TAXES AND CAPITAL STRUCTURE: A STUDY OF EUROPEAN COMPANIES
We analyse the impact of tax policy on firms' leverage ratios in a balanced panel of 129 medium-sized listed European companies from 1993 to 2005. A general model of company leverage is applied within which King's tax ratios are used to capture tax policy changes, controlling for non-tax influences. Leverage measures studied include total, long-term and short-term debt. A generalized method of moments estimator is used to control for endogeneity. The results suggest that tax policy has a significant but small impact on firms' debt ratios and that non-debt tax shields are a substitute for debt in company activities.
1 Introduction: Tax Policy and Capital Structure
In theory, tax policy can have an important impact on corporate financial decisions. The literature originates with Modigliani and Miller (1963) who point out that if companies can deduct debt interest before arriving at taxable profits, a wedge is driven between the after-tax costs of equity and of debt, and this creates an exception to their famous irrelevancy theorem (Modigliani and Miller, 1958). Modigliani and Miller's (1963) paper implies that the value of a firm increases linearly with leverage, suggesting that value maximization requires 100 per cent debt financing. Evidently, firms do not behave in this way: increased debt increases the risk of bankruptcy, and this leads to the tax-cum-bankruptcy model in which optimal company leverage is determined by a trade-off between the tax advantages of debt and the prospective costs of bankruptcy or financial distress (Kraus and Litzenberger, 1973).
Corporate income is also taxed in the hands of investors through taxes on interest, dividends and capital gains. Since investors care about their after-tax returns, yields on corporate debt and equity will adjust as tax rates change. Furthermore, there are inherent nonlinearities in tax systems: in the rate structure, and in the treatment of different investors (such as tax-exemptinstitutions). Miller (1977) argues that such nonlinearities alone can explain the existence of an interior optimum firm-specific capital structure. King (1977) also emphasizes the interaction between the personal and corporate tax systems and proposes a hierarchy of funding among debt, equity and retentions based solely on tax considerations.1 However, as Auerbach (2002) emphasizes, in a model in which debt, equity and retentions are all determined simultaneously, tax effects on leverage depend also on dividend payout policy and constraints on tax arbitrage. Firms can also benefit from non-debt tax shields such as depreciation. DeAngelo and Masulis (1980) point out that non-debt tax shields are tax substitutes for debt interest deductions and thus may moderate the tax benefits of debt. They show that this reinforces Miller's (1977) argument for a tax-determined optimal capital structure. More recent theoretical research suggests a more integrated view of tax and capital structure, based on the tax-cum-bankruptcy model, but with a clearer account of the costs of financial distress (Strebulaev, 2007).
There is a vast empirical literature on company financing decisions, in which tax is just one element in a general model of company financing. Recent reviews of this literature include Prasad et al. (2005) and Frank and Goyal (2008). Studies that focus specifically on tax and company financing typically find a positive relationship between a firm's leverage and its marginal tax rate (MacKie-Mason, 1990; Givoly et al., 1992; Graham, 1996a; Sarkar and Zapatero, 2003). Non-debt tax shields are mostly found to have the predicted negative impact on leverage (MacKie-Mason, 1990; Givoly et al., 1992; Shenoy and Koch, 1996; Graham and Tucker, 2006), but some studies find a positive or insignificant impact. Moore (1986) explains this by a ‘collateral effect’: non-debt tax shields are mostly tangible assets that can be used as collateral to secure debt, obtain more favourable terms for borrowing and therefore to increase leverage. There is also a ‘tax-exhaustion effect’ (MacKie-Mason, 1990): non-debt tax shields should only affect leverage in so far as they affect firms' marginal tax rates on interest deductions. Therefore, increases in non-debt tax shields for firms far away from tax exhaustion are unlikely to affect debt policy.2 Studies of the impact of tax on capital structure have been reviewed by Graham (2008) who concludes that, in general, taxes do affect corporate financial decisions, but the magnitude of the effect is ‘not large’, and certainly less than might be expected on the basis of theory.
There are three main methods of analysing the impact of tax on capital structure: cross-section or panel regression (Rajan and Zingales, 1995), event studies (Mikkelson and Partch, 1986) and simulation (Graham, 2000). Each method has its problems. Probably the main difficulty with the regression approach is the lack of suitable summary measures of a complex corporate tax schedule for use in the model. Statutory tax rates generally change infrequently and are common across firms, thus limiting the extent to which it is possible to calculate firm-specific responses to tax changes. Effective average tax rates generally bear little relation to the tax rates that individual firms actually face (Graham, 2008).
In this paper we adopt a new approach to the measurement of tax policy using a regression method with panel data. We study the impact of tax policy on a balanced panel of 129 non-financial firms from 11 European countries from 1993 through 2005. This epoch is of great interest because all the European Union (EU) countries experienced tax regime changes, and in countries such as Ireland there was a concerted policy of corporate tax reduction. By pooling firms with tax homes in different countries we are able to exploit both time-series and cross-sectional variations in statutory tax rates to a much greater extent than is normally possible in such studies. Since most previous studies focus on a single country, it is also of interest to examine how far it is possible to trace the cross-country impact of tax changes. There are obvious difficulties in a cross-country analysis including comparability of accounting information and identification of the tax rates that are relevant to each company, given the extent of cross-border activity and the use of transfer pricing and capital transactions with subsidiaries (tunnelling) to minimize international tax obligations. These are general issues in any tax research, but are perhaps more acute in a cross-country study.
We deal with the issue of comparability by using company data from Worldscope (available from http://www.thomsonfinancial.com) which is generally accepted to be the most comprehensive available international company database (Ulbricht and Weiner, 2005). We identify tax rates using statutory tax rates for each jurisdiction, on the grounds that statutory rates correspond most closely to the instruments of tax policy. In comprehensive simulations of US corporate tax schedules, Graham (1996b) and Plesko (2003) have shown that the top statutory tax rate is the best proxy for the true marginal tax rate for most firms. This suggests that statutory tax rates can provide a reasonable measure of corporate tax incentives. The final issue is that of cross-border tax-minimizing transactions. The scope for these is the greatest within large multinational companies, and our sample therefore consists exclusively of medium-sized quoted companies in the EU whose scope for cross-border tax-minimizing transactions, particularly those affecting capital structure, is either small or negligible. A further advantage of studying middle-sized companies is that Gordon and Lee (2001) provide evidence that the impact of tax on the financing activities of smaller US companies is greater than that for larger companies. It will therefore be interesting to compare their results with ours for the EU. Hall et al. (2004) studied a sample of small and medium-sized EU companies and found substantial unexplained cross-country variations in their capital structures. Cross-country differences in taxation are one obvious possible source of these unexplained variations. Finally, in contrast to many previous studies (such as Hall et al., 2004), we do not limit our model to a few variables but instead we embed the analysis of tax policy within a more general model of company leverage, controlling for a range of non-tax factors which are generally recognized to help determine leverage.
