The macroeconomic determinants of private equity investment: a European comparison

Abstract

A strong private equity (PE) market is a cornerstone for commercialization and innovation in modern economies. However, substantial differences exist in the relative amounts raised and invested in PE across European countries. We investigate the macroeconomic determinants of PE investment in Europe, focusing on the comparison between Central and Eastern European (CEE) and Western European countries. Our estimations are based on a data set running from 2001 to 2011 that covers 16 countries. Applying robust estimation techniques, we identify a ‘robust’ set of determinants of PE activity in both regions. We find similarities as well as differences in the driving forces of PE investments in Western European and CEE countries. Our results suggest that economic activity, the inflation rate, equity market capitalization, unit labour costs, the unemployment rate as well the the institutional and legal environment are significant determinants of PE activity.

Keywords:

• private equity
• extreme bounds analysis

JEL Classification:

• C23
• C52
• E22
• G24

Acknowledgements

The authors are grateful to Eric Girardin, Katja Neugebauer, Dorothea Schäfer and participants at the FINESS Advisory and Steering Committee Meeting for helpful comments.

Notes

¹ PE is equity capital provided to enterprises not quoted on a stock market. PE includes the following investment stages: venture capital (VC), growth capital, replacement capital, rescue/turnaround and buyouts. In particular, VC refers to equity investments made for launch (seed), early development (start-up) or expansion (later-stage venture) of business and hence provides equity funding to younger, small and relatively high-risk companies featuring strong growth potential. On the other hand, leveraged buyouts and restructuring deals target more mature firms where substantial gains in operational efficiency are expected to materialize.

² Empirical evidence on the economic impact of PE is provided, for example, in the work of Levine (1997), Hellmann and Puri (2000), Kortum and Lerner (2000), Belke et al. (2003) and Fehn and Fuchs (2003).

³ CEE comprises the countries of Bulgaria, Croatia, the Czech Republic, Estonia, Hungary, Latvia, Lithuania, Poland, Romania, Slovakia and Slovenia.

⁴  European Venture Capital Association (EVCA) PE investments ‘market’ statistics, i.e., aggregation of figures according to the location of the portfolio company. At European level, this relates to investments in European companies regardless of the location of the PE firm.

⁵ The European GDP is the sum of GDP for all European countries included in the statistics; Europe includes: Austria, Baltic countries (Estonia, Latvia, Lithuania), Belgium, Bulgaria, Czech Republic, Denmark, Finland, France, Germany, Greece, Hungary, Ireland, Italy, Luxembourg, Netherlands, Norway, Other CEE (Bosnia-Herzegovina, Croatia, Macedonia, Montenegro, Serbia, Slovenia, Slovakia), Poland, Portugal, Romania, Spain, Sweden, Switzerland, Ukraine, United Kingdom.

⁶ Five countries – Poland, Hungary, the Czech Republic, Ukraine and Romania – accounted for more than 80% of the total investment amount.

⁷ In contrast to Jeng and Wells (2000), Schertler (2003) finds that liquidity of stock markets significantly affects early-stage investments rather than expansion-stage VC investments. These opposing results could be originated by differences in the specification of the proxy variables.

⁸ Jeng and Wells (2000) find that labour market rigidities do not significantly influence total VC but affects negatively the early stage of VC investment.

⁹ This finding is in line with Bliss (1999), Karsai et al. (1998) and Chu and Hisrich (2001).

¹⁰ One might argue that in this context factor analysis is an alternative to EBA. However, principle components analysis makes the interpretation of the impact of individual coefficients less clear. Moreover, since we want to allow for differences in the determinants between Western and CEE countries, we would have to calculate separate factors for both regions, for this our data set is too small.

¹¹ Their exclusion would result in an omitted variable bias.

¹² Sala-i-Martin (1997) and Levine and Renelt (1992) suggest to limit k to a maximum of three.

¹³ Sala-i-Martin (1997) proposes to aggregate the individual CDF(0)’s by applying a weighting scheme that is proportional to the (integrated) likelihoods of each regressions, such that models that are more likely to be the true model get more weight. However, this goodness of fit measure is not optimal for two reasons. First, in case of unbalanced panels due to missing observations in some explanatory variables, which is the case in our data set, regressions based on a larger set of observations will get more weight, since they have a (spuriously) better fit. Second, as argued by Sturm and de Haan (2005), weights constructed in this way are not equivariant for linear transformations in the dependent variable, such that changing scales will result in different outcomes. We will therefore calculated the unweighted averages of the individual CDF(0)’s.

¹⁴ The countries included in the estimation panel are selected according to poolability tests.

¹⁵  We employ total PE investments 'market' statistics, which include all investment stages: VC, buyouts and restructuring. Unfortunately, the data available is not detailed enough to differentiate between investment stages as it is done, e.g., by Jeng and Wells (2000) and Da Rin et al. (2006)

¹⁶  EVCA has recently started to collect data on other CEE countries but these data series are either too short or data of several countries are pooled together, e.g., the Baltic states. The limited data availability hampers an extention of our data set. Moreover, Czech Republic, Hungary and Poland are the major recepients of PE investments, e.g., in 2006 PE to these three countries accounted for more than 80% of total investments to all CEE countries.

¹⁷  We tested the variables for stationarity using the panel unit root test of Im et al. (2003). It suggests that the assumption of stationarity is a reasonable one. The results are available from the authors upon request.

¹⁸  Compare e.g., Gompers and Lerner (1998); Romain and Pottelsberghe de la Potterie (2004a), Groh and Liechtenstein (2009). As a robustness test, we have also estimated EBA without sorting GDP growth and per capita GDP in the set of ‘standard regressors’. Our results confirmed that these two variables are highly robust and significant regressors.

¹⁹  We have also repeated the estimates choosing and . The results did not differ substantially from the results when setting . The estimation results are available from the authors on request.

²⁰ In addition, we have also included as a robustness test each of the remaining explanatory variables to the regression, but they all turned out to be highly insignificant. Thus, this confirms the findings from EBA and we can be sure that we are not missing any important information from the set of available regressors.

²¹ De Hoyos and Sarafidis (2006) point out that the Friedman tests are biased in cases where cross-sectional dependence is characterized by alternating correlations in the residuals. The Frees’ test is not subject to this drawback and is therefore more reliable.

²² Estimating our equation as a system of seemingly unrelated regression equations (SURE) and then estimate the system by generalized least squares (GLS) techniques is not a feasible option with our data set, since our time horizon T is shorter than the panel dimension N. Beck and Katz (1995) show that they estimtion approach has good small sample properties.