- Email: [email protected]
Tax effect on Spanish SME optimum debt maturity structure
Journal of Business Research 64 (2011) 649–655
Contents lists available at ScienceDirect
Journal of Business Research
Tax effect on Spanish SME optimum debt maturity structure José López-Gracia ⁎, Reyes Mestre-Barberá Universitat de València, Facultad de Economía, Departamento de Contabilidad, Campus dels Tarongers, 46071-Valencia, Spain
a r t i c l e
i n f o
Article history: Received 1 February 2010 Received in revised form 1 March 2010 Accepted 1 June 2010 Available online 10 July 2010 Keywords: Small and medium-sized enterprises Debt maturity structure Dynamic model Tax effect System-GMM
a b s t r a c t This paper analyzes the influence of the tax effect on small and medium-sized enterprise (SME) debt maturity structure. This study builds a dynamic adjustment model which endogenizes optimum structure and assumes the existence of adjustment costs. Using Spanish data, the model is estimated using a systemGMM regression to a complete panel (11,028 firms) covering 1997–2004. SMEs adjust to their target at a speed of about 37% annually, the equivalent of employing about 20 months to cover only half of the existing gap. This rate is lower than those reported in other similar papers studying large companies with publicly tradable equity. © 2010 Elsevier Inc. All rights reserved.
1. Introduction The empirical evidence on debt maturity focuses mainly on large firms with publicly tradable equity (Barclay and Smith, 1995; Stohs and Mauer, 1996). As regards taxes (tax approach), the topic this study addresses, research by Brick and Ravid (1985, 1991) and Kane, Marcus and McDonald (1985) figures prominently. Recently, however, the debt maturity behavior of small and medium-sized enterprises (SMEs) has also been studied (García-Teruel and Martínez-Solano, 2007; Scherr and Hulburt, 2001). Empirical research exclusively addressing tax models is rare (e.g., Harwood and Manzon, 2000). Although many theorists focus on debt maturity from the angle of new debt issues, others assume that firms continuously follow an optimal policy, and therefore define the dependent variable in terms of debt maturity structure (Ravid, 1996). The characteristics of our data, discussed below, recommend this second approach. SMEs are the focus of the paper due to the important role they play in the Spanish economy and how different they are from large firms. The different strategies followed by SMEs versus large firms regarding their debt maturity structure are well documented (Scherr and Hulburt, 2001). Typical financial restrictions affecting SMEs due to their financial opacity, higher information asymmetry, and higher business risk may produce significant agency costs of debt. Consequently, SMEs use mostly short term financing to lower agency costs. SME profit volatility tends to be higher, making them an interesting subject to analyze using the tax effect. Barclay and Smith (1995) do not find strong evidence of a tax ⁎ Corresponding author. E-mail addresses: [email protected] (J. López-Gracia), [email protected] (R. Mestre-Barberá). 0148-2963/$ – see front matter © 2010 Elsevier Inc. All rights reserved. doi:10.1016/j.jbusres.2010.06.001
effect on debt maturity in large firms, probably because of these firms' lower earnings variability. Our paper specifically analyzes the tax effect on the debt maturity structure of SMEs, a subject that, to our knowledge, previous research has not addressed. This research focuses on the Spanish economy, which recorded much faster growth over the sample period, 1997–2004, than its European counterparts. SMEs were the main contributors to this growth. SMEs accumulated high levels of debt, taking advantage of low interest rates and tax relief. The Spanish tax code provided some advantages targeted at small business, including incentives to make new investments and hire new employees. Our sample consists of an eight-year panel on 11,028 firms. This research assumes a certain survivorship bias in estimation results due to the high rate of bankruptcy of small firms (Céspedes et al., 2010). This study proposes a dynamic adjustment model that makes it possible to confirm whether SMEs adopt an optimum debt maturity structure and examine the speed at which they adjust to their target. Capital structure research often employs this approach (Flannery and Rangan, 2006; López-Gracia and Sogorb-Mira, 2008 and Serrasqueiro and Nunes, 2009), but not so maturity structure (Antoniou et al., 2006; Ozkan, 2000). As optimum debt structure is not observable, the model endogenizes optimum debt structure by replacing it with a vector of observable explanatory variables. The advantage of this approach is that our model captures the adjustment costs firms face in seeking their optimum level. Otherwise, estimates would erroneously reflect that firms do not face such costs and that the financial markets they use are friction-free, thus producing bias. This paper extends previous research on SME debt maturity, particularly that of Scherr and Hulburt (2001) and García-Teruel and Martínez-Solano (2007). While the former focuses on the determinants of debt maturity through cross-sectional data (US firms), the latter
650
J. López-Gracia, R. Mestre-Barberá / Journal of Business Research 64 (2011) 649–655
analyzes a data panel of Spanish manufacturing firms and considers a trade-off between risk and return as per Jun and Jen (2003). In contrast, this paper focuses on the tax effect on debt maturity and uses a dynamic analysis approach. This research contributes to the current literature in several specific ways. First, our sample includes all SMEs from a relatively new database. Second, this research focuses on the tax effect on debt maturity, which has been seldom applied to SMEs. Third, our empirical methodology consists of estimating a dynamic adjustment model, which is also uncommon in the debt maturity literature. Finally, the research applies a novel estimation technique, systemGMM regression, which improves the efficiency of estimates, makes it possible to estimate the speed of adjustment to optimal debt maturity, and incorporates industry effects. The main results indicate that the model fits the data well and that SMEs seem to adopt an optimum debt maturity structure, which they converge to slowly due to the high adjustment costs they face. Average adjustment speed is estimated at around 37%, the equivalent of taking some 20 months to cover half the existing gap. This adjustment rate is slower than the 50% estimated for large firms with publicly tradable debt (Antoniou et al., 2006; Ozkan, 2000). The effective tax rate is highly significant and both the interest rate gap and interest rate volatility also have a significant impact on debt maturity. 2. Theory and relevant background According to Brick and Ravid (1985), choosing between short and long term debt becomes important when the interest rate curve is not flat and the long term interest rates are taken as an accurate indication of future short term interest rates. Thus, when the yield curve is positively sloped, investors conclude that firms will pay higher interest on long term debt than on short term debt. Firms anticipate the tax deductions that debt provides when they choose longer maturities. In a later paper, Brick and Ravid (1991) extend the model to include interest rate uncertainty. The need to refinance short term debt at an unknown interest rate results in long term financing being the optimum strategy, regardless of whether the term structure is positively sloped or flat. The cost of interest rate uncertainty could exceed the advantages of short term debt when the yield curve is negatively sloped and, consequently, long term debt would be the optimum strategy. In addition, since only profitable firms pay taxes and tax relief is fixed or subject to certain limits, the current value of tax deductions on short term debt interest will decrease in proportion to the variability of such interest payments. Consequently, when short term interest rate volatility is high, firms will choose long term debt, as tax deductions for the interest paid remain fixed (Emery et al., 1988). Kane et al. (1985) indicate that optimum debt maturity should increase if the effective tax rate drops, such that the tax deductions exceed the transaction costs that firms accrue every year. Optimum maturity increases if firm value volatility decreases, as the firm would not have to readjust its capital structure so regularly. Finally, Scholes and Wolfson (1992) point out that firms subject to a high marginal tax rate will choose long term debt, as they are more capable of using the tax deductions the interest on this type of financing provides. The marginal tax rate is difficult to calculate, but tax deductions other than those obtained through debt, also called alternative tax shields, commonly proxy them. The firms that enjoy such shields have a lower incentive to use the tax benefits of debt. This study is based on the expected positive relationship between both the term structure of interest rates and short term interest rate volatility and debt maturity structure, on the one hand, and the expected negative relationship between both the effective tax rate and the volatility of firm value and the debt maturity structure, on the other. Furthermore, this research contemplates the effect of the marginal tax rate on debt maturity structure and controls for some other factors, as discussed below.
3. Empirical specifications and estimation methodology 3.1. Static model Although our research is mainly based on a dynamic model, the study first introduces a more general (static) model that most scholars have used. This model takes the following form: Yit = β0 + ∑βj Xjit + vi + uit
ð1Þ
Yit is the debt maturity structure observed in the current (t) period for company i. Xj is a vector of j characteristics that vary among firms and over time, β0 is a constant and βj is an associated vector of coefficients. Eq. (1) also incorporates specific or fixed firm effects (vi) and, therefore, does not include industry dummies. Furthermore, two of the characteristics included in Eq. (1) do not vary from one firm to another, only over time. As a result, this model can not include time dummies as the main goal is to analyze the separate effect of these variables. Finally, uit is an error term. Eq. (1) is estimated using fixed effects for comparative reasons alone. The intention is to provide a benchmark for comparison with previous research. Second, the regression is a reference to compare to the systemGMM technique. Note that Eq. (1) assumes that adjustment costs are not relevant while moving towards optimum debt maturity, representing the traditional approach. Dependent variable In previous research, debt maturity structure is measured as (i) the value-weighted average maturity of the firm's various debt issues (Easterwood and Kadapakkam, 1994) and (ii) the long term debt to total debt ratio (Antoniou et al., 2006). Although the latter measure is easier to operationalize it is less precise, as all long term debts are aggregated, regardless of each issue's term. The firms in this paper are SMEs whose stock is not traded on public exchanges. Relatively poor data disclosure forces us to measure debt maturity structure as the proportion of total debt represented by long term debt due in more than one year. Table 1 defines the variables, along with their notation and expected sign. Explanatory variables of debt maturity structure (1). Term structure of interest rates Firms supposedly have a larger proportion of long term debt when the term structure of interest rates slopes upwards. However, Antoniou et al. (2006) do not confirm this effect. This paper approaches the term structure of interest rates (denoted term) as the spread between the yields on ten-year Spanish public debt and one-year Spanish Treasury bills. Each of these two rates is obtained as monthly averages from the Bank of Spain. This measure varies over time, but is constant for all firms. Another way to analyze the variable term is as a dummy variable. Defined as a continuous variable, any fluctuation in the term spread, no matter how small, is expected to affect debt maturity structure. However, this may not necessarily be so (Newberry and Novack, 1999). When the spread is particularly wide, the effect could be opposite to that when it is narrow. For this reason, the variable term is also defined as a trend proxy (denoted dum_term) that takes noncontinuous values. (2). Short term interest rate volatility Long term debt allows firms to maintain high interest tax deductions even when short term rates are volatile. Hence, we expect short term interest rate volatility (denoted interestvolatility) to have a positive effect on debt maturity structure. Cai, Fairchild and Guney
J. López-Gracia, R. Mestre-Barberá / Journal of Business Research 64 (2011) 649–655 Table 1 Definition of variables and expected relationships. Variable
Definition Long term debt Long term debt + Short term debt
Debt maturity structure (longterm%) Sign Term structure of interest rates (term)
+
dum_term
+
Volatility of short term + interest rates (interestvolatility) dum_interestvolatility + Effective tax rate (taxrate1) taxrate1×dum_shield
Effective tax rate (taxrate2) taxrate2×dum_shield
Volatility of firm value (firmvolatility) Ratio of debt (debt) Size (size) Growth opportunities (growth) Asset maturity (assetmaturity)
− −
− −
−
Difference between the interest rates of 10-year Spanish public debt and the 1-year Spanish Treasury bills. Dummy variable that equals 1 if term N 1.61 and 0 otherwise. Standard deviation (at any year) of the 12-month interbank interest rate. Dummy variable that equals 1 if interestvolatility N 0.34 and 0 otherwise. Income tax Operating cash flow Interaction term between taxrate1 and the dummy variable shield which equals 1 if shield≤0.034 and 0 otherwise. The variable shield is defined as the ratio of depreciation to total fixed assets. Income tax Profit before taxes Interaction term between taxrate2 and the dummy variable shield which equals 1 if shield≤0.031 and 0 otherwise. The variable shield is defined as the ratio of depreciation to total fixed assets. Standard deviation of operational cash flow Mean of total assets
Long term debt + Short term debt +/ Total assets − + Ln (total assets) − +
Salest Salest1 Fixed assets Fixed assets + × Depreciation Total assets Receivables Receivables × + Total assets Total revenues Inventories Inventories Other assets × + + Total assets Sales Total assets
+
(2008) fail to confirm the expected positive effect on debt maturity. The variable interestvolatility is calculated as the annual standard deviation of the monthly 12-month interbank rate. Therefore, this variable is constant across firms, but varies over time. It is possible that the effect of this variable on debt maturity could be hidden when defined as continuous, yet could be much stronger in the specific years when interest rates are particularly volatile. Hence, the model also uses a dummy variable called dum_interestvolatility.
