Does trade credit play a signalling role? Some evidence from SMEs microdata

Suppliers would also have a “liquidation advantage” over banks, as the former would be more able, in case of firms’ default, to obtain a greater liquidation value from collateralized inputs (Fabbri and Menichini 2010). On this issue, see also Mian and Smith (1992), Petersen and Rajan (1997), Longhofer and Santos (2003), Frank and Maksimovic (2004), and Atanasova (2012).

On a very close issue see Aktas et al. (2012).

In Italy, like in other OECD countries, the weight of trade debt has been traditionally very important for firms in general (Rajan and Zingales 1995; Demirgüç-Kunt and Maksimovic 2001; Giannetti 2003; De Blasio 2005; Marotta 2005) and for SMEs in particular (De Socio 2010). As reported by De Socio (2010), in the years 2004–2007 trade credit accounted for 25 % of Italian SMEs total assets, compared to an average of 16.8 % for the main European countries. On the importance of trade credit for SMEs see also, among others, Mian and Smith (1992) and Biais and Gollier (1997).

In the following we do not review the “non-financial” theories on trade credit, which provide explanations on the existence and use of suppliers credit based, for instance, on transaction costs minimization (e.g. Ferris 1981), price discrimination (e.g., Schwartz and Whitcomb 1979; Brennan et al. 1988; Mian and Smith 1992), product quality guarantees (e.g., Smith 1987; Lee and Stowe 1993; Long et al. 1993; Emery and Nayar 1998), inventory management (Bougheas et al. 2009), and long-term seller/customer relationships (e.g. Summers and Wilson 2002). We refer to Petersen and Rajan (1997), Ng et al. (1999) and Giannetti et al. (2008) for some reviews on trade credit theories.

In the literature, the Meltzer’s (1960) hypothesis is often referred to as the redistributional view on trade credit.

Some of the earliest contributions providing support to the Meltzer’s (1960) idea are from Brechling and Lipsey (1963), Jaffee and Modigliani (1969), Jaffee (1971), Herbst (1974), and Duca (1986).

Empirical evidence that firms might use more trade credit as bank credit is unavailable emerges also from the cross-country study of Fisman and Love (2003), who use the data and the methodology of Rajan and Zingales (1998). For other cross-country analyses on the role of trade credit (during financial crises), see Love et al. (2007) and Coulibaly et al. (2012). Finally, for studies not supporting the substitution hypothesis we refer to Gertler and Gilchrist (1993), Eichenbaum (1994), and Oliner and Rudebush (1995, 1996).

A similar conclusion is reached by Petersen and Rajan (1994, p. 32): “Clearly, our evidence that trade creditors lend when institutional lenders do not suggests that they have collateral, incentives related to the product they are selling, sources of leverage over the firm, or information that the institutions do not possess.” Petersen and Rajan (1997) also argue that suppliers appear to be more efficient (relative to banks) in liquidating firms’ collateral goods, as vendors may take advantage of their customers’ network to resell, at lower cost, the assets of a defaulted buyer. Furthermore, by lending to credit constrained firms, suppliers can invest in customers relationships aiming at taking advantage of future profitable projects.

Finally, the model offers an interpretation of some stylized facts, such as the heterogeneity of trade credit patterns among countries and across time. For instance, where creditors’ rights are better protected, cash tends to be less diverted, hence trade credit should tend to be less employed, as shown by Demirgüç-Kunt and Maksimovic (2001).

Even though public firms are more transparent than small private firms, Saito and Bandeira (2010) reach a similar conclusion considering Brazilian public companies. Indeed, according to their study, trade credit seems to signal a firm’s good quality, facilitating access to bank credit for public firms. Thus, the reputation hypothesis appears confirmed.

Besides, their model predicts that in most cases, the substitution effect is counter-cyclical, being larger during slow-growth periods than during rapid-growth periods. Using a balanced panel data of listed Chinese companies, and addressing problems of endogeneity, the authors find strong evidence in favor of both the substitution and the counter-cyclicality hypotheses.

Nevertheless, one cannot exclude that banks may still observe the signal conveyed by trade credit even after many years of continuous relation. This could be because, besides the informational advantage mentioned above, suppliers may have more incentives to monitor the financial situation of the firm, since customer concentration is likely higher in the case of suppliers than in the case of banks. Banks could be aware of the aforementioned circumstances and keep trying to take advantage of sellers’ information.

For other analyses concerning the duration of bank relationships see Degryse and Van Cayseele (2000); Lehmann and Neuberger (2001); Ongena and Smith (2001).