Previous research suggests that the amount of leverage and its determinants vary significantly, depending on the precise definition of leverage adopted (Rajan and Zingales, 1995; Bevan and Danbolt, 2002). Furthermore, it can be argued that decomposing debt into its different components enables a finer test of the trade-offs inherent in the tax-cum-bankruptcy model (Song, 2005). Therefore, we consider and compare different definitions of leverage, distinguishing between a broad and a narrower definition, and between total debt, long-term debt and short-term debt. This taxonomy enables us to check how far the model is consistent with different theories of company financing, as well as to investigate the robustness of the results, a particularly important factor in a setting in which it is not possible to control fully for cross-country differences in the reporting of financial and other variables in the model. The model is estimated using the panel generalized method of moments (GMM) procedure proposed by Arellano and Bover (1995) to control for endogeneity of some of the explanatory variables.
The rest of the paper is organized as follows. In Section 2, we discuss the tax and company accounts data used in the analysis; in Section 3 we set out the model and describe the variables used; Section 4 discusses the estimation and testing procedures; the main results are discussed in Section 5; and Section 6 summarizes and concludes. An appendix describes the statutory tax variables used in the analysis.
Graham and Tucker (2006) find that it is large firms with foreign subsidiaries that are most likely to engage in dedicated tax-sheltering transactions such as transfer pricing and tunnelling. To limit this problem, we study middle-sized quoted EU companies. We selected all firms from Worldscope within the tier ranked 501–1500 in the EU according to their market value. Such firms are likely to have fewer internationally distributed plants and therefore less opportunity for transfer pricing and tunnelling. We sought a balanced panel using a relatively long time series of companies from a reasonable cross-section of EU countries. Beginning with 1000 companies, we deleted financial companies, those with missing data and those whose sales or earnings before interest and taxes (EBIT) were non-positive in any year. This left 129 non-financial companies that reported balance sheets, income and cash flow statements every year from 1993 through 2005. The sample firms are from 11 EU countries and a wide range of industries (Tables 1 and 2). Data for each calendar year in the sample include companies reporting at any time in that year. All the sample countries have a tax year that coincides with the calendar year except for the UK (April to March), and most sample companies have a December report date; a few companies have a June, July, September or October report date; and some UK firms report in March. To identify more precisely the relationship between company financing and its determinants, all the macroeconomic data used in the analysis are matched up to the report date of each company.3 The main source of corporate and personal tax data is the European Tax Handbook (Kesti, 1993–1995, 2000–2005; Kesti and Andersen, 1996–1998; Kesti et al., 1999).4 The calculation of the tax variables is explained in Section 3.
Country Composition of Sample Companies
Industry Composition of Sample Companies
|01||Agricultural Production—Crops||1||44||Water Transportation||3|
|13||Oil and Gas Extraction||1||45||Transportation by Air||1|
|15||General Building Contractors||7||47||Transportation Services||1|
|16||Heavy Construction, Excluding Building||3||49||Electric, Gas and Sanitary Services||8|
|17||Special Trade Contractors||1||50||Wholesale Trade—Durable Goods||7|
|20||Food and Kindred Products||10||51||Wholesale Trade—Non-durable Goods||5|
|23||Apparel and Other Textile Products||3||52||Building Materials and Garden Supplies||1|
|25||Furniture and Fixtures||1||53||General Merchandise Stores||2|
|26||Paper and Allied Products||2||54||Food Stores||1|
|27||Printing and Publishing||4||55||Automotive Dealers and Service Stations||1|
|28||Chemicals and Allied Products||4||57||Furniture and Home Furnishings Stores||1|
|30||Rubber and Miscellaneous Plastics Products||4||58||Eating and Drinking Places||3|
|32||Stone, Clay and Glass Products||4||59||Miscellaneous Retail||4|
|33||Primary Metal Industries||1||70||Hotels and Other Lodging Places||1|
|34||Fabricated Metal Products||3||72||Personal Services||1|
|35||Industrial Machinery and Equipment||7||73||Business Services||5|
|36||Electronic and Other Electric Equipment||4||75||Auto Repair, Services and Parking||1|
|37||Transportation Equipment||7||79||Amusement and Recreation Services||2|
|38||Instruments and Related Products||6||80||Health Services||1|
|39||Miscellaneous Manufacturing Industries||1||87||Engineering and Management Services||6|
3 The Model and Variables
3.1 The Model
The model to be estimated is
where ynt is the leverage ratio; is a vector of determinants that vary across firms and over time; and unt is a composite residual which in its most general form is
where µn is the unobservable firm effect, ηt is the time-specific component and νnt is the remaining disturbance. We emphasize at the outset that this is a static model of leverage. Recent research has focused on the dynamic adjustment process (Ozkan, 2001; Bartholdy and Mateus, 2003), but we do not include dynamics for four reasons. First there is little agreement on a general theory of capital structure, and certainly no accepted theory of the dynamic adjustment process; a dynamic model could be rejected or incorrectly accepted because of a combination of mis-specified static and dynamic models. Second, there is a fundamental difference between the static trade-off and dynamic target adjustment models of capital structure. The latter does not consist of the former with a first-order lag in leverage appended, but they are too often treated this way (Frank and Goyal, 2008). In this paper, the focus is on a static model especially as the time dimension in the data is short (though longer than is common in comparable studies). Third, the standard GMM estimator for dynamic models (Arellano and Bond, 1991) involves a loss of information in that time-invariant effects cannot be identified. This is a serious problem in corporate finance studies, for company-specific effects are known to be important in financing decisions. Finally, we do not intend to test a ‘new’ theory of finance but to identify the importance of tax relative to other determinants of leverage. Accordingly a static approach appears more appropriate.