651
Anticipation of interest tax deductions is more important for firms with high marginal tax rates and, therefore, low alternative tax shields from sources other than debt and debt maturity (Harwood and Manzon, 2000). To account for the strength of a firm's alternative tax shield, the model introduces a new dummy variable (denoted dum_shield) which interacts with the effective tax rate (taxrate1 and taxrate2). This interaction term is called taxrate1×dum_shield or taxrate2×dum_shield. (4). Volatility of the firm value If firm value is highly volatile, the risk of insolvency increases and so do expected bankruptcy costs. To reduce these costs, firms must frequently change their capital structure and use short term debt, which entails lower restructuring costs. In contrast, if firm value is less volatile, managers will use long term financing, as less frequent capital structure rebalancing is necessary. Stohs and Mauer (1996) confirm this hypothesis, whereas Körner (2007) finds this relationship insignificant. Firm value volatility (denoted as firmvolatility) for each year is the standard deviation of operating cash flow between that year and all previous years, divided by average total assets during those years. (5). Control variables First, the model includes the debt ratio because differences between firms' levels of borrowing, and consequent tax deductions, could be related to their debt maturity structures (Stohs and Mauer, 1996). Second, the research incorporates firm size. The transaction costs of debt, inversely related to firm size, play an important role when SMEs choose their debt maturity structure. Likewise, the existence of scale economies in debt contracts, as well as asymmetric information (agency costs), make it difficult for SMEs to access certain types of long term financing. The model therefore expects firm size to have a positive effect (Titman and Wessels, 1988). Third, the research considers growth opportunities. According to Myers (1977), shortening debt maturity so that it matures before an investment option can be exercised solves the problem of sub-optimal investment decisions. This problem can be more serious in SMEs which have high growth opportunities. The model expects a negative relationship between debt maturity and growth opportunities. Finally, the research controls for asset maturity because debt maturity and asset maturity should be matched in order to reach financial equilibrium (Morris, 1976). Notations and definitions of all control variables are shown in Table 1. 3.2. Dynamic model At this stage, the analysis assumes that firms partially adjust their debt maturity structure on an annual basis and, therefore, approach their optimum structure gradually (Antoniou et al., 2006). The speed at which firms approach their target depends on the adjustment costs they face, as well as on the costs of remaining unbalanced. A dynamic adjustment model could be as follows:
(3). Effective tax rate The effective tax rate affects the choice of debt maturity due to interest being tax deductible. When the effective tax rate is low, the tax relief stemming from debt decreases and, therefore, longer term debt is necessary for tax deductions to exceed the transaction costs that firms accrue each year. Daniševská (2002) confirms this hypothesis, whereas Ozkan (2000) confirms the opposite sign. The effective tax rate is measured as the income tax to operating cash flow ratio (taxrate1). The model also defines an effective tax rate (taxrate2) as the ratio of income tax expense to pretax profit. In this second variable, the rate itself already includes the effect of borrowing (Graham, 1996). Firms reporting losses are dropped from the sample when this variable is used, to avoid misinterpretation of the effective tax rate.
Yit − Yit−1 = μ ðY*it − Yit−1 Þ
ð2Þ
where Y⁎ it, unknown, denotes the optimum debt maturity structure for firm i in the current period t, Yit and Yit − 1 are the maturity structures observed in t and t − 1, respectively and μ is the speed of adjustment towards optimum structure, which is assumed constant. The coefficient μ normally takes a value between 0 and 1. Thus, each year firms progress a proportion μ of the distance between their target debt at the end of the year and their actual structure the previous year. If μ N 1, firms over-adjust and if μ b 0, firms deviate from their target. Rearranging Eq. (2), the model obtains Yit = ð1−μ Þ:Yit−1 + μ Y*it
ð3Þ
652
J. López-Gracia, R. Mestre-Barberá / Journal of Business Research 64 (2011) 649–655
Eq. (3) captures the adjustment process in which firms converge to their optimum debt structure, which due to being unknown, must be substituted by an estimated value. Therefore, a vector of explanatory variables replaces Y⁎ in Eq. (3) as follows: Y*it = β0 + ∑βj Xjit
ð4Þ
where Xj is a vector of j characteristics that vary among firms and over time and which have been examined in subsection 3.1 above, β0 is a constant term and βj is the vector of associated coefficients. By substituting Eq. (4) in Eq. (3), rearranging the terms and bearing in mind the panel data methodology, Yit = ð1−μ Þ · Yit−1 + μY*it + vi + vs + uit = ð1−μ Þ · Yit−1 + μ · β0 + ∑βj Xjit + vi + vs + uit = μβ0
ð5Þ
+ ð1−μ Þ⋅Yit−1 + μ∑βj Xjit + vi + vs + uit where uit is an error term, vi represents the individual effects which capture firm heterogeneity and are fixed over time and vs are industry dummies to control the possible effect of the sector a firm belongs to. As discussed in Eq. (1) above (static model), some of the explanatory variables of Y⁎ (variables term and interestvolatility) vary over time, but remain constant for all firms. Consequently, the model omits the temporary dummy variables that reflect possible general macroeconomic effects in order to estimate their value accordingly. Industry dummies are introduced only when using system GMM, as this method makes a double estimation (in levels and first differences, see Roodman, 2008). Eq. (5) is the principal model for this research. Eq. (5) is first estimated by using the first-differenced GMM estimator of Holtz-Eakin, Newey and Rosen (1988) and Arellano and Bond (1991), which provides an adequate procedure for obtaining asymptotically efficient estimators and controls for both individual heterogeneity (fixed effects) and potential endogeneity problems. Following Arellano and Bover (1995) and Blundell and Bond (1998), the research uses the system GMM estimator, which is based on instruments in first differences for equations in levels and instruments in levels for equations in first differences. In addition, this estimator has much smaller finite sample bias and is much more accurate when estimating autoregressive parameters using panels with a large number of cross-section units and a small number of time periods, as is the case with our data (Bond, 2002). 4. Data and descriptive analysis of variables 4.1. Data The data for this research are drawn from the SABI database (Sistema de Análisis de Balances Ibéricos for Spain), which covered 932,204 firms. This database is compiled by Bureau van Dijk (BvD, http://www.bvdep. com) which provides financial information on Spanish firms obtained mainly from their annual accounts. Spanish corporations must report their basic financial statements annually to the Mercantile Register, which makes the information available to the public. The sample firms in this research comply with the European Union criteria for SMEs (European Union Recommendation, 2003/361/CE). The sample excludes firms with less than 10 workers (micro firms) and firms that were not Public Limited Corporations (plc) due to their financial information being low quality. The sample discards firms belonging to the financial or insurance sectors and also those that had undergone bankruptcy proceedings during the sample period. Additionally, the sample removes the firms whose data were missing or internally inconsistent. These requisites hold for all eight years from 1997 to 2004, which is the sample period and also coincides with the years available in the database when the sample was selected. The sample eliminates outliers outside the interval μ ± 3σ for most of the
explanatory variables. Finally, the sample comprises complete panel data on 11,028 firms with 88,224 observations. The observations from 1997 are lost during construction of variables. Using the Standard Industrial Classification of Economic Activities 1993, Royal Decree 1560/ 1992 as the benchmark nomenclature, the sample is divided into nine groups. Manufacturing firms constitute 46.3% of the sample, followed by Trade and Construction with 27.5% and 12.3% respectively. 4.2. Descriptive analysis of variables Table 2 contains summary statistics for the dependent and explanatory variables. All data are nominal (not real or deflated). The average value of total assets for sample firms is 4,501,943 Euro. Firms have an average debt ratio of 60% and an average effective tax rate (taxrate1) of about 15%. Firms showing a debt ratio of 99% are not excluded as this situation may be temporary. The variable term structure of interest rates (term) fluctuates from 0.91 in 2000 (minimum) to 1.95 in 2004 (maximum), averaging 1.47. Regarding the dependent variable (longterm%), 20% of the sample has no long term debt, which is consistent with our initial assumption of SMEs facing special difficulties accessing long term financing. A few firms show almost 100% long term debt. The distribution of the dependent variable by year (not shown) consistently averaged around 17%. Thus, Spanish SMEs use relatively little long term debt, unlike the 48% in samples used by Scherr and Hulburt (2001). 5. Analysis of results 5.1. Static model estimation In order to make the estimates of the static model (1) more robust, four different specifications are used and Table 3 presents the results. The Hausman test calls for the fixed effects or intragroup model (robust version) to be used rather than the random effect model. Wald's test of joint significance of regressors rejects the null hypothesis of all the parameters being equal to zero in all regressions, which guarantees the suitability of the static model. The results of estimation (1) in Table 3 fail to confirm the expected relationships for the term structure of interest rates (term) and short term interest rate volatility (interestvolatility). Regarding the variable term, this adverse result could be due to SMEs having greater incentives to embark on riskier projects when gaining access to long term financing and creditors' aversion to lending long term when the term structure slope is positive. The result for interestvolatility could reflect how unimportant loan market turmoil is for SMEs in light of their marked borrowing limitations. Hence, SME debt maturity policy would not be affected by refinancing difficulties stemming from interest rate volatility, but by the effective level of financial restrictions for these firms. The coefficient for the effective tax rate (taxrate1) is negative, as expected. Thus, SMEs increase their debt maturity when the effective tax rate is low. All the model specifications reject the conjecture that firm value volatility (firmvolatility) has a significant negative effect on debt maturity. Therefore, SME debt maturity is not related to restructuring or bankruptcy costs. The coefficients of the model's four control variables (debt, size, growth and assetmaturity) are significant in all specifications. The research also explores other ways of analyzing the term structure of interest rates (term) and short term interest rate volatility (interestvolatility). As indicated in subsection 3.1, the model replaces the variable term for another trend proxy that takes noncontinuous values. The values and trend of this variable are:
Year Term
98 1.03
99 1.71
00 0.91
01 1.20
02 1.61
03 1.93
04 1.95
J. López-Gracia, R. Mestre-Barberá / Journal of Business Research 64 (2011) 649–655 Table 2 Summary statistics. Number of observations: 88,224 (11,028 firms) Variables
Mean
Standard deviation
Median
Minimum
Maximum
longterm% term interestvolatility taxrate1 taxrate2a firmvolatilityb debt size growthb assetmaturity
0.17 1.47 0.34 0.15 0.3 0.04 0.6 7.77 1.1 4.42
0.19 0.37 0.13 0.27 0.11 0.03 0.21 1.1 0.25 6.3
0.09 1.50 0.35 0.14 0.32 0.03 0.63 7.68 1.07 2.52
0 0.91 0.11 −12.18 −1.04 0 0.002 3.78 0.02 0.01
0.99 1.95 0.54 12.05 1.65 0.17 0.99 12.6 7.15 91.85
Notes: Period of analysis 1997–2004. a taxrate2 statistics use only 68,808 observations due to variable construction requirements. b firmvolatility and growth statistics use only 77,196 observations due to variable construction requirements.
