Determinants of debt maturity structure across firm size

ABSTRACT This study examines the empirical determinants of debt maturity structure across the size of Spanish firms. Our evidence offers support for the relevance of growth opportunities, size, asymmetric information and asset to maturity to explain debt maturity structure. The paper also provides evidence regarding the differences in explanations according to firm size. It is shown that debt maturity in small firms is higher when the slope of the interest rate term structure increases and for very low-risk and very risky firms.


INTRODUCTION
Capital structure research has traditionally focused on explaining the incentives that lead large public corporations to choose particular fi nancing policies. There has been less research on other features of debt fi nancing, such as debt maturity structure. Theories of debt maturity have focused on the roles of agency costs (Myers, 1977;Barnea et al., 1980), asset maturity (Myers, 1977;Stohs and Mauer, 1996), asymmetric information (Flannery, 1986;Diamond, 1991;Berger et al., 2005) and taxes (Brick andRavid, 1985, 1991;Lewis, 1990) (1) .
In line with these theories, empirical analysis has identifi ed several factors that can affect a fi rm's choice of debt maturity structure. These factors include the fi rm's options for growth as a measure of agency costs of debt, the maturity of existing assets, the level of asymmetric information, and the effective income tax rate of the fi rm. Barclay and Smith (1995), Stohs and Mauer (1996), Guedes and Opler (1996), and Ozkan (2000) have provided empirical evidence about the determinants of maturity structure of debt for large fi rms. Other papers have focused on small fi rms, such as Scherr and Hulburt (2001), and Berger et al. (2005) (2) . The evidence provided by these papers is mixed. Barclay and Smith (1995), Guedes and Opler (1996) and Ozkan (2000) provide strong support for the hypothesis that fi rms with more growth opportunities in their investment sets tend to have more shorter-term debt. In line with Diamond's (1991) prediction, Barclay and Smith (1995), Stohs and Mauer (1996) and Scherr and Hulburt (2001) fi nd evidence of a nonmonotonic relationship between debt maturity structure and credit quality as a consequence of the adverse selection problem (3) . The evidence also provides strong support for the maturity-matching hypothesis, which predicts that fi rms match the maturity of their debt to that of their assets.
Several papers have examined the infl uence of the country's fi nancial systems and institutional aspects on debt maturity structure. For example, Antoniou et al. (2006) analyze the determinants of the debt maturity structure of French, German and UK fi rms, fi nding that the impact of fi rm-specifi c factors on debt maturity is country dependent. Along similar lines, Demirgüç- Kunt and Maksimovic (1999) examine the maturity of debt in 30 countries and highlight the relevance of the effectiveness of the legal system, the level of activity of the stock market, and the size of the banking sector as determinants of debt maturity. González and González (2008) and Hernández-Cánovas and Koëter-Kant (2008) fi nd results consistent with the infl uence of the bank system on debt maturity. González and González (2008) show that bank concentration positively infl uences debt maturity for a sample of listed fi rms in 39 countries. The results obtained by Hernández-Cánovas and Koëter-Kant (2008) reveal that stronger (1) See Ravid (1996) for a survey of debt maturity.
(2) This paper only tests the implications of the models of Flannery (1986) and Diamond (1991) concerning the effects of asymmetric information.
(3) Barclay and Smith (1995) and Stohs and Mauer (1996) use a bond rating variable based on a fi rm's S&P bond rating, while Scherr and Hulburt (2001) use Altman's Z score (1968) as a measure of default risk, seeing as the debt of small fi rms is not rated. fi rm-bank relationships lengthen the maturity of bank loans for a sample of SMEs from 19 European countries.
For Spain, Cuñat (1999) has provided evidence on the determinants of debt maturity structure for a sample of 227 listed fi rms in the period [1983][1984][1985][1986][1987][1988][1989][1990][1991][1992][1993][1994]. His results show that fi rms with higher growth opportunities shorten the maturity of their debt signifi cantly, although there is no evidence in favor of signaling or tax models. Furthermore, bigger fi rms and those with a greater degree of government participation present a higher maturity. Although this author also analyses the existence of an effect of size in the determinants of debt maturity, he uses only listed fi rms. García-Teruel and Martínez-Solano (2010) study the effects of ownership structure on debt maturity of listed fi rms. Additionally, two papers have analyzed the determinants of debt maturity in Spanish SMEs. García-Teruel and Martínez-Solano (2007) analyze the debt maturity structure of small and medium-sized fi rms in terms of the risk and return trade-off associated with the use of short-term debt (Jun and Jen, 2003). López-Gracia and Mestre-Barberá (2011) analyze the infl uence of the tax effect on SME debt maturity structure.
Within this context, the present paper examines the factors that Spanish fi rms take into consideration when choosing the maturity of their debt, analyzing whether the validity of the agency cost hypothesis, the signaling hypothesis, the maturity-matching hypothesis and the tax hypothesis varies with fi rm size. There have been studies analyzing the explanations of debt maturity in small fi rms, such as Scherr and Hulburt (2001) for US fi rms and García-Teruel and Martínez-Solano for Spanish fi rms, although neither of these papers considered large fi rms. The main contribution of the present paper is that of analyzing the relevance of these explanations jointly for a large sample of small, medium-sized and large fi rms. As far as we know, there are no papers that have tested the validity of these explanations on a single sample that includes both large and small fi rms. Considering a single sample allows us to test the existence of a different validity for each of the explanations proposed in the literature in accordance with fi rm size.
The results highlight the relevance of growth opportunities, size, signaling and asset maturity explanations in explaining debt maturity. The main differences when considering fi rm size are the existence of lower validity with respect to the predictions of Diamond's (1991) model for small fi rms, and higher debt maturity in smaller fi rms when the slope of the interest rate term structure increases.
The rest of the paper is organized as follows. Section 2 discusses the validity of theoretical arguments as determinants of debt maturity. Section 3 describes the database and methodology employed. Section 4 discusses the empirical results, while Section 5 tests the robustness of our results to the presence of endogeneity. Finally, Section 6 concludes the paper.

