trade credit

Access to Institutional Finance and the Use of Trade CreditAuthor(s): Christina AtanasovaSource: Financial Management, Vol. 36, No. 1 (Spring, 2007), pp. 49-67Published by: Wiley on behalf of the Financial Management Association InternationalStable URL: http://www.jstor.org/stable/30133749 .

Accessed: 12/01/2015 00:31

Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at .http://www.jstor.org/page/info/about/policies/terms.jsp

.JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range ofcontent in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new formsof scholarship. For more information about JSTOR, please contact [email protected].

.

Wiley and Financial Management Association International are collaborating with JSTOR to digitize, preserveand extend access to Financial Management.

http://www.jstor.org

This content downloaded from 111.68.99.250 on Mon, 12 Jan 2015 00:31:15 AMAll use subject to JSTOR Terms and Conditions

http://www.jstor.org/action/showPublisher?publisherCode=black

http://www.jstor.org/action/showPublisher?publisherCode=fma

http://www.jstor.org/stable/30133749?origin=JSTOR-pdf

http://www.jstor.org/page/info/about/policies/terms.jsp


Access to Institutional Finance and the

Use of Trade Credit

Christina Atanasova*

I develop a conceptual framework for analyzing the effect of the availability of institutional loans on firms' demand for supplier (trade) finance. I test for the existence of credit constraints and their effect on corporate financing policies. My empirical results support the hypothesis that trade credit is taken up by firms as a substitute for institutionalfinance at the margin when they are credit constrained. Further, in line with studies on the credit channel of monetary policy transmission, Ifind an increased reliance on trade credit by financially constrained firms during periods of tight money.

For most firms, trade credit is an essential element of their operations. In developed countries, the majority of firms rely heavily on trade credit extension as a source of finance. In a Federal Reserve Board study, Elliehausen and Wolken (1993) note that in 1987, accounts payable constituted 20% of all non-bank, non-farm, small businesses' liabilities and 15% of all large firms' liabilities. On the other hand, accounts receivable represents one of the main assets on most corporate balance sheets. Therefore, an important aspect of trade credit is the two-way nature of the transaction. Many companies, particularly those at intermediate points in the value chain, use trade credit as customers and provide it as suppliers. Thus, trade credit represents a substantial component of both corporate liabilities and assets.

Alongside this obvious economic importance, trade finance should be considered by policy makers because of its ability to affect the outcome of policy interventions. For example, Davis and Yeoman (1974) show evidence that large UK firms used trade credit to cushion themselves from tight monetary policy in the late 1960s.

In this paper, I study the interdependence between the two major sources of short- to medium- term corporate funds: 1) institutional loans (bank debt of short- to medium-term maturity, lines of credit, etc.) and 2) trade credit. Trade credit is a more expensive financing alternative to conventional loans because suppliers have a higher direct cost of funds. For example, for suppliers, these higher costs can take the form of inefficiencies in the collection of payments, but financial intermediaries enjoy cost advantages due to specialization.

Previous studies offer numerous examples of how, for some firms, financial market imperfections may create dependence on trade credit as a source of funds. Petersen and Rajan (1997) and Nilsen (2002) argue that firms that have no access to markets for traded long-term securities or commercial paper rely on trade credit for financing during economic downturns and monetary policy contractions. Deloof and Jegers (1999) provide empirical evidence that the amount of trade credit used is

I am indebted to Alexander Triantis and the anonymous referee for constructive comments and suggestions, which led to

significant improvement of this paper I would like to thank the participants at the 34th FMA conference in Denver, Kevin Reilly, Keith Glaister and Robert Hudson for helpful discussions and ICC for providing the data.

* Christina Atanasova is a Lecturer at the University of York, York, UK. Financial Management * Spring 2007 * pages 49 - 67



50 Financial Management * Spring 2007

determined by the availability of internal funds and is an important alternative not only for short- term bank debt but also for long-term financial debt.

Antov (2005) and Alphonse, Ducret, and Severin (2004), on the other hand, argue that the availability of institutional finance increases with the level of trade credit. Antov examines firms' choice to use trade credit and in particular the way in which the availability and level of institutional loans affect that choice. His finding that higher levels of trade credit are associated with higher levels of conventional loans suggests that there exists synergy arising from combining supplier and bank credit. Alphonse et al. provide similar evidence. They argue that trade credit is used as a quality signal that helps firms acquire reputation and improve their access to institutional finance. Both of these studies, however, base their analyses on cross-sectional data and assume a static environment and the absence of macroeconomic shocks.

The two lines of research are not mutually exclusive: firms use trade credit because they are denied access to institutional finance (a demand effect), but also trade credit granted by suppliers facilitates access to institutional loans (a supply effect). I examine the hypothesis that over time, variations in the use of trade credit are affected by the availability of cheaper substitutes, such as bank loans, and that this substitution effect is likely to be stronger during recessions and periods of tight monetary policy. In my analysis, I include macroeconomic variables that control for the prevailing monetary policy conditions and capture the supply side variation in institutional loans.

It is difficult to disentangle the supply and demand effects in a cross-sectional data analysis. However, in the context of panel data, I observe that over time, a certain class of borrowers systematically increases its use of trade credit when the level of institutional finance declines. This finding suggests a strong demand side effect.

I hypothesize that when credit constraints are binding, trade credit is a substitute form of short- term financing to conventional institutional loans. That is, when firms are constrained, I expect that the use of trade credit as a substitute source of funds for institutional loans is decreasing in bank loans, but when firms are unconstrained, I expect this substitution effect to be weaker, because firms with alternative sources of funds will avoid the more expensive trade credit. My empirical findings provide strong support for this proposition.

Testing for the effects of credit constraints on the choice of financing alternatives raises difficult identification problems. In previous studies, the basic strategy has been to divide the sample firms into groups that may face different degrees of information and agency problems. Most of these classifications are based on some time-invariant prior criterion, and the researcher estimates separate equations for each subsample.

However, there are several problems with this approach. First, it separates firms into groups on the basis of a single indicator that may not be a good proxy for credit quality and access to institutional finance. In general, the use of a single indicator prevents the researcher from controlling for the many factors that influence firm's borrowing ability. Second, in the conventional strategy, whether a firm belongs to the financially constrained or unconstrained group is determined exogenously and is fixed over the entire sample period. This approach is far too restrictive, since it does not allow for firms to switch between the constrained and the unconstrained groups over time.

