Internationalisation of technology-oriented firms in Germany and the UK

In this article, only firms that export their products or services are regarded as internationally active firms. This means that only internationalisation on the sales market is considered. Of course, firms may import investment goods or components, i.e. internationalise on the procurement market. Although the latter way of internationalisation may also be important for small high-tech firms, it is neglected in this article.

See, in particular, Bernard and Jensen (2004) and Bernard and Wagner (2001).

Formulas and notation used in this article are identical to those used by Bernard and Jensen (2004) and Bernard and Wagner (2001).

This is the simplest way of introducing sunk costs. Alternatively, it can be assumed that current profits are also affected by the export status from periods prior to the immediately preceding interval (cf. Roberts and Tybout 1997), or we can introduce exit costs from the foreign market. The basic argumentation remains unchanged.

Bernard and Jensen (2004) provide a detailed discussion of the role of different econometric methods when estimating the binary choice framework given in Eq. 5. They stress that the fixed effects estimator in a linear probability model is almost surely biased downward but gives a lower bound of the effect of previous export status. Using data of German manufacturing firms in Lower Saxony, Bernard and Wagner (2001) get similar results for the effect of sunk costs. The lagged export status increases the probability of exporting today by 38% (linear probability fixed effects model), or 68% (random effects probit model), respectively.

Models of technology diffusion, among them epidemic learning models, rank, stock, and order effect models are described in detail in Stoneman (1983) and Karshenas and Stoneman (1993).

Oviatt and McDougall (1994), for example, regard new technology-based firms that internationalise quickly as firms with an intangible, knowledge-based competitive advantage. This perspective is consistent with the RBV of the firm. The reasoning of RBV models that firm-specific assets determine the internationalisation behaviour of firms is also similar to the role of ownership advantages in Dunning’s OLI (ownership, location, internalisation) framework, also called the “eclectic paradigm” (Dunning 1993). According to Dunning himself, the eclectic paradigm intends to explain “what are” rather than, in the normative sense, “what should be” a firm’s international business activities.

As Germany’s largest credit rating agency, Creditreform has the most comprehensive database of German firms at its disposal. Creditreform provides data on German firms to the Centre for European Economic Research (ZEW) for research purposes. Dun & Bradstreet is the UK equivalent.

Subsidiaries, de-mergers or firms that were founded as a management buy-out (MBO) or buy-in (MBI) were excluded from the analysis.

The first survey is described in detail in Bürgel et al. (2004). This report also includes numerous descriptive and econometric analyses of this unique data set.

In his influential study, Little (1977) used a definition of NTBFs which includes firms as old as 25 years. In contrast, the first survey this article is based on considered only firms that were 10 years of age or younger at the time it was taken, which is in line with more recent studies of NTBFs (see, e.g. Storey and Tether 1996).

According to the analysis of Prantl (2002), those firms indicated as “dead” by Creditreform have almost certainly left the market. The reverse, however, is not true: Voluntary firm closures are often recorded by Creditreform after a considerable delay, causing the number of closed firms to be underestimated.

Table 6 in the appendix also shows the number of still-living firms in 2003. Since the number of mismatches in the population is indeterminable, possible mismatches are not considered in Table 6 in the appendix.

In fact, in contrast to the first survey where no sector bias was found, the ICT-hardware sector is underrepresented in the German as well as in the UK-based sample. Conversely, the health/life sciences sector (engineering sector) is overrepresented in the German (UK-based) sample.

The effect of internationalisation on firm survival has been examined, for example, by Bernard and Jensen (1999).

It is important to note that the dependent variable is the transition probability and not the individual firm. Provided that there are no missing values for the independent variables, a single firm will enter the log-likelihood function twice: with the transition probability from the start-up period to 1997 and with the transition from 1997 to 2003. I have also extended this model by allowing for the possibility of random effects, one for each firm and for each type of transition as proposed by Van et al. (2004). The linear index the transition probabilities depend on now becomes \(x_{it} \beta _{jj^{\prime}} +\sigma_{jj^{\prime}} u_{ijj^{\prime}} \) . The terms \(\sigma_{jj^{\prime}} u_{ijj^{\prime}} \) are assumed to be mutually independent and independent of x, with mean 0 and variance \(\sigma_{jj^{\prime}}^2.\) The random variable \(u_{ijj^{\prime}} \) has been assumed to be standard normal distributed. The parameters \(\sigma _{jj^{\prime}} \) have to be estimated. This model has been estimated by simulated maximum likelihood. However, a likelihood ratio test of the restricted model (i.e. with all \(\sigma_{jj^{\prime}} \) set to 0) against the unrestricted model cannot reject the null hypothesis (LR χ²(4) = 2.836;   [Prob > χ²] = 0.586). Apparently, the large set of firm-specific variables in my estimation equation (see below) is able to discriminate between exporters and non-exporters, i.e. a large part of firm-specific heterogeneity is, in fact, observed. Therefore, in the following I will restrict my analysis to a model without simulated heterogeneity.

Nominal exchange rates were taken from the historical exchange rate database of Oanda Corp., available at http://​www.​oanda.​com. Consumer price indices used to calculate real exchange rates were taken from Global Economic Data of EconStats, available at http://​www.​econstats.​com

The results were obtained using the statistical software package STATA, version 8.2 SE. The estimator of the rare events logit, programmed by Michael Tomz, Gary King, and Langche Zeng, is available at http://​GKing.​Harvard.​Edu (cf. King and Zeng 2001). For analysing the results, I further used the STATA based programme CLARIFY, written by Michael Tomz, Jason Wittenberg, and Gary King, and also available at http://​GKing.​Harvard.​Edu (cf. King et al. 2000).

The nominal exchange rate of the euro and the pound to the most important foreign currency, the US dollar, did indeed change significantly during the period from 1997 to 2003. However in 1997, the average rate was nearly the same as the average rate in 2003.

Bernard and Jensen (2004) tested for geographical spillover effects in the US manufacturing industry, but did not find any evidence of the existence of spillover effects that were hypothesised to reduce entry costs. According to the authors, this may result from the exclusive selection of large manufacturing plants.

In this section, it often seems reasonable to compare the empirical results of this article with the results of the cross-sectional analyses of the first survey. The results of the first survey are given in Bürgel et al. (2004). I refer to this report, although I do not mention it explicitly in each case.

Setting all independent variables, including number of employees, to their mean, the predicted probability of remaining an exporter in the rare event logit model is 93.61%. In comparison, applying the conventional logit model, i.e. without the correction described in Sect. 4, results in a predicted probability of 98.70%. Thus, in contrast to the unbiased estimator (i.e. the rare event logit model), the conventional logit model overestimates the probability of the more frequent event, that is, the probability of staying in the foreign market.

Due to the limited number of observations in our dataset, it is however impossible to determine econometrically whether these differences are a result of a varying endowment with firm-specific resources or whether the influence of the exogenous factors differs between German and UK-based firms.