I. Introduction

Monetary policy actions such as changes in the Federal funds rate and the money supply concern individual firms, because such policy actions affect their cash flows and cost of external funds. To date, the related literature has focused on the transmission mechanism of monetary policy at the aggregate level, which has long been one of the key issues in macro and monetary economics (see Bernanke and Blinder, 1992; Ramey, 1993). There is another important dimension—the behavior of liquid asset holdings at the firm level, which is of special importance in corporate finance (John, 1993; Opler and others, 1999). In this paper, we combine these two dimensions to explain and estimate the direct effects of monetary policy on corporate liquidity demand by using fully disaggregated quarterly panel data.

Conventional money demand literature implies that monetary policy can affect money demand by the economy or firms indirectly through changes in the interest rate. The theory of firms’ precautionary demand for money (see Miller and Orr, 1966) posits that monetary policy may affect money demand by influencing cash flow uncertainty. Also, Laidler (1993, p. 187) insightfully points out that considerable turbulence in the conduct of monetary policy may affect the stability of money demand. Nonetheless, the literature suggests no direct link between money demand and monetary policy.

In contrast, the monetary transmission literature has examined how monetary policy affects the supply of bank loans (Bernanke and Blinder, 1988; Kashyap and Stein, 1994) and the liability of firms (Friedman and Kuttner, 1993; Christiano and others, 1996). For example, Christiano and others (1996), using aggregated data on flow of funds, find that upon a tight monetary policy shock, the business sector initially raises net funds. Also, time series tests on the lending channel show that, following tighter monetary policy, bank loans often rise initially and then decline with a lag, roughly coinciding with aggregate spending (Bernanke and Blinder 1992; Kashyap and Stein, 1994; Bernanke and Gertler, 1995). Bernanke and Blinder, (1992) and Morgan (1998) conjecture that this sluggish response of aggregate loans partly reflects the fact that many bank loans are contractual commitments.² Morgan provides evidence that loans under commitment contracts are unaffected or even accelerated after tighter policy, whereas those not under commitment contracts are lowered. Choi and Oh (2000) suggest that upon tighter policy, investors increase money holdings in the short run when the money supply is endogenous.

At the firm level, John (1993) argues that firms under stronger financial stress hold more cash than other firms do. Another motive for holding liquid assets, as stated by Opler and others (1999), is to enable firms to continue investing when cash flows are low and outside funds are expensive. These studies suggest that, in tight financial environments, agents strive to keep liquidity reserves to pursue good investment projects³. Such behavior is consistent with the familiar suggestion of market observers: after the economic expansion nears its peak, “as central banks raise interest rates to quell inflation, cash is king” (The Economist, April 21–27, 2001). However, few studies have been conducted on the effects of monetary policy on firms’ financial decisions.

In a simple model, we incorporate an important feature of the credit view that bank loans and bonds are imperfect substitutes (Bernanke and Blinder, 1988) and that commercial loans can be provided to firms on demand through loan commitments (Morgan, 1998; Holmstrom and Tirole 1998; Kashyap and others 1999). We also introduce inventories as close substitutes for liquid assets, since they are more marketable than other physical assets and are readily reversible (Carpenter and others, 1994), and as a factor of production (Ramey, 1989). We propose that monetary policy affects corporate liquidity demand directly through a separate channel—what we call “the loan commitment channel.” Upon persistent monetary policy shocks, firms make substitutions between sources of funds for intertemporal liquidity management, taking advantage of loan commitments and sluggish movements in loan rates. Specifically, since the loan rate often responds to tighter policy more slowly than does the market rate, the spread between the loan rate and the commercial paper (CP) rate initially drops. Expecting higher fund-raising costs in the near future, firms with access to credit lines committed by banks accumulate liquidity reserves by drawing on credit lines on a longer term basis while proportionately reducing the CP issuance. As credit lines are exhausted and the spread begins to rise, firms may temporarily increase their CP issuance, but they eventually reduce net funding of liquidity reserves. Since firms differ in their access to loan commitments and CP issuance, monetary policy has heterogeneous effects on corporate liquid assets.

Guided by the implications of our model, we specify regression equations that focus on the management of liquid assets, on the basis of the identification scheme that policy has contemporaneous impacts on liquidity, but when policy reacts to those variables, it does so with a lag. We estimate panel regressions to determine how corporate liquid assets (defined as cash plus short-term investments) respond to motivating factors, especially monetary policy shocks. The data for the period 1975:1–97:4 are taken from Compustat files for two distinct groups of firms: S&P 500 and non-S&P firms, for comparison purposes. We also examine the implications of firm size or creditworthiness for the heterogeneous effects of monetary policy on liquid assets.

Our empirical analysis indicates the following. First, upon tighter monetary policy, S&P 500 firms initially increase their liquid assets before reducing them, whereas non-S&P firms reduce them more quickly. This finding from firm-level data corroborates the response pattern found for big and small firms’ demand for net funds (Christiano and others, 1996). Our finding suggests that firms’ responses to tighter policy shocks generate positive policy-induced shocks to money demand. Tighter policy calls for higher interest rates. Thus, this study helps identify an important source at the firm level for the liquidity puzzle, which states that a positive contemporaneous correlation exists between monetary aggregates and interest rates (Strongin, 1995). Second, liquid asset holdings decrease with inventories, supporting the view of inventories as buffers for internal finance, and variations in liquid assets across firms are explained by firm characteristics including liquidation costs, Tobin’s q-ratio, the debt-equity ratio, individual interest costs, and debt ratings.

The paper is organized as follows. Section II provides a background for the monetary policy effect on corporate liquidity demand using a simple model. Section III describes the empirical identification scheme and specifies regressions that are based on our model and extended to control for firm-specific factors. Section IV describes the data and measuring variables. Section V presents estimated results for corporate liquid asset demand, along with some findings from the aggregated data. Section VI provides robustness checks, and Section VII concludes the paper.