3.2 Dependent Variable: Leverage Ratios
We use six different measures of company leverage based on book values (Table 3). We distinguish total debt, long-term debt and short-term debt. Long-term debt represents all interest-bearing financial obligations, excluding amounts due within one year. Short-term debt is debt payable within one year including the current portion of long-term debt and sinking fund requirements of preferred stock or debentures. It is likely that the leverage-related costs of short-term debt and long-term debt are different, so firms may have different policies with regard to short-term and long-term borrowing. Tax and non-debt tax shields may also be expected to have different effects on long- and short-term borrowing. We scale the different debt measures by total assets, and also by the more narrowly defined debt + equity + reserves.
Measures of Leverage
|LR2||Debt/(debt + total shareholder's equity + non-equity reserves)|
|LTD1||Long-term debt/total assets|
|LTD2||Long-term debt/(debt + total shareholder's equity + non-equity reserves)|
|STD1||Short-term debt/total assets|
|STD2||Short-term debt/(debt + total shareholder's equity + non-equity reserves)|
3.3 Modelling Tax Systems
Three main methods have been used to represent tax policy in models such as (1). King (1977) shows that a firm's choice between debt, equity and retained earnings depends on three key ratios based on the after-tax income retention rates for these three sources of funds. Firms will choose the source of funds that has the highest retention rate. Statutory tax rates may be used to calculate King's tax ratios and, in principle, these can then be used as explanatory variables in the model. Examples of this approach include Chowdhury and Miles (1989) and Green and Murinde (2008). King's ratios are bilateral (i.e. relative) tax prices that are strictly valid only if one form of financing is given. As King himself and inter aliaAuerbach (2002) emphasize, in a setting in which debt, equity and retentions are determined simultaneously, tax effects on leverage depend also on dividend payout policy and constraints on tax arbitrage. A further concern with using King's ratios is that few statutory rate structures can be summed up in a single marginal tax rate. Moreover, statutory tax rates are true marginal rates only under restrictive assumptions, particularly because of the existence of non-debt tax shields. A firm that can newly use a deduction or that loses a deduction may face a marginal tax rate anywhere between zero and 100 per cent, and there can be substantial inter-firm variation in true marginal tax rates (Graham, 1996a). Finally, in a short panel, there is often little or no time variation in tax rates, and obviously none at all in a cross-section from a single country.
Therefore, a more common practice is to use the average tax rate, calculated as the ratio of taxes paid to pre-tax earnings, as an explanatory variable (e.g. Booth et al., 2001).5 The problem with this approach is that the average tax rate is the endogenous outcome of the present and previous year's corporate business decisions on investment, financing, write-offs, loss carry-backs, loss carry-forwards etc. (Graham et al., 1998). Since all these are determined partly by the firm itself, the average tax rate cannot be interpreted as a measure of tax policy. A similar problem is presented by non-debt tax shields, which are typically estimated from company accounts using relevant variables such as depreciation (Titman and Wessels, 1988). Reported depreciation may bear little relation to tax depreciation, and other non-debt tax shields such as loss carry-overs can only be estimated indirectly, if at all. Just as with the average tax rate, since all these are determined partly by the firm, estimated non-debt tax shields also cannot be interpreted as measures of tax policy.
More recent research has involved a simulation approach to estimate company-specific marginal tax rates based on the tax statutes and allowing inter alia for the carry-forward and carry-back of losses. These tax rates are then used in standard leverage regressions. Examples include Graham (1996a: USA) and Alworth and Arachi (2001: Italy). However, these estimates invariably require substantial pre- and post-sample data to calculate each year's marginal tax rates, and we do not have an adequate time span of such data for medium-sized EU firms. There is also a subtle difficulty in interpreting calculated firm-specific marginal tax rates: such rates are actually ex post rates, as they are functions of realized profits. However, company decisions may have been based on marginal tax rates assuming a different level of forecast profits from that which subsequently materialized. Furthermore, these rates do not allow for tunnelling which is likely to be an endemic problem in interpreting the accounts of large international companies. Overall, therefore, it is not clear that calculated marginal tax rates are substantially more reliable than statutory rates as a measure of the marginal tax rates that firms believed they faced when making financial decisions, and therefore as a guide to the effects of policy. Furthermore, Graham (1996b) and Plesko (2003) suggest that, leaving aside loss carry-overs, the top statutory tax rate is the best proxy for the true marginal rate for most (US) firms.6
The upshot of these considerations is that we are unlikely ever to find a unique mapping from the whole tax code to the tax actually paid by every individual firm. Therefore, we would argue that statutory tax rates (measured by King's tax ratios) and variables that measure the effects of the tax code (effective tax rates and non-debt tax shields) are all relevant in understanding the effects of tax policy. Furthermore, statutory tax rates are unique in providing an unambiguous and essentially exogenous measure of tax policy. A central advantage of using cross-country data in this context is that we can exploit time-series and cross-sectional variation in tax rates. Our sample includes over 60 distinct observations of relative tax prices derived from statutory tax rates, far more than would be obtained from a single country data set.