THEORETICAL BACKGROUND AND HYPOTHESES
Four non-mutually exclusive hypotheses have been put forward to explain corporate debt maturity structure: asset maturity, agency costs, asymmetric information and ARTÍCULOS DOCTRINALES taxes. In what follows, we summarize these explanations, which are then tested by the empirical model.

ASSET MATURITY
Firms match their debt maturities to their asset maturities. If the maturity of debt is shorter than that of assets, the fi rm may not have suffi cient cash available to pay its fi nancial obligations when they are due. However, if debt has a longer maturity, debt payments remain due when the cash fl ows from assets cease. Matching the maturities of assets and debt reduces these risks. Myers (1977) argues that the underinvestment problem can be mitigated by matching the maturity of liabilities and assets. Therefore, a positive relationship is expected between debt maturity and asset maturity.

AGENCY COSTS
The agency costs of debt may infl uence corporate debt maturity bearing in mind that outstanding debt may create incentive problems for shareholders. There are two such incentive problems: underinvestment and risk-shifting (Ravid, 1996). When a fi rm has future options for growth via a profi table investment opportunity set, the existence of risky debt in the capital structure means that the benefi ts from undertaking profi table investment projects will go only partly to shareholders. Debtholders will share the benefi t, because the probability of default is reduced by the investment projects. As the benefi t goes partly to debtholders, shareholders have incentives to reject positive net present value projects, thus leading to what is known as the underinvestment problem (4) . Myers (1977) argues that a fi rm may control this underinvestment incentive by shortening the effective maturity of its debt so that debt matures before growth options are exercised. This explanation of debt maturity based on agency costs suggests that fi rms whose value depends to a large extent on investment opportunities have an incentive to borrow short-term. Several papers have provided favorable evidence for this relationship, such as Barclay and Smith (1995), Guedes and Opler (1996) and Ozkan (2000).
The risk-shifting problem consists in the incentives of shareholders to substitute a risky project for a less risky one whose losses they do not bear, but whose gains accrue solely to shareholders (Black and Scholes, 1973;Jensen and Meckling, 1976). Agency problems between shareholders and debtholders may be particularly severe for small fi rms as a consequence of underinvestment incentives and risk shifting (Pettit and Singer, 1985;Smith and Warner, 1979). Like Myers (1977), Barnea et al. (1980) suggest that these problems may be reduced by issuing shorter-term debt. These arguments thus suggest that debt maturity varies directly with fi rm size. Barclay and (4) See Diamond and He (2010) for an in-depth analysis on the effects of debt maturity on the equity incentives to undertake both current and future investments and to identify the forces that determine overhang. SPANISH JOURNAL OF FINANCE AND ACCOUNTING. Vol. XLII, n.º 158 · April-June 2013 Smith (1995), Stohs and Mauer (1996) and Ozkan (2000) provide results in line with a positive relationship between size and debt maturity. Diamond (1991) provides a model to explain why risky fi rms with long-term projects might use short-term debt under the existence of asymmetric information. Firms with favorable private information and low-risk (high credit ratings) may choose shortterm debt at relatively low interest rates because the refi nancing risk is small. Firms with favorable private information and intermediate risk may choose long-term debt at a higher rate to reduce their greater liquidity risk of being unable to refi nance the debt if they choose short-term debt. Since short-term borrowing exposes fi rms to the risk of excessive liquidations, fi rms with high-risk (low credit ratings) prefer long-term debt so as to reduce this refi nancing risk. Firms with higher default risk may be unable to borrow long-term because of the high probability of bad projects. Thus, Diamond's (1991) model predicts debt maturity to have a nonmonotonic relationship with risk ratings. Very low-risk fi rms and very risky fi rms borrow short term and fi rms with intermediate risks are more likely to borrow long term.

ASYMMETRIC INFORMATION
Several studies analyze the relationship between debt maturity and risk ratings. Barclay and Smith (1995) show a nonmonotonic relationship between debt maturity and bond ratings. Firms with higher bond ratings tend to have more short-term debt than those with lower bond ratings. Firms without bond ratings have more short-term debt. Stohs and Mauer (1996) and Scherr and Hulburt (2001) also provide results in line with Diamond's (1991) predicted nonmonotonic relationship. More recently, Berger et al. (2005) also provide support to the predictions of Diamond's (1991) model for low-risk fi rms. In effect, maturity is an upward-sloping function of risk ratings. However, their evidence for high-risk fi rms confl icts with the predictions of Diamond's model insofar as high-risk fi rms do not present signifi cantly different maturities to intermediate-risk fi rms. Brick andRavid (1985, 1991) argue that the expected value of tax benefi ts depends on the maturity of debt when the term structure of interest rates is not fl at. If the yield curve is upward sloping, fi rms increase their value by increasing the amount of longterm debt. A term structure of interest rates with a positive slope implies that, under the unbiased expectations theory, the interest expense from issuing long-term debt is greater than the expected interest expense from rolling short-term debt in early years, and will be lower in later years. For that reason, the benefi ts of debt are accelerated using long-term debt. Likewise, short-term debt increases fi rm value if the yield curve has a negative slope. Consequently, a positive relationship can be expected between the term structure of interest rates and the proportion of long-term debt according to the tax explanation of debt maturity. ARTÍCULOS DOCTRINALES Determinants of debt maturity structure across fi rm size REVISTA ESPAÑOLA DE FINANCIACIÓN Y CONTABILIDAD. Vol. XLII, n.º 158 · abril-junio 2013 Kane et al. (1985) develop a model in which the optimal debt maturity structure involves a trade-off between bankruptcy and debt issue fl otation costs and the perperiod tax advantage of debt fi nancing. In this context, the maturity of debt should rise if the effective tax rate decreases, the fl otation cost increases and the volatility of fi rm value decreases. The empirically testable hypothesis is that a fi rm's debt maturity increases as its effective tax rate decreases.