II. A Conceptual Background: The Model

The model illustrates in a minimalist fashion that firms change their liquidity holdings through changes in bank loans and CP issues upon monetary policy shocks and that inventories and liquid assets are substitutes in liquidity management. Based on a partial equilibrium approach, we present a simple model that describes firms, banks, and the Federal Reserve (“Fed”), but leaves the rest of the economy unspecified. The model is a modification of Ramey’s (1993), which is rooted in Bernanke and Blinder’s (1988) formulation of the lending channel. Unlike Ramey, we specify two-period loans that banks provide on demand, inventories in the production function, and the Fed’s policy rule reflecting interest rate targeting.

The production function of the representative firm is given by

where Y_t is real output, K_t is the capital stock, N_t is labor, M_t/P_t is real money balances, V_t, is the inventory stock, and λ_t is a technology shock. We assume that the capital stock is rented from consumers as in Ramey (1993). Since the firm must pay for some factors of production before receiving revenue, liquidity facilitates the production process. Thus, the effective real liquidity, M_t/P_t +ϕV_t, where ϕ reflects that inventories are imperfect substitutes for money as buffers for internal finance, is included in the production function.⁴ Carpenter and others (1994) argue that because inventories are a large proportion of a firm’s assets and readily reversible, a relatively small percentage change in inventories can largely offset changes in cash flow. Also, inventories are a factor of production, as emphasized by Ramey (1989), which means that as the inventory stock falls, it becomes more costly for firms to sacrifice inventories at the margin.

We assume that the firm’s profit is always distributed to the shareholders. To reflect that loans have longer maturity than CP, we assume that the maturities of CP and loans are one period and two periods, respectively.⁵ Hence, the firm obtains money by borrowing from banks, in the form of two-period bank loans, L₂, or by floating CP, B, in the open market. We also assume that there are no other uses or sources of funds.⁶ That is, M_t=L_2,t+L₂,_t-1+B_t.

Two features are imposed on the cost of finance. First, according to the lending channel view (see Bernanke and Blinder, 1988; Kashyap and Stein, 1994), bank loans and CP are imperfect substitutes, and external finance by CP incurs additional costs. To reflect this feature, we introduce a quadratic cost function assuming that the additional costs rise with the amount of CP issues. Second, the loan rate is the effective rate, consisting of official and implicit costs.⁷ The loan rate is assumed to be higher than the market rate but not prohibitively high to ensure that both bank loans and CP are positive, that is, M_t> B_t>0.

The real profit of the representative firm in period t is given by

where R_t is the nominal rate of return on the Treasury bills, ρ_2,t is the two-period loan rate, r_t is the real rental rate on K, and w_t is the real wage. The real price of inventories is unity, and the cost parameter of issuing CP, θ, is a positive fraction. At the end of period t, the firm pays back the one-period CP with interest, returns the loans borrowed in period t-1 with interest, and distributes its profits to the shareholders.

The firm maximizes the expected present value of real profits as follows

where β is the discount factor, π_t is the inflation rate, m_t is the real liquid assets (M_t/P_t), and the balance-sheet identity, M_t=L_2,t+L_2,t-1+B_t, is used to substitute for b_t(=B_t/P_t) in the equation. The firm chooses the optimal levels of liquid assets, period-t loans, CP, inventories, and labor and capital inputs, given their relative prices, interest rates, and period t–1 loans.

Among the efficiency conditions, by equating the marginal productivity of real liquidity to its opportunity cost and the marginal benefit of the two-period real loans to its marginal cost, we have the following two conditions:

Define the spread between the two-period loan rate and the market rate over two periods as

where π_t+1 is assumed to have a negligible effect on the spread. Taken together, equations (4)–(6) can be used to obtain the liquid asset demand function⁸.

Substituting equation (6) into equation (5) and solving the resulting equation for b_t results in

where $Q_{t+i}={\prod }_{k=1}^i{(-\beta )}^i{(1+{{\pi }}_{t+k})}^{-1}$ and the transversality condition, $\underset{T\to \infty }{{lim}}{E}_t{{\beta }}^T{B}_{t+T}/P_t=0$, is imposed. We consider a production function given by

Substituting equation (7) and $\partial Y_t/\partial m_t=\alpha Y_t/(m_t+{\varphi }{V}_t)$ into equation (4), we obtain the demand for liquid assets by the representative firm as the following:

Corporate liquid asset demand increases with the volume of real output and decreases with the interest rate. Unlike the conventional equation for money demand, this one also depends on the current spread, the expected weighted sum of future spreads, and inventories.

Now we specify the determination of the spread from the interaction between the loan rate set by the banks and the Fed’s policy rule. We assume that banks provide loans on the basis of a loan commitment contract.⁹ According to Bernanke and Blinder (1988), tighter monetary policy reduces loans and increases the spread. In practice, however, the loan rate often responds to changes in financial markets with a lag (Forbes and Mayne, 1989).¹⁰ To capture that the loan rate adjusts to the market rate movement with inertia, we specify the loan rate process as

That is, in the short run, banks can insulate their lending activities from shocks to bank reserves by issuing certificates of deposits (CDs) or by paring their net holdings of securities. However, over the long term, they are unable to fully insulate their lending activities from the shocks that persist over several periods.¹¹

We account for the Fed’s tendency to influence the interest rate (for a survey on the related literature, see Clarida and others, 1999). For simplicity, we assume that the market rate is directly affected by the Fed’s policy stance or a policy shock, μ_t,

where $\overline{R}$ is a constant and ${\vartheta }_t$ is a stochastic disturbance caused by nonpolicy factors. The policy stance follows an autoregressive process, μ_t=νμ_t-1+ξ_t, where 0 < ν < 1 and ξ_t is an independently distributed error. Although simple, equation (10) reconciles a general feature that the interest rate can be changed through a liquidity effect (Christiano and Eichenbaum, 1992; Choi, 1999) or through direct control of the funds rate (Taylor, 1993). It can also be considered as a simplified version of Taylor’s rule that allows responses to the economywide output gap, inflation, and the lagged interest rate due to interest rate smoothing (Clarida and others, 1999).