To deal with potential misspecification problems caused by the issues discussed above, I follow the studies on credit market disequilibria that measure the impact of credit constraints directly. I use an endogenous switching regression model that allows me to derive the probability that a firm faces credit constraints directly from the distribution of the firm's characteristics. I estimate the model on the basis of individual firm-level data for a period of 20 years. When the time span of the sample is long, it is important that firms are allowed to switch between groups over time.



Atanasova * Institutional Finance and the Use of Trade Credit 51

Since I do not know a priori which firms are subject to credit constraints, I estimate a switching model with sample separation unknown. I assume that each firm at each point in time operates in one of two possible regimes: a credit-constrained regime or a credit-unconstrained regime.'

An advantage of my method is that it allows me to analyze how substitution in corporate financing alternatives is likely to vary over time with the prevailing macroeconomic environment and with the monetary policy conditions in particular. I use the patterns of borrowing constraints imposed by financial intermediaries to distinguish between demand and supply side effects. I find that when monetary policy is tight, the average probability of being in the constrained regime rises and the credit constrained firms increase their use of trade credit. This finding suggests that monetary policy may affect firms' access to institutional loans. Although contractionary monetary policy has an important effect on the supply of institutional credit by limiting its availability to a certain class of borrowers, it does not affect the demand for loans in the same direction. In general, the demand for credit will increase due to the reduced cash flow caused by a decline in product demand. By observing these patterns over time, I can begin to make inferences about the "cycle" of corporate financing.

The paper is organized as follows. Section I provides a review of the theoretical and empirical literature on the motives for trade credit use. Section II presents my switching regression model of the demand for trade credit and describes the estimation technique for the model. Section III describes the data set and presents some descriptive statistics. Section IV reports the estimation results for the model. Section V considers several checks to the robustness of the estimation results. Section VI examines the effect of the prevailing macroeconomic conditions, and in particular monetary policy stance, on the severity of credit constraints and corporate financing choice. Section VII concludes with a summary of the main findings.

I. Motives for Trade Credit Use

Long, Malitz, and Ravid (1993) point out that theoretical studies on trade credit have developed in different but not necessarily mutually exclusive directions. Although different theories attempt to explain the existence of trade credit (see Ng, Smith, and Smith, 1999), there is little systematic research on which firms are most reliant on trade credit and when.

A. Transaction Motive

Many firms use at least some trade credit as a normal part of their transaction cycle. The transaction theory of trade credit is mentioned in Schwartz (1974) and discussed in detail in Ferris (1981). For some firms, the primary benefit of trade credit is as a cash management tool. By delaying the payment for purchases, a firm may be able to match the timing of cash receipts from sales with the cash outlays for the cost of goods sold. The transaction theory predicts that this cost saving motive for trade credit use will help explain a significant part of the amount of accounts payable on firms' balance sheets.

At the time when the product is purchased, trade credit can also play a role in a firm's quality control efforts. Smith (1987) and Long, Malitz, and Ravid (1993) argue that the use of trade credit allows a firm to verify product quality before paying. These types of benefits suggest that

'I use these terms loosely. The constrained regime includes not only the case in which firms are literally rationed, but also when the premium on external finance is prohibitively high or the managers dislike the monitoring and control properties of debt.




the use and perceived importance of trade credit may be related to a firm's industry. If the complexity of the items purchased by a firm increases, the firm's transaction demand for trade credit may increase as well.

B. Financing Motive

The finance motive focuses on the customer as initiator of the post purchase "credit extension." Ferris (1981) argues that trade credit becomes less an instrument of trade and more an instrument of finance as the length of credit period increases, with the seller acting as a financial intermediary. Trade credit extension then becomes a type of short-term loan between seller and buyer that is tied to the exchange of goods in terms of value and timing (Franks et al., 1985).

Theory links the financing motive to credit market imperfections, which may cause financial institutions (the major source of business credit) to restrict credit to their customers. Although granting trade credit exposes the firm to financial risks, the supplier may be willing to offer financing to constrained borrowers because the firms have broader interests than the financial transaction. The supplier may benefit in the longer run by helping a struggling customer stay in business and therefore make future sales. The supplier is often in a better position to obtain information about the creditworthiness of a buyer than a financial intermediary. Contact from the selling process can facilitate the monitoring of customers on an ongoing basis and, as Smith (1987) suggests, suppliers can also use two-part terms to obtain information on credit worthiness. Biais and Gollier (1997) and Jain (2001) argue that suppliers have private information about their customers that banks do not. Thus, savings in monitoring costs and the informational advantage may explain why some firms provide trade credit. Habib and Johnsen (1999) suggest suppliers have repossession advantages when redeploying the asset sold and Wilner (2000) argues that restructuring debt advantages explain why trade credit is being offered.

Previous empirical work on the financing motive of trade credit provides mixed evidence. Long, Malitz, and Ravid (1993) find no support for the financing motive: less creditworthy nonfinancial firms do not apply to more creditworthy firms for financing due to credit constraints. Ng, Smith, and Smith (1999) obtain similar results, but Antov (2005) and Alphonse, Ducret and Severin (2004) find that firms with high levels of trade credit have high levels of institutional loans.

In contrast, Nilsen (2002) in the case of the U.S., and Biais, Hillion, and Malecot (1995) for France, show that small firms, which are more likely to be credit rationed, rely heavily on trade credit when credit market conditions deteriorate. Also, Petersen and Rajan (1994, 1995) find that firms that are less likely to be bank credit constrained tend to rely less on trade credit. It is possible that the ambiguous empirical evidence on the financing motive of trade credit use is due to the static and dynamic misspecifications caused by the conventional classifications used to split sample firms into constrained and unconstrained (e.g., small versus large firms).

When financial institutions restrict credit, those firms depending on intermediated finance are forced to use trade credit. Given that trade credit terms remain constant over time (see Ng, Smith, and Smith, 1999), during recessions and monetary contractions, trade finance may become a relatively cheaper source of funds for some firms. It is this imperfect substitutability of trade credit and institutional finance that allows me to observe systematic differences in the use of trade credit. Therefore, the increase in the demand for trade credit as a source of finance during monetary contractions provides a test for the existence of a credit channel of monetary policy transmission. These arguments are consistent with Laffer's (1970) suggestion that trade credit should be considered a part of money supply.




