درباره اثر متقابل اختلاف مالی و هزینه های تعدیل سرمایه ثابت: شواهدی از یک پانل شرکت های آلمانی

This paper analyzes the interaction of financial frictions and non-convex adjustment costs. Non-convex adjustment costs imply that firm-level investment is lumpy. Firms invest infrequently but each investment is large. This allows financial variables to influence investment along two margins. They can alter the size of the stock of capital a company wishes to hold in the long run or they can influence the frequency at which investment projects are undertaken. The empirical analysis of this paper reveals that finance has nearly no long-run influence on the stock of capital in a sample of German companies. By contrast, the influence of finance on investment decisions is substantial. Consequently, finance primarily affects investment frequencies and financial factors and fundamental capital productivity strongly interact in the determination of investment.

Empirical research on investment still lacks a conclusive summary of the determinants of investment both at the micro-level as well as for the aggregate. Seemingly well-established is only the fact that the neoclassical, q-theoretic model investment with quadratic adjustment costs has difficulties in explaining empirical patterns of investment (see Caballero, 2000). Still, this finding leaves open the qualitative question of which of the assumptions of the neoclassical model lead to its empirical failure. Moreover, it calls to quantify the extent to which each deviation from the neoclassical model actually helps to explain investment data. Beginning with Fazzari et al. (1988), the empirical literature has emphasized the role of financial factors in company-level investment. More recently, attention has been drawn to the role of non-convexities in investment technology.1 This paper looks at the interaction of these deviations from the neoclassical model and employs an error correction, in other words cointegration, approach to investment in the empirical analysis. In particular, we focus on the exact way in which finance affects investment, if it does so at all. Bayer (2006a) finds that financial frictions may alter target levels of capital as well as the probability of infrequent investment in a setting with fixed adjustment costs. We apply the error-correction model and two-step econometric technique outlined in that paper to a sample of German companies. In difference, the employed estimation strategy takes into account at all steps of the estimation that both the financial situation of a company as well as the intensity of its fundamental investment incentives are endogenous. Thus, we are modifying and improving upon the technique proposed in Bayer (2006a). In the first step of the estimation we determine the difference (‘gap’) between the actual stock of capital and the stock of capital a company would like to hold if there were no adjustment costs in the current period. This capital gap (technically: the error term) can be understood as the ‘mandated investment’. The second step of the analysis then focusses on the investment process itself, which is understood as the adjustment process of capital that subsequently closes the capital gap. A distinct feature of our error-correction approach is that it can be derived from a structural model of investment both under the neoclassical null hypothesis of quadratic adjustment costs as well as under the alternative hypothesis of fixed adjustment costs and capital market frictions.2 Because of this feature, it allows us to simultaneously analyze both fixed costs and capital market frictions. In particular, the approach enables us to differentiate between short- and long-term influences of financial variables, because we sequentially estimate target levels of capital and adjustment dynamics. The estimation hence reveals whether abundant financial resources alter investment rates mainly by directly shifting average investment rates, or by changing the investment process in a more complex manner in interaction with fundamental investment incentives. For adjustment costs themselves, inference mainly draws from the estimate of the adjustment paths of capital, i.e. investment. Within the error-correction approach to investment, the difference in model alternatives manifests in a different form of the error-correction process. Under quadratic adjustment costs and a unit root in productivity, the error correction is linear. If adjustment costs are non-convex instead, e.g. fixed, then higher-order terms of the cointegration error become significant and the adjustment speed varies with the size of the error term, i.e. gap between the desired and actual stock of capital. While our approach is borrowed from Bayer (2006a), we modify and extend it to cope with a broader class of endogeneity issues. The non-parametric analysis in Bayer (2006a) is replaced by a parametric non-linear error-correction model. This allows us to address endogeneity of the regressors in the investment regression. This endogeneity stems from the presence of residual autocorrelation in investment rates and if not controlled for, it will bias the analysis. For this reason we make use of the GMM estimator developed by Blundell and Bond (1998) which takes into account both the problem of an endogenously determined financial status for the estimation of the capital gap as well as the problem of lagged dependency of mandated investment in the investment regression. Parallel to Bayer (2006a), Whited (2006) shows as well – here based on the estimation of hazard rates – that finance influences the frequency of investment activity. She finds that financially constrained firms in the US exhibit investment spikes much less often than unconstrained ones. Bayer (2006a) provides similar empirical evidence from UK data on the basis of a non-linear error-correction model. The influence of finance on adjustment speed was introduced in a theoretical contribution by Holt (2003). All three papers highlight the potential importance of the interaction of financial frictions and fixed adjustment costs. Our estimation results lend support to this hypothesis too. Our analysis take into account that Cooper and Willis (2004) argue that inference drawn from an error-correction or ‘gap’ approach to investment has to be treated with care. They show that the model is sensitive to deviations from its assumptions, in particular the assumption of non-stationary productivity. Without pre-testing this assumption one may draw misleading conclusions from the estimation of a gap model, hence. The core of Cooper and Willis’ argument is that a measurement-error problem results if productivity has below unit-root serial correlation. The present paper takes this issue into account as follows. First, we estimate unobservable productivity by exploiting firm-level data on employment, wages, and sales. We follow the method developed in Cooper and Haltiwanger (2006) to minimize the measurement error in the first place. Secondly, we show that both productivity and capital at the firm level exhibit a unit root in our data and are cointegrated.3 Consequently, the cointegration error identifies the gap between the desired and the actual stock of capital, so that Cooper and Willis’ (2004) criticism does not apply. Therefore the gap approach with its (non-linear) error-correction model yields an interpretable estimate of the investment function irrespective of the actual form of adjustment costs.4 The empirical analysis of this paper uses accounting data from German companies that stems from the ‘Bonner Datenbank’. These data contain balance sheet information for a large sample of companies from 1960 to 1997. A distinct feature of the data is thus the long period of observation for most firms in the sample. Only this makes the error-correction model a feasible and sensible approach. For these data, we find that capital stocks follow capital profitability closely and almost as predicted by the neoclassical model. Moreover, if we control for endogeneity of finance, the financial status of the company has no influence on its long-term capital decision; again as predicted by the neoclassical investment theory. The equity ratio, our measure of the financial status of a company, is insignificant in the explanation of the level of capital that a company employs. Accordingly, the estimated long-term elasticity of capital with respect to the equity ratio is close to zero. In short, the neoclassical model is a good summary of capital decisions in the long run. For the short run, i.e. for investment, however, the findings differ partly from the predictions of a standard neoclassical investment model. On the one hand, we find that the gap between desired capital – as determined by profitability – and actual capital can explain a substantive part of the variation in investment, as predicted by the neoclassical model. On the other hand investment is a moderately convex function of the gap and exhibits moderate residual autocorrelation. Higher-order terms of the capital gap significantly influence investment, so that adjustment costs are found to be non-convex. Finally, while finance has no significant influence on target levels of capital, it influences the investment process, so that finance only plays a role in the short run.5 More finance speeds up adjustment of capital to its target level. This implies that the impact of finance on investment varies strongly with the size of the capital gap. A good financial status is complementary to a large gap in shaping investment decisions in the sense that only if there is a large capital gap, i.e. a strong fundamental investment incentive, then a change in the financial situation has an impact on investment. Figuratively speaking, finance is the grease in the investment process but it is not the fuel. It eases adjustment to the target level of capital, but from the estimation of the long-run relationship we know it does not alter that target. Since finance has no predictive power for capital decisions in the long run, the results for investment cannot be attributed to the failure to measure capital productivity correctly. If finance contained information about future investment prospects that was not contained in the productivity measure, this would be reflected in a significant correlation of finance and the level of capital. The remainder of this paper is organized as follows: Section 2 develops the theoretical grounds for the empirical analysis. The section sketches an extension of the gap approach to cover the influence of financial frictions. It also reviews the recent debate between Cooper and Willis (2004) and Caballero and Engel (2004) on the gap approach and hence focuses our attention on the most critical steps and parts of the analysis. Section 3 introduces the method used by the present paper to measure productivity, and to estimate the long-run optimal stock of capital and the investment equation. Section 4 gives a brief description of the data that has been used. Section 5 presents the empirical results of both the regression for the optimal stock of capital and the investment regression. Finally, Section 6 concludes and a data appendix follows.