Equations (9) and (10) imply that ρ_2,t can be expressed as

where $\underset{T\to \infty }{{lim}}{(h_1-h_2)}^T{\rho }_{2,t-T}=0$ is imposed since |h₁–h₂ |< 1. The loan rate is a function of lagged shocks only, and the sign of the policy shock effects depends on the lag length and the weight given to the previous loan rate and (h₂) the speed of convergence to a close alignment with the market rate (h₂). Abstracting from the lagged values of an unobservable error, ${\vartheta }_t$, the spread can be rewritten as: g_t=g(μ_t, μ_t-1, μ_t-2, μ_t-3,…).¹² If the inflation rate is approximated to a constant, the spread decreases with the current shock and depends on past shocks.

Equations (10) and (11) imply that an increase in μ_t in period t raises R_t and R_t+1 while leaving ρ_2t unchanged. Thus, tighter policy reduces the spread in the short run, providing firms with an incentive to draw on their lines of bank credit.¹³ From period t+1, the spread will rebound as ρ_2t catches up with the market rate (oscillations around the original level can occur if h₂>h₁). In fact, a tight policy shock calls for a brief fall in the spread, which is followed by increases in the spread as shown by Friedman and Kuttner (1993). Equation (8) implies that upon tighter policy the firm increases its liquid assets by increasing its current bank loans while proportionately reducing CP issues. Thus, the demand for liquid assets can be expressed as

In the face of tighter policy, the firm increases the demand for liquid assets contemporaneously.¹⁴The signs of the effects of lagged policy shocks on the demand for liquid assets, however, depend on the length of lag and the adjustment speed of the loan rate.

Also, the relation that describes inventories from the first-order conditions is given by

Hence, liquid assets are negatively correlated with inventories, while the inventory-output and liquidity-output ratios become steady in the long run.

III. The empirical specifications

IV. The Data and Measuring Variables

We construct two different data sets from Compustat for the period 1975:1–97:4. To link the panel data and time series consistently, the calendar (not fiscal) year date for each firm’s data is used. For the main data set, we consider S&P 500 firms from the S&P industrial or transportation index list (Compustat annual item no. 276 = 10 or 49) in any of the three years 1978,1987, and 1997. We exclude financial firms that may have different motivations in portfolio management (for example, to meet capital requirements) and utility firms that do not report liquid assets. A firm is included in our sample if its liquid assets, sales, and inventories are reported and positive for more than 12 consecutive quarters. After these screenings, we are left with an unbalanced panel that contains 659 firms in 53 industries as indicated by the two-digit standard industrial classification (SIC) code. The alternative data set is comprised of 689 non-S&P firms in 46 industries, which are selected to match the S&P firms closely except that firms in this set are smaller in size (but not too small). Appendix A provides details on the sample.

Liquid assets are defined as cash plus short-term investments (Compustat quarterly item no. 36) as in Opler and others (1999).²⁰ Figure 1 depicts the cross-section averages of liquid assets, sales, and inventories in real terms from the S&P 500 data set (Panels A and B) and the non-S&P data (Panels C and D). The variables are measured in logs or in ratios (normalized by the beginning-of-quarter total assets). The cross-section average of liquid assets tends to grow less rapidly than that of sales, possibly reflecting the progress of financial technology and more accurate predictions of revenues and payments, both of which enable firms to save liquidity holdings. The stagnant inventory stock may reflect advances in inventory management. An indication of the downward shift in sales and inventories in the early 1980s, which is more pronounced in the non-S&P data, possibly reflects the downsizing of U.S. corporations.

Cross-Section Averages of Liquid Assets, Sales, and Inventories

Citation: IMF Working Papers 2001, 177; 10.5089/9781451858877.001.A001

• Download Figure
• Download figure as PowerPoint slide

View Full Size

Cross-Section Averages of Liquid Assets, Sales, and Inventories

Citation: IMF Working Papers 2001, 177; 10.5089/9781451858877.001.A001

• Download Figure
• Download figure as PowerPoint slide

Cross-Section Averages of Liquid Assets, Sales, and Inventories

Citation: IMF Working Papers 2001, 177; 10.5089/9781451858877.001.A001

• Download Figure
• Download figure as PowerPoint slide

Aggregate time series are from FRED at the Web site of the Federal Reserve Bank of St. Louis. The consumer price index (CPI) is used to convert current dollars into 1992 dollars. The monetary policy stance in period t is measured by the change in the Federal funds rate, ΔFFR_t, or by the negative value of the mix of nonborrowed reserves and one-period-lagged total reserves, MNBRX_t. As shown by Bernanke and Blinder (1992), Strongin (1995), and Christiano and others (1996), both measures are good indicators of the Fed’s policy stance. Figure 2A depicts the two quarterly measures of the policy stance and the market rate measured by the three-month Treasury bill rate. A higher value in both measures represents a tighter policy. We use the market rate as a proxy for the opportunity cost of liquidity holdings because the own rates of return on liquid assets are unavailable. Figure 2B depicts the monthly spread between the prime rate (one of the most important indicators of lending rates) and the market rate along with ΔFFR_t. The spread has always been positive and often mirrors ΔFFR_t.

Monetary Policy Stance Measures and Spread

Citation: IMF Working Papers 2001, 177; 10.5089/9781451858877.001.A001

Note: Vertical solid lines denote NBER business cycle peaks, and vertical dash-dotted lines denote business cycle troughs.

• Download Figure
• Download figure as PowerPoint slide

View Full Size

Monetary Policy Stance Measures and Spread

Citation: IMF Working Papers 2001, 177; 10.5089/9781451858877.001.A001

Note: Vertical solid lines denote NBER business cycle peaks, and vertical dash-dotted lines denote business cycle troughs.

• Download Figure
• Download figure as PowerPoint slide

Monetary Policy Stance Measures and Spread

Citation: IMF Working Papers 2001, 177; 10.5089/9781451858877.001.A001

Note: Vertical solid lines denote NBER business cycle peaks, and vertical dash-dotted lines denote business cycle troughs.

• Download Figure
• Download figure as PowerPoint slide

To show how corporate liquid assets move along with a policy tightening that precedes a recession, Figure 3 depicts distributed movements of policy stance measures and liquidity before and after the onset of the 1982 and 1991 recessions for six quarters. The policy stance becomes tighter as the economy nears the peak and looser as it deepens into recessions (panels A and B). The change in the liquid assets-total assets ratio rises before a recession and then shows a J- shape adjustment after the business cycle peak (panel C). The J-shape adjustment takes place perhaps because liquidity holding is more costly with higher interest rates in the early stage of the recession but becomes less costly with lower interest rates as the recession deepens.

