Growth effects of corporate balance sheet adjustments in the EU

Abstract

This paper investigates the impact of active balance sheet adjustments in the non-financial corporate sector on economic growth in the EU. We jointly model firms’ ability to reduce their balance sheet imbalances and a growth equation in an instrumental variables (IV) panel context. This enables us to explicitly consider the contemporaneous interaction between corporate balance sheet adjustment and growth, which can otherwise bias inference. Our main findings inter alia suggest that periods of active corporate deleveraging are associated on average with lower output growth compared to periods when no adjustment takes place. Moreover, a decline in corporate debt overhang supports output growth. Model-based simulations are in line with these findings.

Appendices

Appendix: Data and sources

Econometric estimations presented in Sect. 2 are based on an annual unbalanced dataset for 25 EU Member States (i.e. apart from Luxemburg, Malta and Sweden) over the period 1995–2015. Annual sectoral financial variables at national currency (at current prices) and on a non-consolidated basis are drawn from the OECD Database on Financial Accounts (National Accounts) and Eurostat Financial Balance Sheets (nasa_10_f_bs). For households and non-financial corporations, debt is defined as the sum of debt securities and loans. Remaining annual sectoral variables are taken from the OECD Non-financial accounts by sectors and Eurostat Non-financial transactions (nasa_10_nf_tr).

The main macroeconomic variables (real GDP per capita, real effective exchange rate, CPI) are drawn from the AMECO Database and Eurostat. Share prices are taken from the OECD Monthly and Monetary Financial Statistics and are deflated by CPI. Credit growth to non-financial corporations is calculated as the annual percentage change of total loans to corporates. Data on systemic banking crises come from the systemic banking crises database of Laeven and Valencia (2012). A banking crisis is defined as systemic when there are significant signs of financial distress in the banking sector as well as policy intervention measures in response to significant losses in the banking sector. World GDP per capita growth is taken from World Bank National Accounts database. Finally, data on the short-term interest rate and the cyclically adjusted primary budget balance (as a % of potential GDP) come from the OECD Economic Outlook No.99 (Fig. 1).

Fig. 1

Evolution of NLB and NLB-to GDP ratio per country (1995–2015)

Statistical tests on IV estimates

A battery of statistical tests is performed to ensure IV/2SLS model identification, robust inference in light of weak instruments and to test for the endogeneity of the regressors that would support the use of IV estimators. More specifically, in order to estimate Eqs. (1a, 1b) and (2), they must be identified. A necessary (though not sufficient) condition is that the order condition is met, namely the number of endogenous regressors should be less than the instruments used. The rank order condition must also be satisfied, meaning that the (excluded) instruments must be correlated with the endogenous regressors. In the opposite case, the bias in the estimated IV coefficients gets larger, and the IV estimator loses consistency (see, e.g. Baum et al. 2007; Hahn and Hausman 2002; Wooldridge 2003).

Against this background, we report the Kleibergen and Paap (2006) rk LM statistic that tests for model under-identification. Under the null hypothesis, the set of (excluded) instruments is not correlated with the endogenous regressors. Therefore, accepting the null hypothesis suggests that the model cannot be identified and estimated. Moreover, even if the model is identified, a weak correlation of the instruments with the endogenous regressors can still lead to biased IV coefficients (see also Baum et al. 2003). We account for the weak instruments problem by examining the significance of the (excluded) instruments in the first-stage IV regressions. The latter are reduced form regressions of the endogenous variables on the instrument set. Commonly, the weak instruments problem is accounted for if the F statistic of testing the joint significance of the instruments in the first-stage regression is higher than 10 (Staiger and Stock 1997). For the IV-GMM estimations, we report the Sanderson-Windemeijer test which is a weak instruments F-test in the first-stage regression. As an additional control of the weak instruments problem, the statistical significance of the individual coefficients of the instruments in the first-stage regressions is assessed.

As discussed, the use of IV estimators requires that the instruments are also uncorrelated with the error term. This can be formally tested if the number of instruments is larger than the number of endogenous regressors, namely if the equation is over-identified. For testing the over-identifying restrictions, we report the Hansen (1982) J statistic. Accepting the null hypothesis implies that the orthogonality condition is met, namely instruments are uncorrelated with the error term.

In the IV/2SLS context, output growth and balance sheet adjustments are treated as endogenous regressors in the system. This is formally tested by conducting a C test (or “GMM distance”) test of the exogeneity of these regressors. The null hypothesis is that the specified endogenous variables can be treated as exogenous. For instance, inability to accept the null when testing for the potential endogeneity of output growth in the equation of balance sheet adjustments suggests that growth should not be treated as exogenous to corporate deleveraging; there is some degree of simultaneity between the two variables.