Tax policy variables are set out in Tables 4–7: Table 4 summarizes the countries' tax systems; Table 5 gives formulas for tax retention rates suggested by the work of King (1977);7Table 6 shows (sample) empirical tax rates and bilateral ratios of retention rates (King's ratios) for 2000; formulas for King's ratios are listed in Table 7 along with the other explanatory variables of the model. More details about the derivation of King's ratios are given in the Appendix. A simplification in King's procedure is that the tax costs of equity and retained earnings are allocated to dividend tax and capital gains tax, respectively.8 A further practical issue is the choice of tax rates with which to summarize a complex tax system and thus to calculate the numerical values of King's ratios. Initially, we calculated retention rates for each country using the highest tax rates for each class of income, particularly having regard to the results of Graham (1996b) and Plesko (2003),9 and then estimated the model using these data. High-rate investors typically have great flexibility to move equity and debt cash flows between income and capital gains, but this affects the statute under which the tax is paid and not per se the applicable tax bracket. It should be borne in mind that the top tax rate does not correspond to the top income bracket in all countries. Therefore, as a check, we re-estimated our model using tax rates at the top income bracket, and although inevitably there were some quantitative changes in our findings, the qualitative results remained unchanged.10
Summary of Tax Systems 1993–2005
|France||1993–2004: Imputation system|
|2005–: Classical system|
|Germany||1993–2000: Two-rate system|
|2001–: Classical system|
|Greece||Dividend exemption system|
|Ireland||1993–April 1999: Imputation system|
|April 1999–: Classical system|
|Italy||1993–2003: Imputation system|
|2004–: Classical system|
|Portugal||1993–2001: Imputation system|
|2002–: Partial exemption system|
|UK||1993–April 1999: Imputation system (Advance Corporation Tax)|
|April 1999–: Imputation system|
Retention Rates under Different Tax Systems
|New issues (dividends)||(1 − c)(1 − mD)|
|Retained earnings (capital gains)||(1 − c)(1 − z)||(1 − c)(1 − z)||(1 − cu)(1 − z)|
|Debt||1 − mI||1 − mI||1 − mI|
Statutory Tax Rates and King's Ratios: 2000
Definitions of Explanatory Variables
|Capital structure theory|
| 1||Asset tangibility||+||TAN||Fixed assets/total assets|
| 2||Asset intangibility||−||INTAN||Intangibles/total assets|
| 3||Size||±?||SIZE||Ln(sales/consumer price index)|
| 4||Growth opportunities||±?||GROW||Price/book value|
| 5||Profitability||±?||PROF||EBIT/total assets|
| 6||Liquidity||−||LIQD||Current assets/current liabilities|
| 7||Industry classification||−||IND||= 1, for SIC codes from 3400 to 4000 (manufacturing)|
= 0, otherwise
| 8||Dividend dummy||+||NODIV||= 1, if the firm does not pay dividends|
= 0, otherwise
| 9||Volatility||−||VOLA||Standard deviation (EBIT/total assets)|
|10||Debt–equity margin||+||TXDVE||Classical:||(1 − mI)/[(1 − c)(1 − mD)] − 1|
|Imputation:||(1 − mI)(1 − s)/[(1 − c)(1 − mD)] − 1|
|Two-rate:||(1 − mI)(1 + cd − cu − s)/[(1 − cu)(1 − mD)] − 1|
|11||Debt–retentions margin||±?||TXDVR||Classical:||(1 − mI)/[(1 − c)(1 − z)] − 1|
|Imputation:||(1 − mI)/[(1 − c)(1 − z)] − 1|
|Two-rate:||(1 − mI)/[(1 − cu)(1 − z)] − 1|
|12||Average tax rate||±||TAXR||= income taxest/pre-tax incomet−1 if positive|
= 0 otherwise
|13||Non-debt tax shields||−||NDTS||Depreciation/total assets|
|14||Term||−||TERM||Long-term interest rate–short-term interest rate|
We have expressed King's conditions as ratios of retention rates. It is equally possible to express them as differences between retention rates. In either case, only two of the three conditions are independent. However, since the differences are linear, only two can enter a regression whereas, in theory, all three ratios could appear in the model, assuming that the impact of tax rates on leverage takes this specific nonlinear form. Chowdhury and Miles (1989) use all three ratios, presumably on this basis. We are agnostic about the exact form of the conditions but, since our focus is on leverage, and to limit possible over-parameterization of the model, we use only the debt–equity and debt–retention ratios as regressors.
Notwithstanding its limitations, we also use the average tax rate (the ratio of corporate tax paid to pre-tax corporate income) as a measure of the impact on firms of the tax system as a whole. If this is a forward-looking rate used in financial decisions, we would expect a positive relationship with leverage, provided firms have tax-shelterable income. However, a high effective tax rate could reflect high profitability, or past low leverage for reasons unrelated to tax. Thus the relationship to leverage could be negative, as indeed Booth et al. (2001) found.
According to DeAngelo and Masulis (1980) non-debt tax shields and debt interest are tax-shelter substitutes but, unlike debt, non-debt tax shields will not increase the probability of bankruptcy. Thus, non-debt tax shields will always be preferred to debt, and we would therefore expect non-debt tax shields to have a negative impact on leverage. However, there may also be a difference between their impact on short-term debt and long-term debt. Song (2005) finds that non-debt tax shields are not related to the total debt ratio. They do have the expected negative effect on long-term debt, but a positive effect on short-term debt, perhaps because such debt finances working capital which increases with the ownership of tangible assets associated with higher depreciation. Our measure of non-debt tax shields is the standard: the ratio of book depreciation to total assets.
3.4 Control Variables
Our control variables (listed in Table 7) model the impact of non-tax factors on leverage, and are based on standard models used in earlier studies. We particularly include variables aimed at mitigating ambiguities in the tax effects, especially the collateral and tax-exhaustion effects.
3.4.1 Asset Tangibility. Tangible assets can be used as collateral to secure debt and hence tangibility helps control for the collateral effect associated with non-debt tax shields. A company's capacity to secure its debt in the event of financial distress and therefore its leverage rate should increase with the proportion of tangible assets on the balance sheet. Trade-off theory also predicts a positive relationship between leverage and the proportion of tangibles. However, Grossman and Hart (1983) argue that the agency costs of excessive management perquisites are higher for firms with low collateral. Managers of highly levered firms are less able to consume excessive perquisites because bond-holders monitor such firms more closely. Thus shareholders would like firms with fewer tangibles to be more highly geared, and such firms may voluntarily choose higher debt levels. In addition, if non-debt tax shields are measured with error, tangibles may pick up the substitution effect between debt and non-debt tax shields instead of the collateral effect and again have a negative sign. Empirically, a positive relationship has been confirmed by Rajan and Zingales (1995), Shenoy and Koch (1996) and many others; but recent empirical studies suggest that the relationship between tangibility and leverage also depends on the measures of debt applied. For example, Chittenden et al. (1996), Bevan and Danbolt (2002) and Song (2005) find that tangibility is positively correlated with long-term debt but negatively related to short-term debt.
3.4.2 Asset Intangibility. Since tangible assets exclude liquid assets they may not in fact be a good measure of a company's ability to repay debt in the event of financial distress (Titman and Wessels, 1988). Arguably, asset intangibility is less ambiguous as it measures a company's inability to secure its debt, and hence should be negatively related to leverage. Few studies use asset intangibility as a control but we believe it is a useful supplement to the tangibility measure.