Distributed Movements of Policy Stance Measures and Aggregated Liquidity near the Onset of Recessions.

Citation: IMF Working Papers 2001, 177; 10.5089/9781451858877.001.A001

Notes: The horizontal axes represent the number of quarters before (with a negative sign) and after the onset of the 1982 recession (81:4) and the 1991 recession (90:3). For each recession, the curve for a group of firms is the locus of the group mean of j-quarter-ahead variable. Panel C depicts the annualized change in liquidity $(M_t/A_t^N-M_{t-4}/A_{t-4}^N)$. The curves without symbols are for S&P firms and those with symbols are for non-S&P firms.

• Download Figure
• Download figure as PowerPoint slide

View Full Size

Distributed Movements of Policy Stance Measures and Aggregated Liquidity near the Onset of Recessions.

Citation: IMF Working Papers 2001, 177; 10.5089/9781451858877.001.A001

Notes: The horizontal axes represent the number of quarters before (with a negative sign) and after the onset of the 1982 recession (81:4) and the 1991 recession (90:3). For each recession, the curve for a group of firms is the locus of the group mean of j-quarter-ahead variable. Panel C depicts the annualized change in liquidity $(M_t/A_t^N-M_{t-4}/A_{t-4}^N)$. The curves without symbols are for S&P firms and those with symbols are for non-S&P firms.

• Download Figure
• Download figure as PowerPoint slide

Distributed Movements of Policy Stance Measures and Aggregated Liquidity near the Onset of Recessions.

Citation: IMF Working Papers 2001, 177; 10.5089/9781451858877.001.A001

Notes: The horizontal axes represent the number of quarters before (with a negative sign) and after the onset of the 1982 recession (81:4) and the 1991 recession (90:3). For each recession, the curve for a group of firms is the locus of the group mean of j-quarter-ahead variable. Panel C depicts the annualized change in liquidity $(M_t/A_t^N-M_{t-4}/A_{t-4}^N)$. The curves without symbols are for S&P firms and those with symbols are for non-S&P firms.

• Download Figure
• Download figure as PowerPoint slide

We measure variables to capture firm characteristics as follows. Since the quarterly data of R&D expenditures are not available before 1989, we use the annual data of the R&D ratio. We treat firms with non-available figures as firms with no R&D expenditures as in Opler and others As a proxy for Tobin’s q-ratio, we use the market-to-book ratio, [book value of total assets + common stock (market value - book value)]/(book value of total assets), following the finance literature (see Yermack, 1995; Opler and others, (1999). We make an adjustment to the market- to-book ratio for cross-section averages for each period to remove the longer-term trend and cyclical variations. The debt-equity ratio and individual interest cost ratio are also adjusted for the cross-section averages to control for cyclical variations and policy effects. The beginning-of- quarter value of the debt-equity ratio is used to avoid a possible simultaneity between liquid assets and debts.

Table 1a presents the summary statistics of the main variables for the main data set (see Appendix A for the non-S&P data set). On average, liquid assets are about one-fourth of annual sales and 8 percent of total assets. Liquid assets and inventories show more variability than sales do, and they are more closely aligned with assets than with sales as shown by their standard deviations in the ratio (0.104 and 0.139 versus 0.211). Firm-characteristic variables such as the R&D-sales ratio and Tobin’s q-ratio are widely dispersed. As shown in Table 1b, the levels of sales, inventories, and liquid assets are highly correlated, consistent with equation (13). In terms of the ratio to total assets, liquid assets are not as closely related to sales as inventories are, implying liquidity management in terms of asset portfolio. Tobin’s q-ratio is positively correlated with the R&D-sales ratio possibly because the R&D expenditures reflect to some extent the firm’s growth opportunities (Mehran, 1995), whereas it is negatively correlated with the debt-equity ratio.²¹ The positive correlation between the individual interest costs and the debt-equity ratio suggests that higher debts render the firm’s fund-raising more costly.

Table 1A.

Descriptive Statistics for Main Variables

Notes: Distributions of variables are computed with the S&P 500 data (659 firms in 53 industries) for the 1975:1–97:4 period. The CPI is used to convert all current dollars into 1992 dollars in measuring all real variables.

^aFigures in square brackets are statistics of the variable adjusted for cross-section average.

^bDebt-equity and interest cost ratios are in logarithms.

Table 1A.

Descriptive Statistics for Main Variables

Variable	N	Mean	Standard deviation	Minimum	1st quarter	Median	3rd quarter	Maximum
ln mj,t	46,699	4.128 [0.000]a	1.634 [1.626]	-5.990 [-10.321]	3.112 [-1.003]	4.183 [0.052]	5.160 [1.032]	9.539 [5.236]
ln Sj,t	46,699	6.106 [0.000]	1.331 [1.320]	-4.853 [-0.816]	-5.293 [-10.834]	6.166 [0.060]	6.974 [0.867]	10.394 [4.326]
ln Vj,t	45,567	5.318 [0.000]	1.462 [1.461]	-5.870 [-11.094]	4.526 [-0.790]	5.432 [0.116]	6.281 [0.962]	9.401 [3.987]
$M_{j,t}/{\widehat{A}}_{j,t}^N$	45,929	0.077 --	0.104 --	0.000 --	0.017 --	0.042 --	0.100 --	3.698 --
$S_{j,t}^N/{\widehat{A}}_{j,t}^N$	45,929	0.328 --	0.211 --	0.001 --	0.205 --	0.294 --	0.397 --	3.133 --
$V_{j,t}^N/A_{j,t}^N$	44,843	0.183 --	0.139 --	0.000 --	0.069 --	0.161 --	0.265 --	1.154 --
R&D /sales	46,699	0.022 --	0.258 --	0.000 --	0.000 --	0.004 --	0.023 --	54.54 --
Tobin’s q	42,501	1.539 [0.004]	0.937 [0.896]	0.245 [-1.422]	1.023 [-0.457]	1.250 [-0.236]	1.707 [0.156]	21.31 [19.40]
Debt /equity b	45,097	0.194 [-0.001]	0.744 [0.726]	-4.193 [-4.238]	-0.230 [-0.405]	0.185 [0.002]	0.578 [0.379]	8.860 [8.498]
Interest costb	41,980	-4.694 [-0.002]	0.676 [0.657]	-13-47 [-8.486]	-4.941 [-0.230]	-4.577 [0.120]	-4.297 [0.380]	-1.489 [3.481]

Notes: Distributions of variables are computed with the S&P 500 data (659 firms in 53 industries) for the 1975:1–97:4 period. The CPI is used to convert all current dollars into 1992 dollars in measuring all real variables.