TAXES
Little favorable evidence has been reported for the tax hypothesis, receiving mixed support in Stohs and Mauer (1996). As predicted, these authors fi nd a negative relation between tax rate and debt maturity, although there is no evidence that the debt maturity structure is positively related to the slope of the term structure. Barclay and Smith (1995), Guedes and Opler (1996) and Ozkan (2000), on the other hand, fi nd no evidence for the tax hypothesis. For Spain, López-Gracia and Mestre-Barberá (2011) offer strong evidence for the tax explanation of debt maturity structure for a sample of small and medium-sized fi rms.

ECONOMETRIC SPECIFICATION AND DATABASE
We use the following model to investigate the determinants of the debt maturity structure of fi rms: where DEBTMAT is the debt maturity of fi rm i in the year t and the determinants are agency costs (GROWTH and SIZE), credit quality (DEFAULT RISK), maturity of assets (ASSETMAT), and taxation (TERM PREMIUM and TAX EXP).

2006
Σ t=1995 is a set of dummy time variables for each year that capture any unobserved fi rm-invariant time effect not included in the regression, v i is the fi rm effect, which is assumed constant for fi rm i over t, and ε it is the error term.
Most of the literature on debt maturity has adopted a similar framework to this paper. However, other papers assume that fi rms have long-run optimal debt maturity structures and examine the speed at which they adjust to their target. Dynamic adjustment models have frequently been used to study capital structure (among others Shyam-Sunder and Myers, 1999;Frank and Goyal, 2003;Welch, 2004;Flannery and Rangan, 2006;and González and González, 2008), and, to a lesser extent, maturity structure (Ozkan, 2000;Antoniou et al., 2006;and López-Gracia and Mestre-Barberá, 2011).
To test the empirical determinants of debt maturity, we use a sample of Spanish fi rms during the period 1995-2006. The data come from SABI and consist of fi nancial statement data and ratios for over a million Spanish companies. We select nonfi nancial corporations (fi rms with SIC codes 6000-6999 have been excluded) that have (1) more than 10 employees, and (2) data throughout the 12-year sample period to construct the variables used. We exclude fi rms reporting zero debt. Finally, the sample is made up of 39,603 corporations and 246,344 observations, although the number of observations varies depending on the variables used.
SABI reports the amount of long-term debt payable in one year. To measure the maturity structure of a fi rm's debt (DEBTMAT), we examine the percentage of the fi rm's total debt (long-term debt plus debt in current liabilities) that has a maturity of more than one year. Prior studies have used various measures of debt maturity, considering either a balance sheet approach or an incremental approach. Examples of the balance sheet approach are Scherr and Hulburt (2001), Barclay and Smith (1995) or Ozkan (2000). Scherr and Hulburt (2001) use two specifi cations, long-term debt payable after one year to total debt, as in this paper, and weighted-average debt maturity, reporting that differences in results between the two specifi cations of debt maturity are minor; Barclay and Smith (1995) use the percentage of long-term debt payable after three years to total debt; while the dependent variable in Ozkan (2000) is the ratio of debt that matures in more than fi ve years to total debt. Guedes and Opler (1996) and Berger et al. (2005), on the other hand, use an incremental approach. Guedes and Opler (1996) consider the maturity of debt issues as the dependent variable; while Berger et al. (2005) use a sample on new loans to small businesses. The argument for using the maturity of new issues is that some questions about the determinants of debt maturity, such as signaling models of maturity choice, can only be properly tested using the incremental approach. However, the incremental approach is not well suited to testing theories that relate asset maturity to the average of the maturities of the fi rm's existing liabilities, since the term-to-maturity of an individual issue only provides information about incremental fi nancing choices.
We use a balance sheet approach in the present paper. This is driven primarily by the nature of the sample. Seeing as we are concerned with whether there are differences among the determinants of corporate debt maturity structure according to fi rm size, we need a dependent variable that can be measured for fi rms of any size. The balance sheet approach allows us to use a measure of debt maturity common to all fi rms. Moreover, the debate concerning the use of a balance sheet or incremental approach is of less importance in Spain. Garcia-Teruel and Martínez-Solano (2007) show a high proportion of short-term debt with respect to total debt. For their sample of 11,533 small and medium-sized fi rms, 80.81% of total debt is short-term debt. We also fi nd a high percentage of short-term debt in our sample, as will be seen in Table 1. If debt is mainly short term, the limitations related to maturity structure approximation based on the ratio of long-term debt to total debt are less relevant.
Growth options have usually been proxied by the market-to-book ratio (Barclay and Smith, 1995;Guedes and Opler, 1996;Stohs and Mauer, 1996;and Ozkan, 2000). Seeing as we consider non-listed fi rms, it is not possible to measure a fi rm's growth opportunities by the market-to-book ratio, as is usual in other papers. One way of measuring a fi rm's growth opportunities (GROWTH) is to assess its past growth, assuming that fi rms that grow faster also have greater opportunities for future growth. We measure the ratios of current sales to prior sales (SALESGROWTH) and current ARTÍCULOS DOCTRINALES Determinants of debt maturity structure across fi rm size REVISTA ESPAÑOLA DE FINANCIACIÓN Y CONTABILIDAD. Vol. XLII, n.º 158 · abril-junio 2013 assets to prior assets (ASSETGROWTH) to capture past growth. We also use the ratio of depreciation to total assets (DEPREC_TA) to measure the weight of tangible assets, as in Scherr and Hulburt (2001), which are expected to be negatively related to growth opportunities (5) . Size, on the other hand, is measured as the natural logarithm of fi rm assets (LNASSETS) in constant 1995 thousands of euros.
The predictions of Diamond's (1991) model have usually been tested by using bond ratings to measure default risk, as in Barclay and Smith (1995), Guedes and Opler (1996), and Stohs and Mauer (1996). As our database includes small fi rms whose debt is not rated, we have used Altman's Z score (1968) as a measure of default risk (DEFAULT RISK) following Scherr and Hulburt (2001). Altman's Z score is computed using fi ve accounting ratios, with high values indicating a low probability of default: where: X 1 = (current assets-current debt) / total assets; X 2 = retained earnings/total assets; X 3 = earnings before interest and taxes / total assets; X 4 = equity / total debt; X 5 = sales / total assets.
For the empirical test of the maturity-matching hypothesis, asset maturity (ASSETMAT) is computed by means of the ratio between property, plant and equipment and the annual depreciation (PPE_DEPREC). The idea underlying this measure is that longer maturity assets will depreciate at a slower rate (6) . A similar defi nition has been used by Ozkan (2000).
To test the tax hypothesis, we measure the term structure of interest rates as the difference between the month-end yield on ten-year government bonds and the monthend yield on six-month government bonds (TERM PREMIUM). The data are obtained from the database provided by the Central Bank of Spain. To measure the effective tax rate, we use the ratio of income tax expense to total assets (TAX EXP) (Guedes and Opler, 1996).
We have split the sample into small, medium-sized and large enterprises applying the criteria of fi rm size defi ned by the European Union in the Commission Recommendation of 3. rd April 1996 (96/280/EC) (7) . A small fi rm is defi ned as an enterprise that has (5) Another alternative is to consider the ratio of R&D expenses to total sales to measure growth-oriented investments. The number of fi rms for which we have this information drops considerably, providing only 11,886 observations. The results for this variable are not signifi cant. (6) Similar results are obtained when the variable PPE_DEPREC is multiplied by the ratio between property, plant and equipment and total assets. (7) This criteria was the one in force during the period covered by our study. On 6. th May 2003, the Commission adopted a new Recommendation (2003/361/EC) regarding the defi nition of SMEs which replaced Recommendation 96/280/EC as from 1. st January 2005. We have found that the results do not vary with the new classifi cation of fi rm size. fewer than 50 employees but more than 10, and has either an annual turnover not exceeding seven million euros or an annual balance-sheet total not exceeding fi ve million euros. Medium-sized fi rms are defi ned as enterprises that have between 50 and 249 employees, and have either an annual turnover not exceeding 40 million euros, or an annual balance-sheet total not exceeding 27 million euros. Firms that exceed these limits are considered large enterprises. Table 1 presents mean values for debt maturity and independent variables in Panel A, and the differences among subsamples in Panel B. The fi rst aspect worth highlighting is the low long-term debt of the sample of Spanish fi rms. Barclay and Smith (1995) report a percentage of total long-term debt of around 70%, versus 23.28% for the sample under study shown in table 1. The division of the sample into small, mediumsized and large fi rms does not refl ect major differences in the percentage of total debt according to fi rm size, although these differences are signifi cant when we compare small and medium-sized versus large fi rms. The percentage of long-term debt is signifi cantly larger in small fi rms compared to large and medium-sized fi rms.