3.4.3 Growth Opportunities. Firms with substantial growth opportunities will use less debt so as to mitigate agency problems (Myers, 1977). Managers of leveraged firms have an incentive to engage in asset substitution and transfer wealth from creditors to shareholders, which increases the interest rate on bonds demanded by investors and deters borrowing. Firms with high growth prospects are more likely to pass over profitable investment opportunities if they are highly levered and therefore will prefer equity over debt. Thus, a negative relation is expected between firm leverage and growth opportunities (measured by the market/book ratio). Myers (1977) also argues that agency problems can be mitigated if firms issue short-term debt rather than long-term debt. Hence, a positive relationship might be expected between growth opportunities and short-term debt. However, it can also be argued that the relationship between leverage and growth could be positive. Higher growth opportunities imply a higher demand for external funds. Pecking order theory predicts that firms prefer debt to equity, which implies a positive relation between growth opportunities and leverage.
3.4.4 Firm Size. Firm size is often thought of as an inverse proxy for the probability of bankruptcy (Rajan and Zingales, 1995). Since large firms tend to be more diversified and less likely to fail than smaller companies (Titman and Wessels, 1988), they can issue debt at relatively lower cost. Hence, large firms may be expected to have higher debt ratios. Alternatively, size can be seen as a proxy for information asymmetries between firm insiders and capital markets (Drobetz and Fix, 2003). Large firms are more closely observed by analysts and investors and have less asymmetric information problems. They should therefore be more able to issue equity, implying a negative relationship between leverage and size. However, Titman and Wessels (1988) suggest that if this relationship exists, it should be associated with small firms using more short-term debt to finance their investment, since smaller firms have higher transaction costs to issue long-term debt or equity. The two most common measures of firm size are the natural log of (real) net assets or (real) sales. Green and Murinde (2008) suggest that sales are less likely to be contaminated by idiosyncratic asset structures or reporting procedures.
3.4.5 Firm Risk. The volatility of income is clearly relevant to a firm's default risk. DeAngelo and Masulis (1980) argue that investors are less able to forecast the future earnings of firms with high earnings volatility using public information, and they therefore demand a higher risk premium to provide debt. In addition, more volatile cash flows increase the probability that earnings may fall below the debt service commitment. Firms with high volatility would avoid issuing debt when seeking funds, and so a negative relationship between earnings volatility and leverage is expected. Risk is often measured by the standard deviation of returns over some past time period, but this method throws away potentially valuable time-series data. We proceed instead by estimating the following time-series regression for each company from 1993 to 2005:
The absolute values of the residuals from these regressions are consistent estimates of the conditional standard deviations of EBIT/(total assets), and we use these to measure firm risk.
3.4.6 Profitability. Pecking order theory suggests that firms prefer to finance new investment from retained earnings and then debt, with new equity as a last resort. Since the availability of internal funding depends on profits, a negative correlation is expected between profitability and leverage. On the other hand, Ross (1977) suggests that high levels of debt financing can be used by managers to signal an optimistic future of a firm to the market, because debt financing ensures that managers are disciplined to making efficient investment decisions and are not pursuing individual objectives which may increase the probability of bankruptcy. Therefore, high profitability may be associated with a high debt level.
3.4.7 Liquidity. Pecking order theory suggests that firms require financial slack to avoid resort to external finance (Myers, 1977). The liquidity (quick assets) ratio is a measure of this slack and is expected to be negatively related to leverage.
3.4.8 Dividend Dummy. It is generally agreed that dividends give a credible signal of firm quality because they cost bad firms more than good firms in terms of reduced investment (Miller and Rock, 1985). Ross (1977) argued that debt also has a positive signalling effect. If debt and dividends are both signals of firm quality, they could be substitutes for one other (Frydenberg, 2004). Therefore it can be argued that a dividend-paying firm is expected to use less debt. However, pecking order theory would suggest the opposite relationship: given a firm's cash flow, higher dividends imply higher external (debt) financing (Frank and Goyal, 2008). Following Graham (2000), we use a 0–1 dummy to capture the effect of dividend payments on capital structure.
3.4.9 Industry Classification. Factors that influence firms' capital structure may be common within industrial groups. Firms in the same industry often face common product and factor markets and thus similar economic conditions. They are more likely to face the same investment opportunities and have similar capital requirements (Harris and Raviv, 1991). Titman and Wessels (1988) find that firms manufacturing or using equipment that requires specialized servicing and spares have less debt, presumably because they have higher liquidation costs. We follow their idea and include a dummy variable equal to unity for manufacturing firms and zero otherwise.
3.4.10 Term Spread. Finally we include a country-specific measure of the term spread as an indicator of macroeconomic conditions.11 An increase in the term spread is typically associated with expected future inflation (Mishkin, 1990). Expected inflation tends to increase the debt ratio through a tax effect: higher future inflation increases the real value of current tax deductions. However, the term spread is also positively related to the current and future cost of borrowing (Fama, 1990), and thus controls for cross-sectional variations in the expected level of interest rates. In the cross-section, therefore, a higher term spread is likely to be associated with lower leverage and particularly with a lower long-term debt ratio. The effect on short-term borrowing may be smaller or even positive.
The appropriate estimation technique for (1) depends on the structure of the error term (unt) and correlation between the error components and the explanatory variables of the model. The simplest method is to treat firm and time effects as fixed, assuming the regression intercept varies across firms and over time. This ‘within’ estimator is consistent, but it cannot be used to estimate time- or company-invariant effects since these are swept out by the within transformation applied in estimation, thus wasting useful information contained in the relations among individual means (Baltagi, 2005). The random effects estimator is more efficient and enables identification of time- and company-invariant effects. However, it is only appropriate if the unobservable firm and time effects are independent and identically distributed (IID) random variables that are uncorrelated with the explanatory variables; otherwise it is inconsistent. Therefore, we first consider if there is evidence of firm and/or time effects, and if so, whether they are correlated with the explanatory variables. We also test for heteroskedasticity. We see from Table 8 that there are significant firm effects for all leverage measures; but there is less evidence of time effects, and this suggests that we can concentrate on a model that incorporates only firm effects:
|1||Firm effects||F(128, 1535)||22.62***||19.38***||18.39***||16.14***||12.59***||12.45***|
|2||Time effects||F(12, 1650)||1.51||1.05||1.98**||1.75*||2.02**||1.98**|
However, there is obviously considerable evidence of heteroskedasticity for all the debt ratios.