^aFigures in square brackets are statistics of the variable adjusted for cross-section average.

^bDebt-equity and interest cost ratios are in logarithms.

View Table

Table 1A.

Descriptive Statistics for Main Variables

Variable	N	Mean	Standard deviation	Minimum	1st quarter	Median	3rd quarter	Maximum
ln mj,t	46,699	4.128 [0.000]a	1.634 [1.626]	-5.990 [-10.321]	3.112 [-1.003]	4.183 [0.052]	5.160 [1.032]	9.539 [5.236]
ln Sj,t	46,699	6.106 [0.000]	1.331 [1.320]	-4.853 [-0.816]	-5.293 [-10.834]	6.166 [0.060]	6.974 [0.867]	10.394 [4.326]
ln Vj,t	45,567	5.318 [0.000]	1.462 [1.461]	-5.870 [-11.094]	4.526 [-0.790]	5.432 [0.116]	6.281 [0.962]	9.401 [3.987]
$M_{j,t}/{\widehat{A}}_{j,t}^N$	45,929	0.077 --	0.104 --	0.000 --	0.017 --	0.042 --	0.100 --	3.698 --
$S_{j,t}^N/{\widehat{A}}_{j,t}^N$	45,929	0.328 --	0.211 --	0.001 --	0.205 --	0.294 --	0.397 --	3.133 --
$V_{j,t}^N/A_{j,t}^N$	44,843	0.183 --	0.139 --	0.000 --	0.069 --	0.161 --	0.265 --	1.154 --
R&D /sales	46,699	0.022 --	0.258 --	0.000 --	0.000 --	0.004 --	0.023 --	54.54 --
Tobin’s q	42,501	1.539 [0.004]	0.937 [0.896]	0.245 [-1.422]	1.023 [-0.457]	1.250 [-0.236]	1.707 [0.156]	21.31 [19.40]
Debt /equity b	45,097	0.194 [-0.001]	0.744 [0.726]	-4.193 [-4.238]	-0.230 [-0.405]	0.185 [0.002]	0.578 [0.379]	8.860 [8.498]
Interest costb	41,980	-4.694 [-0.002]	0.676 [0.657]	-13-47 [-8.486]	-4.941 [-0.230]	-4.577 [0.120]	-4.297 [0.380]	-1.489 [3.481]

Notes: Distributions of variables are computed with the S&P 500 data (659 firms in 53 industries) for the 1975:1–97:4 period. The CPI is used to convert all current dollars into 1992 dollars in measuring all real variables.

^aFigures in square brackets are statistics of the variable adjusted for cross-section average.

^bDebt-equity and interest cost ratios are in logarithms.

Table 1B.

Correlation Coefficients between Main Variables

Notes: Pearson correlation coefficients (p-values in parentheses) between variables are computed with the S&P 500 data (659 firms) for the 1975:1–97:4 period. Figures for the log specification are in the lower triangle and those for the ratio specification are in the upper triangle. Debt–equity and interest cost ratios are in logarithms and adjusted for cross-section averages.

Table 1B.

Correlation Coefficients between Main Variables

	$M_{j,t}/{\widehat{A}}_{j,t}^N$	$S_{j,t}^N/{\widehat{A}}_{j,t}^N$	$V_{j,t}^N/A_{j,t}^N$	R&D /sales	Tobin’s q	Debt /equity	Interest cost
ln mj,t		0.07 (0.00)	-0.04 (0.00)	0.38 (0.00)	0.36 (0.00)	-0.35 (0.00)	-0.26 (0.00)	$M_{j,t}/{\widehat{A}}_{j,t}^N$
ln Sj,t	0.61 (0.00)		0.57 (0.00)	-0.08 (0.00)	0.15 (0.00)	-0.07 (0.00)	-0.20 (0.00)	$S_{j,t}^N/{\widehat{A}}_{j,t}^N$
ln Vj,t	0.48 (0.00)	0.84 (0.00)		0.00 (0.40)	0.05 (0.00)	-0.13 (0.00)	-0.10 (0.00)	$V_{j,t}^N/A_{j,t}^N$
R&D /sales	0.02 (0.00)	-0.07 (0.00)	-0.01 (0.02)		0.26 (0.00)	-0.23 (0.00)	0.03 (0.00)	R&D /sales
Tobin’s q	0.03 (0.00)	-0.22 (0.00)	-0.22 (0.01)	0.26 (0.00)		-0.40 (0.00)	-0.26 (0.00)	Tobin’s q
Debt /equity	0.00 (0.47)	0.24 (0.00)	0.16 (0.00)	-0.23 (0.00)	-0.40 (0.00)		0.37 (0.00)	Debt /equity
Interest cost	-0.14 (0.00)	-0.00 (0.31)	-0.01 (0.09)	-0.03 (0.00)	-0.26 (0.00)	0.37 (0.00)		Interest Cost
	ln mj,t	ln Sj,t	ln Vj,t	R&D /sales	Tobin’s q	Debt /equity	Interest cost

Notes: Pearson correlation coefficients (p-values in parentheses) between variables are computed with the S&P 500 data (659 firms) for the 1975:1–97:4 period. Figures for the log specification are in the lower triangle and those for the ratio specification are in the upper triangle. Debt–equity and interest cost ratios are in logarithms and adjusted for cross-section averages.

View Table

Table 1B.