II. A Switching Model of Trade Credit

To begin, I conceptualize the link between credit constraints and trade credit demand. I represent firm i's notional demand and supply curves for institutional loans at time t by LD (r, Z,,, u,,) and Ls (r,, Z,, u2,i,), respectively. I use r1 to denote the market interest rate.

Z, is a

vector of observable firm characteristics that determine the demand and supply of loans for firm

i at time t, and u1,, and u2,it are variables that represent the unobservable characteristics for the same firm in the same time period. I denote the firm's excess demand for loans at time t by:

L*it LD(r,,ZitU!~ir)-LS(rrZirUZ.ir). (1)

The excess demand quantity L*, is not observable. I define an indicator variable Li.

if it 0 (2) otherwise.

I am interested in the determinants of the probability that there is excess demand for loans, i.e., Pr(L*, > 0). I assume that credit availability is a function of the firm's characteristics, and that for firm i at time t I can write the excess demand quantity as L*,, = Z' y + u3,i, where

, is a

parameter vector (to be estimated) and u3,t

is the innovation term that captures unobservable

qualities of borrowers. Then Pr(L*,,

> 0) = Pr(Z',1 y + u3,it >0). (I note that if u3,it is Gaussian with mean zero and variance one, this formulation leads to a standard probit model.) The unconstrained firms receive the desired quality of bank credit (L, = 0), but for the constrained firms I observe that the maximum credit granted is LS, since LDi, > Li.

I denote the quantity of trade credit obtained by the constrained and unconstrained firms by TCC, and TC' ,, respectively. In general, for firm i at time t I can write the expected level of trade credit as:

E(TCUII I L, = 0) = X ', /u

-+ ULS, + E(E, I L, = 0) (3)

or

E(TCC, I = 1) = X

/c +c + = 1). (4)

In the two Equations (3) and (4), XA, is a vector that includes the two types of observed heterogeneity that determine the level of trade credit. One is a group offirm-specific characteristics, and the other is a group of variables specific to particular supplier practices common for firms in the same industry. The characteristics that are not observable to the econometrician, but that affect the level of trade credit used, fall into the error terms El,it and

62,it. According to my proposition, when credit constraints are binding, trade credit is a substitute to conventional institutional loans. I expect that the amount of trade credit used by the constrained firms will decrease with the amount of short-term institutional loans received, i.e., 6c < 0; and that there should be no such substitution effect for the unconstrained firms, i.e., 86U < 0. If trade credit complements conventional loans, then I expect 6c > 0 and 36U > 0.

To obtain a measure of trade credit usage, I use the ratio of accounts payable to total assets. (I note that all stock variables denote end-of-period values.) This ratio gives the percentage of total assets that is financed by trade credit, and thus represents a firm's reliance on interfirm credit. For similar reasons, I scale institutional loans (LS,) by total assets.

In the empirical specification of the switching model, firm i at time t operates in the unconstrained regime with a demand for trade credit equation defined by:

A =__

X ,,

f it + if Z'1it

Y+u3,it

<0 (5) TA, , --




or the firm operates in the constrained regime with a demand equation defined by:

Ai' = X 'it 6C C +

E2,it if Z Y + u3,it

2 0. (6) TA, ,

TA

A. Model Estimation

I estimate a switching regression model in which I obtain observations from two different regression regimes, but the separation into these regimes is unobservable a priori. I assume that the vector of error terms

(y,;,, e2,it u3,it)' is jointly iid Gaussian with mean zero and covariance

matrix 1, where (3x3) -0ij

} i,j = 1,2,3. The nonzero covariance between E,,, E2,it and

u3,it (endogenous switching) allows for possible correlation between the shocks to the demand for trade credit with the shocks to firms' financial and other characteristics. Although I cannot observe which regime the firm is in, I can specify the probability with which each regime occurs as follows:

Pr[A, ITA,, =(AP.,/TA,, =

pr[ Z',

y+u3,it <0] (7)

'- (7) =

Pr[u3,it <-Z ', r] = (-Z 'it Y)

Pr[Ai,/TAp,,i,

= (AjP, / TA,,_

)c = PrZ', u,it 0] (8)

= 2 -Z ', y] =

The likelihood function for each observation AP, / TAi

,_ is a weighted average of the conditional density functions of l,it and

2,it with weights Pr u3,, <-Z',, y] and Pr[u3,it -Z', y]. The likelihood function is given by:

it --O(

I

[/'/,i,

< -Z

'i, Y])

'it 'it Y)]

1 -

Z (9)

011 )

where b(.) and P(.) are the normal probability density and the cumulative distribution functions, respectively.

q(Ei, .) denotes the conditional density, and

q(e,1, 0 ) denotes the marginal density

function. The second equality of Equation (9) uses the fact that a joint density equals the product of the

conditional density times the marginal density and the properties of the bivariate normal distribution. Like the probit model, I cannot estimate 7 and

-33 separately, so without loss of

generality I normalize 33 = 1. Also, -12

does not appear in Equation (9) and thus is not estimable.

For a panel of N firms with T observations for firm i, the log-likelihood function for all the observations is given by:

NT

L =

log(i,t). (10)

i=1 t=l

I estimate the parameter vectors /", 3" and y and the parameters 6" and 6" by maximizing the log-likelihood function through the EM algorithm of Dempster, Laird, and Rubin (1977).




If I knew which regime generated each data point, the problem would be easy to solve. To start the iterative procedure, I split firms into two regimes according to their size. (I classify firms with total assets above/below the median total assets for given year as large/small in that year.) When the procedure eventually converges, I overwrite the initial guess variable to contain the probabilities that a given observation adheres to the first (unconstrained) regime.

Given an initial guess of the partition of the observations into the two regimes, the algorithm estimates the classification vector, i.e., the probability that a given observation is in the unconstrained regime (the expectation step). I obtain this estimate by reweighing the probabilities based on the errors of the observations in the two regression Equations (5) and (6). I then use the updated probabilities to weight observations in each of the two equations (the maximization step). As noted above, this iterative procedure eventually converges to the Maximum Likelihood estimates (MLEs) of the parameters.

To ensure robustness to heteroskedasticity and within firm (cluster) correlation, I compute bootstrapped standard errors as in Douglas (1996). I draw firms (clusters) with replacement rather than drawing observations with replacement. I note that the general bootstrapping procedure of drawing observations with replacement does not necessarily control for heteroskedasticity and within-firm clustering. On the other hand, bootstrapping by firms takes account of these problems.

