Monetary Policy and Credit Flows

Transcription

1 Monetary Policy and Credit Flows Timothy Bianco University of Kentucky September 15, 2017 Abstract This paper investigates the effect of monetary policy on the reallocation of credit. Following Herrera, Kolar, and Minetti s (2011) methodology, I measure credit flows using Compustat data. Through a vector autoregression approach, I analyze the role of monetary policy on credit flows by testing existing credit channel theories. I find that monetary policy s impact on credit flows is consistent with the collateral constraint, searching for yield, and risk perceptions channels of monetary policy. PRELIMINARY AND INCOMPLETE. JEL codes: E32, E44, E51, G21, G31, G32 1

2 1 Introduction In neoclassical macroeconomic models without financial frictions, credit markets are an afterthought; banks act as an inconsequential intermediary between owners and ultimate users of capital (Jorgenson, 1963 and Tobin, 1969). In such models, credit markets are fully efficient by assumption and capital flows to its most productive use. In reality, credit markets are not without frictions and they could impact economic activity severely. In models without financial frictions, monetary policy impacts real economic activity through the cost-of-capital channel alone (Bernanke, 2007), however more recent models incorporate financial frictions with expanded channels of monetary policy transmission. Hence, policymakers do not only influence market interest rates through the amount of reserves and bank capital in the banking system, 1 but also through the impact on borrowing firms costs of external financing 2 and lending institutions perceptions of risk. 3 While there is a large body of literature investigating the effect of monetary policy on credit market outcomes, 4 an underexplored aspect is policy s impact on the reallocation of credit among borrowing firms. Existing theory implies that the composition of credit will adjust following a macroeconomic shock. 5 As credit to some firms contracts and gets reshuffled to other firms, assessment of credit market conditions through inspection of aggregate net credit changes alone is incomplete and potentially misleading. For instance, if credit 1 See Bernanke and Blinder (1988) and Kashyap and Stein (1995). 2 See Bernanke and Gertler (1989), Bernanke and Gertler (1995), Kiyotaki and Moore (1997), and Bernanke, Gertler, and Gilchrist (1999). 3 See Borio and Zhu (2012), Adrian and Shin (2010), Bruno and Shin (2015), Gambacorta (2009), and Dell Ariccia, Laeven, and Suarez (2017). 4 See Ramey (2016) for an overview of recent studies. 5 See for instance, models of homogeneous firms with heterogeneity investment project profitability (Matsuyama, 2007) or heterogeneity in firm quality (Bernanke, Gertler, and Gilchrist, 1996). 2

3 extended to one firm increases by 10% but that implies a 10% decrease in credit to another then the net credit change is zero. Yet, credit has been reshuffled between these two firms, implying a reallocation of credit. Such reallocation can have important implications for future capital formation and growth (see, e.g. Herrera, Kolar and Minetti 2015). Thus, a careful assessment of monetary policy s impact on credit markets ought to account for this reallocation. The objective of this paper is to explore the effect of monetary policy shocks on credit reallocation. Following Herrera, Kolar, and Minetti (2011), I use Compustat North America data to compute credit flows for borrowing firms. Using Compustat has several advantages. First, credit extended by nonbanks like private equity firms, investment banks, or insurance companies are included. Second, Compustat data contains relevant characteristics of the borrowing firm which can be used to dig deeper into the dynamic response of credit flows among firms that differ, for instance, in their degree of financial dependence or profitability. The ability of parsing the data along different characteristics, in turn, allows me to evaluate the relevance of alternative channels of monetary policy transmission. For instance, credit flows may be larger and more fluid for firms highly dependent on external financing. Therefore, the reallocation of credit among such firms may be of a larger consequence for economic activity than those capable of generating cash flow internally. An additional benefit of Compustat data in analyzing credit flows is that total debt is divided into short-term and long-term components. 6 Typically, short-term debt is used to provide cash flow to firms rather than financing long-term investment projects. However, short-term, non-intermediated debt can 6 Short-term debt includes bank acceptances, commercial paper, the current portion of long-term debt, etc. Long-term debt includes loans, bonds, lines of credit, etc. maturing in more than one year. 3

4 be relevant source of bridge financing (see Kahl, Shivdasani, and Wang 2015). Further, as Guedes and Opler (1996) document, borrowing (i.e. debt issuance) at very short and long maturities are common features of large, highly rated firms, which produce a large portion of U.S. output. Nevertheless, there are disadvantages to Compustat data. First, the data cannot distinguish between the type of borrowing, other than whether it is short- or long-term debt. Second, the database only includes information on publicly traded firms, which tend to be large, more developed firms. With Compustat, credit movement amongst small, non-publicly traded firms are ultimately excluded. However, to the extent that the firms covered by Compustat carry out most of the investment and produce a large portion of tehe US output, they provide a good testing ground for exploring the impact of monetary policy on credit flows. The topic of credit reallocation was first addressed by Dell Ariccia and Girabaldi (2005) using bank lending data. They find that credit expansion and contraction tend to co-move, specifically among banks of comparable size, loan type, and location. Herrera, Kolar, and Minetti (2011) draw similar conclusions by analyzing firm borrowing rather than bank lending. They find credit reallocation among borrowing firms to be intense, volatile, procyclical, and highly concentrated among firms of similar size, location, and industry. A connected study by Craig and Haubrich (2013) investigates the substantial bank consolidation in the 1990s and the response of credit creation, destruction, and reallocation in and out of recessions. They observe that credit creation is higher during expansions and that credit destruction is higher during recessions. 7 These studies either do not consider the impact of monetary policy on the reallocation or credit or do so only in a tangential manner. 7 Contessi, DiCecio, and Francis (2014) analyze credit reallocation on both sides - lending and borrowing. 4

5 In this paper, I show that monetary policy s impact on credit flows is large and idepends on specific firm characteristics and length of debt maturity. This is consistent with certain predictions of the balance sheet and risk-taking channels of monetary policy. That is, in response to monetary easing, credit tends to flow to firms whom are collateral constrained or perceived as relative riskier to the lender. This paper is organized as follows. Section 2 describes the construction and trends in the credit flow and monetary policy measures. Section 3 summarizes the theoretical background and documents the heterogeneity in credit flows. Section 4 outlines the empirical framework and discusses the impact of monetary policy s influence on credit flows. Section 5 discusses implications for future growth and productivity and Section 6 concludes. 2 Data construction and description 2.1 Credit flows As in Herrera, Kolar and Minetti (2011), I compute measures of inter-firm credit flows starting from the balance sheet for all publicly traded U.S. firms reported in Compustat North America. Firms in the finance, insurance, and real estate industry groups are removed from the sample given that the aim of this study is the impact of monetary policy on the (less studied) firms that demand credit, instead of those that create credit. I follow Herrera, Kolar, and Minetti s (2011) definition and measurement of credit flows in most aspects. In particular, (i) the unit of observation is the firm, as Compustat does not provide data on firms individual projects; (ii) I exclude accounts payable by suppliers from the 5

6 measure of debt; (iii) I exclude firms for whom the ratio of end-of-period gross capital to end-of-period capital exceeds 120% to control for existing firms that enter the data-set; 8 (iv) only exits due to merger, acquisition, liquidation, or bankruptcy are treated as credit subtractions. Previous research by Herrera, Kolar, and Minetti (2011) analyzes annual credit flows between 1954 and In this paper, I employ quarterly data spanning the period between 1974:Q1 and 2016:Q2. The use of quarterly data, instead of annual, allows for clearer identification of monetary policy shocks. Furthermore, the lengthy period of time covered by Compustat data allows for the use of structural VAR tools such as historical decomposition when studying the impact of monetary policy (see Kilian and Lütkephol, 2017). The quarterly rate of debt growth, g it, for firm i in quarter t is given by g it = debt it debt it 1 (debt it + debt it 1 )/2 (1) This transformation measures a symmetric and bounded growth rate around zero, thus allowing for a unified treatment of continuing, newborn and dying firms (Davis and Haltiwanger, 1992; Herrera, Kolar and Minetti, 2011). In particular, g it [ 2, 2] where 2 corresponds to debt growth of firms that died in the current year whereas 2 corresponds to debt growth of newborn firms. With this rate of growth, I then proceed to compute aggregate credit creation and destruction for a set of firms s in quarter t as the weighted sum of the rates of growth for expanding or entering firms and the weighted sum of the rate of debt growth for contracting 8 See Ramey and Shapiro (1998) for the use of a similar criteria applied to flows of physical capital and Herrera, Kolar, and Minetti (2011) for a detailed description. 6