Correlation Coefficients between Main Variables

	$M_{j,t}/{\widehat{A}}_{j,t}^N$	$S_{j,t}^N/{\widehat{A}}_{j,t}^N$	$V_{j,t}^N/A_{j,t}^N$	R&D /sales	Tobin’s q	Debt /equity	Interest cost
ln mj,t		0.07 (0.00)	-0.04 (0.00)	0.38 (0.00)	0.36 (0.00)	-0.35 (0.00)	-0.26 (0.00)	$M_{j,t}/{\widehat{A}}_{j,t}^N$
ln Sj,t	0.61 (0.00)		0.57 (0.00)	-0.08 (0.00)	0.15 (0.00)	-0.07 (0.00)	-0.20 (0.00)	$S_{j,t}^N/{\widehat{A}}_{j,t}^N$
ln Vj,t	0.48 (0.00)	0.84 (0.00)		0.00 (0.40)	0.05 (0.00)	-0.13 (0.00)	-0.10 (0.00)	$V_{j,t}^N/A_{j,t}^N$
R&D /sales	0.02 (0.00)	-0.07 (0.00)	-0.01 (0.02)		0.26 (0.00)	-0.23 (0.00)	0.03 (0.00)	R&D /sales
Tobin’s q	0.03 (0.00)	-0.22 (0.00)	-0.22 (0.01)	0.26 (0.00)		-0.40 (0.00)	-0.26 (0.00)	Tobin’s q
Debt /equity	0.00 (0.47)	0.24 (0.00)	0.16 (0.00)	-0.23 (0.00)	-0.40 (0.00)		0.37 (0.00)	Debt /equity
Interest cost	-0.14 (0.00)	-0.00 (0.31)	-0.01 (0.09)	-0.03 (0.00)	-0.26 (0.00)	0.37 (0.00)		Interest Cost
	ln mj,t	ln Sj,t	ln Vj,t	R&D /sales	Tobin’s q	Debt /equity	Interest cost

Notes: Pearson correlation coefficients (p-values in parentheses) between variables are computed with the S&P 500 data (659 firms) for the 1975:1–97:4 period. Figures for the log specification are in the lower triangle and those for the ratio specification are in the upper triangle. Debt–equity and interest cost ratios are in logarithms and adjusted for cross-section averages.

V. Empirical Results

A. Some Findings from Aggregated Time Series

To determine whether our identification scheme is consistent with the actual data, we conduct Granger causality tests using aggregated time series for the 1975:1–97:4 period. One may consider the reverse causality: the monetary policy stance is tightened to mop up excess liquidity contemporaneously as liquid asset demand rises. To deal with this issue, we estimate a policy reaction function under the identifying assumption that policy responds to the aggregated corporate liquidity up to the last quarter, but not to contemporaneous liquidity. We regress ΔFFR on eight lags of ΔFFR, eight lags of the cross-section average of corporate liquid assets scaled by the beginning-of-period total assets, a time trend, and seasonal dummies. The null hypothesis of no Granger causality from corporate liquidity to the policy stance is not rejected by the F-test: the test statistic is F(8,62) = 0.25 (p-value = 0.587). On the other hand, to test the causality that the policy stance causes corporate liquidity, we estimate a regression that uses the current value of the cross-section average of corporate liquid assets as the dependent variable in place of ΔFFR in the policy reaction function above. The null hypothesis of no Granger causality from the policy stance to corporate liquid assets is strongly rejected: F(8,62) = 3.90 (p-value = 0.001). Therefore, the causality test results are consistent with our identification scheme.²²

Next, we attempt to see whether the responses of loans and CPs to tighter monetary policy are consistent with the loan commitment channel explanation.²³ Since the data for CPs and loans are unavailable at the firm level, we use their aggregated time series compiled from FRED. We estimate a monthly vector autoregression (VAR) model that contains five variables in the following order: the change in the Federal funds rate, the change in the three-month CP rate, the spread between the prime rate and the CP rate, the (annualized) growth of loans, and the growth of CPs. The sample period is from 1975 :01 to 1987:09, the ending date of which is dictated by the availability of the CPs of nonfinancial companies.²⁴

Figure 4 shows the impulse responses of the variables to a positive, 1 percent point shock in the change in the funds rate. The CP rate corresponds closely to the policy shock. The spread response shows an initial fall followed by a hump-shaped curve, consistent with Friedman and Kuttner’s (1993) finding. With the initial fall in the spread, the loan growth rises whereas the CP growth shows little change. As the spread rises over the original level after a few months, the loan growth decreases whereas the CP growth increases. These responses are consistent with the loan commitment channel. Upon tighter policy, firms increase loans until the option of committed loans is exhausted²⁵. As tighter policy gains momentum and loan rates catch up with market rates, firms start increasing the share of CP issuance to fill up the shortage in their liquidity reserves (see Kashyap and others, 1993). As the financial market distress deepens, eventually, firms reduce external funds.

Impulse Responses to a Policy Tightening Shock Using a Monthly 5-Variable VAR Model

Citation: IMF Working Papers 2001, 177; 10.5089/9781451858877.001.A001

Note: The solid line represents the point estimate, and the dotted lines denote one-standard error bands computed from bootstrapping with 1000 replications.

• Download Figure
• Download figure as PowerPoint slide

View Full Size

Impulse Responses to a Policy Tightening Shock Using a Monthly 5-Variable VAR Model

Citation: IMF Working Papers 2001, 177; 10.5089/9781451858877.001.A001

Note: The solid line represents the point estimate, and the dotted lines denote one-standard error bands computed from bootstrapping with 1000 replications.

• Download Figure
• Download figure as PowerPoint slide

Impulse Responses to a Policy Tightening Shock Using a Monthly 5-Variable VAR Model

Citation: IMF Working Papers 2001, 177; 10.5089/9781451858877.001.A001

Note: The solid line represents the point estimate, and the dotted lines denote one-standard error bands computed from bootstrapping with 1000 replications.