TABLE 1 DESCRIPTIVE STATISTICS
The table presents the mean values of the dependent and independent variables and the differences in these variables among subsamples. TOTAL DEBT is the ratio of total debt (short-and long-term debt) to total assets. DEBT MAT is the percentage of the fi rm's total debt (long-term debt plus debt in current liabilities) that has a maturity of more than one year. SALES GROWTH is the ratio of current sales to prior sales. ASSET GROWTH is the ratio of current assets to prior assets. DEPREC_TA is the ratio of depreciation to total assets. LNASSETS is the natural logarithm of fi rm assets in constant 1995 thousands of euros. DEFAULT RISK is Altman's Z score. PPE_DEPREC is the ratio between property, plant and equipment and annual depreciation. TERM PREMIUM is the difference between the month-end yield on ten-year government bonds and the month-end yield on six-month government bonds. TAX EXP is the ratio of income tax expense to total assets. Firms have been split into small, medium-sized and large enterprises applying the criteria of fi rm size defi ned by the European Union in the Commission Recommendation of 3. rd April 1996 (96/280/EC). *** , ** and * represent signifi cance at the 1%, 5% and 10% level, respectively. The sample under study contains extreme values of SALESGROWTH and ASSETGROWTH, especially for large fi rms. We exclude from the analysis the observations with SALES GROWTH and ASSET GROWTH above the 99. th percentile. After excluding these observations, the mean values for SALES GROWTH (ASSET GROWTH) are thus 10.35% (11.69%), 9.02% (11.07%), 12.27% (12.72%), and 14.65% (13.28%) respectively for the total sample, small, medium-sized and large fi rms. According to the differences shown in panel B for the DEFAULT RISK variable, large and medium-sized fi rms have higher default risk than small fi rms, these differences being signifi cant. Table 2 shows the relation between Altman's Z score and debt maturity for the total sample and for small, medium-sized and large fi rms. As we can see in this table, the values of the ratio of total debt to total assets and the ratio of earnings before interest and taxes plus depreciation expenses to total assets corroborate the classifi cation of default risk obtained for Altman's Z score. It can be observed that the level of total debt decreases with fi rm solvency, while profi tability grows as solvency improves according to Altman's Z score. The fi rms belonging to the lower solvency group (Z < 0) present a mean proportion of total debt to assets of 123.15% and a profi tability of -25.05%. However, for the group made up of the more solvent fi rms (Z ≥ 7), the level of total debt is situated at 35.14% and profi tability at 19.95%. Moreover, the ratio of total debt to assets is monotonically decreasing according to Altman's Z score, whereas profi tability is monotonically increasing. This shows the validity of Altman's Z score as a proxy of the fi nancial strength of Spanish fi rms.