7 or exiting firms, respectively. Specifically, aggregate credit creation for a group of firms s in quarter t (P OS st ) is defined as P OS st = g it >0,i s t g it ( debtit debt st ). (2) Similarly, credit destruction for a group of firms s in quarter t (NEG st ) is defined as NEG st = ( ) debtit g it. (3) debt g it <0,i s st t The first panel of Figure 1 plots credit creation and destruction for all publicly traded firms from the first quarter of 1974 through the second quarter of These two credit flow measures follow dissimilar patterns. Credit creation was highest in the late 1980s, a period when credit destruction was below its mean growth rate. This rise was the effect of considerable increases in debt, specifically long-term debt in the Transportation Equipment Manufacturing sector (NAICS 336). 9 Leading to the official start of the late 2000s recession, credit destruction fell roughly from 8 to 2 percent as credit creation increased to nearly 8 percent. From these pre-recession peaks, credit creation subsequently fell from 7.8 percent in the fourth quarter 2007 to 2.2 percent in the second quarter Also, credit destruction from 2.4 percent at the start of the recession to 4.3 percent in the first quarter I compute gross credit reallocation as the sum of credit creation and destruction (SUM st = P OS st + NEG st ) and net credit change by subtracting credit destruction from credit creation (NET st = P OS st NEG st ), shown in the second panel of Figure 1. Analysis of net 9 In the first quarter of 1988, General Motors was the transportation equipment sector s largest borrower and their long-term debt increased by a factor of four that quarter. Whereas large increases in debt were experienced by other sectors in the late 1980s, these spikes were principally due to credit creation of large firms in this sector. 7

8 credit growth alone may provide an incomplete depiction of the market because it masks the reallocation of credit. For instance, credit reallocation was high in the late-1990s and mid-2000s, which were periods of moderate net credit. I measure excess credit reallocation as the credit reallocation in excess of what is required to accommodate net credit changes (EXC st = SUM st NET st ). In a credit market, large movement in credit reallocation can occur due to large increases or decreases in either credit creation or destruction. destruction is unchanged. For example, suppose credit creation increases 10%, but credit In this scenario, credit reallocation is 10%, but no credit was truly reallocated from one borrower to another. In other words, the change in NET and SUM are equivalent. In a more fluid credit market, excess credit reallocation rises with simultaneously expanding and contracting credit. As the second panel of Figure 2 shows, excess credit reallocation is consistently non-zero and there was in upward trend in this measure until the late 1990s. Since then, excess credit reallocation has been on a downward trend. Table 1 provides further detail on these five credit flow measures by decade. In line with the findings of Herrera, Kolar, and Minetti (2011), throughout the decades, credit reallocation is large and it far exceeds what is needed to accommodate net credit changes. This table also shows that credit creation and destruction are quite volatile. While the volatility of credit creation is larger than that of credit destruction, the volatility of credit creation has diminished since its peak in the 1980s. The same holds for credit reallocation and excess credit reallocation. It is important to note that the magnitudes of these flows are not directly comparable with those by Herrera, Kolar, and Minetti (2011) because I use quarterly rather than annual measures. Further, reliable quarterly data is only available 8

9 from Compustat starting in the early 1970s. See the Appendix for a description of credit flows using annual growth rates that more closely resemble their credit flow measures. 2.2 Monetary policy measure and debt maturity Empirical investigations into the effect of monetary policy on economic activity have commonly identified the federal funds rate as the monetary policy instrument. However, from December 2008 until December 2015 the federal funds rate was effectively at the zero lower bound (ZLB), thus limiting the use the instrument to stimulate the economy and invalidating its use as the monetary policy variable in structural vector autoregressions (SVARs). An alternative measure of the monetary policy stance at the ZLB has been proposed by Wu and Xia (2016). These authors develop an approximation to the forward rate in the multifactor shadow rate term structure model, which can be used to replace the effective federal funds rate in SVARs. They find that unconventional monetary policy had a non-trivial impact on economic activity once the ZLB was reached. This measure has gained popularity for the exploration of unconventional monetary policy in recent years. Figure 2 plots the effective federal funds rate along with the shadow rate during the ZLB period. 10 In analysis, I employ the effective federal funds rate as our measure of monetary policy, but replace it with the Wu-Xia shadow rate during the ZLB period. Figures 3 and 4 plot the long- and short-term credit flow measures, respectively, along with the monetary policy rate. Long-term credit creation gradually increased in the 1980s as the policy rate was declining and it also increased throughout the 1990s, a period of a 10 Similar models that aim to capture monetary policy during the ZLB include Krippner (2012) and Bauer and Rudebusch (2016). See the Appendix for a comparison of these measures. 9

10 moderate policy rate. Prior to the early 2000s recession, long-term credit creation fell as the policy rate decreased and remained low until the monetary tightening of the mid-2000s. Prior to the late 2000s recession, long-term credit creation was nearly 10 percent, but fell roughly 2.5 percent during the into the ZLB period. Long-term credit destruction, on the other hand, was relatively stable throughout the period. It was highest in the mid-2000s during the peak of this period s monetary tightening. Short-term credit creation was highest in the late 1980s, a period of monetary easing. It also increased in the late 1990s through 2007, before the monetary easing of the late 2000s. Once the policy rate passes the ZLB threshold, short-term credit creation increases and remains high throughout the ZLB period. Short-term credit destruction tends to be much larger and volatile. Short-term credit destruction increased rapidly during the monetary easing of the late 2000s, but this followed a period of already increasing short-term credit destruction since the early-1990s. Relative to the policy rate, the net credit increases following monetary easing and this was apparent during the easing of the late 2000s for both short- and long-term credit. Longterm credit reallocation steadily increased in the 1980s and 1990s as monetary policy tended to ease, relative to the early 1980s. It was not until prior to the early 2000s recession that long-term credit reallocation started to noticeably decline. This coincided with the aggressive monetary easing of the mid-2000s. It spiked again at the height of the tightening prior to the late 2000s recession. Short-term credit reallocation, was highest during the 1970s and early 1980s. Since the early 1990s, however, following a large decline in shortterm credit reallocation, it has been slowly increasing. These patterns are suggestive of a decline in fluidity of short-term credit, which is consistent with the decline in labor fluidity 10

11 documented by Davis and Haltiwanger (2014). Long-term excess credit reallocation trended upward until the late 1990s, a period of easy monetary policy. Prior to the early 2000s recession, long-term excess credit reallocation began to steadily decline and this continued well into the ZLB period. This contrasts with short-term excess credit reallocation, which increased beyond the early 2000s recession. 2.3 Macroeconomic variables For the empirical analysis, the VAR includes, in addition to these credit flow and monetary policy measures, real GDP, the unemployment rate, and consumer price index. Real GDP comes from the Bureau of Economic Analysis and the unemployment rate and consumer price index comes from the Bureau of Labor Statistics. The monetary policy rate is the effective federal funds rate on the last trading day of the year from Federal Reserve s H.15 release in normal times and the Wu-Xia shadow rate, sourced from the Federal Reserve Bank of Atlanta, during the ZLB period. 3 The role of monetary policy and the flow of credit A traditional mechanism by which monetary policy is transmitted to real economic activity is through the supply of loans. That is, during time of expansionary monetary policy, banks have more reserves to lend to firms, leading to capital formation. Accordingly, following an easing shock, credit creation is expected to increase. Following analysis on job reallocation by Davis and Haltiwanger (2001), I consider this an aggregate channel, as it is a prediction from traditional macroeconomic modeling. This contrasts with allocative channels, which 11