• Download Figure
• Download figure as PowerPoint slide

B. Regressions for Corporate Liquid Asset Demand

An important issue in the literature on corporate money demand is whether economies of scale exist for cash balances.²⁶ Using the Compustat annual files, Mulligan (1997a) finds that the sales elasticity for cash is 0.83–0.85, implying the presence of economies of scale for cash balances. Using a broader sample of Compustat firms, Mulligan (1997b) finds that the sales elasticity of liquid assets (0.78) is close to the sales elasticity for cash (0.75). In line with these studies, we start with pretest regressions that regress liquid assets on sales alone or on sales and inventories as a substitute for liquid assets for 1975:1–97:4. Log transformation is applied to these variables for the pragmatic virtue of yielding constant elasticities. To control for the financial technology progress, we use time dummies following Mulligan (1997a) or adjust the variables for the cross-section averages. All aggregate variables, associated with cyclical fluctuations and longer-term trends, are swamped in the quarter-year dummies or cross-section averages of the variables. We also estimate regressions by controlling for fixed-firm effects.

Table 2 summarizes the pretest regression results. The estimated sales elasticity is little affected by the choice between the level variables with quarter-year dummies (columns 2–5) and the variables demeaned for the cross-section averages (columns 6–9). The sales elasticity is 0.72–0.76 when the real liquid assets are regressed on sales and 0.79–0.86 when the real liquid assets are regressed on both sales and inventories. Note that regressions controlling for fixed- firm effects have lower estimates for sales elasticity. Standard errors are computed using White’s correction for heteroskedastic errors in Table 2 and in what follows.

Table 2.

Pretest Regressions for Liquid Asset Demand

Notes: The regressions were performed with the S&P 500 data (659 firms) for the 1975:1–97:4 period. ${\overline{R}}^2$ excludes variance explained by the fixed-firm effects. Standard errors (in parentheses) are computed using White’s correction for heteroscedasticity.^**indicates significance at the 1% level,^*indicates significance at the 5% level, and ^† indicates significance at the 10% level.

Table 2.

Pretest Regressions for Liquid Asset Demand

Independent Variable	Dependent variable: ln mj,t
	Level				Deviation from cross-section average
ln Sj,t	0.755** (0.005)	0.719** (0.012)	0.863** (0.009)	0.788** (0.023)	0.754** (0.005)	0.713** (0.012)	0.861** (0.009)	0.785** (0.023)
${ln}{\widehat{V}}_{j,t}$	-	-	-0.100** (0.008)	-0.105** (0.017)	-	-	-0.099** (0.008)	-0.108** (0.017)
Quarter-year effects	Yes	Yes	Yes	Yes	No	No	No	No
Fixed-firm effects	No	Yes	No	Yes	No	Yes	No	Yes
${\overline{R}}^2$	0.381	0.381	0.382	0.382	0.376	0.376	0.377	0.377
N	46,698	46,698	44,717	44,717	46,698	46,698	44,717	44,717

Notes: The regressions were performed with the S&P 500 data (659 firms) for the 1975:1–97:4 period. ${\overline{R}}^2$ excludes variance explained by the fixed-firm effects. Standard errors (in parentheses) are computed using White’s correction for heteroscedasticity.^**indicates significance at the 1% level,^*indicates significance at the 5% level, and ^† indicates significance at the 10% level.

View Table

Table 2.

Pretest Regressions for Liquid Asset Demand

Independent Variable	Dependent variable: ln mj,t
	Level				Deviation from cross-section average
ln Sj,t	0.755** (0.005)	0.719** (0.012)	0.863** (0.009)	0.788** (0.023)	0.754** (0.005)	0.713** (0.012)	0.861** (0.009)	0.785** (0.023)
${ln}{\widehat{V}}_{j,t}$	-	-	-0.100** (0.008)	-0.105** (0.017)	-	-	-0.099** (0.008)	-0.108** (0.017)
Quarter-year effects	Yes	Yes	Yes	Yes	No	No	No	No
Fixed-firm effects	No	Yes	No	Yes	No	Yes	No	Yes
${\overline{R}}^2$	0.381	0.381	0.382	0.382	0.376	0.376	0.377	0.377
N	46,698	46,698	44,717	44,717	46,698	46,698	44,717	44,717

Notes: The regressions were performed with the S&P 500 data (659 firms) for the 1975:1–97:4 period. ${\overline{R}}^2$ excludes variance explained by the fixed-firm effects. Standard errors (in parentheses) are computed using White’s correction for heteroscedasticity.^**indicates significance at the 1% level,^*indicates significance at the 5% level, and ^† indicates significance at the 10% level.

Now we estimate the baseline regression models. To introduce aggregate variables, we control only for seasonal effects with industry-quarter dummies instead of using quarter-year dummies or demeaning variables by cross-section averages. Fixed-firm effects are still controlled. We consider two alternative specifications: the log and the ratio as in equations (14) and (15). For each specification, we perform two regressions using each of the two policy stance measures, the change in the Federal funds rate, ΔFFR, and the negative value of the mix of nonborrowed reserves and total reserves, MNBRX (see Table 3). In the log specification, regression (a) includes the market rate and the current policy stance, in addition to sales and inventory variables, and regression (b) additionally includes the lagged policy stances and a time trend. The regressors in the ratio specification correspond to those in the log specification except that regression (a), too, includes a time trend to account for a trend in sales and inventory ratios.

Table 3.

Baseline Regressions for Corporate Liquid Asset Demand

Notes: The regressions were performed with the S&P 500 data (659 firms) for the 1975:1-97:4 period. The market rate (R_t) and Federal funds rate (FFR_t) are fractions (annual percentage/100). All regressions include both the quarter-industry effects and fixed firm effects. ${\overline{R}}^2$ excludes variance explained by the fixed-firm effects. Standard errors (in parentheses) are computed using White’s correction for heteroscedasticity. ** indicates significance at the 1% level, * indicates significance at the 5% level, and ^† indicates significance at the 10% level.

Table 3.

