setup.

Finally, the econometric specification is closely related to the recent branch of the empirical literature on estimating impulse response functions using local projections (LPs henceforth) following Jorda` (2005). This rapidly evolving literature shows that LPs are not only more flexible in estimating dynamic responses than standard SVAR models by imposing fewer restrictions (see, e.g., Ramey (2016)), but also that the impulse responses are the same as the ones from VARs (see, e.g., Plagborg-Møller & Wolf (2021)). Recently, Montiel Olea & Plagborg-Møller (2021) prove that lag-augmented LP models, i.e., models that include sufficient lags of the variables in the regressions as controls, yield robust estimates even at longer horizons or if the data is highly persistent. Additionally, Montiel Olea & Plagborg-Møller (2021) show that by including enough lags as controls, i.e., in a lag-augmented LP model, there is no need to compute Heteroskedasticity and Autocorrelation Robust (HAR) standard errors and that the simple Eicker-Huber-White heteroskedasticity-robust standard errors are sufficient since, under weak assumptions, the regression scores are serially uncorrelated. Our approach to estimate dynamic responses of key banks’ balance sheet items is based on the “external instruments” insights obtained by Mertens & Ravn (2013) and Stock & Watson (2012, 2018), given that we instrument the changes in the policy rate with our informationally-robust MPIs when estimating our lag-augmented LP models.

The remainder of the paper is organized as follows: Section 2 briefly describes our data, and section 3 describes the construction of our informationally-robust monetary policy instruments. In section 4 we present the empirical setup using lag-augmented LP-IV models. The results of our quantitative exercise are given in section 5. Section 6 presents a battery of robustness checks and some further extensions. Finally, section 7 concludes.

Keep reading here.