12 alter the flow of credit to correct the mismatch between actual and desired credit positions of lenders and borrowers. Macroeconomic theory implies that monetary policy works through allocative channels that arise due to firm heterogeneity. A brief discussion of these channels and how policy influences credit flows follows. 3.1 Theoretical underpinnings The balance sheet channel T here are three aspects to consider under the balance sheet channel - the cost of external financing, collateral constraints, and cash flows. The two former aspects follow Bernanke, Gertler, and Gilchrist (1996), among others. If a firm s borrowing is constrained by the collateral they can pledge, they pay an external finance premium 11 above the market interest rate. Expansionary monetary policy lowers external financing costs by the impact on market interest rates and asset values. The latter eases collateral constraints, thereby lowering external finance premiums. Following monetary easing, net credit should increase as external financing becomes cheaper. Theory is clear that this is the result of an increase in credit creation, particularly for collateral constrained firms. However, theory appears to be silent regarding policy s impact on credit destruction. Credit destruction may increase as monetary policy enables firms to better meet interest and principal debt payments or decrease as loan defaults become less probable. When the change in net credit is positive, the increase in credit creation cannot be offset by an increase in credit destruction. Credit reallocation increases when both 11 Such as an agency premium that arises endogenously as the shadow value of relaxing the collateral constraint in Bernanke, Gertler, and Gilchrist (1996). 12

13 credit creation and destruction rise. When monetary policy increases credit creation, the sign of the impact on excess credit reallocation will correspond to that of credit destruction. The third aspect of the balance sheet channel is related to firms cash flow (Mishkin, 1996). Interest expense on variable rate loans positively co-moves with the monetary policy rate. Therefore, cash flows increase when monetary policy rate fall for borrowing firms. This lowers the overall need for external financing to meet interest obligations. In other words, credit creation ought to decrease, particularly for high debt service firms. Credit destruction may increase or decrease for the same reasons as discussed for the easing collateral constraints channel. The impact on net credit and credit reallocation depends on the impact on credit destruction, but excess credit reallocation will decrease The risk-taking channel Monetary policy may impact economic activity through the risk-taking channel. 12 A lender s credit risk tolerance and the ease by which they price credit risk is sensitive to monetary policy. For example, during the 2000s, interest rates in the U.S. were considered by many as too low for too long (Taylor, 2009). Even Federal Reserve officials expressed concern over the dangers of keeping policy rates unnecessarily low in the aftermath of the financial crisis (Bullard, 2015). Excessively low interest rates for an extended period may cause asset price bubbles and financial sector instability as banks engage in excessive risk-taking. In times of monetary easing, lenders may engage in searching for yield in which the allocation of loans in a lender s portfolio shifts toward high risk, high return loans (Rajan, 2005). Nominal return targets are slow to adjust downward when market interest rates 12 See Borio and Zhu (2012), Adrian and Shin (2010), Bruno and Shin (2015), Gambacorta (2009), and Dell Ariccia, Laeven, and Suarez (2017). 13

14 fall. To achieve previous returns, lenders are incented to reallocate the risk composition toward riskier borrowers. An implication of this channel is that credit creation increases in aggregate. Credit destruction likely increases as lenders reshuffle credit to reallocate their risk compositions. When both measures increase, credit reallocation and excess credit reallocation also increase. Net credit rises when increases in credit creation are not offset by a sufficiently large increase in credit destruction. I expect the increase in credit creation to be larger for firm perceived as relatively risky, and credit destruction to increase for those perceived as less risky. The second component of the risk-taking channel is the risk perception mechanism (Borio and Zhu, 2012). This mechanism is related to the balance sheet effect, although it operates through fundamental changes in lenders risk perceptions rather than through collateral constraints and agency premia. Monetary easing inflates asset prices and decreases volatility, which tends to lower the lender s perception of credit risk (Gambacorta, 2009). In other words, following monetary easing, the marginal borrower, whom may not have previously met lending standards, may appear creditworthy though their relative riskiness is unchanged. This mechanism implies that credit creation increases in response to monetary easing, although the response of credit destruction is again unclear. Net credit increases when the change in credit destruction does not offset the increase in credit creation. Credit reallocation will increase when credit destruction does not decrease by a larger magnitude than credit creation s increase. The sign impact on excess reallocation will correspond to that of credit destruction. The final mechanism is described by Dell Ariccia, Laeven, and Suarez (2017) as the riskshifting channel. Bank liabilities, such as wholesale funding and bank deposits, become less 14

15 costly when monetary policy lowers market interest rates. This increases lenders margins, and induces lenders to shift their allocation of loans away from high risk lending. As opposed to the searching for yield channel, this will lead to an increase in credit for relatively less risky borrowing. To accommodate shifting loan portfolios, credit destruction is more likely to increase, specifically for risky firms. The aggregate effect will be an increase in credit reallocation and excess credit reallocation. 3.2 Measuring heterogeneity To examine the importance of these channels in the transmission of monetary policy shocks to credit flows, I first must effectively group firms by relevant characteristics. To investigate the relevance of the balance sheet channel, firms are classified by their degree of financial constraints and debt service. 13 For the risk-taking channel, firms should be grouped by lenders risk perceptions of the borrowing firm; yet, information on these perceptions is not available. Instead, I group firms by different characteristics that might affect risk: default probability, asset size, financial dependence, and debt service. Table 2 shows the number of quarter-firm observations in each grouping. While certain groups of firms may have similar characteristics, these categories do highlight different aspects. To illustrate, of the 151,379 observations of high default probability across time, the majority of these are low debt service and non-financially dependent firms, but also small firms. The former two are characteristics of non-financially constrained firms and the latter is a characteristic of financially constrained firms. Here, I briefly discuss the measures of heterogeneity and describe asymmetries in credit flows of these groups. 13 High debt service firms may also be financially constrained. 15

16 3.2.1 Financial constraints and perceived riskiness Researchers have attempted to quantify firms financial constraints in equity and debt markets in recent years. 14 These measures are typically created by analyzing firms annual reports (Kaplan and Zingales, 1997), structural modeling (Whited and Wu, 2006), whether a firm has a credit rating (Almeida, Campello, and Weisbach, 2004) or whether they pay a dividend (Fazzari, Hubbard, and Petersen, 1988; Almeida and Campello, 2007). Farre- Mensa and Ljungqvist (2015) find support that firms classified as financially constrained by such measures, to a greater extent, have characteristics consistent with young firms in the growth phase of their life cycle rather than being truly financially constrained. Further, they find empirical support that firms close to default behave as if they are truly financially constrained. 15 Therefore, I group firms by their default probability. This not only identifies firms more highly constrained in credit markets, which is relevant for the balance sheet channel, but also those that are perceived as relatively risky borrowers, which is relevant for the risk-taking channel. Corporate default risk is constructed as in Merton (1974) whereby DD it = Distance to default it = log( E it+f it F it ) + r it 0.5σit 2 (4) σ it where E it = prccq cshoq 10 3 (5) 14 See Farre-Mensa and Ljungqvist (2015) for a survey of the literature regarding the measurement of financial constraints. 15 They also find support that private firms behave as if they are financially constrained. Compustat contains income statement and balance sheet data only for publicly traded firms, therefore, I do not group firms by whether they are publicly traded. 16