An important question is whether the estimated endogenous switching model is statistically significant relative to a linear single-equation model. Although the single-equation model is nested within the switching model, under the null hypothesis of linearity there are several unidentified parameters of the switching model. This problem complicates the calculation of the degrees of freedom. In addition, the asymptotic likelihood ratio statistic does not have the conventional chi-square distribution. However, Monte Carlo tests suggest that setting the degrees of freedom equal to the number of constraints plus the number of unidentified parameters yields a conservative test when I use the chi-square distribution.

From the likelihood density function I calculate both the unconditional and conditional probabilities of each sample firm being in the constrained regime. The unconditional probability does not take into account the information about the amount of trade credit

used, and is simply Pr[u3,it -Z'i, 7]. The conditional probability takes into account this information and updates the unconditional probability according to Bayes' rule. The conditional probability that a firm faces credit constraints given the information about the amount of trade credit used is:

Pr u3,it 2 -Z '

l A], / =

(-2, 3,it

For brevity of exposition, I report only the unconditional probability of being in the constrained regime.

III. Description of Data and Variables

In my empirical work, I use panel data on UK firms provided by ICC. The ICC database covers the entire population of limited-liability UK firms. I select a random sample of 2,000 firms over a 20-year period, from 1981 to 2000, as a stratified sample from the ICC SIC codes. The sample firms in this study are of all sizes and industries excluding firms in sectors such as financial




intermediation and other financial services. The sample consists of 1,839 (91.9%) private limited companies and 162 (8.1%) public limited companies. Unlike in the US, where only quoted firms are required to file their quarterly or annual accounts, UK firms must disclose their accounts even if they are not traded on the stock exchange.

A. Vector of Variables Z

In the switching function, in addition to age, industry, and year dummies, the vector Z, includes a set of variables that are indicators of financial strength. The financial variables measure a firm's ability to raise capital and offer collateral. I measure liquidity by the level of profits (PROF/TA) and the quick ratio (current assets over current liabilities, QR). I use assets (logTA) and tangibility (depreciation over assets, DEPR/TA) as my proxy for the firm's ability to collateralize its debt. To control for the capital structure of the firm, I include the values of the firm's leverage ratio (LR=DEBT/TA) and of the coverage ratio (operating income-to-interest ratio, CR). I also include in the switching function a status dummy variable PD,, which equals one if ith company is public and zero if it is private. I expect that private firms will find it more difficult to acquire external finance than will publicly held firms, both because of information opacity and limited access to public capital markets. To investigate whether financial weakness is particularly important for the borrowing ability of private firms, I allow for the interaction between the financial ratios and the status dummy.

Theoretical factors suggest that a firm is more likely to face credit constraints when its level of profits, assets, and tangibility are low; when its quick and coverage ratios are low; and when its leverage ratio is high (Petersen and Rajan, 1997). Private firms are more likely to face information and agency problems. Older firms tend to be more diversified, less prone to bankruptcy, have a better track record, and therefore should suffer less severe agency costs. Thus, I expect the coefficients of age and status dummy to be negative.

Including year dummies in the switching function captures changes in the general financial and macroeconomic conditions that affect all firms in the same way. In years characterized by recessions or tight monetary policy, I expect the coefficient of the year dummy to be larger. In addition to the observed firm-specific variables explicitly included in the model, I may want the probability of facing credit constraints to depend on unobservable firm-specific, time-invariant factors. If I treat these factors as fixed effects, I can introduce 2,000 firm-specific dummies in Z and X. Although this method of handling fixed effect is possible in principle, in practice it is not feasible since it leads to a huge loss of degrees of freedom. To be inverted in the maximization procedure, the matrix is of dimension (N+K), where N=2000 and K is the dimension of either Z or X. The inclusion of industry dummies, for each two-digit SIC code serves to capture these effects.

I note that the negative of the sum of the coefficients of the year and industry dummies can be interpreted as a threshold. If a linear combination of the financial variables and age exceeds this threshold, then I expect the firm to be in the constrained regime in the sense that Z',it ~ 2 0. I also note that the ex ante probability of being in the constrained regime is

Pru 3,, it - 'j 1 - (-Z 'i, y).

B. Vector of Variables X

In the trade credit regressions, the vector X, includes a set of observable firm-specific variables, industry-wide averages, and industry and year dummies. The demand for trade credit attributable to the financing motive is associated with the cost and availability of substitute sources of funds.




A pecking-order behavior (Myers and Majluf, 1984) would suggest that internally generated funds are higher in the pecking order than trade credit. Therefore, the firm's ability to generate cash internally will decrease its demand for trade credit. I calculate cash flow as profits plus depreciation and use the cash flow-to-asset ratio (CF/TA) as my proxy for the firm's reliance on internally generated funds.

The general demand for funds depends on the expected future profitability. I expect firms with higher expected future profitability to have a higher proportion of their assets financed with accounts payable. I use the sales-to-assets ratio (SALES/TA) as my proxy for future profitability. (I note that the results are not materially different when I use sales growth as my proxy for future profitability instead of the SALES/TA ratio.)

Another dimension of the financing motive for trade credit demand is the maturity of the planned investment. Finance studies suggest a maturity-matching approach (Deloof and Jegers, 1999). Therefore, firms with more short-term assets will have a higher demand for short-term credit in general and trade credit in particular. To capture this effect I add three categories of short-term assets: 1) inventory-to-assets (INV/TA), 2) accounts receivable-to-assets (AR/TA), and 3) cash holdings-to-assets (CASH/TA) ratios.

The industry-wide averages control for the effect of common supplier practices. I define these averages as follows. The demand for trade credit attributable to the transaction motive depends on the firm's uncertainty in transactions with its suppliers. Industries associated with a higher variability of inventory levels should have a higher average volume of purchases from suppliers as their inventories are depleted, and therefore a higher demand for trade credit. The standard deviation of inventory turnover (SD(TURN)) represents the variability or uncertainty in transactions with suppliers.2 I also include the industry-wide average of accounts payable over total assets (AP/TA), since the greater the industry practice in relying on trade credit, the more a firm operating in that industry will use it.

Overall business conditions can have a more direct effect on short-term financing decisions, so I also include year dummies in the trade credit equation. Further, I include industry dummies for each two-digit SIC code. The inclusion of industry dummies captures some of the influence of the unobservable firm-specific effects.