According to these figures, the cut-off point that makes it possible to identify the years in which the interest rate gap is wide, is 1.61. Therefore, in 1999, 2003 and 2004, long term interest rates are clearly higher than short term rates. This difference is not as noticeable in the rest of the years. The model defines a dummy variable called dum_term: dumXterm = 1 if term N 1:61 and 0 otherwise: Likewise, the model substitutes for the variable interestvolatility another dummy called dum_interestvolatility: dum– interestvolatility = 1 if interestvolatility N 0:34 ðmeanÞ and 0 otherwise
where the mean 0.34 is calculated on the 88,224 observations from the whole sample. Estimation (2) of Table 3 presents the results for the new specifications. The prediction that the interest rate spread and interest rate volatility would positively affect debt maturity structure holds entirely using the new variables dum_term and dum_interestvolatility. In addition, the results of estimation (2) confirm all forecasts, with the exception of that for firm value volatility.
653
The research now explores the relationship between the effective tax rate and debt maturity by dividing firms into two groups, according to whether they have strong alternative tax shields. Following Chen (2004) and Titman and Wessels (1988), the model uses depreciation as a proxy for alternative tax shields. A dummy variable called dum_shield separates firms, as follows: dum_shield= 1 if shield ≤ 0.034 (median) and 0 otherwise, where shield is the ratio of depreciation to total fixed assets and is calculated on the whole sample. Next, the variable taxrate1 is replaced in the model with an interaction term taxrate1 and the dum_shield. This term, taxrate1×dum_shield, equals the value of the effective tax rate if they have low alternative tax shields, and 0 if firms have high alternative tax deductions. The coefficient for this interaction variable represents the difference in the effect tax rate's impact on debt maturity for firms with weak alternative tax shields versus firms with strong alternative tax shields. The expected relationship between taxrate1×dum_shield and the dependent variable is therefore negative. Column (3) of Table 3 displays the results of this new specification, which confirm expectations. The variable taxrate1 is separately included to obtain a complete effect, although it is insignificant. Thus, SME debt maturity is negatively related to the effective tax rate when firms lack strong alternative tax shields. The predictions confirm results obtained in the previous specifications. As a robustness check, the model is tested using the second definition of the effective tax rate (taxrate2): the ratio of income tax to profit before taxes. Column (4) of Table 3 shows these results, fully consistent with those obtained earlier. 5.2. Dynamic model estimation Table 4 presents the estimation results of difference-GMM and system-GMM regressions of dynamic model (5). Both provide results for the two different specifications used for effective tax rate (taxrate1 and taxrate2). The estimation technique applies a robust second-stage version of the GMM estimator, although results from the robust first-stage procedure are similar (not shown). The model treats the explanatory variables as endogenous, with the exception of the dummies. The Wald test of joint significance of regressors (Wald-1) verifies the overall validity of the explanatory variables in the model (all four specifications have a p-value equal to zero). The result of the Wald test of joint significance of industry variables (Wald-2) for system-GMM
Table 3 Fixed effect regressions of debt maturity structure during 1997–2004. Static model (1): Yit = β0 + ∑ βj Xjit + vi + uit Explanatory variables
Expected relationship
term dum_term interestvolatility dum_interestvolatility taxrate1 taxrate2 × dum_shield taxrate2 taxrate2×dum_shield firmvolatility debt size growth assetmaturity constant
+ + + + − − − − − +/− +/− − +
Within R2 Wald test (F statistic) Hausman's test χ2 Firms Observations Note: P-values in parentheses. ⁎ Significant at the 1% level. ⁎⁎ Significant at the 5% level.
(1)
(2) −0.0016 (0.291) −0.0045 (0.288) −0.0087 (0.000)⁎
0.0089 0.3325 0.0502 −0.0267 0.0057 −0.4134
(0.742) (0.000)⁎ (0.000)⁎ (0.000)⁎ (0.000)⁎ (0.000)⁎
0.1426 562.14 (0.000) 1794.40 (0.000) 11,028 77,196
(3) 0.0047 (0.000)⁎ 0.0061 (0.000)⁎ −0.0089 (0.000)⁎
0.0066 0.3268 0.0531 −0.0268 0.0057 −0.4425
(0.806) (0.000)⁎ (0.000)⁎ (0.000)⁎ (0.000)⁎ (0.000)⁎
0.1429 567.88 (0.000) 109.85 (0.000) 11,028 77,196
(4) 0.0047 (0.000)⁎
0.0045 (0.000)⁎
0.0060 (0.000)⁎ −0.0029 (0.275) −0.0096 (0.012)⁎⁎
0.0049 (0.002)⁎
0.0057 0.3274 0.0532 −0.0269 0.0057 −0.4437
(0.831) (0.000)⁎ (0.000)⁎ (0.000)⁎ (0.000)⁎ (0.000)⁎
0.1431 504.21 (0.000) 211.54 (0.000) 11,028 77,196
−0.0057 −0.0700 −0.0272 0.3235 0.0515 −0.0254 0.0056 −0.4261
(0.324) (0.000)⁎ (0.344) (0.000)⁎ (0.000)⁎ (0.000)⁎ (0.000)⁎ (0.000)⁎
0.1434 389.93 (0.000) 658.17 (0.000) 8,601 60,207
654
J. López-Gracia, R. Mestre-Barberá / Journal of Business Research 64 (2011) 649–655
Table 4 GMM regressions of debt maturity structure during 1997–2004. Dynamic model (5): Yit = μβ0+ (1−μ).Yit−1 + μ∑ βj Xjit + vi + vs + uit Difference-GMM Explanatory variables
Expected relationship
Yit−1 dum_term dum_interestvolatility taxrate1 taxrate1×dum_shield taxrate2 taxrate2×dum_shield firmvolatility debt size growth assetmaturity constant
+ + + − − − − − +/− +/− − +
Wald-1 Wald-2 Autocorrelation-1 Autocorrelation-2 Sargan test Instruments Firms Observations
(1)
System-GMM (2)
0.5946 0.0059 0.0045 0.0743 −0.0932
−0.1289 0.2209 0.0399 −0.0470 0.0028 −0.3457
(0.000)⁎ (0.001)⁎ (0.128) (0.057)⁎⁎⁎ (0.018)⁎⁎
(0.245) (0.000)⁎ (0.004)⁎ (0.006)⁎ (0.004)⁎ (0.019)⁎⁎
1921.01 (0.000) −30.332 (0.000) 0.52711 (0.5981) 45.8238 (0.6027) 61 11,028 55,140
(3) 0.5934 (0.000)⁎ 0.0056 (0.002)⁎ 0.0055 (0.059)⁎⁎⁎
−0.0324 −0.0480 −0.1491 0.1437 0.0327 −0.0209 0.0038 −0.2597
(0.281) (0.043)⁎⁎ (0.175) (0.021)⁎⁎ (0.015)⁎⁎ (0.200) (0.003)⁎ (0.053)⁎⁎
2263.98 (0.000) −28.081 (0.000) 0.82664 (0.4084) 58.85597 (0.6584) 76 8,601 43,005
(4) 0.7079 0.0053 0.0047 0.0443 −0.0554
0.0020 0.1524 0.0260 −0.0221 0.0047 −0.1499
(0.000)⁎ (0.000)⁎ (0.029)⁎⁎
0.6328 (0.000)⁎ 0.0050 (0.001)⁎ 0.0050 (0.031)⁎⁎
(0.167) (0.114)
(0.980) (0.000)⁎ (0.000)⁎ (0.124) (0.000)⁎ (0.353)
4252.55 (0.000) 30.29 (0.0002) −33.276 (0.000) −1.1643 (0.2443) 109.7465 (0.0502) 107 11,028 66,168
−0.0126 −0.0484 −0.0448 0.1280 0.0282 −0.0041 0.0038 0.1606
(0.646) (0.004)⁎ (0.588) (0.000)⁎ (0.000)⁎ (0.775) (0.000)⁎ (0.349)
4517.46 (0.000) 18.38 (0.0185) −28.582 (0.000) 0.90313 (0.3665) 102.2163 (0.4754) 122 8,601 51,606
Notes: P-values in parentheses. Estimated industry dummies are not reported. Variables Yit−1, firmvolatility, debt, size, growth and assetmaturity have been instrumented with one lag, whereas taxrate1 and the interaction term taxrate1×dum_shield includes instruments up to three lags (regressions 1 and 3). Yit−1 is instrumented up to two lags, taxrate2 and taxrate2×dum_shield up to the fifth lag and firmvolatility, debt, size, growth and assetmaturity with one lag (regressions 2 and 4). ⁎ Significant at the 1% level. ⁎⁎ Significant at the 5% level. ⁎⁎⁎ Significant at the 10% level.