17 F it = dlcq + 1 dlttq (6) 2 E σ it = [ E + F σ F E,it] + [ E + F ( σ E,it)] (7) where σ E,it is the rolling one-year standard deviation of prccq (stock price), r it is the yearover-year stock return, dlttq is total long-term debt, and cshoq is common shares outstanding. Following Farre-Mensa and Ljungqvist (2015), firms with a probability of default greater than 25 percent from N(-DD), where N is the cumulative standard normal distribution function, are high default probability firms and all others are low default probability firms. Table 3 shows that credit creation, destruction, reallocation, and excess credit reallocation are, on average, larger and more volatile for high default probability firms for both shortand long-term credit. It is important to note that the credit measures are constructed by growth rates and even if credit growth and destruction rates are higher, the dollar amount of credit created or destroyed is lower than total credit flows of low default probability firms. For instance, over this period, the average share of credit for all high default probability firms was 7.2 percent. Further, the default probability threshold is static and depending on overall conditions, the number of firms considered high or low default probability fluctuates. In the fourth quarter of 2008, at the height of the financial crisis, over 21 percent of firms were deemed high default probability as compared to just above 1 percent in the first quarter of Debt service I group firms as in Kudlyak and Sanchez (2017), utilizing the leverage ratio (short-term debt as a percentage of total assets) to measure debt service. I classify a firm as high debt service 17

18 their leverage ratio is in the top quartile of firms in each quarter, and as low debt service otherwise. 16 Kudlyak and Sanchez (2017) find that during the financial crisis, low leverage firms short-term credit fell relatively more than high leverage firms on net. This is consistent with an alternative explanation by Calomiris and Himmelberg (1995) who argue that low leverage firms are financially constrained because external financing is overly expensive to obtain in response to adverse shocks such as the financial crisis. From Table 3, short-term credit creation and destruction, on average are larger for low debt service, although the net credit change is similar. Therefore short-term credit reallocation and excess credit reallocation are relatively high for low debt service firms at 28.8 and 24.8 percent, compared to and percent, respectively for high debt service firms. This implies that the short-term credit market is much more fluid for low debt service firms. Conversely, all long-term credit flow measures are larger for high debt service firms. In both short- and long-term credit, credit flows are more volatile for high debt service firms Firm size In relation to the balance sheet channel, the size of a firm ought to coincide with the degree to which they are collateral constrained. This was also shown empirically by Bernanke, Gertler, and Gilchrist (1996) and Gertler and Gilchrist (1994). The latter analyzes manufacturing firms by size, find that following monetary tightening, that credit flows from small to large firms. When the topic was re-examined by Kudlyak and Sanchez (2017) and they find that small firms whose credit contracted relatively more during the financial crisis. As commonly done in the literature, I classify firms whose assets lie above a threshold 16 Kudlyak and Sanchez (2017) limit their analysis to manufacturing firms as in Gertler and Gilchrist (1994). 18

19 as large and all others are small. Following Kudlyak and Sanchez (2017), I utilize a cutoff of one billion 2014 dollars. This coincides with previous studies that analyze aggregated data from the Quarterly Financial Report for Manufacturing, Mining, and Wholesale Trade (QFR), whose highest asset class is one billion dollars. Table 3 shows credit flow measures for these two groups of firms. Credit creation and destruction are higher for small firms, noting again that the dollar amount of credit is likely higher, on average for large firms. As a result, credit reallocation is larger for small firms as well as excess credit reallocation being larger. This holds for both short- and long-term credit Financial dependence In general, external financing is costlier than generating cash flow internally through operations, so firms that are highly dependent on external financing are more likely to be impacted by changing credit market conditions. Through the balance sheet channel, if financially dependent firms are more likely financially constrained (Rajan and Zingales, 1998; Kudlyak and Sanchez, 2017) then they are susceptible to shocks in the same manner. A widely used measure of financial dependence comes from Rajan and Zingales (1998) in their influential study of the link between financial dependence and economic growth. This is capital spending less cash flow from operation as a percentage of capital spending. Utilizing Kudlyak and Sanchez s (2017) cutoffs, I classify firms as financially dependent if the ratio is in the top quartile, and non-financially dependent otherwise. On average, as shown in Table 3, without exception, all credit flow measures are larger for financially dependent firms. In fact, short-term credit reallocation is larger on average, than any other subset in this study at percent, compared to percent for non-financially 19

20 constrained firms. Long-term credit reallocation is also large compared to non-financially constrained firms and other classifications at percent. 4 Monetary policy shocks and credit flows 4.1 Empirical strategy To analyze monetary policy s impact on credit flows, I utilize a SVAR, represented as A 0 Y t = a + n A j Y t j + u t. (8) j=1 where the vector, Y t, includes a block of macroeconomic variables (real GDP, unemployment, consumer prices), the monetary policy rate (the effective federal funds rate supplemented by Wu and Xia s (2016) shadow rate during the ZLB period), and a block of credit flow measures (credit destruction and creation), with this ordering. The A matrices are matrices of coefficients, a is a vector of constants, and u t is a vector of structural innovations. This methodology is comparable to Craig and Haubrich (2013), who specify a VAR with a block of macroeconomic variables, the federal funds rate, and a block of credit flows. However, the specification used in this paper differs in two aspects. First, to account for the price puzzle, whereby consumer prices responds unexpectedly to a monetary shock in monetary policy literature, I follow Estrella (2015). I impose a zero restriction on the coefficient of the first quarter lag of the monetary policy rate on consumer prices. In addition, the A 1 0 matrix is assumed to be lower triangular. That is, monetary policy does not directly 20

21 impact the current or the following period s consumer prices. 17 Estrella s (2015) method is motivated by theoretical analogs of identities for aggregate supply, aggregate demand, and a monetary policy rule. Because I impose the restriction on the lagged coefficient matrix, ordinary least squares equation by equation does not produce efficient estimates of the VAR parameters. Hence, I estimate the VAR via iterated seemingly unrelated regression (SUR), which is equivalent to full information maximum likelihood (Hamilton, 1994). Second, I order credit destruction before credit creation in Y t by the assumption that credit creation responds contemporaneously to credit destruction, but not vice versa. This assumption is motivated by the likely ability of borrowers to respond quickly to a credit destruction shock, although adjustment of credit destruction to a credit creation shock would take time to implement. That is, firms can quickly draw upon existing credit, but the impact of the shock on credit destruction through the maturation or repayment of existing credit is likely to take longer to materialize. Yet, the results are robust to the alternate ordering. This VAR is estimated separately for total, short-term, and long-term credit flows by rotating credit creation and destruction as the final block in Y t. In addition, to study the effect across groups of firms, I rotate an additional block of group credit flows for each. The results of the VAR are summarized below. 17 Alternative approaches to account for the price puzzle vary from including commodity prices to capture internal information on inflation expectations (Christiano, Eichenbaum, and Evans, 1999; Craig and Haubrich, 2013 among others), utilizing a factor-augmented VAR (Bernanke, Boivin, and Eliasz, 2005) to capture monetary policymakers internal information relevant for monetary policy, or including an omitted variable, the output gap, in a misspecified VAR (Giordani, 2004). 21

22 4.2 The effect of monetary policy shocks for all publicly-traded firms To start, the VAR is estimated using credit flow measures of all publicly traded firms. Figure 5 shows the impulse responses of Y t to a one time, unexpected 100 basis point decrease of the shadow federal funds rate (i.e. a monetary easing shock). 18 In response to the shock, real GDP increases with a lag, remaining persistently high while the unemployment rate falls and remains persistently low. While consumer prices fall beyond the first quarter lag, the response is small relative to that of an unconstrained VAR and the statistical significance is eliminated at a 10 percent significance level. Credit creation and credit destruction follow similar patterns in response to a monetary easing shock. Throughout the horizon, these credit flow measures tend to increase, although the magnitude is larger for credit creation. Accordingly, credit reallocation tends to increase throughout the horizon as shown in the first panel of Figure 6. The increase in credit reallocation is statistically significant at quarters 0-1 and 8-13 following the shock. An example of the reallocation process is that firms receive credit to pay off existing loans - a cleansing effect. The impulse responses of net credit change and excess credit reallocation are also plotted in the first panel of Figure 6. The response of net credit is only positive and significant on impact, and the impact on excess credit reallocation is positive and significant on impact, but negative and significant at 2 quarters following the shock. This suggest that the reshuffling of total credit exceeds what is needed to accommodate the net change in credit induced by the monetary policy shock only on impact. 18 The error bands are the 68 percent residual-based wild bootstrap interval. 22