C. Descriptive Statistics

Table I reports means and medians of firm level variables for public and private companies. The t tests show that the means are significantly different at 1% for every firm characteristic except for profit-to-assets, quick, and leverage ratios. The results are robust to using sales instead of assets as the scaling factor and to controlling for every two digit SIC industry code.

Table I clarifies several facts. The first is that public firms are much larger than private firms, as measured by all three indicators of size: volume of sales, value of total assets, and number of employees. Public companies also have lower SALES/TA ratio as well as lower INV/TA and AR/ TA ratios. Public firms appear to rely less on short-term loans and trade credit than do private firms.

The low sales SALES/TA ratio of public firms may reflect a lower cost of capital, which in equilibrium may imply a more capital-intensive production process.

2 It is possible that the greater uncertainty may be associated with a large deviation of the sales to inventory ratio from its "optimal level," which has been changing over time due to the introduction of new inventory management techniques in the early 1990s. I perform tests for possible structural breaks in the trade credit equation. The results of the tests do not suggest the presence of a break. The results are available on request.




Table I: Descriptive Statistics for Public and Private Firms

The sample contains 2,000 UK firms from 1981 to 2000. The mean and median numbers are in British pounds. Standard deviations are reported in parenthesis.

Public Private

Variable Mean Median Mean Median

log(SALES) 17.6 (2.3) 17.3 15.9 (1.6) 15.7 log(TA) 17.8 (2.3) 17.6 15.5(1.8) 15.3 SALES/TA 1.2(1.2) 1.2 2.3 (15.4) 1.6 Employees 6934.8 (19913) 418 467.7 (2582) 85 Short-term Liabilities AP/TA 12.6% (12.2) 10.2% 16.6% (17.2) 12.5% L/TA 11.9% (17.4) 6.9% 21.3% (97.9) 11.8% Short-term assets CASH/TA 6.2% (9.5) 2.5% 8.8% (14.4) 2.1% AR/TA 16.9% (15.3) 14.2% 22.4% (18.2) 20.2% INV/TA 17.9% (16.4) 16.0% 20.4% (18.9) 17.0% Liquidity PROF/TA 0.051 (0.16) 0.058 0.077 (2.79) 0.052 QR 3.8 (98.8) 1.3 3.3 (50.5) 1.3 Capital structure LR 50.7% (40.5) 44.3% 56.7% (251.6) 44.4% CR 46.8 (674) 3.9 23.7 (258) 3.0

Second, public firms hold lower stocks of inventories relative to assets, which may reflect differences in factor costs related to capital market imperfections. Stocks of inventories relax financial constraints, since they are the preferred form of collateral for bank loans. Also, inventories (particularly raw materials, which constitute the bulk of inventories) are easily appraised and liquidated. Thus, firms that depend on institutional credit may find it advantageous to select material-intensive production techniques, or they may be required by lending covenants to maintain a minimum amount of working capital.

IV. Estimation Results

In Table II, I report the estimated parameters of the conventional linear equation model. I report robust standard errors to control for heteroskedasticity and within cluster (firm) correlation.

Column (1) in Table II contains the estimated parameters for the whole sample. I estimate the panel data model with random effects. The results show that the demand for trade credit decreases with institutional loans. The estimation results support both the maturity-matching hypothesis and the pecking-order theory.

The coefficients of cash holdings, inventory and accounts receivable are positive and highly significant. In contrast, internally generated funds have a negative effect on the demand for trade credit. The coefficient of the standard deviation of the industry-wide inventory turnover ratio is positive and highly significant, which suggests that greater uncertainty of transactions




Table II: Estimation of Linear Regression Model

The table shows the linear regression results. The regression is:

i

L = f + + Ei. (T.2.1) TA.,_ TAi.,_

The sample contains 2,000 UK firms from 1981 to 2000. Standard errors robust to heteroskedasticity and within firm clustering are reported in parentheses.

Whole sample Small firms Large firms

L/TA -0.06298 (0.0023)*** -0.10592 (0.0056)*** -0.04883 (0.00246)*** X variables CF/TA -0.0003 (0.0002) -0.11545 (0.0099)*** -0.00066 (0.00246) SALES/TA 0.0058 (0.0005)*** 0.00581 (0.0005)*** 0.01225 (0.00144)*** AR/TA 0.1102 (0.0046)*** 0.144879 (0.0097)*** 0.10033 (0.005)*** INV/TA 0.0331 (0.00512)*** 0.08697 (0.01046)*** 0.0164 (.005615)*** CASH/TA 0.024 (0.0059)*** 0.00976 (0.0117) 0.02772 (0.0063)*** Industry averages SD(TURN) 0.0002 (5.89e-06)*** 0.0001 (7.00e-06) 0.0003 (1.73e-06)*** AP/TA 0.0036 (0.0005)*** 0.003 (0.00977) 0.003 (0.00098)*** Log-likelihood 12591.31 4834.57 8712.10

*** Significant at the 0.01 level.

with suppliers is associated with greater demand for accounts payable. The coefficient of the industry-wide average of accounts payable to assets is also positive and significant.

The second part of Table II reports the estimated parameters of the linear model for the subsamples of small and large firms. I find that for small firms, trade credit is decreasing with institutional finance. The coefficient of institutional loans is larger than the coefficient for the whole sample and is highly significant. For large firms, trade credit is also decreasing with institutional finance. Although the coefficient of loans is smaller, it is still highly significant. This finding is at odds with the fact that researchers assume that large firms have wide access to institutional finance.

Table III presents the estimated parameters for the switching model. The constrained-regime equation resembles the linear model, but with some differences. Most notably, the effect of institutional loans on the demand for trade credit is much stronger, which is also true of the effect of internally generated funds, albeit their effect is only weakly significant. On the other hand, the unconstrained-regime parameter estimate for institutional loans contrasts sharply with the linear model estimates. The coefficient on institutional loans is not statistically significant at conventional levels and is much smaller in magnitude. The rest of the results are similar to the estimates in Table II.

Table III also reports the estimated parameters of the switching function. These results are consistent with the findings of previous studies. In particular, the coefficients on the profits, assets, and tangibility, quick, and coverage ratios are significant and negative, and the coefficient on the leverage ratio is significant and positive. The age coefficient is negative.