Baseline Regressions for Corporate Liquid Asset Demand

Independent variable	Dependent variable: ln mj,t				Dependent variable: $M_{j,t}/{\widehat{A}}_{j,t}^N$
	ΔFFRt		MNBRXt		ΔFFRt		MNBRXt
	(a)	(b)	(a)	(b)	(a)	(b)	(a)	(b)
ln Sj,t	0.694** (0.019)	0.794** (0.024)	0.696** (0.019)	0.796** (0.024)	--	--	--	--
${ln}{\widehat{V}}_{j,t}$	-0.056** (0.015)	-0.112** (0.017)	-0.056** (0.015)	-0.113** (0.017)	--	--	--	--
$S_{j,t}^N/{\widehat{A}}_{j,t}^N$	--	--	--	--	0.107** (0.008)	0.107** (0.008)	0.108** (0.008)	0.108** (0.008)
$V_{j,t}^N/A_{j,t}^N$	--	--	--	--	-0.328** (0.011)	-0.327** (0.011)	-0.327** (0.011)	-0.327** (0.011)
Rt	-1.793** (0.161)	-3.240** (0.179)	-2.234** (0.198)	-3.332** (0.224)	-0.149** (0.011)	-0.145** (0.012)	-0.168** (0.014)	-0.153** (0.016)
μt	1.701** (0.361)	1.681** (0.372)	0.650** (0.134)	0.594** (0.172)	0.096** (0.025)	0.093** (0.026)	0.030** (0.013)	0.035** (0.016)
μt-1	--	0.799* (0.377)	--	-0.030 (0.206)	--	0.019 (0.025)	--	-0.007 (0.017)
μt-2	--	-0.000 (0.357)	--	-0.266 (0.214)	--	-0.017 (0.025)	--	-0.002 (0.015)
μt-3	--	-0.544 (0.359)	–	-0.086 (0.154)	--	-0.052* (0.025)	--	0.000 (0.012)
Tt	--	-0.004** (0.0002)	--	-0.004** (0.0002)	-7.0x10−4** (1.8xl0−5)	-7.0x10−4** (1.8xl0−5)	-7.0x10−4** (1.8xl0−5)	-7.0x10−4** (1.8xl0−5)
${\overline{R}}^2$	0.485	0.487	0.485	0.487	0.183	0.183	0.184	0.184
N	44,717	44,717	44,717	44,717	44,710	44,710	44,710	44,710

Notes: The regressions were performed with the S&P 500 data (659 firms) for the 1975:1-97:4 period. The market rate (R_t) and Federal funds rate (FFR_t) are fractions (annual percentage/100). All regressions include both the quarter-industry effects and fixed firm effects. ${\overline{R}}^2$ excludes variance explained by the fixed-firm effects. Standard errors (in parentheses) are computed using White’s correction for heteroscedasticity. ** indicates significance at the 1% level, * indicates significance at the 5% level, and ^† indicates significance at the 10% level.

View Table

Table 3.

Baseline Regressions for Corporate Liquid Asset Demand

Independent variable	Dependent variable: ln mj,t				Dependent variable: $M_{j,t}/{\widehat{A}}_{j,t}^N$
	ΔFFRt		MNBRXt		ΔFFRt		MNBRXt
	(a)	(b)	(a)	(b)	(a)	(b)	(a)	(b)
ln Sj,t	0.694** (0.019)	0.794** (0.024)	0.696** (0.019)	0.796** (0.024)	--	--	--	--
${ln}{\widehat{V}}_{j,t}$	-0.056** (0.015)	-0.112** (0.017)	-0.056** (0.015)	-0.113** (0.017)	--	--	--	--
$S_{j,t}^N/{\widehat{A}}_{j,t}^N$	--	--	--	--	0.107** (0.008)	0.107** (0.008)	0.108** (0.008)	0.108** (0.008)
$V_{j,t}^N/A_{j,t}^N$	--	--	--	--	-0.328** (0.011)	-0.327** (0.011)	-0.327** (0.011)	-0.327** (0.011)
Rt	-1.793** (0.161)	-3.240** (0.179)	-2.234** (0.198)	-3.332** (0.224)	-0.149** (0.011)	-0.145** (0.012)	-0.168** (0.014)	-0.153** (0.016)
μt	1.701** (0.361)	1.681** (0.372)	0.650** (0.134)	0.594** (0.172)	0.096** (0.025)	0.093** (0.026)	0.030** (0.013)	0.035** (0.016)
μt-1	--	0.799* (0.377)	--	-0.030 (0.206)	--	0.019 (0.025)	--	-0.007 (0.017)
μt-2	--	-0.000 (0.357)	--	-0.266 (0.214)	--	-0.017 (0.025)	--	-0.002 (0.015)
μt-3	--	-0.544 (0.359)	–	-0.086 (0.154)	--	-0.052* (0.025)	--	0.000 (0.012)
Tt	--	-0.004** (0.0002)	--	-0.004** (0.0002)	-7.0x10−4** (1.8xl0−5)	-7.0x10−4** (1.8xl0−5)	-7.0x10−4** (1.8xl0−5)	-7.0x10−4** (1.8xl0−5)
${\overline{R}}^2$	0.485	0.487	0.485	0.487	0.183	0.183	0.184	0.184
N	44,717	44,717	44,717	44,717	44,710	44,710	44,710	44,710

Notes: The regressions were performed with the S&P 500 data (659 firms) for the 1975:1-97:4 period. The market rate (R_t) and Federal funds rate (FFR_t) are fractions (annual percentage/100). All regressions include both the quarter-industry effects and fixed firm effects. ${\overline{R}}^2$ excludes variance explained by the fixed-firm effects. Standard errors (in parentheses) are computed using White’s correction for heteroscedasticity. ** indicates significance at the 1% level, * indicates significance at the 5% level, and ^† indicates significance at the 10% level.

Table 3 summarizes the baseline regression results. The log (columns 2–5) and ratio (columns 6–9) specifications provide qualitatively the same implications about monetary policy effects on liquid asset demand. Estimated coefficients have expected signs and are statistically significant, except for lagged values of the policy stance.²⁷ Most important, tighter policy has a contemporaneous, positive impact on liquid asset holdings. The subsequent impact of tighter policy tends to turn negative in two to three quarters and, in the ratio specification with ΔFFR, becomes marginally significant in three quarters. In all cases, the firms’ liquid asset holdings are significantly lowered with the market rate and the time trend.