23 The second and third panels of Figure 5 plot the responses of long- and short-term credit flows to the monetary easing shock. It is apparent that the results are driven mainly by responses of long-term credit flows. The impact of the monetary easing shock on credit creation and destruction are larger and the positive response of credit destruction is statistically significant beyond impact. Contrary to the responses of long-term credit flows, both creation and destruction of short-term credit tend to fall in response to the shock and their responses are smaller in magnitude. Recall that monetary policy may lead to fewer loan defaults or a higher likelihoods of loan repayment on aggregate. The impact of a monetary easing on credit destruction for the former is negative and the for the latter is positive. As these figures show, for shortterm credit, the effect of monetary easing on limiting loan defaults outweighs the impact on repayment. The opposite holds for long-term credit flows. 4.3 Heterogeneity in responses of credit flows to monetary policy shocks Table 4 provides the impulse responses from a VAR with an additional block for group credit measures. These results highlight a general impact of monetary easing shocks - that total credit reallocation increases in response to such shocks. Of the subsets, credit creation significantly increases for high default probability, high debt service, and financially dependent firms. The largest increase in total credit reallocation was high default probability firms, whose cumulative increase was 2.40 percentage points at a horizon of 8 quarters following a monetary easing shock. This is disproportionately larger than the 0.19 percentage point 23

24 increase for low default probability firms, which is not statistically significant. Other disproportionate responses are financially dependent relative to non-financially dependent firms responses (1.16 and 0.13 cumulative percentage point increases, respectively) and high relative to low debt service firms response (0.73 and cumulative percentage point changes, respectively) at 8 quarters following the shock. As the second and third panels of Table 4 show, most of the responses of total reallocation is driven by the response of long-term credit flows. As for total credit reallocation, the response of long-term credit reallocation for high default probability firms, financially dependent firms, and high debt service firms are larger compared to their counterparts, rising 2.88, 2.04, and 1.47 cumulative percentage points, respectively, 8 quarters following a monetary easing shock. Among subsets, short-term credit reallocation tends to decrease in response to the monetary easing shock. However, this is only statistically significant for low default probability firm at a horizon of 8 quarters. The response of excess credit reallocation is consistently negative. This occurs because, apart from high default probability firms, the response of credit reallocation is negative at both horizons. Therefore, the amount of credit reallocation needed to accommodate is large. Of groups whose long-term credit reallocation increases, excess credit reallocation tends to increase, although these are not statistically different from zero. Of the statistically significant responses of excess credit reallocation, it is statistically significant and negative for small and low debt service firms. 24

25 4.4 Implications for credit channels of monetary policy These results highlight that monetary operates, at least, through the aggregate channel through the positive impact on credit creation. By decomposing the responses of credit reallocation into the contributions from credit creation and destruction, I can shed light on the allocative impact of monetary policy through the aforementioned channels All publicly-traded firms I begin by analyzing the responses of credit flows in aggregate. Recall that in response to the monetary easing shock, credit creation and destruction increase for total and long-term credit. Credit destruction tends to increase and remain persistently high for long-term credit only and fall for short-term credit. Overall, the results of the impulse responses are consistent with the balance sheet channel, particularly for long-term credit markets. As predicted, credit creation increases and the increase in credit destruction tends to be smaller, leading to a net credit increases. Further, except for the second quarter following the shock, credit reallocation and excess credit reallocation persistently increase. These responses are also consistent with the searching for yield and risk-shifting mechanisms. By these mechanisms, I expect credit creation and destruction to increase to accommodate lenders shifting loan portfolios. The risk perceptions mechanism also predicts an increase in credit creation, although the expected response of credit destruction through this mechanism is unclear. By grouping firms and analyzing their impulse responses, I can provide more convincing evidence for these channels. 25

26 4.4.2 Default probability grouping Recall that the searching for yield predicts that credit creation should increase for risky (i.e. high default probability) firms and that credit creation should increase relatively more for risky firms by the risk perceptions model. This prediction is confirmed in total and long-term credit markets because the increase in credit creation is positive and significant for high default probability firms, but small and not statistically significant for low default probability firms. Further, through these channels, theory implies that monetary policy plays an allocative role reshuffling credit between high and low default probability firms. While credit destruction increases for both subsets, they are not statistically different from zero. The risk-shifting channel predicts that lenders will extend relatively more credit to low default probability firms as monetary easing indirectly shrinks lender margins. The evidence presented in Table 4 does not support this as a relevant channel of monetary policy. By the balance sheet channel, following a monetary easing shock, financially constrained firms ought to experience a relatively larger increase in credit creation. Further, while the impact on credit destruction is ambiguous, theory predicts that monetary policy should increase net credit. If default probabilities effectively group firms by financial constraints as Farre-Mensa and Ljungqvist (2015) find, then the responses of total and long-term credit creation, destruction, net credit, credit reallocation, and excess credit reallocation are all consistent with this theory. 26

27 4.4.3 Debt service grouping If high debt service firms are financially constrained, then like high default probability and financially dependent firms, theory would imply that credit creation to increase following a monetary shock that eases financing constraints. The alternative explanation of the impact of monetary policy regards the impact on variable rate interest expense. By this mechanism, a monetary easing shock will lead to a decrease in credit creation for high debt service firms as their variable interest expense falls. From Table 4, the responses of short- and longterm credit creation are not statistically different from zero. Long-term credit destruction increases for high debt service firms and this is likely due to increases in cash flow to better pay down principal. For short-term markets, high debt service firms credit destruction significantly decreases, potentially due to fewer loan defaults Asset size grouping By the searching for yield and risk perceptions channels, following a monetary easing shock, credit creation ought to increase proportionately more for small firms because they are perceived as riskier. The response of total credit creation was 0.38 percentage points for small firms, which is comparable to the 0.28 percentage point increase for larger firms, although the latter is not statistically significant. For long-term credit, the disparity shrinks further, but only the response of small firms credit creation is statistically significant. Therefore, this is consistent with the risk perceptions and searching for yield channels. Because small firms are more likely to be collateral constrained, these results are also consistent with the collateral constraint channel. For short-term credit creation, the responses of credit creation 27

28 and destruction are not statistically significant and therefore do not provide support for any channels Financial dependence grouping The response of financially constrained and unconstrained firms does not provide evidence of the balance sheet channel like impulse responses for other groups have. While credit creation increases for total and long-term credit relatively more for financially constrained firms following a shock, the response of credit creation is statistically insignificant for both group. This evidence is consistent with the results of Kudlyak and Sanchez (2017), who find that the credit of financially constrained did not contract more during the financial crisis. They conclude that the easing of collateral constraints are not relevant for explaining credit markets during the financial crisis. However, this financial dependence ratio may not effectively group firms by their degree of financial constraints. 4.5 Historical decompositions Historical decompositions show how much of the fluctuations of the endogenous variables in the system are caused by the individual structural innovations. Consider the structural VAR in 8. If the VAR is covariance stationary, then the value of Y t is a function of previous structural innovations, which also pre-date the initial period. The effects of old innovations are small, so an innocuous approximation of Y t is t 1 Y t Ŷt = b + θ s u t s. (9) s=0 28

29 The structural MA coefficient matrices, (θ 0, θ 1,..., θ T 1 ), are responses of each element in Y t to a single u t shock at the horizon h = 1, 2,..., H, so Y t=h u t = θ h (10) where θ h is a (KxK) matrix. K is the number of variables and therefore the number of structural innovation series in the VAR. Each element of the θ h is a (HxH) impulse response matrix. Figure 7 shows the cumulative contribution of monetary policy innovations to the creation and destruction of credit for subsets of firms across time. Among high and low default probability firms, these show that monetary policy innovation contributions to credit destruction are similar. However, a stark difference is the magnitude of the monetary innovation contribution to credit creation between these firms. Leading to the financial crisis and late 2008s recession, and during the period of unconventional monetary policy, these innovations contributed to a decrease in credit creation for high default probability firms. It was not until the end of the monetary easing period, that the contribution of monetary policy shocks begun to contribute positively to credit creation for this subset once again. There are substantial disparities between the contribution of monetary policy innovations to financially dependent and non-financially dependent firms. These disparities exist for the contributions for both credit creation and destruction. From this figure, the contribution of monetary policy innovations to credit creation and destruction is large in magnitude, and the contribution to credit creation appears to lag the contribution to credit destruction. These results are muted for non-financially dependent firms. 29