estimates also confirms the joint significance of industry dummies. The second order serial correlation test (Autocorrelation-2) indicates no second order serial correlation in any estimates. The Sargan test of over identifying restrictions (Sargan) generally confirms the validity of the instruments chosen. Only estimation (3) does not confirm the Sargan test, probably due to our use of more observations and instruments (see Bowsher, 2002). As expected, all the estimates in Table 4 display markedly significant coefficients (1− μ), for the lagged dependent variable, which range from 0.59 for both difference-GMM estimations to 0.70 and 0.63, respectively, for system-GMM estimations (3) and (4). System-GMM estimations on lagged dependent variable are higher than their counterpart differenceGMM estimations as the latter may be subject to downward sample bias if (i) the (theoretical) autoregressive parameter is moderately high and the number of time-series observations is moderately small (Blundell and Bond, 1998). These findings on (1− μ) coefficients suggest the existence of an optimum SME debt maturity structure and, at the same time, the presence of high adjustment costs. Using specification (4) as a basis, a coefficient of 0.63 is equivalent to an adjustment speed towards target debt structure (μ) of 0.37. Thus, SMEs on average take approximately 20 months to cover only half the gap between their optimum debt maturity structure and their actual beginning-of-year structure. This time also implies that SMEs consider the cost of remaining unbalanced to be lower than the adjustment costs of reaching their optimum structure. Ozkan (2000) had reported an adjustment rate of 50% for large, publicly tradable firms. These results confirm the initial suspicion that SMEs find it more difficult to adjust to their optimum debt maturity structure. As Table 4 shows, the explanatory variables in the dynamic model (5) have the expected coefficient signs, with the exception of firm value volatility. The GMM estimations, therefore, reinforce the fixed effect estimations. Essentially the same result is obtained from the systemGMM estimations (specifications (3) and (4)). The term structure of interest rates (dum_term) affects debt maturity positively, particularly in the years when the interest rate gap is larger than 1.61. Likewise,
strong evidence supports the expected positive influence of short term interest rate volatility (dum_interestvolatility) on debt maturity, especially in years when interest rates are highly volatile. Debt maturity is negatively related to the effective tax rate (taxrate1×dum_shield and taxrate2×dum_shield), especially when firms enjoy few tax exemptions other than those from debt. To verify possible general effects of the time factor, the research also tests the dynamic model (5) by removing the variables term and interestvolatility (both constant for the firms and variant over time) and by introducing time dummies. The new results obtained (not shown) coincide with those displayed in Table 4 and the time dummies are equally significant. As manufacturing firms represent a large proportion of the sample, the study also carries out additional regressions on this sector separately. This new regression (not shown) resembles the previous results. Finally, the control variables, debt ratio, firm size, growth opportunities and maturity of assets generally confirm the previous fixed effect results from the static model where these variables were highly significant in all specifications. The positive effect of the debt ratio on debt maturity should be interpreted as an attempt by SMEs to reinforce their solvency. The findings confirm that the smallest firms, with weak financial disclosure and relatively high information costs, see their chances of obtaining long term financing restricted. Debt maturity is also negatively related to growth opportunities, suggesting that SMEs try to solve the under-investment problem by shortening debt maturity. Lastly, asset maturity is significant in all specifications, consistent with SMEs attempting to maintain compatibility between asset and liability maturities. 6. Conclusions This paper provides empirical evidence on the determinants of Spanish SME debt maturity structure and the speed at which these firms adjust to their optimum structure. Results show that SMEs sought out optimum debt maturity structure throughout 1997–2004.
J. López-Gracia, R. Mestre-Barberá / Journal of Business Research 64 (2011) 649–655
A dynamic adjustment model fits the data well and confirms its suitability for SMEs. The two system-GMM estimates indicate a speedof-adjustment score of 30% and 37%, respectively, the equivalent of a firm taking 20 months to cover half the distance between its beginning-of-year structure and the optimum structure. This result is lower than the case of large companies with publicly tradable equity, where the few extant studies report figures of about 50%. Thus, SMEs may prefer to remain unbalanced, rather than reach their targets at top speed due to the high adjustment costs they face. The results also indicate that such high adjustment costs could be a consequence of the financial restrictions afflicting SMEs. As regards the factors that determine optimum debt maturity structure from the perspective of the tax approach, the evidence from the system-GMM estimation is robust and indicates a strong positive relationship with the term structure of interest rates and interest rate volatility, as expected. Debt maturity is strongly related to the effective tax rate, particularly when firms have few alternative tax exemptions. The debt ratio, firm size, growth opportunities and asset maturity significantly affect debt maturity. These results are in line with both Scherr and Hulburt (2001) and García-Teruel and Martínez-Solano (2007) on SMEs and offer a new perspective based on tax rationale and the advantages of a dynamic model. Nevertheless, some care must be taken when comparing the results of this research, as they have two possible limitations stemming from (i) the fact that hypotheses are formulated under a general tax approach commonly used on large (traded) firms and (ii) possible survivorship bias caused by the use of a complete data panel. Acknowledgments The authors are grateful to Pedro Martinez-Solano and Alejandro Casino-Martínez for their valuable comments and suggestions made on this manuscript. Helpful observations from Julio Pindado, Félix López, Juan Sanchis, Vicente Pallardó, Cristina Aybar, Francisco Sogorb and two anonymous referees are also appreciated. David Smith (the Associate Editor of JBR) also provided helpful comments and encouragement. Funding from the Regional Government of Generalitat Valenciana is gratefully acknowledged (Project GVPRE/2008/349). References Antoniou A, Guney Y, Paudyal K. The determinants of debt maturity structure: evidence from France, Germany and the UK. Eur Financ Manage 2006;12(2):161–94. Arellano M, Bond S. Some tests of specification for panel data: Monte Carlo evidence and an application to employment equations. Rev Econ Stud 1991;58:277–97. Arellano M, Bover O. Another look at the instrumental-variable estimation of errorcomponents models. J Econometrics 1995;68:29–52. Barclay MJ, Smith Jr CW. The maturity structure of corporate debt. J Finance 1995;L(2): 609–31 June. Blundell RW, Bond SR. Initial conditions and moment restrictions in dynamic panel data models. J. Econometrics 1998;87:115–43.
655
Bond S. Dynamic panel data models: a guide to micro data methods and practice. Portuguese Econ J 2002;1(2):141–62. Bowsher C. On testing overidentifying restrictions in dynamic panel data models. Econo Lett 2002;77(2):211–20. Brick IE, Ravid SA. On the relevance of debt maturity structure. J Finance 1985;XL(5): 1423–37 December. Brick IE, Ravid SA. Interest rate uncertainty and the optimum debt maturity structure. J Financ Quant Anal 1991;26(1):63–81 March. Cai K, Fairchild R, Guney Y. Debt maturity structure of Chinese companies. Pac Basin Finance J 2008;16:268–97. Céspedes J, González M, Molina CA. Ownership and capital structure in Latin America. J Bus Res 2010;63:248–54. Chen JJ. Determinants of capital structure of Chinese-listed companies. J Bus Res 2004;57:1341–51. Daniševská P. Is debt maturity determined by asymmetric information about shortterm or long-term earnings? Social Science Research Network (SSRN). http://ssrn. com/abstract=3027372002 June. Easterwood JC, Kadapakkam PR. Agency conflicts, issue costs, and debt maturity. Q J Bus Econ 1994;33(3):69–80 Summer. Emery DR, Lewellen WG, Mauer DC. Tax-timing options, leverage, and the choice of corporate form. J Financ Res 1988;11:99-110. Flannery MJ, Rangan KP. Partial adjustment toward target capital structures. J Financ Econ 2006;79:469–506. García-Teruel PJ, Martínez-Solano P. Short-term debt in Spanish SMEs. Int Small Bus J 2007;25(6):579–602. Graham JR. Proxies for the corporate marginal tax rate. J Financ Econ 1996;42:187–221. Harwood E, Manzon Jr GB. Tax clienteles and debt maturity. J Am Taxation Assoc 2000;22(2):22–39. Holtz-Eakin D, Newey W, Rosen HS. Estimating vector autoregressions with panel data. Econometrica 1988;56:1371–96. Jun SG, Jen FC. Trade-off model of debt maturity structure. Rev Quant Finance Acc 2003;20:5-34. Kane A, Marcus AJ, McDonald RL. Debt policy and the rate of return premium to leverage. J Financ Quant Anal 1985;20(4):479–99 December. Körner P. The determinants of corporate debt maturity structure: evidence from Czech firms. Czech J Econ Finance 2007;57:142–58. López-Gracia J, Sogorb-Mira F. Testing trade-off and pecking order theories financing SMEs. Small Bus Econ 2008;31(2):117–36 August. Morris JR. On corporate debt maturity strategies. J Finance 1976;XXXI(1):29–37 March. Myers SC. Determinants of corporate borrowing. J Financ Econ 1977;5:138–66 [November]. Newberry KJ, Novack GF. The effect of taxes on corporate debt maturity decisions: an analysis of public and private bond offerings. J Am Taxation Assoc 1999;21(2):1-16. Ozkan A. An empirical analysis of corporate debt maturity structure. Eur Financ Manage 2000;6(2):197–212. Ravid SA. Debt maturity-a survey. Financ Mark Inst Instrum 1996;5(3):1-69 March. Roodman D. How to do xtabond2: an introduction to difference and system GMM in Stata. Center for Global Development Working Paper number 103. Social Science Research Network (SSRN), December 2006 (revised July 2008). http://ssrn.com/ abstract = 982943. Scherr FC, Hulburt HM. The debt maturity structure of small firms. Financ Manage 2001: 85-111 Spring. Scholes MS, Wolfson MA. Taxes and business strategy: a planning approach. Englewood Cliffs, NJ: Prentice Hall; 1992. Serrasqueiro Z, Nunes PM. Non-linear relationships between growth opportunities and debt: evidence from quoted Portuguese companies. J Bus Res 2009, doi:10.1016/j. jbusres.2009.11.003. Stohs MH, Mauer DC. The determinants of corporate debt maturity structure. J Bus 1996;69(3):279–312. Titman S, Wessels R. The determinants of capital structure choice. J Finance 1988;XLIII(1): 1-19 March.
