There is a finding across virtually all micro firm-level data sets that exporting and efficiency are positively correlated. Much work has focused on how this correlation is to be interpreted. Work on countries outside of Africa, reviewed in Bigsten et al (2004), has found more evidence that causation runs from efficiency to exporting - there is self-selectivity into exporting - rather than from exporting to efficiency - implying firms learn from exporting.1 For sub-Saharan Africa Bigsten et al (2004) present evidence for sub-Saharan Africa firms of learning by exporting. In their study there is little evidence for an efficiency effect by which firms self-select into exporting. However they note that "this may be due to the co-linearity between some of the regressors in the export probit". (page 133) The time dimension is short, three years, and with such data it is difficult to assess convincingly the importance of efficiency as a determinant of exporting while allowing for firm heterogeneity and the presence of sunk costs with a dynamic specification.
While the focus of much of the literature to date has been on distinguishing learning from exporting and self-selection into exporting a common finding across all the work on African firms has been a strong correlation between firm size and exporting, SС†derbom and Teal (2000, 2003), Bigsten et al (2004), and van Biesebroeck (2005). The finding in Bigsten et al (2004, page 128) is
that firm size is a robust determinant of export participation across specifications which allow for certain forms of firm heterogeneity and dynamics. In this paper we exploit a data set which has a much longer panel dimension than has been used before to assess the relative importance of selfselection based on efficiency and firm size as determinants of export participation.
One interpretation of the size effect is that it is really a disguised element of efficiency in that larger firms benefit from increasing returns to scale. The evidence to date is strongly against this interpretation. SС†derbom and Teal (2004) use a longer run of the Ghana data, from 1991 to 1997, to show that constant returns to scale is readily accepted by the data. Rankin (2004) has a similar finding for comparative data across the same countries used in this study. A second interpretation of the size effect is that it can be explained by the presence of fixed costs. However this interpretation is difficult to reconcile with the robust significance of the size variable when there are controls for sunk costs in the form of a lagged dependent variable (see Bigsten et al (2004, p. 128)). It also seems to be the case that size is not capturing sectoral differences in technology as size remains highly significant with sectoral controls in the equation. There remains the possibility, which none of the regressions in either SС†derbom and Teal (2003) or Bigsten et al (2004) can rule out, that size is simply a proxy for some time-invariant aspect of the firm correlated with the other regressors. If larger firms simply have more skilled labour and that is the key to exporting then size has no causal role in affecting the ability to export. If this is the case then the association between exporting and size implies that it is not the size of the firm that matters for policy but some unobserved aspect of large firms - the most obvious being their skills broadly interpreted - which matters for understanding the ability of firms to participate in the
export market. To distinguish convincingly between these alternatives the longer time series dimension to our data is vital.
The relative importance of these factors is of importance for understanding the factors that determine the export behaviour of African firms. African firms are in general very small and the focus of small firms on the domestic market ensures that, in aggregate, their growth is limited by the growth of domestic incomes. If size is related to a firm's participation in the export market this may reflect a strategy of successful growth - limited domestic demand ensures that the only way for firms to expand is for them to grow into the export market. As a first step to investigating this hypothesis we ask if changes in firm size, measured as the log of employment, change the probability of a firm participating in the export market in the future.
In the next section we set out our general modelling framework which allows a clear identification of the relative importance of these efficiency and size effects on export participation. In section 3 we present a comparative data set, which includes firms across a wide range of sizes, of five African countries - Ghana, Kenya, Tanzania, Nigeria and South Africa - that allows us to investigate these issues over the period from 1993 to 2003. As we have panel data we can address the issues of time-invariant heterogeneity in a way that has not been possible before for firms in sub-Saharan Africa. In section 4 we present the results of this modelling exercise focusing on allowing for the importance of unobservables in affecting our view as to the determinants of exports. The possible roles of ownership and skills are discussed in section 5. A final section concludes.
Tybout and Westbrook (1995) on Mexico; Clerides, Lach and Tybout (1998) on Mexico, Colombia and Morocco; Kraay (1999) on China; Aw et al. (2000) on the Republic of Korea and Taiwan; Isgut (2001) on Colombia; and Fafchamps et al. (2002) on Morocco.