30 5 Credit flows and investment efficiency So far this paper has focused on the response of firm creadit flows to monetary policy. Yet, ultimately the question of interest for policy makers and academics is whether monetary policy is helps or hinders the reallocation of credit from less productive to more productive firms. Because Compustat does not provide detail on investment projects, I am not able to determine if credit flows to high or low investment projects. However, I can classify firms by their investment efficiency using an index by Galindo, Schiantarelli, and Weiss (2007). This index is constructed as I it = sales it debt it capital it debt kt sales it 1 capital it 1 debt it 1 debt kt 1. (11) This ratio exceeds one when the debt-weighted sales as a percentage of capital (a proxy for investment efficiency) is growing. Herrera, Kolar, and Minetti (2014) utilize this index to test the impact of interstate financial deregulation on states productivity growth through the reallocation of credit. Table 5 shows the mean and standard deviation of this index by subset of firm across time. 19 On average, the lowest investment efficiency subset are high default probability firms, whose mean index was as low as in the 2000s. As our results conclude, monetary easing shocks have a large allocative impact on these firms. From Figure 7, the contribution of monetary policy shocks to high default probability firm credit creation was high in the 1990s, and Table 5 shows that during this time that the mean investment efficiency was Also, low default probability firms mean investment efficiency was during this 19 The top and bottom 1 percent of indexes are trimmed in the calculation of these statistics. 30

31 time, but as seen in Figure 7, the contribution of monetary policy shocks was substantially smaller. I have also shown that monetary easing also leads to increases in credit reallocation for high debt service firms and financially dependent firms. As Table 5 shows, these are firms of higher investment efficiency on average than high default probability firms. These provide mixed support for the ultimate influence of monetary policy on economic activity through the reallocation of credit. 6 Conclusion This paper shows that monetary policy leads to substantial reallocation of credit. The impact is larger for firms of high default probability, high debt service, and financial dependence. This response consistent with the easing of collateral constraints that ensues a monetary expansion as purported by the balance sheet channel. Yet, such behavior is also implied by the searching for yield and risk perceptions mechanisms of the risk-taking channel. Apart from high default probability firms, these mechanisms hold for long-term credit markets. Short-term credit flows tend to fall following a monetary easing shock. Further, I find that while credit reallocation is largest for high default probability firms, these firms are of low investment efficiency on averege. The potential for future research on these topics is fruitful. While recent studies have quantified the impact of unconventional monetary policy during the financial crisis, there is yet to be a thorough analysis of the impact on credit flows, specifically credit reallocation during this time. In addition, lending firms are also subject to the same financial frictions 31

32 discussed in this study. While it is common to focus on the lending or borrowing side of credit, an underexplored topic is the role of these two channels in lender financing and credit reallocation. Propping up these credit market was a key motivation of the unconventional monetary policy during and after the financial crisis. References [1] Adrian, Tobias and Hyun Song Shin (2010). Financial intermediaries and monetary economics Federal Reserve Bank of New York Staff Reports, no [2] Almeida, Heitor and Murillo Campello (2007). Financial constraints, asset tangibility, and corporate investment Review of Financial Studies, 20.5: [3] Almeida, Heitor, Murillo Campello, and Michael Weisbach (2004). The cash flow sensitivity of cash The Journal of Finance, 31.4: [4] Bauer, Michael and Glenn Rudebusch (2016). Monetary policy expectations at the zero lower bound Journal of Money, Credit and Banking, 48.7: [5] Bernanke, Ben (2007). The financial accelerator and the credit channel June 15, 2007, at the The Credit Channel of Monetary Policy in the Twenty-first Century Conference, Federal Reserve Bank of Atlanta, Atlanta, Georgia. [6] Bernanke, Ben and Alan Blinder (1988). Credit, money, and aggregate demand American Economic Review 78.2:

33 [7] Bernanke, Ben, Jean Boivin and Piotr Eliasz (2005). Measuring the effect of monetary policy: a factor-augmented vector autoregressive (FAVAR) approach Quarterly Journal of Economics 120.1: [8] Bernanke, Ben and Mark Gertler (1989). Agency costs, net worth, and business fluctuations American Economic Review, 79: [9] Bernanke, Ben and Mark Gertler (1995). Inside the black box: the credit channel of monetary policy transmission Journal of Economic Perspectives, 9: [10] Bernanke, Ben, Mark Gertler, and Simon Gilchrist (1996). The financial accelerator and the flight to quality The Review of Economics and Statistics, 78.1: [11] Bernanke, Ben, Mark Gertler, and Simon Gilchrist (1999). The financial accelerator in a quantitative business cycle framework Handbook of macroeconomics, [12] Borio, Claudio, and Haibin Zhu (2012). Capital regulation, risk-taking and monetary policy: a missing link in the transmission mechanism Journal of Financial Stability, 8.4: [13] Bruno, Valentina, and Hyun Song Shin (2015). Capital flows and the risk-taking channel of monetary policy Journal of Monetary Economics, [14] Bullard, James (2015). U.S. monetary policy normalization OMFIF City Lecture, March 26, 2015, Frankfurt, Germany. [15] Calomiris, Charles and Charles Himmelberg (1995). Investment banking costs as a measure of access to external finance Unpublished working paper. 33

34 [16] Christiano, Lawrence, Martin Eichenbaum, and Charles L. Evans (1999). Monetary policy shocks: What have we learned and to what end? Handbook of macroeconomics, [17] Contessi, Silvio, Riccardo DiCecio, and Johanna Francis (2014). Aggregate shocks and the two sides of credit reallocation Unpublished working paper. [18] Craig, Ben, and Joseph Haubrich (2013). Gross loan flows Journal of Money, Credit and Banking, : [19] Davis, Steven, and John Haltiwanger (1992). Gross job creation, gross job destruction, and employment reallocation Quarterly Journal of Economics, 107.3: [20] Davis, Steven, and John Haltiwanger (2001). Sectoral job creation and destruction responses to oil price changes Journal of Monetary Economics, 48.3: [21] Dell Ariccia, Giovanni, and Pietro Garibaldi (2005) Gross credit flows The Review of Economic Studies, 72.3: [22] Dell Ariccia, Giovanni, Luc Laeven, and Gustavo Suarez (2017). Bank leverage and monetary policy s risk-taking channel: evidence from the United States Journal Finance, 72.2: [23] Eichenbaum, Martin (1992). Comment on Interpreting the Macroeconomic Time Series Facts: The Effects of Monetary Policy, European Economic Review, 36.5: [24] Estrella, Arturo (2015). The price puzzle and VAR identification, Macroeconomic Dynamics, 19.8:

35 [25] Farre-Mensa, Joan and Alexander Ljungqvist (2015). Do measures of financial constraints measure financial constraints? Review of Financial Studies, 29.2: [26] Fazzari, Steven, Glenn Hubbard, and Bruce Petersen (1988). Financing constraints and corporate investment Brookings Papers on Economic Activity, : [27] Galindo, Arturo, Fabio Schiantarelli, and Andrew Weiss (2007). Does financial liberalization improve the allocation of investment?: Micro-evidence from developing countries Journal of Development Economics, 82.2: [28] Gambacorta, Leonardo (2009). Monetary policy and the risk-taking channel BIS Quarterly Review, December [29] Gertler, Mark and Simon Gilchrist (1994). Monetary policy, business cycles, and the behavior of small manufacturing firms Quarterly Journal of Economics, 109.2: [30] Giordani, Paolo (2004). An alternative explanation of the price puzzle Journal of Monetary Economics, 51.6: [31] Gertler, Mark and Peter Karadi (2015). Monetary policy surprises, credit costs, and economic activity American Economic Journal: Macroeconomics, 7.1: [32] Guedes, Jose and Tim Opler (1996). The determinants of the maturity of corporate debt issues The Journal of Finance, 51.5: [33] Hamilton, James (1994). Time series analysis Vol 2. Princeton: Princeton University press. 35