The Hausman tests show that there is strong evidence of endogeneity between the regressors and the unobserved individual effects. However, endogeneity also occurs more directly in the model: since some of the wnt in (3) are scaled by total assets, they are necessarily endogenous when the left-hand side variable is a ratio of debt to total assets, as in our model. In general, one would expect many of the entries in a firm's balance sheet to be determined simultaneously at a point in time, and since balance sheet entries appear on both the left-hand side and the right-hand side of (3), at least some of the wnt are necessarily endogenous in any version of (3). In addition, the effective tax rate is evidently not independent of any measure of leverage. Endogeneity in the sense set out here is commonly ignored in empirical studies of capital structure. Rajan and Zingales (1995) arbitrarily lag all their explanatory variables one period; other recent studies ignore the issue entirely and treat all the wnt as exogenous (Booth et al., 2001). Green and Murinde (2008) show that properly allowing for this endogeneity in leverage models has an important impact on the size and significance of the estimated coefficients.
To clarify these issues, rewrite (3) in matrix notation as
Here, y′ = (y11, . . . , yN1, . . . , y1T, . . . , yNT); and W is an NT × (G + H) matrix of explanatory variables (N firms and T years) partitioned as W = (X, iT ⊗V): X are (H) firm- and time-varying explanatory variables; V are (G) time-invariant variables; and iT is a T-vector of ones. Also, X = (X1, X2) and V = (V1, V2), such that (X1, V1) are ‘exogenous’ and (X2, V2) are ‘endogenous’. X2t and V2 are correlated with vnt; but X1t are not.12 Lagged values of X2 are assumed to be uncorrelated with vnt: E(v|X1, X2,t−j, V1,µ) = 0 ( j > 0). However, E(v|X1, X2,t, V1,µ) ≠ 0 ( j > 0), and so the popular Hausman–Taylor estimator (Hausman and Taylor, 1981) is inappropriate, because of the endogeneity of the balance sheet variables among the X2,t. Therefore we adopt the method of Arellano and Bover (1995). This is preferred to the Arellano and Bond (1991) estimator as it enables us to identify the time-invariant industry effect in the model. We interpret (4) as T cross-section regressions, each corresponding to a certain year, and pre-multiply this system by the non-singular transformation:
K is any (T − 1) × T matrix of rank T − 1 such that KiT = 0. Arellano and Bover show that is invariant to choice of K, and suggest either the first difference operator or the first T − 1 rows of the within-group operator. We use the latter as it splits the model into T − 1 within-group equations and the Tth (between) equation. Given a matrix of instruments (Z) such that
the H-transformed version of (4) can be estimated by GMM and is given by
where Ω may be estimated from the residuals of a preliminary consistent estimator.
The instruments are chosen as follows.13 As the within-group operator eliminates µ from the first T − 1 equations in the model, all exogenous and predetermined variables (X1, X2,t−j, j > 0, V1) are valid instruments in these equations.14 The Tth equation must be instrumented by variables that are uncorrelated with u, and therefore uncorrelated with the firm effect (µ) and with the idiosyncratic effect (v). This limits the number of endogenous variables whose coefficients can be identified in the model. Provided we can assume that the correlation between X1 and µ is constant over time, removing the time mean from X1 (applying the within-group transformation) creates valid instruments for the Tth equation, since E(µ|X1 − ) = 0; = ∑X1t/T. Therefore in the first T − 1 equations we used as instruments one lag of the predetermined variables (TAN, INTAN, GROW, PROF, LIQD, TAXR, NDTS) and current values of the exogenous variables. Instruments in the Tth equation were the current value and three lags of the time-varying exogenous variables and the firm-specific values of the time-invariant exogenous variable. It seems reasonable to treat TERM, SIZE and VOLA as time-varying exogenous variables; IND is time-invariant and falls under the constant correlation assumption. For estimation purposes, TXDVE, TXDVR and NODIV are treated analogously to IND in that only the current values are used as instruments each year. King's tax ratios vary over time, but not sufficiently to use with 1 + 3 lags; and clearly they are exogenous. NODIV also has insufficient time variation to use with 1 + 3 lags. We argue that NODIV is exogenous because whether to pay or not to pay is a structural decision which would not typically be affected by concurrent adjustments in leverage. Finally, we calculate cluster-robust standard errors to correct for heteroskedasticity (Baltagi, 2005), and report J tests for the validity of the overidentifying restrictions implied by the instrument set (Sargan, 1958). Overall, we would argue that these estimation procedures are substantially more conservative than is common in the literature.
5 Discussion of Results
The results (Table 9) show that GMM estimation provides a substantial degree of efficiency. All but a few of the estimated coefficients are significant at statistically acceptable or high levels. The J tests confirm the validity of the overidentifying restrictions.
5.1 Tax Variables
The coefficients on King's ratios are almost all statistically significant and exhibit a consistent sign pattern as between debt–equity and debt–retentions, and among different types of debt. The debt–equity margin is signed positive throughout as expected and shows that this tax margin did affect financing decisions in the expected way. The debt–retentions margin is signed negative throughout (but smaller in value than the debt–equity margin except for LTD2) which is less expected, but could be explained by the attribution of the capital gains tax to the price of retained earnings. High statutory capital gains tax rates may encourage more efficient asset management and therefore thecross-sectional variation in effective capital gains tax rates may be less than or even inverse to the variation in the statutory rates.
The estimated coefficients on the average tax rate are positively signed for total and long-term debt ratios as might be expected, but negative in the short-term debt equation, suggesting that there are differences between the determinants of long- and short-term debt as we conjectured. The coefficients on non-debt tax shields are all negative, consistent with the DeAngelo–Masulis story that firms with many non-debt tax shields will reduce their use of debt to minimize the probability of bankruptcy. However, the coefficient is appreciably smaller in the short-term debt equation, a result not inconsistent with though weaker than that obtained by Song (2005) who found that non-debt tax shields are complements or unrelated to short-term finance.
5.2 Control Variables
Asset tangibility is uniformly signed negative and significant. As noted earlier, this is consistent with Grossman and Hart (1983), and could also be related to the effect of collateral on non-debt tax shields: TAN could be picking up some effects of non-debt tax shields on leverage and therefore be signed negative. Asset intangibility has generally unexpected (positive) signs, except in one of the short-term debt equations. A possible explanation is that intangibles consist of patents, copyrights, development costs etc., which it can be argued are generally used by firms to generate future investments. Therefore, intangibility may also serve as an indicator of growth prospects. Consistent with pecking order theory, high-growth firms would choose long-term borrowing relative to equity when retained earnings are exhausted.