In this paper the interaction of fixed capital adjustment costs and financial frictions was studied empirically. To do so, a proxy for the profitability of investment was obtained. This proxy explicitly accounted for the technological heterogeneity of firms across sectors. Since capital profitability follows an I(1) process in the analyzed sample and is cointegrated with capital, we have performed a two step non-linear cointegration analysis for capital and investment. From the estimation of the long-term relation, we found that finance is hardly correlated with the choice of capital. Accordingly, the Modigliani–Miller theorem holds, or more precisely, cannot be rejected from a long-term perspective. Plainly speaking, finance does not matter for long-term capital holdings. However, the picture changes substantially if the effect of finance on investment decisions is analyzed taking a short-term perspective. Being more equity financed starkly increases the speed of adjustment of capital to its equilibrium level. In summary, the results for the long- and short-run imply that financial considerations primarily have intertemporal substitution effects for investment. Firms endowed with more financial means do not invest more, but they invest more often and in smaller amounts than firms which have lesser financial means. Figuratively speaking, finance is the grease but not the fuel of investment. However, we need to take into account that most firms in our sample are relatively large, established, publicly listed firms. A priori we would not expect these firms to suffer from substantial (agency) costs of finance in the long run. For a more representative sample of firms the overall picture may differ. The findings of our analysis obviously hinge on the quality of the proxy used for capital profitability and later for mandated investment. The derived measure of mandated investment (gap) can explain a significant fraction of the variation in investment. This suggests that the proxy can be considered as reasonable. Moreover, the econometric issues raised by Cooper and Willis (2004) do not apply to our data either because we find capital profitability to be non-stationary. Therefore, the differences in adjustment speeds at high and low levels of mandated investment can be interpreted structurally as evidence for non-convex adjustment costs, reconfirming our a priori assumption. In particular, we can interpret them structurally, as the differences are not only econometrically but also economically significant. This raises the question of what is to be structurally concluded from our results with respect to economic primitives particularly for finance. Although the estimation technique of the present paper does not recover the parameters of economic primitives themselves, such as the adjustment cost function, and hence does not allow us to draw strong structural conclusions directly, some economic structures are more compatible with our results than others. We find that finance primarily influences the adjustment speed but not the level of the stock of capital, for example. This does not lend much support – at least on an intuitive basis – to models in which managerial discount factors or market interest rates depend strongly on the financial situation of a company.