In line with the discussion in the previous subsection, the estimated coefficient of the status dummy has a negative sign. Of the interaction terms, the coefficient of the leverage ratio is




Table IIl: Estimation of Switching Regression Model

The table shows the switching regression results. The regression equations are:

Ai' X ',,

x u + 8" , +

Elit

if Z'it Y + u3,it

< 0. (T.3.1) TAT TA1

T A X 'it c c + if Y +

u3,it 20. (T.3.2)

The sample contains 2,000 UK firms from 1981 to 2000. Robust bootstrapped standard errors are reported in parentheses.

Switching Model

C regime UC regime

L/TA -0.1251 (0.0114)*** -0.0158 (0.034) X variables CF/TA -0.0369 (0.028) -0.046 (0.0084)*** SALES/TA 0.0189 (0.003)*** 0.003 (0.001)*** AR/TA 0.497 (0.012)*** 0.189 (.0066)*** INV/TA 0.3346 (0.012)*** 0.099 (0.005)*** CASH/TA 0.284 (0.019)*** 0.0001 (0.006) Industry averages SD(TURN) 0.00003 (0.00001)*** 0.00002 (7.31e-06)*** AP/TA 0.0433 (0.021)*** 0.051 (0.0103)***

Z variables Switch Function

Age -0.0121 (0.0002)*** PROF/TA -0.449 (0.022)*** QR -2.799 (0.0036)*** logTA -0.088 (0.002)*** DEPR/TA -7.741 (0.113)*** LR 0.0001 (4.42e-06)*** CR -4.071 (0.009)*** PD -1.581 (0.099)*** PD*QR 1.551 (0.0096)*** PD*LR -0.001 (0.00003)*** PD*CR 0.80 (0.02)***

Log-likelihood 12875.42


significant and negative; the quick and coverage ratios are significant and positive. This result implies that for private firms, high levels of indebtedness and low liquidity increase the probability of facing credit constraints. But for public firms, a high level of debt increases the probability of being in the unconstrained regime. For these firms, having obtained debt in the past is perceived as a good signal. Overall, my results confirm that indebtedness plays a different role for the two types of firms.

Using the likelihood ratio test, the null hypothesis of linearity is rejected at any conventional level of significance. The nonlinear switching regression model fits the data much better than




does the linear model. This is the case when I estimate the model with the whole sample, and when I estimate it with the two subsamples of small and large firms. The switching model indicates that when facing constraints on institutional finance, firms use trade credit as a substitute to mitigate these constraints. The overall results support the important role of credit market imperfections in corporate financing behavior.

V. Further Investigation

The major caveats of our estimation procedure concern the use of industry dummies to control for firm-specific fixed effects, and that some of the determinants of the demand for trade credit are endogenously determined. For instance, since the variable institutional loans is a choice variable for the firm, it is correlated with the innovation term of the trade credit equation. Here, I evaluate the size and significance of the possible bias due to these caveats.

I split the sample into firm-years for which Z ', < 0 (unconstrained) and firm-years for which Z',, 2 0 (constrained). I estimate a random-effect unbalanced panel model for each subsample, using a 2SLS IV procedure, in which the instruments are the exogenous variables (industry-wide averages) and the lagged value of the endogenous regressors (observed firm-specific variables).

Table IV reports the estimation results. There is no material difference between the results in Table IV and those in Table III, although some of the main results appear to be even stronger. This finding implies that the endogeneity of short-term loans and the inclusion of industry dummies does not generate substantial bias in my original results.

Although this procedure is a robustness check on my main results, it is not completely satisfactory. To provide further results for the effect of the unobservable firm-specific heterogeneity, I demean the data by using the within transformation, and re-estimate the switching model. The within- transformation removes the status dummy PD and reduces the variation of the industry averages.

Table V reports the results, which are clearly similar to the estimated parameters in Table III. The unobserved firm specific effects do not appear to have a significant effect on the parameter estimates. This evidence suggests that in contrast to the case in which firm effects are important and so larger firms are generally less constrained, as firms grow larger over time, access to institutional finance for them becomes easier.

VI. Credit Constraints and the Business Cycle

In this section, I provide evidence based on a balanced subsample of firms where the data are available continuously from 1983 to 1999. I choose a balanced sample because my objective here is to assess how macroeconomic conditions, and monetary policy in particular, affect the probability (across firms) that firms face credit constraints. Choosing a balanced sample eliminates the possibility of the average probability changing over time because of a changing sample composition. I choose the length of the new panel to strike a balance between having a period that is long enough to cover substantial business cycle changes, but not so long as to reduce excessively the number of firms in the sample, and therefore the cross-sectional variation in the data. The final sample comprises 676 firms. I estimate the switching regression model using the balanced sample. The estimation results are not materially different.3

3 The estimation results are available on request.



62 Financial Management ' Spring 2007

Table IV: 2SLS Random-Effects IV Estimation

The table shows the instrumental variables regression results. I split the sample on the basis of the estimated switching function and perform panel data analysis for each subsample. The regression equations are:

AgP L s

TA X ',, fu

+ Su + (, if Z ', J < 0. (T.4.1) TA , , TA, A,

The sample contains 2,000 UK firms from 1981 to 2000. The instruments are the exogenous variables (industry-wide averages) and the lagged value of the endogenous regressors (observed firm-specific variables). Standard errors robust to heteroskedasticity and within firm clustering are reported in parentheses.

Constrained firms Unconstrained firms

L/TA -0.063 (0.0023)*** -0.0536 (0.112) X variables CF/TA -0.0005 (0.0025) -0.0523 (0.01745)*** SALES/TA 0.0058 (0.00051)*** 0.0023 (0.0056) AR/TA 0.1094 (0.0047)*** 0.3059 (0.0354)*** INV/TA 0.03437 (0.0053)*** 0.03773 (0.0113)*** CASH/TA 0.0221 (0.0061)*** 0.07222 (0.0217)*** Industry averages SD(TURN) 8.89e-06 (5.97e-06) 2.79e-06 (0.00001) AP/TA 0.602 (0.046)*** 0.393 (0.106)***


In the switching regression model, through both the balance sheet variables and the year dummies included in the switching function, general financial macroeconomic conditions will affect the probability of being in one or the other regime. For example, deteriorating macroeconomic conditions can be reflected in the decreasing collateral value of the firm's assets. Similarly, less cash flow will be available to cover interest payments. Including the year dummies in the switching function captures changes in general macroeconomic conditions that affect all firms in the same way, and that are not accounted for by changes in firm-specific variables. In years characterized by recession or contractionary monetary policy, the theory predicts a larger value of the coefficient of the year dummy, since firms are more likely to face financial constraints.