36 [34] Herrera, Ana Maria, Marek Kolar, and Raoul Minetti (2011). Credit reallocation Journal of Monetary Economics, 58.6: [35] Herrera, Ana Maria, Marek Kolar, and Raoul Minetti (2014). Credit reallocation and the macroeconomy Unpublished Working Paper. [36] Herrera, Ana Maria, Marek Kolar, and Raoul Minetti (2011). Credit reallocation Journal of Monetary Economics, 58.6: [37] Jordà, Òscar (2005). Estimation and inference of impulse responses by local projections American economic review, 95.1: [38] Jorgenson, Dale (1963). Papers and proceedings of the seventy-fifth annual meeting of the American Economic Association American Economic Review, 53.2: [39] Kahl, Matthias, Anil Shivdasani, and Wang (2015). Short-term debt as bridge financing: evidence from the commercial paper market The Journal of Finance, 70.1: [40] Kaplan, Steven and Luigi Zingales (1997). Do investment-cash flow sensitivities provide useful measures of financing constraints? The Quarterly Journal of Economics, 112.1: [41] Kahl, Matthias, Anil Shivdasani, and Wang (2015). Short-term debt as bridge financing: evidence from the commercial paper market The Journal of Finance, 70.1:

37 [42] Kilian, Lutz, and Helmut Lütkephol (2017). Structural vector autoregressive analysis Cambridge University Press. [43] Kashyap, Anil and Jeremy Stein (1995). The impact of monetary policy on bank balance sheets Carnegie-Rochester Conference Series on Public Policy, 42. [44] Kiyotaki, Nobuhiro and John Moore (1997). Credit cycles Journal of Political Economy, 105: [45] Krippner, Leo (2012). Modifying gaussian term structure models when interest rates are near the zero lower bound Reserve Bank of New Zealand Discussion Paper, [46] Kudlyak, Marianna and Juan Sanchez (2017). Revisiting the behavior of small and large firms during the 2008 financial crisis Journal of Economic Dynamic & Control, 77: [47] Matsuyama, Kiminori (2007). Credit traps and credit cycles American Economic Review, 97.1: [48] Merton, Robert (1974). On the pricing of corporate debt: The risk structure of interest rates The Journal of Finance, 29: [49] Mishkin, Frederic (1996). The channels of monetary transmission: lessons for monetary policy NBER Working Paper, w5464. [50] Rajan, Raghuram (2005). Has financial development made the world riskier? NBER Working Paper, w

38 [51] Rajan, Raghuram and Luigi Zingalez (1998) Financial dependence and growth American Economic Review, 88.3: [52] Ramey, Valerie (2016). Monetary policy shocks and their propagation Handbook of Macroeconomics, 2:2, [53] Ramey, Valerie, and Matthew Shapiro (1998). Capital churning Unpublished working paper. [54] Stiglitz, Joseph and Andrew Weiss (1981). Credit rationing in markets with imperfect information American Economic Review, 71.3: [55] Taylor, John (2009). The financial crisis and the policy responses: an empirical analysis of what went wrong NBER Working Paper, w [56] Tobin, James (1969). A general equilibrium approach to monetary theory Journal of Money, Credit and Banking, 1.1: [57] Whited, Toni (1992). Debt, liquidity constraints, and corporate investment: evidence from panel data Journal of Finance, 47.4: [58] Whited, Toni and Guojun Wu (2006). Financial constraints risk Review of Financial Studies, 19.2: [59] Wu, Jing Cynthia and Fan Dora Xia (2016). Measuring the macroeconomic impact of monetary policy at the zero lower bound Journal of Money, Credit, and Banking, :

39 7 Tables Table 1: Descriptive statistics of quarterly credit measures by decade - all firms Average Coefficient of variation NEG POS NET SUM EXC NEG POS NET SUM EXC 1974:Q1-1979:Q :Q1-1989:Q :Q1-1999:Q :Q1-2009:Q :Q1-2015:Q Note: This table reports means and coefficients of variation of credit measures for all publicly traded firms. P OS refers to credit credit creation, NEG is credit destruction, NET is net credit change (NET st = P OS st NEG st), SUM is credit reallocation (SUM st = P OS st + NEG st), and EXC is excess credit reallocation (EXC st = SUM st NET st ). Table 2: Frequency of quarter-firm observations by subset Low debt service Large High debt service Small Non-financially dependent Low default probability 603, , , , , ,540 High default probability 146,675 4,704 27, , ,447 36,932 Low debt service 192, , , ,560 High debt service 50, , ,849 65,912 Large 227,606 15,548 Small 537, ,924 Financially dependent Note: This table shows the number quarter-firm observations of subsets from 1974:Q1-2016:Q2. Firms are subset first by their default probability following Farre-Mensa and Ljungqvist (2015) whereby firms whose default probability exceeds 25 percent at a point in time are considered high default probability firms and all others are low default probability firms. Firms are classified as high debt service if the leverage ratio is in the top quartile of firms in a given quarter. Firms are considered large if their total assets exceed 1 billion 2014 dollars as in Kudlyak and Sanchez (2017). Financially dependent firms are those whose need for external financing as in Rajan and Zingales (1998) is in the top quartile as in Kudlyak and Sanchez (2017). 39

40 Table 3: Descriptive statistics of credit measures by subset Average Total credit Short-term credit Long-term credit NEG POS NET SUM EXC # of obs. NEG POS NET SUM EXC # of obs. NEG POS NET SUM EXC # of obs. All firms , , ,276 High default probability , , ,102 Low default probability , , ,796 [0.000] [0.000] [0.000] [0.000] [0.000] [0.001] [0.000] [0.002] [0.000] [0.024] [0.000] [0.000] [0.000] [0.000] [0.000] High debt service , , ,350 Low debt service , , ,412 [0.000] [0.000] [0.147] [0.000] [0.368] [0.000] [0.000] [0.637] [0.000] [0.000] [0.000] [0.000] [0.148] [0.000] [0.000] Large , , ,880 Small , , ,111 [0.000] [0.000] [0.003] [0.000] [0.000] [0.000] [0.000] [0.033] [0.000] [0.000] [0.000] [0.000] [0.083] [0.000] [0.000] Financially dependent , , ,207 Non-financially dependent , , ,210 [0.000] [0.000] [0.000] [0.000] [0.000] [0.000] [0.000] [0.011] [0.000] [0.000] [0.000] [0.000] [0.000] [0.000] [0.000] Coefficient of variation Total credit Short-term credit Long-term credit NEG POS NET SUM EXC NEG POS NET SUM EXC NEG POS NET SUM EXC All firms High default probability Low default probability High debt service Low debt service Large Small Financially dependent Non-financially dependent Note: This table reports means, p-values of a two-sided t-test for mean equivalence in brackets, number of quarter-firm observations, and coefficients of variation for subset of firms. High default probability firms are those whose default probability exceeds 25 percent at a point in time. High debt service firms are those whose quarterly leverage ratio falls in the top quartile of firms. Large firms total assets exceed 1 billion 2014 dollars. Financially dependent firms are those whose quarterly need for external financing falls in the top quartile of firms. High investment efficiency firms are those whose index of investment efficiency exceeds a value of 1. 40