Panel A. Mean values of variables
As regards the analysis of the relation between debt maturity and default risk, it can be seen in Panel A in table 2 that the mean value of long-term debt decreases from a value of 38.78% for Z values between 0 and 1 to 18.04% for Z values above 7. Furthermore, the mean percentage of long-term debt for Z values below zero (fi rms with very high risk) is 31.38%. These values are partially consistent with Diamond's (1991) model seeing as the average debt maturity is lower for fi rms with high default probabilities and for fi rms with low default probabilities with respect to fi rms with intermediate-range default probabilities. The fi rms with high risk are only those presenting Z values less than one. This relationship between default risk and debt maturity is similar to that obtained by Scherr and Hulburt (2001) for a sample of US small fi rms. The main difference is that fi rms with an Altman Z score higher than two have less percentage of long-term debt than the average (20.65% versus 23.28%), whereas Scherr and Hulburt (2001) show that fi rms with a Z score higher than six are the ones that have less debt than the average (43.5% versus 44.31%).
This fulfi lling of the predictions based on Diamond's model (1991) is maintained when fi rm size is taken into account (Panels B, C and D in table 2). The breakdown of the sample into subgroups according to fi rm size following the criteria of European Commission Recommendation 96/280/EC provides similar results to those already reported for the total sample. Small, medium-sized and large fi rms with low risk have lower than average long-term debt. Very high risk fi rms (Z < 0), on the other hand, have less long-term debt than fi rms that present an intermediate risk situation (0 ≤ Z < 1).  Table 3 reports the correlation matrix. According to the arguments of asset maturity, debt maturity is positively correlated with the ratio between property, plant and equipment and annual depreciation (PPE_DEP). The correlation of debt maturity with default risk is negative, highlighting less long-term debt as the solvency of the fi rm increases. The correlation between debt maturity and the tax rate is negative, in line with the arguments of Kane et al. (1985). The independent variables do not present high correlations with one another.

TABLE 3 CORRELATIONS
The table presents the correlation matrix. DEBTMAT is the percentage of the fi rm's total debt (long-term debt plus debt in current liabilities) that has a maturity of more than one year. SALESGROWTH is the ratio of current sales to prior sales. ASSETGROWTH is the ratio of current assets to prior assets. DEPREC_TA is the ratio of depreciation to total assets. LNASSETS is the natural logarithm of fi rm assets. DEFAULTRISK is Altman's Z score. PPE_DEPREC is the ratio between property, plant and equipment and annual depreciation. TERMPREMIUM is the difference between the month-end yield on ten-year government bonds and the month-end yield on six-month government bonds. TAXEXP is the ratio of income tax expense to total assets. *** , ** and * represent signifi cance at the 1%, 5% and 10% level, respectively.

RESULTS
Debt maturity explanations are tested using panel data. Prior to testing, we use the Breusch-Pagan test (Breusch and Pagan, 1980) to identify the existence of individual effects. The null hypothesis of no unobserved heterogeneity is rejected. In this context, a model capturing individual heterogeneity, as the panel data methodology does, is appropriate. The methodology of panel data presents several benefi ts. It is more informative due to providing more variability, less collinearity among variables, and more degrees of freedom, and is consequently more effi cient. However, the main benefi t is that it corrects for unobserved fi rm-specifi c and time-specifi c effects. The panel data estimation was calculated using fi xed effects, as the Hausman test (1978) rejects the null hypothesis of the lack of correlation between individual effects and

DETERMINANTS OF FIRM DEBT MATURITY
Regressions are estimated using panel data. The dependent variable (DEBTMAT) is the percentage of the fi rm's total debt (long-term debt plus debt in current liabilities) that has a maturity of more than one year. SALESGROWTH is the ratio of current sales to prior sales. ASSETGROWTH is the ratio of current assets to prior assets. DEPREC_TA is the ratio of depreciation to total assets. LN_ASSETS is the natural logarithm of fi rm assets in constant 1995 thousands of euros. DEFAULTRISK is Altman's Z score. SQ(DEFAULT) is the Z score times the absolute value of the Z score. HIGHDEFAULT is a dummy variable that equals one if the fi rm has a negative value for Altman's Z score, and zero otherwise. LOWDEFAULT is a dummy variable that equals one if Altman's Z score is above the 66. th percentile, and zero otherwise. PPE_DEPREC is the ratio between property, plant and equipment and annual depreciation. TERMPREMIUM is the difference between the month-end yield on ten-year government bonds and the month-end yield on six-month government bonds. TAXEXP is the ratio of income tax expense to total assets. T-statistics are in parentheses. *** , ** and * represent signifi cance at the 1%, 5% and 10% level, respectively.