Table VI reports the coefficients of the year dummies in the switching function of the estimated balanced switching model. The table also presents, for each year, the probability of being in the constrained regime, which I obtain by averaging individual probabilities across firms in that year. I interpret this average probability as a more complete summary indicator of the effects of macro shocks, since it incorporates both the effect of year dummies and changes in firm-specific variables. The results in Table VI show that the year-dummy coefficients and the average probability of facing financial constraints are the highest in the recession years 1990, 1991, and 1992.

I investigate whether the time variation in the coefficients of the year dummies captures any changes in the stance of monetary policy. To address this question, I use the Bank of England




Table V: Estimation of the Switching Model with Firm Specific Effects

The table shows the switching regression results with firm specific effects. I demean the observations for each variable in Z (zi, - 2i.) and X (x, - xi.) and re-estimate the switching model. The within-transformation removes the status dummy PD and reduces the variation of the industry averages. The sample contains 2,000 UK firms from 1981 to 2000. Standard errors robust to heteroskedasticity and within firm clustering are reported in parentheses.

Switching Model

AP/TA C regime UC regime

L/TA -0.1609 (0.038)*** -0.0158 (0.0389) X variables CF/TA -0.04379 (0.0057)*** -0.0521 (0.0058)*** SALES/TA 0.01696 (0.00597)*** 0.00172 (0.0006)*** AR/TA 0.16272 (0.00581)*** 0.16057 (0.00593)*** INV/TA 0.0682 (0.00655)*** 0.067267 (0.0067)*** CASH/TA 0.1883 (0.00713)*** 0.0189 (0.00735)*** Industry averages SD(TURN) 1.07e-06 (3.15e-06) 1.19e-06 (3.19e-06) AP/TA 0.0065 (0.0056) 0.00787 (0.0058)


Age -0.0117 (0.0002)*** PROF/TA -0.2405 (0.01635)*** QR -4.1845 (0.016)*** logTA -0.09596 (0.0021)*** DEPR/TA -9.4926 (0.1102)*** LR 0.00019 (0.00004)*** CR -3.8241 (0.0534)*** PD*QR 1.8109 (0.01643)*** PD*LR -0.0006 (0.00004)*** PD*CR 0.00052 (0.00003)***


base rate as an indicator of how restrictive monetary policy is (see Bernanke and Blinder, 1992). I note that the results are not materially different if I use the spread between the T-bill and base rate to measure the stance of monetary policy.

I include the average value of the monthly level of the base rate for the previous year as a regressor in the switching function instead of the year dummies. Table VII reports the estimation results.

The table indicates that the coefficient of the lagged base rate is positive and highly significant. Keeping in mind that the yearly nature of the data does not allow me to capture a complete picture of the dynamic effects of changes in base rate in an adequate manner, the results nevertheless suggest that tight monetary policy increases the probability that firms find themselves in the constrained regime.

Next, I analyze the time pattern of the average probability of facing financial constraints and examine the implication of the results for the existence of a credit channel of monetary policy transmission. Figure 1 shows that there is a clear comovement between the average probability of being in the constrained regime and the base rate, with the former following the latter with a lag. This finding suggests that restrictive monetary policy may work through the effect it has on firms' access to institutional finance.




Table VI: Time Profile of Coefficients on Year Dummies and Average Probability of Facing Financial Constraints

Column (2) reports the coefficients of the year dummies in the switching function of the estimated balanced switching model. Column (3) presents, for each year, the probability of being in the constrained regime, which I obtain by averaging individual probabilities across firms in that year. The sample contains 676 UK firms from 1983 to 1999. Robust bootstrapped standard errors are reported in parentheses.

Year Year Dummy Average Probability 1984 -1.01 (0.03)*** 0.397886 1985 -1.04(0.03)*** 0.398021 1986 -1.05 (0.03)*** 0.400711 1987 -1.05 (0.03)*** 0.399459 1988 -1.06(0.04)*** 0.395344 1989 -1.04(0.04)*** 0.396031 1990 -0.09 (0.02)*** 0.400324 1991 -0.10 (0.02)*** 0.405676 1992 -0.12(0.02)*** 0.400720 1993 -0.96 (0.02)*** 0.399459 1994 -1.02 (0.02)*** 0.395874 1995 -1.02(0.02)*** 0.391206 1996 -1.02 (1.66) 0.392374 1997 -1.01 (0.17)*** 0.391074 1998 -1.00(0.02)*** 0.391704 1999 -1.01 (1.07) 0.392451


Overall, my evidence provides support for the existence of a credit channel of monetary policy transmission.

VII. Conclusion

In this paper I analyze quantitatively the effects of credit market imperfections on corporate short-term financing behavior. The asymmetric information theories predict that these effects depend on various firm characteristics, the overall macroeconomic conditions, and the stance of monetary policy. Firms with weak balance sheet positions and for which asymmetric information problems are more severe will rely on trade credit as a source of funds, despite trade credit being an unattractive substitute to conventional institutional loans.

To test these predictions, I develop a switching model in which I assume that firms operate in either an unconstrained or a constrained regime. I then determine the probability endogenously by a switching function. The endogenous determination of each regime attempts to overcome misspecification problems that arise when I use an a priori criterion to classify firms into constrained and unconstrained regimes.

My empirical formulation allows me to control and test for the multiplicity of factors that may affect firms' access to institutional credit. It also recognizes that the significance of such factors may vary over the business cycle.

I estimate the model using panel data for UK firms. The empirical evidence strongly suggests that trade credit is an important source of funds for constrained firms, but that unconstrained




Table VII: Estimation of the Switching Model with Monetary Policy Variable

The table shows the switching regression results. The sample contains 2,000 UK firms from 1981 to 2000. I include the average value of the monthly level of the base rate for the previous year as a regressor in the switching function instead of the year dummies. BR represents the level of lag base rate. Robust bootstrapped standard errors are reported in parentheses.