The growth opportunities variable is positively signed and significant for all but the short-term debt ratios. As noted earlier this is consistent with a version of pecking order theory. Due to higher demands for funds, high-growth firms will eventually consider external financing when retained earnings are exhausted. If external finance is considered, managers prefer debt to equity, which implies a positive correlation between growth and leverage. The negative signs in the short-term debt equations are more consistent with Myers' (1977) view of pecking order theory, but not with his conjecture that short-term debt would be preferred to mitigate agency problems. The estimated coefficients on profitability are negative and significant for all debt ratios. This too is consistent with pecking order theory: firms prefer to use the surplus generated by profits to finance their investment before raising external debt and equity capital. This result also conforms to the preponderance of previous empirical evidence.
The coefficients on liquidity have significant negative signs for total debt and short-term debt, which lend support to the pecking order view that firms use liquid assets to finance their investment to avoid increasing leverage. However, one of the long-term debt equations has a positive sign, providing a little evidence that shareholders may prefer to see higher long-term debt ratios in more liquid firms so as to reduce free cash flow and bond managers to specific projects. Management could evade shareholder control by financing less profitable projects using internal funds, which are subject to a minimum of external monitoring (Jensen, 1986).
Firm size is positively related to leverage for all debt ratios. This is consistent with the view that size is a proxy for a low probability of default since larger firms are more diversified and therefore are able to borrow more debt than smaller firms without encountering financial distress. Conversely, the coefficients on volatility are uniformly negative, in line with the view that high-volatility firms have greater bankruptcy risk and therefore less debt. Other researchers have found negative but insignificant coefficients on volatility (Titman and Wessels, 1988), underlining the importance of using a consistent and efficient estimation procedure. The coefficients on the dividend dummy are all positive and significant, which is consistent with the signalling model, and with the empirical results of Graham (1999). Again, a notable feature of our results is that the coefficients are all highly significant, re-emphasizing the importance of a proper estimation procedure. The industry dummy varies in sign and significance. Negative signs in the long-term debt equations are consistent with Titman and Wessels's (1988) argument that manufacturing firms use less debt than do non-manufacturing firms; but the signs in the short-term debt equations are positive, suggesting that there may be more use of short-term finance in manufacturing.
Finally, the term spread is signed negative as expected but there are only differences of magnitude and not of sign between short- and long-term debt.
Looking at the results for the six measures of leverage, clearly there are substantial quantitative differences among the coefficients, particularly between long- and short-term debt. However, comparing our results with others who have explicitly studied models of long- and short-term leverage (Bartholdy and Mateus, 2003; Song, 2005), it would seem that our findings emphasize more the similarities rather than the differences between the determinants of the different forms of financing.
5.3 Size of Tax Effects
To evaluate the size of the tax effects we calculated the impact of a 10 per cent (1000 basis points) increase in the rate of corporate profits tax in each country, assuming no change in other rates, including imputation rates. This is done by differentiating the empirical counterpart of (1) country-by-country with respect to the corporate profits tax rate (c) and multiplying the result by 0.1. Because of the nonlinearity in the tax ratios, the impact of a tax change varies across countries, depending on the tax system and initial conditions. Our calculations were based on 2004 tax systems and rates of tax; and it must be emphasized that the sample size is very small for some countries (Table 1). The results (Table 10) must therefore be regarded as ‘ballpark’ estimates. They show that a 10 per cent tax increase would increase debt ratios by small proportions: for the narrower leverage definition (LR2) from a maximum of 3.40 per cent (UK) to a minimum of 0.15 per cent (Ireland), and mostly less than 1 per cent. It should be noted that these calculations assume that imputation systems (France, Spain and UK) did not change the imputation rate along with the basic corporate profits tax rate, and this accounts for much of the difference in magnitude between the three imputation countries and the rest.15
Effect of a 10 per cent Increase in the Corporate Profits Tax Rate
Graham (1996a) found that a 10 per cent increase in the tax rate would raise US company debt ratios on a broadly similar definition to our LR2 by 0.69 per cent per annum16 and Alworth and Arachi (2001) report an effect of between 0.8 per cent and 0.9 per cent per annum using Graham's method on Italian companies. Using a regression approach, Gordon and Lee (2001) found that the debt ratios of smaller US companies would increase by 3.6 per cent, a figure that is close to our own estimate for UK companies. Graham's method involves a regression of the change in debt on taxvariables and leverage determinants, whereas Gordon and Lee use the leverage ratio as the dependent variable, as we do. It is therefore not easy to compare all these results directly. However, it is reasonable to suppose that Graham's companies would grow in size over time and therefore, as a first approximation, the equilibrium change in the debt ratio would be about equal to the calculated rate of 0.69 per cent. If so, the results of Graham and Alworth–Arachi fall within the range we are reporting for medium-sized European companies. Moreover, their results are for classical tax systems and so are fully comparable with our estimates for European classical tax systems. Gordon and Lee's results are rather larger than for most countries in our sample except for France, Spain and UK, which contribute 58 per cent (75) of our sample companies, but as noted above these were also the imputation systems.
The literature on firms' capital structure has established a series of stylized facts relating corporate financial decisions to a variety of independent variables, in most of which the tax factor is ignored or treated as just one element in a general model of company financing. In this paper, we have extended the analysis by taking account of the interaction between corporate and personal taxes under different tax systems by using a panel of 129 non-financial firms from 11 European countries for the period 1993–2005. Different debt ratios based on individual components of debt structure have been applied to mitigate the potential bias induced by the choice of leverage measures. The model was estimated using the Arellano–Bover GMM method to control for endogeneity, and virtually all the coefficient estimates are very well determined. Many of the results for the control variables are in line with previous literature and provide rather more support for pecking order theories than for trade-off theories. We do not find strong evidence of substantial differences between the determinants of long-term and short-term financing decisions. Overall, the results give us confidence in the robustness of our model and sample selection and therefore in our results on the impact of tax on company financing.