Expected sign
(1) (3)  observable variables in all regressions. The estimation that takes into account the possible correlation between individual effects and independent variables is the within-group estimation. Table 4 reports the results of the determinants of debt maturity structure. The coeffi cients estimated on the growth of sales [column (1)] and the growth of assets [column (2)] are signifi cant in a way that is consistent with the agency cost hypothesis. Additionally, the ratio of depreciation to total assets [column (3)] has the expected sign (8) . Thus, fi rms that have more confl icts between shareholders and debtholders use a higher proportion of short-term debt to mitigate these confl icts. According to the coeffi cient of ASSET GROWTH in column (8) in table 4, a one standard deviation increase in growth of assets raises debt maturity by 7.17%. The prediction that the debt maturity structure decreases as the proportion of growth options in the fi rm's investment opportunity set increases is also obtained for large and small Spanish fi rms respectively by Cuñat (1999) and García-Teruel and Martínez-Solano (2007). Favorable evidence for the negative relationship between growth opportunities and debt maturity has been also provided in other institutional contexts by Barclay and Smith (1995), Barclay et al. (2003), Guedes and Opler (1996) or Ozkan (2000).
The coeffi cients on size (LNASSETS) are positive in all the estimations. This means that larger fi rms have longer debt maturity structures. This result is consistent with the role of short-term debt in reducing agency problems between shareholders and debtholders that might be particularly severe for small fi rms. Barclay and Smith (1995) and Stohs and Mauer (1996) also provide evidence along these lines. To measure the economic signifi cance of the infl uence of fi rm size on debt maturity structure, we estimated the percent change in the dependent variable that results from a one standard deviation change in the explanatory variable. Considering the coeffi cient on size in column (7) in Table 4, a one standard deviation increase in size raises debt maturity by 37.69%.
In column (1) in table 4, we report a signifi cant and positive relation between fi rm quality and debt maturity. In columns (2) and (3), however, we obtain a negative and signifi cant coeffi cient for DEFAULT RISK variable. Seeing as an increase in Z corresponds to a reduction in default probability, this negative coeffi cient is in line with the use of longer-term debt when the default risk increases.
The proxy for the asset maturity of fi rms, i.e., the ratio of net property, plant and equipment to annual depreciation expense (PPE_DEP), presents a positive and signifi cant coeffi cient. Firms with longer-lived assets use longer-maturity debt. The economic signifi cance of the infl uence of asset maturity on debt maturity structure is 6.90%. This implies support for the maturity-matching hypothesis. Cuñat (1999) shows weak evidence on fi rms matching their debt maturities to their asset maturities for a sample of Spanish listed fi rms.
We fi nd mixed evidence for the tax hypothesis. On the one hand, the tax hypothesis predicts an inverse relationship between debt maturity structure and the ratio of taxes (8) We also obtain a positive coeffi cient using a straightforward measure of the tangibility of assets, such as the ratio between property, plant and equipment and total assets. paid to assets (TAX EXP). In line with this prediction, the coeffi cient of TAX EXP shown in table 4 is negative and signifi cant. From the coeffi cient in column (7), a one standard deviation in the ratio of taxes paid to assets decreases debt maturity by 6.42%.
On the other hand, there is no clear evidence that debt maturity is positively related to the slope of term structure (TERM PREMIUM). The relationship between TERM PREMIUM and DEBTMAT is positive and signifi cant only when growth opportunities are measured as the ratio of current sales to prior sales (SALES GROWTH). In this case, this fi nding would imply favorable evidence for this hypothesis. However, when growth opportunities are measured by ASSET GROWTH or DEPREC_TA, there is a negative and signifi cant association between the term premium and debt maturity (9) . As a consequence of these contradictory results, our fi ndings might be seen as providing modest support for the tax hypothesis.
In columns (4) to (6) in table 4, we test the predictions of Diamond's (1991) model. To do so, we include the square of Altman's Z score [SQ(DEFAULT)] (10) in the estimations. We expect a positive relation between debt maturity and default risk and a negative relation of debt maturity with the SQ(DEFAULT) variable. The results for these variables shown in columns (4) to (6) underscore a negative coeffi cient of DEFAULT RISK and a positive coeffi cient of the SQ(DEFAULT) variable. These signs are the opposite of what is expected following Diamond (1991). As can be seen in table 2, our sample shows two important characteristics regarding the relationship between the Z score and debt maturity: (1) the increase in debt maturity for intermediate-risk fi rms proposed by Diamond (1991) only applies to Z scores less than one; and (2) a fl attening in the relationship between debt maturity and the Z score for values of Z higher than 4. The negative coeffi cient of the DEFAULT RISK variable thus captures the negative relationship between the Z score and debt maturity for values of Z higher than 1 and the positive coeffi cient of SQ(DEFAULT) captures the fl attening of this relationship for Z score values higher than 4.
In view of this result, we also test the implications of Diamond's (1991) model by building two dummy variables according to the level of default risk. The fi rst dummy variable is HIGH DEFAULT, which takes a value of one if the fi rm has a negative value for Altman's Z score, and zero otherwise. LOW DEFAULT is the second dummy variable and equals one if Altman's Z score is above the 66. th percentile, and zero otherwise (11) .
The results are shown in columns (7) to (9) in Table 4. The two dummy variables are signifi cant and negative. These coeffi cients reveal that fi rms belonging to the category with the lowest credit score as well as those with the highest credit score borrow on a shorter term. This result is consistent with the nonmonotonic relationship between debt maturity and default risk and provides evidence that is favorable to the implications (9) This negative relationship between term premium and debt maturity was also the result obtained by Guedes and Opler (1996). (10) Bearing in mind that the Z score may have a negative sign, in order to calculate the square of Z score we multiply the Z score by the absolute value of Z (Scherr and Hulburt, 2001). (11) The value of the 66 th percentile is 2.84, while an Altman Z score equal to zero corresponds to a percentile of 1.85%. ARTÍCULOS DOCTRINALES Determinants of debt maturity structure across fi rm size REVISTA ESPAÑOLA DE FINANCIACIÓN Y CONTABILIDAD. Vol. XLII, n.º 158 · abril-junio 2013 of Diamond's (1991) model. The coeffi cients in column (7) indicate that, all else being equal, debt maturity structure decreases by 2.28% and 8.68% respectively for a one standard deviation in HIGH DEFAULT and LOW DEFAULT.
The fi ndings for growth opportunities, maturity of assets and tax explanation are similar to those discussed previously when testing the predictions of Diamond's (1991) model in columns (4) to (9).