Contents lists available at ScienceDirect
Journal of Business Research
Tax effect on Spanish SME optimum debt maturity structure José López-Gracia ⁎, Reyes Mestre-Barberá Universitat de València, Facultad de Economía, Departamento de Contabilidad, Campus dels Tarongers, 46071-Valencia, Spain
a r t i c l e
i n f o
Article history: Received 1 February 2010 Received in revised form 1 March 2010 Accepted 1 June 2010 Available online 10 July 2010 Keywords: Small and medium-sized enterprises Debt maturity structure Dynamic model Tax effect System-GMM
a b s t r a c t This paper analyzes the influence of the tax effect on small and medium-sized enterprise (SME) debt maturity structure. This study builds a dynamic adjustment model which endogenizes optimum structure and assumes the existence of adjustment costs. Using Spanish data, the model is estimated using a systemGMM regression to a complete panel (11,028 firms) covering 1997–2004. SMEs adjust to their target at a speed of about 37% annually, the equivalent of employing about 20 months to cover only half of the existing gap. This rate is lower than those reported in other similar papers studying large companies with publicly tradable equity. © 2010 Elsevier Inc. All rights reserved.
1. Introduction The empirical evidence on debt maturity focuses mainly on large firms with publicly tradable equity (Barclay and Smith, 1995; Stohs and Mauer, 1996). As regards taxes (tax approach), the topic this study addresses, research by Brick and Ravid (1985, 1991) and Kane, Marcus and McDonald (1985) figures prominently. Recently, however, the debt maturity behavior of small and medium-sized enterprises (SMEs) has also been studied (García-Teruel and Martínez-Solano, 2007; Scherr and Hulburt, 2001). Empirical research exclusively addressing tax models is rare (e.g., Harwood and Manzon, 2000). Although many theorists focus on debt maturity from the angle of new debt issues, others assume that firms continuously follow an optimal policy, and therefore define the dependent variable in terms of debt maturity structure (Ravid, 1996). The characteristics of our data, discussed below, recommend this second approach. SMEs are the focus of the paper due to the important role they play in the Spanish economy and how different they are from large firms. The different strategies followed by SMEs versus large firms regarding their debt maturity structure are well documented (Scherr and Hulburt, 2001). Typical financial restrictions affecting SMEs due to their financial opacity, higher information asymmetry, and higher business risk may produce significant agency costs of debt. Consequently, SMEs use mostly short term financing to lower agency costs. SME profit volatility tends to be higher, making them an interesting subject to analyze using the tax effect. Barclay and Smith (1995) do not find strong evidence of a tax ⁎ Corresponding author. E-mail addresses: [email protected] (J. López-Gracia), [email protected] (R. Mestre-Barberá). 0148-2963/$ – see front matter © 2010 Elsevier Inc. All rights reserved. doi:10.1016/j.jbusres.2010.06.001
effect on debt maturity in large firms, probably because of these firms' lower earnings variability. Our paper specifically analyzes the tax effect on the debt maturity structure of SMEs, a subject that, to our knowledge, previous research has not addressed. This research focuses on the Spanish economy, which recorded much faster growth over the sample period, 1997–2004, than its European counterparts. SMEs were the main contributors to this growth. SMEs accumulated high levels of debt, taking advantage of low interest rates and tax relief. The Spanish tax code provided some advantages targeted at small business, including incentives to make new investments and hire new employees. Our sample consists of an eight-year panel on 11,028 firms. This research assumes a certain survivorship bias in estimation results due to the high rate of bankruptcy of small firms (Céspedes et al., 2010). This study proposes a dynamic adjustment model that makes it possible to confirm whether SMEs adopt an optimum debt maturity structure and examine the speed at which they adjust to their target. Capital structure research often employs this approach (Flannery and Rangan, 2006; López-Gracia and Sogorb-Mira, 2008 and Serrasqueiro and Nunes, 2009), but not so maturity structure (Antoniou et al., 2006; Ozkan, 2000). As optimum debt structure is not observable, the model endogenizes optimum debt structure by replacing it with a vector of observable explanatory variables. The advantage of this approach is that our model captures the adjustment costs firms face in seeking their optimum level. Otherwise, estimates would erroneously reflect that firms do not face such costs and that the financial markets they use are friction-free, thus producing bias. This paper extends previous research on SME debt maturity, particularly that of Scherr and Hulburt (2001) and García-Teruel and Martínez-Solano (2007). While the former focuses on the determinants of debt maturity through cross-sectional data (US firms), the latter
650
J. López-Gracia, R. Mestre-Barberá / Journal of Business Research 64 (2011) 649–655
analyzes a data panel of Spanish manufacturing firms and considers a trade-off between risk and return as per Jun and Jen (2003). In contrast, this paper focuses on the tax effect on debt maturity and uses a dynamic analysis approach. This research contributes to the current literature in several specific ways. First, our sample includes all SMEs from a relatively new database. Second, this research focuses on the tax effect on debt maturity, which has been seldom applied to SMEs. Third, our empirical methodology consists of estimating a dynamic adjustment model, which is also uncommon in the debt maturity literature. Finally, the research applies a novel estimation technique, systemGMM regression, which improves the efficiency of estimates, makes it possible to estimate the speed of adjustment to optimal debt maturity, and incorporates industry effects. The main results indicate that the model fits the data well and that SMEs seem to adopt an optimum debt maturity structure, which they converge to slowly due to the high adjustment costs they face. Average adjustment speed is estimated at around 37%, the equivalent of taking some 20 months to cover half the existing gap. This adjustment rate is slower than the 50% estimated for large firms with publicly tradable debt (Antoniou et al., 2006; Ozkan, 2000). The effective tax rate is highly significant and both the interest rate gap and interest rate volatility also have a significant impact on debt maturity. 2. Theory and relevant background According to Brick and Ravid (1985), choosing between short and long term debt becomes important when the interest rate curve is not flat and the long term interest rates are taken as an accurate indication of future short term interest rates. Thus, when the yield curve is positively sloped, investors conclude that firms will pay higher interest on long term debt than on short term debt. Firms anticipate the tax deductions that debt provides when they choose longer maturities. In a later paper, Brick and Ravid (1991) extend the model to include interest rate uncertainty. The need to refinance short term debt at an unknown interest rate results in long term financing being the optimum strategy, regardless of whether the term structure is positively sloped or flat. The cost of interest rate uncertainty could exceed the advantages of short term debt when the yield curve is negatively sloped and, consequently, long term debt would be the optimum strategy. In addition, since only profitable firms pay taxes and tax relief is fixed or subject to certain limits, the current value of tax deductions on short term debt interest will decrease in proportion to the variability of such interest payments. Consequently, when short term interest rate volatility is high, firms will choose long term debt, as tax deductions for the interest paid remain fixed (Emery et al., 1988). Kane et al. (1985) indicate that optimum debt maturity should increase if the effective tax rate drops, such that the tax deductions exceed the transaction costs that firms accrue every year. Optimum maturity increases if firm value volatility decreases, as the firm would not have to readjust its capital structure so regularly. Finally, Scholes and Wolfson (1992) point out that firms subject to a high marginal tax rate will choose long term debt, as they are more capable of using the tax deductions the interest on this type of financing provides. The marginal tax rate is difficult to calculate, but tax deductions other than those obtained through debt, also called alternative tax shields, commonly proxy them. The firms that enjoy such shields have a lower incentive to use the tax benefits of debt. This study is based on the expected positive relationship between both the term structure of interest rates and short term interest rate volatility and debt maturity structure, on the one hand, and the expected negative relationship between both the effective tax rate and the volatility of firm value and the debt maturity structure, on the other. Furthermore, this research contemplates the effect of the marginal tax rate on debt maturity structure and controls for some other factors, as discussed below.
3. Empirical specifications and estimation methodology 3.1. Static model Although our research is mainly based on a dynamic model, the study first introduces a more general (static) model that most scholars have used. This model takes the following form: Yit = β0 + ∑βj Xjit + vi + uit
ð1Þ
Yit is the debt maturity structure observed in the current (t) period for company i. Xj is a vector of j characteristics that vary among firms and over time, β0 is a constant and βj is an associated vector of coefficients. Eq. (1) also incorporates specific or fixed firm effects (vi) and, therefore, does not include industry dummies. Furthermore, two of the characteristics included in Eq. (1) do not vary from one firm to another, only over time. As a result, this model can not include time dummies as the main goal is to analyze the separate effect of these variables. Finally, uit is an error term. Eq. (1) is estimated using fixed effects for comparative reasons alone. The intention is to provide a benchmark for comparison with previous research. Second, the regression is a reference to compare to the systemGMM technique. Note that Eq. (1) assumes that adjustment costs are not relevant while moving towards optimum debt maturity, representing the traditional approach. Dependent variable In previous research, debt maturity structure is measured as (i) the value-weighted average maturity of the firm's various debt issues (Easterwood and Kadapakkam, 1994) and (ii) the long term debt to total debt ratio (Antoniou et al., 2006). Although the latter measure is easier to operationalize it is less precise, as all long term debts are aggregated, regardless of each issue's term. The firms in this paper are SMEs whose stock is not traded on public exchanges. Relatively poor data disclosure forces us to measure debt maturity structure as the proportion of total debt represented by long term debt due in more than one year. Table 1 defines the variables, along with their notation and expected sign. Explanatory variables of debt maturity structure (1). Term structure of interest rates Firms supposedly have a larger proportion of long term debt when the term structure of interest rates slopes upwards. However, Antoniou et al. (2006) do not confirm this effect. This paper approaches the term structure of interest rates (denoted term) as the spread between the yields on ten-year Spanish public debt and one-year Spanish Treasury bills. Each of these two rates is obtained as monthly averages from the Bank of Spain. This measure varies over time, but is constant for all firms. Another way to analyze the variable term is as a dummy variable. Defined as a continuous variable, any fluctuation in the term spread, no matter how small, is expected to affect debt maturity structure. However, this may not necessarily be so (Newberry and Novack, 1999). When the spread is particularly wide, the effect could be opposite to that when it is narrow. For this reason, the variable term is also defined as a trend proxy (denoted dum_term) that takes noncontinuous values. (2). Short term interest rate volatility Long term debt allows firms to maintain high interest tax deductions even when short term rates are volatile. Hence, we expect short term interest rate volatility (denoted interestvolatility) to have a positive effect on debt maturity structure. Cai, Fairchild and Guney
J. López-Gracia, R. Mestre-Barberá / Journal of Business Research 64 (2011) 649–655 Table 1 Definition of variables and expected relationships. Variable
Definition Long term debt Long term debt + Short term debt
Debt maturity structure (longterm%) Sign Term structure of interest rates (term)
+
dum_term
+
Volatility of short term + interest rates (interestvolatility) dum_interestvolatility + Effective tax rate (taxrate1) taxrate1×dum_shield
Effective tax rate (taxrate2) taxrate2×dum_shield
Volatility of firm value (firmvolatility) Ratio of debt (debt) Size (size) Growth opportunities (growth) Asset maturity (assetmaturity)
− −
− −
−
Difference between the interest rates of 10-year Spanish public debt and the 1-year Spanish Treasury bills. Dummy variable that equals 1 if term N 1.61 and 0 otherwise. Standard deviation (at any year) of the 12-month interbank interest rate. Dummy variable that equals 1 if interestvolatility N 0.34 and 0 otherwise. Income tax Operating cash flow Interaction term between taxrate1 and the dummy variable shield which equals 1 if shield≤0.034 and 0 otherwise. The variable shield is defined as the ratio of depreciation to total fixed assets. Income tax Profit before taxes Interaction term between taxrate2 and the dummy variable shield which equals 1 if shield≤0.031 and 0 otherwise. The variable shield is defined as the ratio of depreciation to total fixed assets. Standard deviation of operational cash flow Mean of total assets
Long term debt + Short term debt +/ Total assets − + Ln (total assets) − +
Salest Salest1 Fixed assets Fixed assets + × Depreciation Total assets Receivables Receivables × + Total assets Total revenues Inventories Inventories Other assets × + + Total assets Sales Total assets
+
(2008) fail to confirm the expected positive effect on debt maturity. The variable interestvolatility is calculated as the annual standard deviation of the monthly 12-month interbank rate. Therefore, this variable is constant across firms, but varies over time. It is possible that the effect of this variable on debt maturity could be hidden when defined as continuous, yet could be much stronger in the specific years when interest rates are particularly volatile. Hence, the model also uses a dummy variable called dum_interestvolatility.
