Investment, default propensity score and cash flow sensitivity in six EU member states: evidence based on firm-level panel data

ABSTRACT

Using a panel data set covering six European countries and $119700$ firms ($582153$ observations) over the period $2008$–$2013$, we analyse the cash flow sensitivity of investment spending. As most of the firms are not listed at stock exchanges, a balance sheet-based approximation on Tobin’s Q is used to indicate investment opportunities. We analyse internal and external liquidity constraints and their effect on investment decisions. In the literature, external constraints are most often indicated by simple accounting-based items/ratios. As the adequacy of the a priori indicator, reflecting the external constraints, is crucial, we contribute in proposing a more sophisticated approach. We estimate propensities to default using adapted random forests. In our descriptive analysis, we find strong evidence for the u-shape of the investment curve. However, after controlling for investment opportunities we find no increased cash flow sensitivity of investment, neither for a priori externally nor for internally constrained firms. Hence, our results hint for the absence of liquidity constraints. We attribute these towards the rather expansionary monetary policy since the financial crisis.

KEYWORDS:

• Investment function
• Q-theory
• investment-cash flow sensitivities
• default propensity score
• financing constraints

JEL CLASSIFICATION:

• C33
• D92
• G32

Disclosure statement

No potential conflict of interest was reported by the authors.

Notes

¹ For the regression analysis we further separated ‘Manufacturing’ ($01$–$33$) in the subsectors ‘Consumer Products’ ($01$–$12$), ‘Textile Goods’ ($13$–$15$), ‘Wooden Products’ ($16$–$18$), ‘Raw Materials’ ($19$–$25$), ‘End Products’ ($26$–$32$) and ‘Repair and Installation’ ($33$).

² For capital stock we use tangible fixed assets.

³ Lewellen and Badrinath (1997) proposed a promising algorithm for calculating reliable capital stock (at replacement) values. However, as this approach requires long time series (about 30 years) it is not applicable to our data set. Even the less data demanding approach proposed by Behr and Bellgardt (2002) is not applicable due to the rather short time dimension of our panel data set.

⁴ In Table 8 we show the estimated average lifetimes of the capital goods. We find that capital goods tend to last longest in sector ‘Energy and Infrastructure’ and shortest in sector ‘Services’.

⁵ As sales are not reported for British firms, we use operating revenue turnover as a proxy. The correlation between sales and operating revenue turnover in all other countries is close to one.

⁶ Not all of the mentioned conditions necessarily lead to bankruptcy but are definitely alarming for potential credit grantors and therefore increase the firm’s financial restriction.

⁷ In our case, $B=100$ turned out to be a suitable compromise between runtime and precision (Breiman and Cutler (2004); Verikas, Gelzinis, and Bacauskiene (2010, 331)).

⁸ To speed up the parameter estimation, we parallelized our calculation using R’s snowfall package.

⁹ The use of nominal values has practical reasons as well. There is neither firm-level price information to deflate the nominal values, nor do all variables (e.g. cash flow) have the structure of a product of prices multiplied by quantities. Therefore, the use of constant prices might lead to unrealistic figures and misleading results. Because firm-level price data are not available, one might consider using sectoral price information, which is available for some of the variables. However, this procedure could only be applied to some variables ($I$, $K$) and not to others ($CF$, market value of the capital stock).

¹⁰ The system GMM estimator proposed by Arellano and Bover (1995)/Blundell and Bond (1998) is model theoretically more efficient, but for our regression equation, the Sargan test indicated a non-valid instrument matrix for all countries.

¹¹ D’Espallier and Guariglia (2015, 6) cite several other papers which are in this strand of argument.

¹² For non-manufacturing firms the Sargan test indicates that the instrument matrix is not valid throughout.