DETERMINANTS OF FIRM DEBT MATURITY ACCORDING TO SIZE
Tables 5 and 6 show the fi ndings regarding the determinants of debt maturity structure according to fi rm size. The sample has been split applying the criteria defi ned by the European Union in Commission Recommendation 96/280/EC. Firstly, the basic model tested in Table 4 for the entire sample is applied in Table 5 to the three subsamples of small, medium-sized and large fi rms. Secondly, the results for the different validities of the explanations are shown in table 6, where these are analyzed using interaction variables between the independent variables and the dummy variable SMALL. In columns (1) to (3), the dummy variable SMALL takes the value of 1 if the fi rm is a small-sized fi rm according to EU criteria, and zero otherwise. In columns (4) to (6), the dummy variable SMALL takes the value of 1 if the fi rm is a small or medium-sized fi rm according to EU criteria, and zero otherwise. These interaction terms allow us to analyze whether the determinants of debt maturity are equally valid in small fi rms versus medium-sized and large fi rms. Due to the fact that table 6 provides a test of the signifi cance of the differences according size, only the results obtained in columns (1) to (3) in Table 6 are discussed.
The results for the independent variables of the basic model are similar to those reported in table 4. Evidence favorable to the expected relationships is found for the agency costs of debt, asymmetric information and maturity-matching hypotheses and mixed evidence for the tax hypotheses. Since these fi ndings have been highlighted previously, our comments focus here on the interaction terms.
Our fi ndings show that the main difference when comparing fi rms according to their size is the existence of a different effect of asymmetric information in smaller fi rms versus medium-sized and large fi rms. The variables SMALL*HIGH DEFAULT and SMALL*LOW DEFAULT present a positive and signifi cant coeffi cient. This sign underscores the presence of greater long-term debt for small fi rms that have very high or very low risk. Thus, although the predictions of Diamond's (1991) model are fulfi lled in smaller fi rms, they are fulfi lled to a greater extent in large fi rms. This difference disappears almost completely when considering the differential effect of small and medium-sized fi rms versus large fi rms [columns (4) to (6)].
The coeffi cients of the variables SMALL*SALES GROWTH [column (4)] and SMALL*ASSET GROWTH [columns (2) and (5)] are consistent with a lesser validity of agency costs explanations in small versus large fi rms. However, the positive coeffi cient of SMALL*DEPREC_TA [columns (3) and (6)] suggests that smaller fi rms with more investment in tangible assets have longer-term debt, thus providing favorable

DETERMINANTS OF FIRM DEBT MATURITY ACCORDING TO SIZE
Regressions are estimated using panel data. The dependent variable (DEBTMAT) is the percentage of the fi rm's total debt (long-term debt plus debt in current liabilities) that has a maturity of more than one year. SALESGROWTH is the ratio of current sales to prior sales. ASSETGROWTH is the ratio of current assets to prior assets. DEPREC_TA is the ratio of depreciation to total assets. LNASSETS is the natural logarithm of fi rm assets in constant 1995 thousands of euros. HIGHDEFAULT is a dummy variable that equals one if the fi rm has a negative value for Altman's Z score, and zero otherwise. LOWDEFAULT is a dummy variable that equals one if Altman's Z score is above the 66th percentile, and zero otherwise. PPE_DEPREC is the ratio between property, plant and equipment and annual depreciation. TERMPREMIUM is the difference between the month-end yield on ten-year government bonds and the month-end yield on six-month government bonds. TAXEXP is the ratio of income tax expense to total assets. Firms have been split into small, medium-sized and large enterprises applying the criteria of fi rm size defi ned by the European Union in the Commission Recommendation of 3rd April 1996 (96/280/EC). T-statistics are in parentheses. *** , ** and * represent signifi cance at the 1%, 5% and 10% level, respectively.   Regressions are estimated using panel data. The dependent variable (DEBTMAT) is the percentage of the fi rm's total debt (long-term debt plus debt in current liabilities) that has a maturity of more than one year. SALESGROWTH is the ratio of current sales to prior sales. ASSETGROWTH is the ratio of current assets to prior assets. DEPREC_TA is the ratio of depreciation to total assets. LNASSETS is the natural logarithm of fi rm assets in constant 1995 thousands of euros. HIGHDEFAULT is a dummy variable that equals one if the fi rm has a negative value for Altman's Z score, and zero otherwise. LOWDEFAULT is a dummy variable that equals one if Altman's Z score is above the 66. th percentile, and zero otherwise. PPE_DEPREC is the ratio between property, plant and equipment and annual depreciation. TERMPREMIUM is the difference between the month-end yield on ten-year government bonds and the month-end yield on six-month government bonds. TAXEXP is the ratio of income tax expense to total assets. In columns (1) to (3), SMALL is a dummy variable that takes the value of 1 if the fi rm is a small fi rm, and zero otherwise. In columns (4) to (6), SMALL is a dummy variable that takes the value of 1 if the fi rm is a small or medium-sized fi rm, and zero otherwise. Firms have been split into small, medium-sized and large enterprises applying the criteria of fi rm size defi ned by the European Union in the Commission Recommendation of 3. rd April 1996 (96/280/EC).T-statistics are in parentheses. *** , ** and * represent signifi cance at the 1%, 5% and 10% level, respectively.