651
Anticipation of interest tax deductions is more important for firms with high marginal tax rates and, therefore, low alternative tax shields from sources other than debt and debt maturity (Harwood and Manzon, 2000). To account for the strength of a firm's alternative tax shield, the model introduces a new dummy variable (denoted dum_shield) which interacts with the effective tax rate (taxrate1 and taxrate2). This interaction term is called taxrate1×dum_shield or taxrate2×dum_shield. (4). Volatility of the firm value If firm value is highly volatile, the risk of insolvency increases and so do expected bankruptcy costs. To reduce these costs, firms must frequently change their capital structure and use short term debt, which entails lower restructuring costs. In contrast, if firm value is less volatile, managers will use long term financing, as less frequent capital structure rebalancing is necessary. Stohs and Mauer (1996) confirm this hypothesis, whereas Körner (2007) finds this relationship insignificant. Firm value volatility (denoted as firmvolatility) for each year is the standard deviation of operating cash flow between that year and all previous years, divided by average total assets during those years. (5). Control variables First, the model includes the debt ratio because differences between firms' levels of borrowing, and consequent tax deductions, could be related to their debt maturity structures (Stohs and Mauer, 1996). Second, the research incorporates firm size. The transaction costs of debt, inversely related to firm size, play an important role when SMEs choose their debt maturity structure. Likewise, the existence of scale economies in debt contracts, as well as asymmetric information (agency costs), make it difficult for SMEs to access certain types of long term financing. The model therefore expects firm size to have a positive effect (Titman and Wessels, 1988). Third, the research considers growth opportunities. According to Myers (1977), shortening debt maturity so that it matures before an investment option can be exercised solves the problem of sub-optimal investment decisions. This problem can be more serious in SMEs which have high growth opportunities. The model expects a negative relationship between debt maturity and growth opportunities. Finally, the research controls for asset maturity because debt maturity and asset maturity should be matched in order to reach financial equilibrium (Morris, 1976). Notations and definitions of all control variables are shown in Table 1. 3.2. Dynamic model At this stage, the analysis assumes that firms partially adjust their debt maturity structure on an annual basis and, therefore, approach their optimum structure gradually (Antoniou et al., 2006). The speed at which firms approach their target depends on the adjustment costs they face, as well as on the costs of remaining unbalanced. A dynamic adjustment model could be as follows:
(3). Effective tax rate The effective tax rate affects the choice of debt maturity due to interest being tax deductible. When the effective tax rate is low, the tax relief stemming from debt decreases and, therefore, longer term debt is necessary for tax deductions to exceed the transaction costs that firms accrue each year. Daniševská (2002) confirms this hypothesis, whereas Ozkan (2000) confirms the opposite sign. The effective tax rate is measured as the income tax to operating cash flow ratio (taxrate1). The model also defines an effective tax rate (taxrate2) as the ratio of income tax expense to pretax profit. In this second variable, the rate itself already includes the effect of borrowing (Graham, 1996). Firms reporting losses are dropped from the sample when this variable is used, to avoid misinterpretation of the effective tax rate.
Yit − Yit−1 = μ ðY*it − Yit−1 Þ
ð2Þ
where Y⁎ it, unknown, denotes the optimum debt maturity structure for firm i in the current period t, Yit and Yit − 1 are the maturity structures observed in t and t − 1, respectively and μ is the speed of adjustment towards optimum structure, which is assumed constant. The coefficient μ normally takes a value between 0 and 1. Thus, each year firms progress a proportion μ of the distance between their target debt at the end of the year and their actual structure the previous year. If μ N 1, firms over-adjust and if μ b 0, firms deviate from their target. Rearranging Eq. (2), the model obtains Yit = ð1−μ Þ:Yit−1 + μ Y*it
ð3Þ
652
J. López-Gracia, R. Mestre-Barberá / Journal of Business Research 64 (2011) 649–655
Eq. (3) captures the adjustment process in which firms converge to their optimum debt structure, which due to being unknown, must be substituted by an estimated value. Therefore, a vector of explanatory variables replaces Y⁎ in Eq. (3) as follows: Y*it = β0 + ∑βj Xjit
ð4Þ
where Xj is a vector of j characteristics that vary among firms and over time and which have been examined in subsection 3.1 above, β0 is a constant term and βj is the vector of associated coefficients. By substituting Eq. (4) in Eq. (3), rearranging the terms and bearing in mind the panel data methodology, Yit = ð1−μ Þ · Yit−1 + μY*it + vi + vs + uit = ð1−μ Þ · Yit−1 + μ · β0 + ∑βj Xjit + vi + vs + uit = μβ0
ð5Þ
+ ð1−μ Þ⋅Yit−1 + μ∑βj Xjit + vi + vs + uit where uit is an error term, vi represents the individual effects which capture firm heterogeneity and are fixed over time and vs are industry dummies to control the possible effect of the sector a firm belongs to. As discussed in Eq. (1) above (static model), some of the explanatory variables of Y⁎ (variables term and interestvolatility) vary over time, but remain constant for all firms. Consequently, the model omits the temporary dummy variables that reflect possible general macroeconomic effects in order to estimate their value accordingly. Industry dummies are introduced only when using system GMM, as this method makes a double estimation (in levels and first differences, see Roodman, 2008). Eq. (5) is the principal model for this research. Eq. (5) is first estimated by using the first-differenced GMM estimator of Holtz-Eakin, Newey and Rosen (1988) and Arellano and Bond (1991), which provides an adequate procedure for obtaining asymptotically efficient estimators and controls for both individual heterogeneity (fixed effects) and potential endogeneity problems. Following Arellano and Bover (1995) and Blundell and Bond (1998), the research uses the system GMM estimator, which is based on instruments in first differences for equations in levels and instruments in levels for equations in first differences. In addition, this estimator has much smaller finite sample bias and is much more accurate when estimating autoregressive parameters using panels with a large number of cross-section units and a small number of time periods, as is the case with our data (Bond, 2002). 4. Data and descriptive analysis of variables 4.1. Data The data for this research are drawn from the SABI database (Sistema de Análisis de Balances Ibéricos for Spain), which covered 932,204 firms. This database is compiled by Bureau van Dijk (BvD, http://www.bvdep. com) which provides financial information on Spanish firms obtained mainly from their annual accounts. Spanish corporations must report their basic financial statements annually to the Mercantile Register, which makes the information available to the public. The sample firms in this research comply with the European Union criteria for SMEs (European Union Recommendation, 2003/361/CE). The sample excludes firms with less than 10 workers (micro firms) and firms that were not Public Limited Corporations (plc) due to their financial information being low quality. The sample discards firms belonging to the financial or insurance sectors and also those that had undergone bankruptcy proceedings during the sample period. Additionally, the sample removes the firms whose data were missing or internally inconsistent. These requisites hold for all eight years from 1997 to 2004, which is the sample period and also coincides with the years available in the database when the sample was selected. The sample eliminates outliers outside the interval μ ± 3σ for most of the
explanatory variables. Finally, the sample comprises complete panel data on 11,028 firms with 88,224 observations. The observations from 1997 are lost during construction of variables. Using the Standard Industrial Classification of Economic Activities 1993, Royal Decree 1560/ 1992 as the benchmark nomenclature, the sample is divided into nine groups. Manufacturing firms constitute 46.3% of the sample, followed by Trade and Construction with 27.5% and 12.3% respectively. 4.2. Descriptive analysis of variables Table 2 contains summary statistics for the dependent and explanatory variables. All data are nominal (not real or deflated). The average value of total assets for sample firms is 4,501,943 Euro. Firms have an average debt ratio of 60% and an average effective tax rate (taxrate1) of about 15%. Firms showing a debt ratio of 99% are not excluded as this situation may be temporary. The variable term structure of interest rates (term) fluctuates from 0.91 in 2000 (minimum) to 1.95 in 2004 (maximum), averaging 1.47. Regarding the dependent variable (longterm%), 20% of the sample has no long term debt, which is consistent with our initial assumption of SMEs facing special difficulties accessing long term financing. A few firms show almost 100% long term debt. The distribution of the dependent variable by year (not shown) consistently averaged around 17%. Thus, Spanish SMEs use relatively little long term debt, unlike the 48% in samples used by Scherr and Hulburt (2001). 5. Analysis of results 5.1. Static model estimation In order to make the estimates of the static model (1) more robust, four different specifications are used and Table 3 presents the results. The Hausman test calls for the fixed effects or intragroup model (robust version) to be used rather than the random effect model. Wald's test of joint significance of regressors rejects the null hypothesis of all the parameters being equal to zero in all regressions, which guarantees the suitability of the static model. The results of estimation (1) in Table 3 fail to confirm the expected relationships for the term structure of interest rates (term) and short term interest rate volatility (interestvolatility). Regarding the variable term, this adverse result could be due to SMEs having greater incentives to embark on riskier projects when gaining access to long term financing and creditors' aversion to lending long term when the term structure slope is positive. The result for interestvolatility could reflect how unimportant loan market turmoil is for SMEs in light of their marked borrowing limitations. Hence, SME debt maturity policy would not be affected by refinancing difficulties stemming from interest rate volatility, but by the effective level of financial restrictions for these firms. The coefficient for the effective tax rate (taxrate1) is negative, as expected. Thus, SMEs increase their debt maturity when the effective tax rate is low. All the model specifications reject the conjecture that firm value volatility (firmvolatility) has a significant negative effect on debt maturity. Therefore, SME debt maturity is not related to restructuring or bankruptcy costs. The coefficients of the model's four control variables (debt, size, growth and assetmaturity) are significant in all specifications. The research also explores other ways of analyzing the term structure of interest rates (term) and short term interest rate volatility (interestvolatility). As indicated in subsection 3.1, the model replaces the variable term for another trend proxy that takes noncontinuous values. The values and trend of this variable are:
Year Term
98 1.03
99 1.71
00 0.91
01 1.20
02 1.61
03 1.93
04 1.95
J. López-Gracia, R. Mestre-Barberá / Journal of Business Research 64 (2011) 649–655 Table 2 Summary statistics. Number of observations: 88,224 (11,028 firms) Variables
Mean
Standard deviation
Median
Minimum
Maximum
longterm% term interestvolatility taxrate1 taxrate2a firmvolatilityb debt size growthb assetmaturity
0.17 1.47 0.34 0.15 0.3 0.04 0.6 7.77 1.1 4.42
0.19 0.37 0.13 0.27 0.11 0.03 0.21 1.1 0.25 6.3
0.09 1.50 0.35 0.14 0.32 0.03 0.63 7.68 1.07 2.52
0 0.91 0.11 −12.18 −1.04 0 0.002 3.78 0.02 0.01
0.99 1.95 0.54 12.05 1.65 0.17 0.99 12.6 7.15 91.85
Notes: Period of analysis 1997–2004. a taxrate2 statistics use only 68,808 observations due to variable construction requirements. b firmvolatility and growth statistics use only 77,196 observations due to variable construction requirements.