As argued by an anonymous referee, the amount of short-term debt used by firms does not necessarily correspond with the debt granted by banks, which, mirroring the firm's creditworthiness, represents the amount likely to be influenced by the trade credit signal. Lacking information on the amount granted, used bank debt can be considered a good proxy for granted debt if their ratio tends to be constant. However, even though the amount of debt granted remains unchanged, bank debt used may vary (together with trade credit) because of an expansion (or contraction) of working capital, and the consequent variation in the firm’s financing need. Analogously, the variation in the average number of days that customers employ to pay invoices (ΔDAYSREC) may affect both trade and bank debt. Indeed, banks and suppliers may consider an increase in days receivables as a negative signal on firms' customer's quality and/or on its credit policy, jeopardizing firm’s creditworthiness. On the other hand, it has to be recognized that an increase in days receivables can imply liquidity problems, resulting in a higher demand of credit. Thus, in our analysis, the sign of this regressor remains an open empirical question.

As a large body of research indicates (e.g., Petersen and Rajan 1995b; Bonaccorsi di Patti and Gobbi 2001), SME's bank financing relies essentially on local credit markets, which in Italy—according to the National Antitrust Authority—corresponds to the existing administrative provinces.

This procedure controls for unobserved heterogeneity by adopting a fixed effects model. Thus, the unobserved components are allowed to be correlated with the explanatory variables. Moreover, the idiosyncratic errors may have serial dependence of unspecified form. In our test, the selection and main equations include the same control variables with the exception of (the log of) population, and the tax incentives dummy (equal to 1 if firms received tax incentives and 0 otherwise). These latter variables are assumed to affect only the selection process, and their exclusion from the main equation allows us to better identify the model. Besides, when we test this assumption by including them in both equations, their estimated coefficients are never statistically significant in the main equation.

That is as variables potentially correlated with past values of the idiosyncratic error, but not correlated to its present and future values.

Moreover, to focus on firms less open in terms of financial information, and therefore more dependent on bank and trade credit, we drop those listed on the Stock Exchange. However, in Sect. 5.1, as a sensitivity check, we re-estimated Eq. (1) without operating this exclusion.

The reason why, in employing this criterion, we considered entry at the end of 2000—instead of entry at the end of each survey—is that in some cases the values of DURAT observed at the end of the three trienniums appeared to be inconsistent. When the 2000 record was missing, we considered the 2003 entry as the base-value, while—when both 2000 and 2003 figures were missing—the 2006 value was employed as the reference point. In all cases, when the value of DURAT became negative, we treated the observation as missing.

As illustrated below, the estimates of Table 3 are obtained by using all available instruments, except in column 3, where, by modifying this choice, we check the robustness of our findings to the set of instruments employed.

Incidentally, in column 2, many control variables appear statistically significant, with signs generally consistent with those expected. For instance, SIZE, PLEDGEASS, ΔWORKCAP and GROWTH are positively related with BANKDEBT, while CASHFLOW, RESERVE and BOND display a negative sign.

Such a discrepancy between individual and joint significance is usually interpreted as a symptom of multicollinearity (see Brambor et al. 2006) induced by the inclusion of an interaction term. As Brambor et al. (2006, p. 70) highlight, “even if there really is high multicollinearity and this leads to large standard errors on the model parameters, it is important to remember that these standard errors are never in any sense ‘too’ large, they are always the ‘correct’ standard errors. High multicollinearity simply means that there is not enough information in the data to estimate the model parameters accurately and the standard errors rightfully reflect this”.

To be precise, in column 3, when considering endogenous regressors, lags 2 through 4 of the levels are used as instruments for the transformed data, and lag 1 of the differences is used for the levels data. When dealing with predetermined variables, lags 1 through 4 of the levels are employed as instruments for the transformed data, and lag 0 of the instrumenting variable in differences is employed for the levels data. Exogenous variables instrument themselves.

By including this proxy we take into account that the use of both bank and trade debt is dependent on the constraints faced by firms on the credit market. In particular, according to the substitution hypothesis, credit-constrained firms tend to use more trade credit than others. Incidentally, the Unicredit surveys provide information also on another indicator of credit rationing, a dummy coded 1 if a firm demanded more credit than it actually obtained, and zero otherwise. However, since this variable is defined only for firms that replied yes to the question used to code RAT, it is available for 2,210 observations of our sample, while the proxy we employ is defined on 14,277 observations.

When this was not the case, the imputation was not carried out. Since we acknowledge that the imputation above described may cause measurement errors, while the criterion followed for DURAT (described in Sect. 3) appears logical, we prefer to omit MAIN and NBAN in our main equation, and use DURAT as proxy for the strength of bank-firm relationships in Eq. (1). For the same reason we omit RAT from the benchmark model, and use it, together with MAIN and NBAN, to carry out the robustness check we are discussing.

For contributions on the role of trade credit during financial crises see, for instance, Love et al. (2007), Yang (2011b), and Bastos and Pindado (2012).