The cyclical behaviour of inventories: European cross-country evidence from the early 1990s recession

Abstract

This paper employs data for a panel of firms from France, Italy and the UK to study the effect of the recession of the early 1990s on inventory investment, controlling for cyclical fluctuations at the firm level. The results clearly show some common patterns across countries, pointing to the relevance of financial factors (namely, the level of leverage) in propagating initial recessionary shocks. Moreover, Italian firms, especially if ‘small and young’, seem more likely to suffer from a reduction in the value of collateralizable assets possibly originated by restrictive policy actions.

Acknowledgements

The authors acknowledge financial support from the Fondation Banque de France and from MIUR. Luigi Benfratello and Claudio Campanale provided excellent research assistance. They also thank Bob Carpenter, Alberto Dalmazzo, Roberto Golinelli, Giovanna Nicodano, Henri Pagès, Fabio Schiantarelli, Patrick Sevestre and seminar participants at the University of Bologna and Siena, and at the Banque de France for useful comments and suggestions on previous drafts of this paper.

Notes

Alternatively, other studies use time-series data on firms disaggregated by geographical region (Carlino and De Fina (1998) for the USA) or by sector (Dedola and Lippi (2000) for the USA and four large European economies).

Also Bond et al. (1997) provide a cross-country study of a panel of European firms, analysing the role of financial factors in affecting firms’ investment behaviour, but with no specific focus on recessionary episodes.

Higher frequency data (e.g. quarterly as in the empirical analysis of US firms by Carpenter et al, 1994, 1998), though more desirable to capture the interaction between cyclical inventory fluctuations and financial factors, are unavailable for most European countries.

The Appendix provides more details on the treatment of the data and some discussion of our sample building strategy.

In preliminary work, also a sample of German firms had been constructed from the same data source. Data for only 325 firms are available and the size distribution is not comparable with the other three countries in the sample (e.g. the median German firm has real sales three times as large as the median UK firm). It was therefore decided to exclude German firms from the investigation.

This may be partly due to the presence for about 50% of UK firms only of their consolidated balance sheets (see the Appendix for details).

The dimensional split among firms may be obtained using other criteria: e.g. by using the median of the real sales distribution in each country as the threshold or by defining those firms in the upper third of the size distribution as ‘large’ and those in the lower third as ‘small’ (in the following empirical analysis both such alternative split criteria were applied with qualitatively very similar results).

Moreover, splitting the sample by firm size as a way of identifying crucial effects is a widely used technique in the literature on investments and financial constraints (Hubbard, 1998; Schiantarelli 1995).

Devereux and Schiantarelli (1989) define as ‘old’ the UK quoted firms with at least 12 years of age; Carpenter and Rondi (2000) classify as ‘old’ those Italian firms with more than 15 years of age.

Furthermore, as suggested by Carpenter and Rondi (2000), the peculiar ownership structure of a large fraction of Italian firms, based on long-lasting family control, represents a constraint on firms’ growth, reducing the correlation between size and age.

Moreover, the lack of data on ‘cash’ held by firms in the data set precludes the computation of a reliable measure of liquid assets. For the same reason, it was not possible to construct a measure of ‘net leverage’ by subtracting cash and other liquid assets to both total debt and total liabilities.

If forward-looking firms are subject to common shocks to sales, so that current sales matter also because they predict the future, the time effect α _t must also pick up a common component in future expected sales not captured by the coefficients on current and lagged sales.

It is well known that in dynamic panel data models where the autoregressive parameter is large and the number of time series (N) is small the first difference GMM estimator suffers from finite sample bias and poor precision, as shown in simulation studies, e.g. Blundell and Bond (1998). However, this is unlikely to be a problem in the present case since N is very large, ranging from 1560 to 2254 in the aggregate estimates, and inventories are known not to be a very persistent phenomenon (Ramey and West, 1999). In fact, Blundell and Bond (1998) find that the sample bias becomes negligible for N = 500 and a true autoregressive parameter around 0.8.

Furthermore, the subsample of ‘large and young’ UK firms shows a larger coefficient on leverage compared to the other subsamples. However, this result has to be taken with caution given both the limited sample size (159 firms) and the suspicious deformity in the magnitude of other coefficients.