According to these figures, the cut-off point that makes it possible to identify the years in which the interest rate gap is wide, is 1.61. Therefore, in 1999, 2003 and 2004, long term interest rates are clearly higher than short term rates. This difference is not as noticeable in the rest of the years. The model defines a dummy variable called dum_term: dumXterm = 1 if term N 1:61 and 0 otherwise: Likewise, the model substitutes for the variable interestvolatility another dummy called dum_interestvolatility: dum– interestvolatility = 1 if interestvolatility N 0:34 ðmeanÞ and 0 otherwise
where the mean 0.34 is calculated on the 88,224 observations from the whole sample. Estimation (2) of Table 3 presents the results for the new specifications. The prediction that the interest rate spread and interest rate volatility would positively affect debt maturity structure holds entirely using the new variables dum_term and dum_interestvolatility. In addition, the results of estimation (2) confirm all forecasts, with the exception of that for firm value volatility.
653
The research now explores the relationship between the effective tax rate and debt maturity by dividing firms into two groups, according to whether they have strong alternative tax shields. Following Chen (2004) and Titman and Wessels (1988), the model uses depreciation as a proxy for alternative tax shields. A dummy variable called dum_shield separates firms, as follows: dum_shield= 1 if shield ≤ 0.034 (median) and 0 otherwise, where shield is the ratio of depreciation to total fixed assets and is calculated on the whole sample. Next, the variable taxrate1 is replaced in the model with an interaction term taxrate1 and the dum_shield. This term, taxrate1×dum_shield, equals the value of the effective tax rate if they have low alternative tax shields, and 0 if firms have high alternative tax deductions. The coefficient for this interaction variable represents the difference in the effect tax rate's impact on debt maturity for firms with weak alternative tax shields versus firms with strong alternative tax shields. The expected relationship between taxrate1×dum_shield and the dependent variable is therefore negative. Column (3) of Table 3 displays the results of this new specification, which confirm expectations. The variable taxrate1 is separately included to obtain a complete effect, although it is insignificant. Thus, SME debt maturity is negatively related to the effective tax rate when firms lack strong alternative tax shields. The predictions confirm results obtained in the previous specifications. As a robustness check, the model is tested using the second definition of the effective tax rate (taxrate2): the ratio of income tax to profit before taxes. Column (4) of Table 3 shows these results, fully consistent with those obtained earlier. 5.2. Dynamic model estimation Table 4 presents the estimation results of difference-GMM and system-GMM regressions of dynamic model (5). Both provide results for the two different specifications used for effective tax rate (taxrate1 and taxrate2). The estimation technique applies a robust second-stage version of the GMM estimator, although results from the robust first-stage procedure are similar (not shown). The model treats the explanatory variables as endogenous, with the exception of the dummies. The Wald test of joint significance of regressors (Wald-1) verifies the overall validity of the explanatory variables in the model (all four specifications have a p-value equal to zero). The result of the Wald test of joint significance of industry variables (Wald-2) for system-GMM
Table 3 Fixed effect regressions of debt maturity structure during 1997–2004. Static model (1): Yit = β0 + ∑ βj Xjit + vi + uit Explanatory variables
Expected relationship
term dum_term interestvolatility dum_interestvolatility taxrate1 taxrate2 × dum_shield taxrate2 taxrate2×dum_shield firmvolatility debt size growth assetmaturity constant
+ + + + − − − − − +/− +/− − +
Within R2 Wald test (F statistic) Hausman's test χ2 Firms Observations Note: P-values in parentheses. ⁎ Significant at the 1% level. ⁎⁎ Significant at the 5% level.
(1)
(2) −0.0016 (0.291) −0.0045 (0.288) −0.0087 (0.000)⁎
0.0089 0.3325 0.0502 −0.0267 0.0057 −0.4134
(0.742) (0.000)⁎ (0.000)⁎ (0.000)⁎ (0.000)⁎ (0.000)⁎
0.1426 562.14 (0.000) 1794.40 (0.000) 11,028 77,196
(3) 0.0047 (0.000)⁎ 0.0061 (0.000)⁎ −0.0089 (0.000)⁎
0.0066 0.3268 0.0531 −0.0268 0.0057 −0.4425
(0.806) (0.000)⁎ (0.000)⁎ (0.000)⁎ (0.000)⁎ (0.000)⁎
0.1429 567.88 (0.000) 109.85 (0.000) 11,028 77,196
(4) 0.0047 (0.000)⁎
0.0045 (0.000)⁎
0.0060 (0.000)⁎ −0.0029 (0.275) −0.0096 (0.012)⁎⁎
0.0049 (0.002)⁎
0.0057 0.3274 0.0532 −0.0269 0.0057 −0.4437
(0.831) (0.000)⁎ (0.000)⁎ (0.000)⁎ (0.000)⁎ (0.000)⁎
0.1431 504.21 (0.000) 211.54 (0.000) 11,028 77,196
−0.0057 −0.0700 −0.0272 0.3235 0.0515 −0.0254 0.0056 −0.4261
(0.324) (0.000)⁎ (0.344) (0.000)⁎ (0.000)⁎ (0.000)⁎ (0.000)⁎ (0.000)⁎
0.1434 389.93 (0.000) 658.17 (0.000) 8,601 60,207
654
J. López-Gracia, R. Mestre-Barberá / Journal of Business Research 64 (2011) 649–655
Table 4 GMM regressions of debt maturity structure during 1997–2004. Dynamic model (5): Yit = μβ0+ (1−μ).Yit−1 + μ∑ βj Xjit + vi + vs + uit Difference-GMM Explanatory variables
Expected relationship
Yit−1 dum_term dum_interestvolatility taxrate1 taxrate1×dum_shield taxrate2 taxrate2×dum_shield firmvolatility debt size growth assetmaturity constant
+ + + − − − − − +/− +/− − +
Wald-1 Wald-2 Autocorrelation-1 Autocorrelation-2 Sargan test Instruments Firms Observations
(1)
System-GMM (2)
0.5946 0.0059 0.0045 0.0743 −0.0932
−0.1289 0.2209 0.0399 −0.0470 0.0028 −0.3457
(0.000)⁎ (0.001)⁎ (0.128) (0.057)⁎⁎⁎ (0.018)⁎⁎
(0.245) (0.000)⁎ (0.004)⁎ (0.006)⁎ (0.004)⁎ (0.019)⁎⁎
1921.01 (0.000) −30.332 (0.000) 0.52711 (0.5981) 45.8238 (0.6027) 61 11,028 55,140
(3) 0.5934 (0.000)⁎ 0.0056 (0.002)⁎ 0.0055 (0.059)⁎⁎⁎
−0.0324 −0.0480 −0.1491 0.1437 0.0327 −0.0209 0.0038 −0.2597
(0.281) (0.043)⁎⁎ (0.175) (0.021)⁎⁎ (0.015)⁎⁎ (0.200) (0.003)⁎ (0.053)⁎⁎
2263.98 (0.000) −28.081 (0.000) 0.82664 (0.4084) 58.85597 (0.6584) 76 8,601 43,005
(4) 0.7079 0.0053 0.0047 0.0443 −0.0554
0.0020 0.1524 0.0260 −0.0221 0.0047 −0.1499
(0.000)⁎ (0.000)⁎ (0.029)⁎⁎
0.6328 (0.000)⁎ 0.0050 (0.001)⁎ 0.0050 (0.031)⁎⁎
(0.167) (0.114)
(0.980) (0.000)⁎ (0.000)⁎ (0.124) (0.000)⁎ (0.353)
4252.55 (0.000) 30.29 (0.0002) −33.276 (0.000) −1.1643 (0.2443) 109.7465 (0.0502) 107 11,028 66,168
−0.0126 −0.0484 −0.0448 0.1280 0.0282 −0.0041 0.0038 0.1606
(0.646) (0.004)⁎ (0.588) (0.000)⁎ (0.000)⁎ (0.775) (0.000)⁎ (0.349)
4517.46 (0.000) 18.38 (0.0185) −28.582 (0.000) 0.90313 (0.3665) 102.2163 (0.4754) 122 8,601 51,606
Notes: P-values in parentheses. Estimated industry dummies are not reported. Variables Yit−1, firmvolatility, debt, size, growth and assetmaturity have been instrumented with one lag, whereas taxrate1 and the interaction term taxrate1×dum_shield includes instruments up to three lags (regressions 1 and 3). Yit−1 is instrumented up to two lags, taxrate2 and taxrate2×dum_shield up to the fifth lag and firmvolatility, debt, size, growth and assetmaturity with one lag (regressions 2 and 4). ⁎ Significant at the 1% level. ⁎⁎ Significant at the 5% level. ⁎⁎⁎ Significant at the 10% level.