41 Table 4: Cumulative impulse response of credit flows to a monetary easing shock Total credit After 4 quarters After 8 quarters NEG POS NET SUM EXC NEG POS NET SUM EXC All firms High default probability ** 1.73* 2.40** 0.67 Low default probability High debt service * Low debt service *** ** Large Small Financially dependent 0.45* ** * Non-financially dependent Long-term credit After 4 quarters After 8 quarters NEG POS NET SUM EXC NEG POS NET SUM EXC All firms High default probability ** 1.21** 1.33** ** 2.18** 2.88** 0.69 Low default probability High debt service 0.40** * *** * 1.31 Low debt service -0.13* *** ** Large Small * * * * Financially dependent 0.40* Non-financially dependent ** Short-term credit After 4 quarters After 8 quarters NEG POS NET SUM EXC NEG POS NET SUM EXC All firms * High default probability Low default probability ** -0.82* *** High debt service *** *** Low debt service * ** Large ** ** Small * * Financially dependent * -1.45** ** * Non-financially dependent *** *** Note: This table shows the cumulative percentage point response of credit measures to an expansionary monetary policy shock (an unanticipated, negative 100 basis point decline in the shadow federal funds rate as measured by Wu and Xia (2016)). The impulse responses are derived from a vector autoregression (VAR) that includes a block of macroeconomic variables, the shadow federal funds rate, and these two seasonally adjusted credit flow measures. The VAR includes a constant and two lags as chosen by BIC. The ***, **, *, and indicate 99, 95, 90, and 68 percent statistical significance using the residual-based wild bootstrap interval method with 2,000 repetitions. High default probability firms are those whose default probability exceeds 25 percent at a point in time. High debt service firms are those whose quarterly leverage ratio falls in the top quartile of firms. Large firms total assets exceed 1 billion 2014 dollars. Financially dependent firms are those whose quarterly need for external financing falls in the top quartile of firms. 41

42 Table 5: Heterogeneity of investment efficiency High default probability Low default probability Mean Std. Dev. Mean Std. Dev. 1974:Q1-1979:Q :Q1-1989:Q :Q1-1999:Q :Q1-2009:Q :Q1-2016:Q Financially dependent Non-financially dependent Mean Std. Dev. Mean Std. Dev. 1974:Q1-1979:Q :Q1-1989:Q :Q1-1999:Q :Q1-2009:Q :Q1-2016:Q High debt service Low debt service Mean Std. Dev. Mean Std. Dev. 1974:Q1-1979:Q :Q1-1989:Q :Q1-1999:Q :Q1-2009:Q :Q1-2016:Q Large Small Mean Std. Dev. Mean Std. Dev. 1974:Q1-1979:Q :Q1-1989:Q :Q1-1999:Q :Q1-2009:Q :Q1-2016:Q Note: This table provides the mean and standard deviation of the investment efficiency index following Galindo, Schiantarelli, and Weiss (2007) by subsets of groups across time. The index is the debt-weighted sales as a percentage of capital and the top and bottom one percent are trimmed in calculation of these statistics. High default probability firms are those whose default probability exceeds 25 percent at a point in time. High debt service firms are those whose quarterly leverage ratio falls in the top quartile of firms. Large firms total assets exceed 1 billion 2014 dollars. Financially dependent firms are those whose quarterly need for external financing falls in the top quartile of firms. 42

43 8 Figures Figure 1: Total credit measures - all publicly traded firms Note: P OS refers to credit creation, NEG is credit destruction, NET is net credit change (NET st = P OS st NEG st), SUM is credit reallocation (SUM st = P OS st + NEG st), and EXC is excess credit reallocation (EXC st = SUM st NET st ) for total credit for all firms. Shaded bars indicate NBER recessions. Figure 2: The shadow rate and effective federal funds rate Note: The effective federal funds rate comes from the Federal Reserve Board s H.15. release and the Wu-Xia shadow rate comes from the Federal Reserve Bank of Atlanta. Shaded bars indicate NBER recessions. 43

44 Figure 3: The policy rate and long-term credit flow measures Note: The solid lines are credit creation (P OS) credit destruction (NEG), net credit change (NET = P OS NEG), credit reallocation (SUM = P OS +NEG), and excess credit reallocation (EXC = SUM NET ) for long-term credit.the monetary policy rate is the effective federal funds rate from the Federal Reserve Board s H.15. release and the Wu-Xia shadow rate from the Federal Reserve Bank of Atlanta during the zero lower bound period, plotted as a dotted line. Shaded bars indicate NBER recessions. Figure 4: The policy rate and short-term credit flow measures Note: The solid lines are credit creation (P OS) credit destruction (NEG), net credit change (NET = P OS NEG), credit reallocation (SUM = P OS+NEG), and excess credit reallocation (EXC = SUM NET ) for short-term credit.the monetary policy rate is the effective federal funds rate from the Federal Reserve Board s H.15. release and the Wu-Xia shadow rate from the Federal Reserve Bank of Atlanta during the zero lower bound period, plotted as a dotted line. Shaded bars indicate NBER recessions. 44

45 Figure 5: Impulse responses to an expansionary monetary shock Note: These graphs plot impulse responses to an expansionary monetary policy shock (an unanticipated, negative 100 basis point decline in the shadow federal funds rate as measured by Wu and Xia (2016)). The impulse responses are derived from a vector autoregression (VAR) that includes a block of macroeconomic variables, the shadow federal funds rate, and two seasonally adjusted credit flow measures. The VAR includes a constant and two lags as chosen by BIC. The bands represent 68 percent interval using the residual-based wild bootstrap method with 2,000 repetitions. 45

46 Figure 6: Impulse responses to an expansionary monetary shock Total credit Long-term credit Short-term credit Note: These graphs plot impulse responses to an expansionary monetary policy shock (an unanticipated, negative 100 basis point decline in the shadow federal funds rate as measured by Wu and Xia (2016)). The impulse responses are derived from a vector autoregression (VAR) that includes a block of macroeconomic variables, the shadow federal funds rate, and two seasonally adjusted credit flow measures for total, short-term, and long-term credit separately. The VAR includes a constant and two lags as chosen by BIC. The dots indicate 68 percent statistical significance using the residual-based wild bootstrap interval method with 2,000 repetitions. 46

47 Figure 7: Historical contribution monetary policy shocks to credit flows These graphs shows the cumulative historical contribution of monetary shocks to credit flow measures. These results are derived from a vector autoregression (VAR) that includes a block of macroeconomic variables, the shadow federal funds rate, two seasonally adjusted aggregate credit flow measures, and two seasonally adjusted subset credit flow measures. The VAR includes a constant and two lags as chosen by BIC. High default probability firms are those whose default probability exceeds 25 percent at a point in time. High debt service firms are those whose quarterly leverage ratio falls in the top quartile of firms. Large firms total assets exceed 1 billion 2014 dollars. Financially dependent firms are those whose quarterly need for external financing falls in the top quartile of firms. 47

48 Appendix A1. Alternative credit flow measure Credit flow measures are calculated using quarterly growth rates of debt from Compustat North America. This follows Herrera, Kolar, and Minetti (2011) who measure credit flow measures using annual growth rates over a loner horizon. The following table presents averages and coefficients of variation by decade using year-over-year growth rates. These more closely resemble credit flow measures constructed with annual data. Average Coefficient of variation NEG POS NET SUM EXC NEG POS NET SUM EXC 1974:Q1-1979:Q :Q1-1989:Q :Q1-1999:Q :Q1-2009:Q :Q1-2016:Q Note: This table reports means and coefficients of variation for of credit measures for all publicly traded firms. P OS refers to credit credit creation, NEG is credit destruction, NET is net credit change (NET st = P OS st NEG st), SUM is credit reallocation (SUM st = P OS st + NEG st), and EXC is excess credit reallocation (EXC st = SUM st NET st ). A2. Alternative monetary policy rates In the empirical analysis of the effect of monetary policy on credit flows, we utilize the effective federal funds rate in normal times and the Wu-Xia (2016) shadow rate during the zero lower bound. Alternative measures of monetary policy during the zero lower bound are by Krippner (2012) and Bauer and Rudebusch (2016). The following first plots these monetary policy rates along with the effective federal funds rate. The following graphs presents the impulse responses using these alternative measures. 48

49 Sources: Federal Reserve Board; Federal Reserve Bank of Atlanta; Federal Reserve Bank of San Francisco; Reserve Bank of New Zealand. The shaded bar indicates the NBER recession. VAR using policy rate by Krippner (2012) VAR using YZ(3)-25 policy rate by Bauer and Rudebusch (2016) 49