SMALL
(1) (2) (3) (4) (5)  evidence for the hypothesis that smaller fi rms with more growth opportunities control suboptimal investment incentives by shortening the maturity of their debt.
The differential effect of the interest rate term structure on smaller fi rms reveals that these fi rms consider term structure to be an important determinant of debt maturity.
Only the debt maturity of smaller fi rms varies directly with the slope of the term structure of interest rates. For large fi rms, term structure has a negative infl uence on debt maturity. There is no clear differential effect, however, of tax expenses on debt maturity according to fi rm size.
The estimations do not provide strong support for the maturity-matching hypothesis or the tax-based hypothesis as explanations that vary between fi rms according to their size. As regards the maturity matching hypothesis, the results in Table 6 only show the existence of a differential effect for smaller fi rms in some estimations, highlighting the greater validity of matching the maturity of assets and debt to reduce the risks of refi nancing and liquidity in small fi rms.

TABLE 7 TEST OF ROBUSTNESS. DETERMINANTS OF FIRM DEBT MATURITY
Regressions are estimated using panel data. The dependent variable (DEBTMAT) is the percentage of the fi rm's total debt (long-term debt plus debt in current liabilities) that has a maturity of more than one year. SALESGROWTH is the ratio of current sales to prior sales. ASSETGROWTH is the ratio of current assets to prior assets. DEPREC_TA is the ratio of depreciation to total assets. LN_ASSETS is the natural logarithm of fi rm assets in constant 1995 thousands of euros. DEFAULT_FITTED is the predicted value of DEFAULT RISK using as instruments the fi rst lag of Z score, profi tability and total debt. SQ(DEFAULT_FITTED) is DEFAULT_FITTED times the absolute value of DEFAULT_FITTED. PPE_DEPREC is the ratio between property, plant and equipment and annual depreciation. TERMPREMIUM is the difference between the month-end yield on ten-year government bonds and the month-end yield on six-month government bonds. TAXEXP is the ratio of income tax expense to total assets. T-statistics are in parentheses. *** , ** and * represent signifi cance at the 1%, 5% and 10% level, respectively.

ROBUSTNESS
It could be argued that the Z score may be subject to endogeneity. For example, Barclay and Smith (1995) fi nd a positive correlation between leverage and debt maturity. Thus, the level of long-term debt could be a determinant of the default risk of a fi rm. For this reason, we now carry out an analysis to consider the potential endogeneity of the DEFAULT RISK variable. Table 7 reports the results of the determinants of debt maturity structure for the fi rst six columns in table 4, as these are the estimations which directly include the DEFAULT RISK variable.
We address the concerns regarding the potential endogeneity of the Z score using instrumental variables estimation. As instruments for the Z score, we chose the fi rst lag of the Z score, profi tability and total debt. We perform a Durbin-Wu-Hausman test of overidentifying restrictions for each regression (Davidson and Mackinnon, 1993). The test verifi es the null hypothesis that the introduction of instrumental variables has no effect on the estimates of the regression's coeffi cients. We performed the Durbin-Wu-Hausman test on each of the fi rst six regressions in table 4, the results of which are reported in the bottom line of table 7. The test is rejected at the one percent level in all the estimations. Therefore, the predicted value for DEFAULT RISK (DEFAULT_FITTED) is included instead of the observed value in the regressions. The SQ(DEFAULT_FITTED) is the DEFAULT_FITTED times the absolute value of this variable.
In all the regressions, we fi nd that the introduction of the fi tted value for DEFAULT RISK variable does not alter the results presented previously. The sign and signifi cance of the coeffi cients remain similar to the results in table 4. Therefore, the results obtained are robust as regards the issue of endogeneity.

CONCLUSIONS
In this paper, the empirical determinants of a fi rm's debt maturity structure are examined for a sample of 38,993 non-fi nancial Spanish fi rms over the period 1995-2006. Our results show the relevance of growth opportunities, size, asymmetric information, and asset maturity in explaining debt maturity. Our fi ndings are not signifi cantly different to those reported for US fi rms. The main difference with respect to the institutional environment arises from the maturity of debt and not from the determinants. Spanish fi rms present a lower ratio of long-term debt to total debt compared to US fi rms.
We fi nd strong evidence in line with the agency cost approach that debt maturity is used to control confl icts of interest between shareholders and debtholders. On the one hand, smaller fi rms tend to use shorter-term debt. On the other, debt maturity is inversely related to proxies for growth opportunities. We obtain evidence consistent with Diamond's (1991) prediction of a nonmonotonic relationship between debt maturity structure and probability of default, although this is only true for very specifi c values of the Z score. We also fi nd evidence in favor of the asset maturity explanation, as in Scherr and Hulburt (2001) and Stohs and Mauer (1996) for US fi rms or Ozkan (2000) for UK fi rms.
Like Cuñat (1999), we provide evidence in favor of the growth opportunities and asset maturity hypotheses and no clear support for tax arguments, the difference residing in the infl uence of asymmetric information. Cuñat (1999) fi nds that smaller fi rms tend to use shorter-term debt, as do we. However, this evidence is obtained considering only listed fi rms. García-Teruel and Martínez-Solano (2007) show more solvent fi rms using higher proportional short-term debt, as do we. However, we also show that fi rms with the lowest credit score borrow on a shorter term. Our results show a positive infl uence of the term structure of interest rates on debt maturity, as do López-Gracia and Mestre-Barberá (2011).
Given that the main contribution of the paper consists in analyzing the different validity of the empirical determinants of debt maturity structure for small, medium-sized and large fi rms, we likewise provide evidence on the differences in the explanations according to fi rm size. We show that the term structure of interest rates and the probability of default are the determinants that have a differential infl uence between small and large fi rms. Debt maturity in smaller fi rms is higher when the slope of the interest rate term structure increases and for very low-risk and very risky fi rms. This last result implies that the prediction of Diamond's (1991) model is fulfi lled to a greater extent in large fi rms than in smaller fi rms.