The impact of foreign takeovers: comparative evidence from foreign and domestic acquisitions in Germany

Abstract

This study provides the first evidence of foreign takeover effects on the performance of acquired firms in Germany that considers a general takeover effect through the comparison with domestic takeovers. A propensity score matching approach combined with a difference-in-differences (DiD) estimator were performed with new high-quality panel data for manufacturing enterprises, provided by German official statistics. The results indicate a negative impact of foreign takeovers on employment and no productivity improvements for the period 2007–2009. This evidence contradicts existing empirical evidence for Germany which suggests significant productivity improvements and no changes in terms of employment. These findings are of particular interest to Germany as one of the most important FDI inflow destinations worldwide. They contribute to the foreign ownership performance premium literature as well as improving the understanding of foreign acquisition consequences, a subject of utmost topicality.

Keywords:

• post-acquisition performance
• foreign takeover
• manufacturing
• treatment analysis
• Germany

JEL Classification:

• F21
• F23
• G34

Notes

¹ For an overview of the foreign performance premium literature, see Bellak (2004) and Barba Navaretti and Venables (2004). For Germany see for example Temouri et al. (2008), Mattes (2010) and Weche Gelübcke (2013a).

² Not to mention the applied econometric method, variables and differences in data quality.

³ For example, Conyon et al. (2004) and McGuckin and Nguyen (1995) find profitability, wages and productivity rising generally after acquisitions in the UK.

⁴ Recent evidence for Germany can be found, for example, in Andrews et al. (2009) and Weche Gelübcke (2013b). The majority of international studies find that foreign investors select overperforming takeover targets, such as Hagemejer and Tyrowicz (2012) for Poland, Harris and Robinson (2002) for the UK, Huttunen (2007) for Finland, Salis (2008) for Slovenia and Oberhofer et al. (2012) for 16 European countries. Lichtenberg et al. (1987) and Chen and Robert (1997) for the US, as well as Gioia and Thomsen (2004) for Denmark, find a preference for underperforming targets. Castellani and Zanfei (2004) and Karpaty (2007) find no support for a selection of particularly over- or underperforming targets.

⁵ For a more detailed description, see e.g., Caliendo and Kopeinig (2008) and Guo and Fraser (2010). Matching and estimations were performed by using the Stata ado file psmatch2 by Leuven and Sianesi (2003). There is a set of alternative econometric approaches and an ongoing discussion on which is preferable, but all in all, if applied appropriately, results should not differ markedly (for an overview and discussion, see e.g., Imbens and Wooldridge (2009), Angrist and Pischke (2009), and Blundell and Dias (2000)).

⁶ To eliminate determining factors which affect takeover targets and nontargets unequally (e.g., restrictions to external finance that may be more harmful to firm performance in times of a post-takeover restructuring process), this study compares foreign takeovers to domestic takeovers. However, this comparison must happen in an indirect way because the number of domestic takeovers is not sufficient to perform a meaningful matching estimation. Hence, estimations are performed separately for each group of takeovers with nontakeovers as the control group and afterwards compared to each other.

⁷ The notation is based on Caliendo and Kopeinig (2008).

⁸ The reasoning behind this is the investors’ preference for targets at either the upper or lower bound of the performance range, the so-called cherry-picking and lemon-grabbing. Evidence for these selection mechanisms is provided by Weche Gelübcke (2013b) likewise for the German manufacturing sector, although with different data.

⁹ Of course, it is highly possible that company acquisitions had been anticipated and therefore indirectly shaped the pre-acquisition performance. This has to be kept in mind generally when takeovers are analysed in a treatment analysis framework.

¹⁰ The terms foreign-controlled, foreign-owned and foreign are used interchangeably in this text.

¹¹ Indirect control refers to the fact that enterprise A is controlled by enterprise B and both are domestic companies but enterprise B is, in turn, controlled from an entity abroad. Then, enterprise A will also be foreign-controlled. Effective minority control is stated when several minority owners with shares of more than 50% in sum act in concert.

¹² All computations were programmed in Stata 12 and carried out within the Research Data Center of the statistical office Berlin-Brandenburg.

¹³ It needs to be emphasized here that these categories cover different types of takeovers and that it may matter if, for instance, an independent entity or a group head is the target firm. It is indeed possible to distinguish these different types in the data but the number of observations in the subcategories would be too small to apply econometric methods properly and also the confidentiality of the micro-data would hamper a separate analysis.

¹⁴ The exclusion of potentially false ownership changes probably puts more weight on larger takeover targets, as the predominant part of identified ownership changes includes domestically owned firms that already belonged to a company network and are thus larger on average. On the one hand, this implies a bias in favour of larger acquisitions; on the other hand, this is a welcome bias in the sense that it reduces the impact of network effects. To be more precise, if an independent firm is taken over by a group head or multinational, changes in post-takeover performance may be due to the new participation in network effects. If the focus is on targets that enjoyed network effects even before the acquisition, this is a welcome step in the direction of comparing like with like.

¹⁵ Given two independent random samples, the nonparametric Kolmogorov–Smirnov test evaluates whether all moments of the two cumulative distribution functions of a performance measure in a case, and , are statistically different from each other and whether one distribution dominates the other (for more details, see Conover, 1999).

¹⁶ However, it remains unclear whether investors solely prefer above-average performing targets. Weche Gelübcke (2013b) presents evidence for the German manufacturing sector, although using other data, whereby investors select both above- and below-average performing targets. Further it seems to be the inverse combination of rentability and profitability that raises the takeover probability rather than individual measures.

¹⁷ Performance effects resulting from ownership changes may take some time before they can be recognized in the data, as shown in the literature review. This argues for the consideration of long-term changes. On the other hand, the determining factors of long-term performance changes may become much harder to handle in a treatment analysis setting. Thus, 2-year changes seem to be appropriate. Anyway, the data at hand do not allow for consideration of a longer period.

¹⁸ These numbers are not inflation-adjusted since the focus is on differences between the treated and untreated groups where the inflation bias is removed.

¹⁹ There are also ATT estimates which cannot be considered as statistically significant on a 90% level or higher, but considering that the data used cover all firms above the threshold of 20 employees, the estimates have informative value per se and it would be misleading to ignore the economically significant productivity increase following domestic takeovers.

²⁰ Besides looking at the results’ sensitivity to different matching algorithms, their sensitivity to slight changes of the assignment model is also an important issue. The selection model, which is described in the previous section, was therefore modified in several ways to evaluate the robustness of estimated ATTs. The results show that the exclusion of the age variable, the inclusion of interaction terms or the replacement of quintile dummies with squared terms can lead to changes in the ATT estimates but these changes do not violate the overall picture drawn from the benchmark estimates. The results of these sensitivity analyses are not reported to save space but are available from the author upon request.

²¹ For example, published in the Weekly Report of the German Institute for Economic Research (DIW) (2010). Moreover, the estimated employment effects in Arndt and Mattes (2010) are also negative in all NN matching specifications but insignificant in terms of bootstrapped SEs where analytical SEs according to Abadie and Imbens (2008) are the preferred method, at least for one NN matching. Moreover, the fact that Arndt and Mattes (2010) use a different time period which is not affected by the 2007/2008 global economic and financial crisis but also reach negative coefficients for post-takeover employment lend support that the negative results from this study are not merely due to the specific time period.