Switching Model

C regime UC regime

L/TA -0.1257 (0.0114)*** -0.0157 (0.034) X variables CF/TA -0.0374 (0.028) -0.046 (0.008)*** SALES/TA 0.0188 (0.003)*** 0.0025 (0.0069)*** AR/TA 0.496 (0.012)*** 0.187 (.0066)*** INV/TA 0.3326 (0.012)*** 0.098 (0.005)*** CASH/TA 0.283 (0.019)*** 0.0003 (0.006) Industry averages SD(TURN) 0.0000231 (0.00001)*** 0.000058 (7.33e-06)*** AP/TA 0.05299 (0.021)*** 0.0542 (0.0101)***


Age -0.0121 (0.0002)*** PROF/TA -0.431 (0.022)*** QR -2.774 (0.0036)*** logTA -0.091 (0.002)*** DEPR/TA -7.781 (0.113)*** LR 0.0001 (4.44e-06)*** CR -4.031 (0.009)*** PD -1.68 (0.0992)*** PD*QR 1.538 (0.0096)*** PD*LR -0.001 (0.00003)*** PD*CR 2.77 (0.02)*** BR 0.0169 (0.002)***


firms avoid using it. Such evidence suggests a strong influence of credit market imperfections on corporate financing behavior.

I also show that low levels of liquidity, profitability, and values of financial ratios increase the probability of facing credit constraints. I also find some evidence that high leverage is associated with greater reliance on trade credit for private firms, but not for public firms. I also find that the likelihood of being in the constrained regime varies over the business cycle with general macroeconomic conditions and the stance of monetary policy.

My study extends previous research on trade credit by considering explicitly the effects of business-cycle dynamics on firms' financing decisions. The credit channel of monetary policy transmission predicts that when monetary policy tightens, the reduction of institutional loans causes some firms to cut spending independently of changes in the cost of capital. I demonstrate that credit-constrained firms increase their use of trade credit, which is an unattractive and expensive substitute for bank credit. This identification from the supply side provides support for a potent channel of monetary transmission.E



66 Financial Management* Spring 2007

Figure 1. Average Probability of Constrained Regime and the Base Rate The figure provides information about the comovement between the average probability of being in the constrained regime and the Bank of England base rate. The left axis of the plot shows the average probability scale. The right axis shows the base rate scale.

Average probability Base rate CD CD

S I I liI I I I I! I 0

~. 2--- Average probability

o -- Base rate

cO o

6

cC o 0

0

00

o,

6 c3) 6

0)

CI)

0

CD

c\J

0

Co

6 1985 1987 1989 1991 1993 1995 1997 1999

References

Alphonse, P., J. Ducret, and E. Severin, 2004, "When Trade Credit Facilitates Access to Bank Finance: Evidence from US Small Business Data," University of Lille Working Paper.

Antov, D., 2005, "Trade Credit and Institutional Financing: Theory and Evidence," Northwestern University Working Paper.

Bernanke, B. and A. Blinder, 1992, "The Federal Funds Rate and the Channel of Monetary Transmission," American Economic Review 82, 901-921.

Biais, B. and C. Gollier, 1997, "Trade Credit and Credit Rationing," Review of Financial Studies 10, 903-937.

Biais, B., P. Hillion, and J. Malecot, 1995, "La Structure Financiere des Entreprises: Une Investigation Empirique sur Donnees Francaises," Economie et Prevision 120, 15-28.

Davis. E and K. Yeoman, 1974, "Company Finance and the Capital Market: A Study of the Effects of Firm Size," University of Cambridge Occasional Paper.

Deloof, M. and M. Jegers, 1999, "Trade Credit, Corporate Groups, and the Financing of Belgian Firms," Journal of Business Finance andAccounting 26, 945-966.




Dempster, A., N. Laird, and D. Rubin, 1977, "Maximum Likelihood from Incomplete Data via the EM Algorithm," Journal of the Royal Statistical Society, Series B 39, 1-38.

Douglas, S., 1996, "Bootstrap Confidence Intervals in a Switching Regression Model," Economic Letters 53, 7-15.

Elliehausen, G. and J. Wolken, 1993, "An Empirical Investigation into Motives for Demand for Trade Credit," Federal Reserve Board Staff Study.

Ferris, S., 1981, "A Transaction Theory of Trade Credit Use," Quarterly Journal of Economics 96, 243-270.

Franks J., J. Broyles, and W. Carlton, 1985, Corporate Finance: Concepts andApplications, Boston, MA, Kent Publishing Company.

Habib, M. and D. Johnson, 1999, "The Financing and Redeployment of Specific Assets," Journal of Finance 54, 693-720.

Jain, N., 2001, "Monitoring Costs and Trade Credit," Quarterly Review of Economics and Finance 41, 89-110.

Laffer, A., 1970, "Trade Credit and the Money Market," Journal ofPolitical Economy 78, 239-267.

Long, M., I. Malitz, and S. Ravid, 1993, "Trade Credit, Quality Guarantees and Product Marketability," Financial Management 22, 117-127.

Myers S. and N. Majluf, 1984, "Corporate Financing and Investment Decisions When Firms Have Information That Investors Do Not Have," Journal ofFinancial Economics 13, 187-221.

Ng C., J. Smith, and R. Smith, 1999, "Evidence on the Determinants of Trade Credit Terms Used in Interfirm Credit," Journal of Finance 54, 1109-1130.

Nilsen, J., 2002, "Trade Credit and the Bank Lending Channel," Journal of Money, Credit and Banking 34, 226-253.

Petersen, M. and R. Rajan, 1994, "The Benefit of Lending Relationships: Evidence from Small Business Data," Journal ofFinance 49, 3-36.

Petersen, M. and R. Rajan, 1995, "The Effect of Credit Market Competition on Firms-Creditor Relationships," Quarterly Journal of Economics 110, 407-443.

Petersen, M. and R. Rajan, 1997, "Trade Credit: Theories and Evidence," Review of Financial Studies 10, 661-91.

Schwartz, R., 1974, "An Economic Model of Trade Credit," Journal ofFinancial and Quantitative Analysis 9, 643-657.

Smith, J., 1987, "Trade Credit and Information Asymmetry," Journal ofFinance 42, 863-873.

Wilner, B., 2000, "The Exploitation of Relationships in Financial Distress: The Case of Trade Credit," Journal ofFinance 55, 153-178.



trade credit

Documents

trade credit extension

lines of credit

use of trade creditauthors

supplier trade finance

institutional finance

constrained firms

majority of firms

large firms liabilities