In considering the implications of our results for the impact of tax on company financing, we would re-emphasize that sample selection was intended to give us the best chance of identifying tax effects through variations in statutory tax rates by concentrating on medium-sized companies from a substantial cross-section of European countries over a relatively long time period for this type of study. The price of this approach is that several countries are represented by a relatively few firms, and this should be borne in mind when interpreting the results. For the tax effects we find that the coefficients on King's tax ratios are mostly significant and consistently and reasonably signed in all the equations. The effective tax rate is significant and positively signed in all but the short-term debt equation, a finding in which we can have confidence given that this variable is instrumented in the estimation procedure. The sign pattern and significance of non-debt tax shields in all the equations provide important new evidence in support of the DeAngelo and Masulis (1980) argument that tax shelters are substitutes for one another and that firms will generally prefer non-debt tax shelters over debt ceteris paribus. The estimated size of the tax effect in most countries is in line with Graham's (2008) ‘not large’ conclusion; and we do not find much support for Gordon and Lee's (2001) result that tax rates play a larger role in the financing decisions of small and medium-sized companies.
An important puzzle raised by these results concerns their implications for theory. The theoretical and empirical tax literature is mostly framed within the tax benefit/bankruptcy cost trade-off model, whereas our empirical approach suggests that non-tax variables affect financing decisions in a way somewhat more consistent with pecking order theory. Clearly more research is needed to reconcile these differences.
Appendix: Calculation of King's Tax Conditions (Table 6)
King (1977) shows that a firm's choice between debt, equity and retained earnings depends on three ratios, each of which is just the ratio of any two retention rates calculated for any particular tax system. Given any one of the three methods of financing, firms will choose among the remaining two that which has a higher retention rate. We follow King and define θ as the degree of tax discrimination between retentions and distributions imposed by different tax systems. θ is the opportunity cost of retained earnings in terms of net dividends forgone (i.e. dividends after payment of all taxes, corporate and personal), and is equal to the additional disposable income which shareholders could receive if one unit of retained earnings were distributed. We also define τ as the basic rate of corporate income tax on undistributed profits.
The Total Tax Bill
The total tax liability of a company and its shareholders equals the tax on profits plus any additional tax which arises because dividends are taxed at a different rate from that levied on retained earnings. If one unit of retained earnings is distributed, shareholders receive θ; 1 − θ goes in tax. Therefore, the additional tax liability per unit of net dividends is (1 − θ)/θ. Defining also T as total tax liability, Y taxable profits, D net dividends, G gross dividends and mD the personal income tax rate on dividend income, then
so (A1) can be rewritten as
We can then use θ and τ to classify different corporate tax systems.
Retention Rate for New Issues
Classical Systems. Companies pay corporate tax at rate c on all taxable profits; shareholders are also liable to personal income tax on dividends received. Total tax liability is
Equating coefficients in (A2) and (A3) gives
Shareholders receive (1 − τ)θ per unit of corporate profits distributed; thus, by substitution, the retention rate for new issues is (1 − c)(1 − mD).
Imputation Systems. Companies pay tax on corporate profits at rate c. Distributed profits are regarded as already having borne personal tax at a rate s, the rate of imputation. Shareholders receive credit for tax paid by the company which may be used to offset their personal tax liability on dividends. Shareholders only need to account for tax on dividends at the difference between their personal income tax rate mD and the rate of imputation. Total tax liability is
Equating coefficients in (A2) and (A4) gives
Using (1 − τ)θ as before, the retention rate for new issues is
The Two-rate System. The two-rate system levies different tax rates on distributed and undistributed profits, with the former lower. Total tax liability is composed of a rate of corporate tax on undistributed profits (cu), a different rate of corporate tax on distributed profits (cd) and the shareholders' rate of income tax on dividends. If shareholders are also granted an imputation credit (s) for the distributed profits in terms of dividends, the total tax liability is
Equating coefficients in (A2) and (A5) gives
Using (1 − τ)θ as before, the retention rate for new issues is
Retention Rate for Retained Earnings
Given the capital gains tax rate (z), the retention rate for retained earnings associated with capital gains tax is (1 − c)(1 − z) for classical and imputation systems. For the two-rate system, the retention rate is (1 − cu)(1 − z).
Retention Rate for Debt
Since the interest payment associated with debt is deducted before corporate profits are taxed, the interest payment is only liable to personal income tax. Therefore, the retention rate for debt under all tax systems is 1 − mI, where mI is the personal income tax rate on interest income.
Devereux (2003) provides a recent exposition of King's model.
Trezevant (1992) controls for the collateral effect, and finds a negative relationship between changes in debt and non-debt tax shields for firms with a high probability of losing the tax deduction.
Company data in Worldscope are reported in the local currency. For the UK and Sweden, we calculated yearly mean exchange rates matched up to each company's report date to convert pounds and krona into euros. Exchange rates were obtained from Datastream (http://www.thomsonfinancial.com).
Some additional information was obtained from OECD Tax Database (available at: http://www.oecd.org/document/60/0,2340,en_2649_37427_1942460_1_1_1_37427,00.html) and KPMG's Corporate Tax Rate Survey (available at: http://www.kpmg.com).
Booth et al. (2001) used the effective tax rates in a set of individual country regressions. However, in a separate cross-country regression they used instead King's debt–equity condition.
Allowing for loss carry-overs, a simple dichotomous (Plesko) or trichotomous (Graham) variable is a good approximation to the ‘true’ simulated marginal rate in the USA.
King (1977) analysed the UK but, as he noted, it is straightforward to extend his calculations to any tax system.
This follows from the simplifying assumption that capital gains are taxed on accruals rather than realizations (Auerbach, 2002).
For debt- and equity-holders, if income is added to the tax base and taxed at graduated rates, then the top marginal personal income tax rates are applied. However, if dividend and interest payments are subject to a lower final withholding tax at flat rates, then the lower rates are used.
We thank an anonymous referee for suggesting this point.
The spread varies across countries but not among firms in the same country.
In our model, the industry dummy is the only time-invariant variable and we assume this to be exogenous.
See Green and Murinde (2008) for further discussion.
Defining Q = (INT − iTiT′/T ⊗ IN) as the within-group operator, QX1 and QX2,t-j are valid alternatives to X1, X2,t-j.
Since the imputation rate is not necessarily set to match the basic corporate profits tax rate, we take the view that it would be arbitrary to change both rates simultaneously.
Graham calculates that 22 per cent rise in the corporate tax rate would raise debt ratios by 1.52 per cent per annum; 0.69 per cent = 1.52 per cent/2.2. There are substantial nonlinearities in Graham's calculations but the 1.52 per cent estimate is based on a linear regression. Therefore, linear rescaling to compare with our results is not unreasonable.