estimates also confirms the joint significance of industry dummies. The second order serial correlation test (Autocorrelation-2) indicates no second order serial correlation in any estimates. The Sargan test of over identifying restrictions (Sargan) generally confirms the validity of the instruments chosen. Only estimation (3) does not confirm the Sargan test, probably due to our use of more observations and instruments (see Bowsher, 2002). As expected, all the estimates in Table 4 display markedly significant coefficients (1− μ), for the lagged dependent variable, which range from 0.59 for both difference-GMM estimations to 0.70 and 0.63, respectively, for system-GMM estimations (3) and (4). System-GMM estimations on lagged dependent variable are higher than their counterpart differenceGMM estimations as the latter may be subject to downward sample bias if (i) the (theoretical) autoregressive parameter is moderately high and the number of time-series observations is moderately small (Blundell and Bond, 1998). These findings on (1− μ) coefficients suggest the existence of an optimum SME debt maturity structure and, at the same time, the presence of high adjustment costs. Using specification (4) as a basis, a coefficient of 0.63 is equivalent to an adjustment speed towards target debt structure (μ) of 0.37. Thus, SMEs on average take approximately 20 months to cover only half the gap between their optimum debt maturity structure and their actual beginning-of-year structure. This time also implies that SMEs consider the cost of remaining unbalanced to be lower than the adjustment costs of reaching their optimum structure. Ozkan (2000) had reported an adjustment rate of 50% for large, publicly tradable firms. These results confirm the initial suspicion that SMEs find it more difficult to adjust to their optimum debt maturity structure. As Table 4 shows, the explanatory variables in the dynamic model (5) have the expected coefficient signs, with the exception of firm value volatility. The GMM estimations, therefore, reinforce the fixed effect estimations. Essentially the same result is obtained from the systemGMM estimations (specifications (3) and (4)). The term structure of interest rates (dum_term) affects debt maturity positively, particularly in the years when the interest rate gap is larger than 1.61. Likewise,
strong evidence supports the expected positive influence of short term interest rate volatility (dum_interestvolatility) on debt maturity, especially in years when interest rates are highly volatile. Debt maturity is negatively related to the effective tax rate (taxrate1×dum_shield and taxrate2×dum_shield), especially when firms enjoy few tax exemptions other than those from debt. To verify possible general effects of the time factor, the research also tests the dynamic model (5) by removing the variables term and interestvolatility (both constant for the firms and variant over time) and by introducing time dummies. The new results obtained (not shown) coincide with those displayed in Table 4 and the time dummies are equally significant. As manufacturing firms represent a large proportion of the sample, the study also carries out additional regressions on this sector separately. This new regression (not shown) resembles the previous results. Finally, the control variables, debt ratio, firm size, growth opportunities and maturity of assets generally confirm the previous fixed effect results from the static model where these variables were highly significant in all specifications. The positive effect of the debt ratio on debt maturity should be interpreted as an attempt by SMEs to reinforce their solvency. The findings confirm that the smallest firms, with weak financial disclosure and relatively high information costs, see their chances of obtaining long term financing restricted. Debt maturity is also negatively related to growth opportunities, suggesting that SMEs try to solve the under-investment problem by shortening debt maturity. Lastly, asset maturity is significant in all specifications, consistent with SMEs attempting to maintain compatibility between asset and liability maturities. 6. Conclusions This paper provides empirical evidence on the determinants of Spanish SME debt maturity structure and the speed at which these firms adjust to their optimum structure. Results show that SMEs sought out optimum debt maturity structure throughout 1997–2004.
J. López-Gracia, R. Mestre-Barberá / Journal of Business Research 64 (2011) 649–655
A dynamic adjustment model fits the data well and confirms its suitability for SMEs. The two system-GMM estimates indicate a speedof-adjustment score of 30% and 37%, respectively, the equivalent of a firm taking 20 months to cover half the distance between its beginning-of-year structure and the optimum structure. This result is lower than the case of large companies with publicly tradable equity, where the few extant studies report figures of about 50%. Thus, SMEs may prefer to remain unbalanced, rather than reach their targets at top speed due to the high adjustment costs they face. The results also indicate that such high adjustment costs could be a consequence of the financial restrictions afflicting SMEs. As regards the factors that determine optimum debt maturity structure from the perspective of the tax approach, the evidence from the system-GMM estimation is robust and indicates a strong positive relationship with the term structure of interest rates and interest rate volatility, as expected. Debt maturity is strongly related to the effective tax rate, particularly when firms have few alternative tax exemptions. The debt ratio, firm size, growth opportunities and asset maturity significantly affect debt maturity. These results are in line with both Scherr and Hulburt (2001) and García-Teruel and Martínez-Solano (2007) on SMEs and offer a new perspective based on tax rationale and the advantages of a dynamic model. Nevertheless, some care must be taken when comparing the results of this research, as they have two possible limitations stemming from (i) the fact that hypotheses are formulated under a general tax approach commonly used on large (traded) firms and (ii) possible survivorship bias caused by the use of a complete data panel. Acknowledgments The authors are grateful to Pedro Martinez-Solano and Alejandro Casino-Martínez for their valuable comments and suggestions made on this manuscript. Helpful observations from Julio Pindado, Félix López, Juan Sanchis, Vicente Pallardó, Cristina Aybar, Francisco Sogorb and two anonymous referees are also appreciated. David Smith (the Associate Editor of JBR) also provided helpful comments and encouragement. Funding from the Regional Government of Generalitat Valenciana is gratefully acknowledged (Project GVPRE/2008/349). References Antoniou A, Guney Y, Paudyal K. The determinants of debt maturity structure: evidence from France, Germany and the UK. Eur Financ Manage 2006;12(2):161–94. Arellano M, Bond S. Some tests of specification for panel data: Monte Carlo evidence and an application to employment equations. Rev Econ Stud 1991;58:277–97. Arellano M, Bover O. Another look at the instrumental-variable estimation of errorcomponents models. J Econometrics 1995;68:29–52. Barclay MJ, Smith Jr CW. The maturity structure of corporate debt. J Finance 1995;L(2): 609–31 June. Blundell RW, Bond SR. Initial conditions and moment restrictions in dynamic panel data models. J. Econometrics 1998;87:115–43.
655
Bond S. Dynamic panel data models: a guide to micro data methods and practice. Portuguese Econ J 2002;1(2):141–62. Bowsher C. On testing overidentifying restrictions in dynamic panel data models. Econo Lett 2002;77(2):211–20. Brick IE, Ravid SA. On the relevance of debt maturity structure. J Finance 1985;XL(5): 1423–37 December. Brick IE, Ravid SA. Interest rate uncertainty and the optimum debt maturity structure. J Financ Quant Anal 1991;26(1):63–81 March. Cai K, Fairchild R, Guney Y. Debt maturity structure of Chinese companies. Pac Basin Finance J 2008;16:268–97. Céspedes J, González M, Molina CA. Ownership and capital structure in Latin America. J Bus Res 2010;63:248–54. Chen JJ. Determinants of capital structure of Chinese-listed companies. J Bus Res 2004;57:1341–51. Daniševská P. Is debt maturity determined by asymmetric information about shortterm or long-term earnings? Social Science Research Network (SSRN). http://ssrn. com/abstract=3027372002 June. Easterwood JC, Kadapakkam PR. Agency conflicts, issue costs, and debt maturity. Q J Bus Econ 1994;33(3):69–80 Summer. Emery DR, Lewellen WG, Mauer DC. Tax-timing options, leverage, and the choice of corporate form. J Financ Res 1988;11:99-110. Flannery MJ, Rangan KP. Partial adjustment toward target capital structures. J Financ Econ 2006;79:469–506. García-Teruel PJ, Martínez-Solano P. Short-term debt in Spanish SMEs. Int Small Bus J 2007;25(6):579–602. Graham JR. Proxies for the corporate marginal tax rate. J Financ Econ 1996;42:187–221. Harwood E, Manzon Jr GB. Tax clienteles and debt maturity. J Am Taxation Assoc 2000;22(2):22–39. Holtz-Eakin D, Newey W, Rosen HS. Estimating vector autoregressions with panel data. Econometrica 1988;56:1371–96. Jun SG, Jen FC. Trade-off model of debt maturity structure. Rev Quant Finance Acc 2003;20:5-34. Kane A, Marcus AJ, McDonald RL. Debt policy and the rate of return premium to leverage. J Financ Quant Anal 1985;20(4):479–99 December. Körner P. The determinants of corporate debt maturity structure: evidence from Czech firms. Czech J Econ Finance 2007;57:142–58. López-Gracia J, Sogorb-Mira F. Testing trade-off and pecking order theories financing SMEs. Small Bus Econ 2008;31(2):117–36 August. Morris JR. On corporate debt maturity strategies. J Finance 1976;XXXI(1):29–37 March. Myers SC. Determinants of corporate borrowing. J Financ Econ 1977;5:138–66 [November]. Newberry KJ, Novack GF. The effect of taxes on corporate debt maturity decisions: an analysis of public and private bond offerings. J Am Taxation Assoc 1999;21(2):1-16. Ozkan A. An empirical analysis of corporate debt maturity structure. Eur Financ Manage 2000;6(2):197–212. Ravid SA. Debt maturity-a survey. Financ Mark Inst Instrum 1996;5(3):1-69 March. Roodman D. How to do xtabond2: an introduction to difference and system GMM in Stata. Center for Global Development Working Paper number 103. Social Science Research Network (SSRN), December 2006 (revised July 2008). http://ssrn.com/ abstract = 982943. Scherr FC, Hulburt HM. The debt maturity structure of small firms. Financ Manage 2001: 85-111 Spring. Scholes MS, Wolfson MA. Taxes and business strategy: a planning approach. Englewood Cliffs, NJ: Prentice Hall; 1992. Serrasqueiro Z, Nunes PM. Non-linear relationships between growth opportunities and debt: evidence from quoted Portuguese companies. J Bus Res 2009, doi:10.1016/j. jbusres.2009.11.003. Stohs MH, Mauer DC. The determinants of corporate debt maturity structure. J Bus 1996;69(3):279–312. Titman S, Wessels R. The determinants of capital structure choice. J Finance 1988;XLIII(1): 1-19 March.