The Determinants of Firm Exit from Exporting: Evidence for the UK

Abstract

This study seeks to understand to what extent new exporters are able to survive in international markets and whether exit from exporting is more likely to be associated with firm-level heterogeneity or more general factors such as trade costs and/or barriers to entry and exit (such as sunk costs). This study presents the first analysis undertaken for a nationally representative group of UK firms on the determinants of exit from exporting, using panel data covering all market-based sectors of the UK during 1997–2003. Our findings suggest that the probability of a firm ceasing to export is directly influenced by its productivity and other attributes associated with firm-level productivity differences (such as size and foreign ownership). Micro-finance factors, such as profitability and the ability to finance through long-term debt, play an additional role. Lastly, sectoral differences (e.g. industrial concentration) also help explain the firm’s exit decision, whilst trade costs lead to a higher probability of exiting from selling internationally.

Key Words:

• Exit from Exporting
• Survival Analysis
• Discrete Time Model
• UK Market Sector

JEL classifications:

• D21
• L25
• O14
• O24

Notes

aIn addition the variables entering the model included 33 industry dummies at the two-digit (SIC 2003) level and 11 Government Office region dummies (as well as composite dummies involving the AGE variable).

bCalculated separately for each three-digit industrial sector.

cSource: Table 3.6 in 2006 UK Input–Output Tables (ONS, 2006).

dDefined as profits (turnover − cost of sales − depreciation − labour costs) divided by total assets.

aBased on OIM standard errors; estimation based on complementary log-log transformation. ^∗∗∗Significant at 1%; ^∗∗significant at 5%; ^∗significant at 10% level.

bTests of model misspecification are based on log relative hazard = . Under the correct specification, β ₁= 1 and β ₂ = 0; thus we test β ₂ = 0. The significance level for rejecting the null is reported here.

aUnweighted FAME data covers the UK.

bSize-bands are in £’000.Source: ARD and FAME databases.

1. See, for instance, Roberts and Tybout (1997) for Colombia, Bernard and Jensen (2004a) for the US, and Gourlay and Seaton (2004) for the UK.

2. Notably, there are a number of studies that concentrate on firm exit from international markets, although only in light of its (often deleterious) impact upon productivity (or efficiency), instead of the factors that drive the decision to stop exporting. Examples of the former include Baldwin and Gu (2003), Bernard and Jensen (2004b), Bernard and Wagner (1997), Clerides et al. (1998), Girma et al. (2003), Requena-Silvente (2005), and most recently Bernard et al. (2007).

3. Most of other empirical work (for various countries, as reviewed earlier) only concentrates on (selected) manufacturing industries, for example Alvarez and López (2008), Das et al. (2007), Wagner (2008).

4. Note Requena-Silvente (2005) also uses the same data source to look at exporting entry and exit for the UK. However, his study only considers selected UK SMEs for the 1994–1998 period, and thus the sample size is less than a tenth of that used for the current analysis.

5. The first year of data (1996) is “lost”, as we include industry growth rates as one of our determining variables (thus 1996 growth cannot be computed without 1995 data). The last year of data (2004) is also omitted as in 2004 all firms “exit” in this year (i.e. the data is right-censored).

6. FAME is a commercial dataset collected by Bureau van Dijk with access available via most UK universities’ subscription. However, the Annual Respondents Database is collected by and held at the Office for National Statistics (ONS) with restricted on-site-only access via application and confidentiality agreement with the ONS.

7. For instance, probit models as estimated in Requena-Silvente (2005), Fariñas and Martín-Marcos (2007), and tobit models are used in Alvarez and López (2008) and Das et al. (2007).

8. Rodríguez (2008) provides an excellent review of the methodology of discrete time survival models in a book chapter.

9. Hence, equation (3) generates results similar to linking the survival function in the continuous time case to the cumulative hazard at all previous times.

10. As Disney et al. (2003) point out, “…with the Cox specification, we cannot enter age directly since it is collinear with the baseline hazard. We could enter age directly if we adopted a parametric specification for the baseline, but we would then be relying on identification of the age effect from the assumed functional form” (p. 105).

11. We also experimented estimating our model for various industry groups separately. Nevertheless, the results obtained from industry-specific estimation are broadly similar to those from pooling all industries together. Thus only results from pooled model are reported here for the sake of brevity.

12. This is consistent with findings by Requena-Silvente (2005), that is, foreign-owned firms in the UK were least likely to exit their export markets.

13. We also included ln (1 + real value of intangible assets) instead of INTASSET, as a check on whether the volume of assets was more important, but the results were less significant.

14. The parameter estimate for [ln OPEN_i] is −0.053 (with an associated z-value of −2.23), while the parameter estimate for [ln (EMP_it− OPEN_i)] is −0.153 is with a z-value of −2.76).

15. For a details description of the ARD, see Griffith (1999), Harris (2005), and Oulton (1997).

16. Efforts have also been made to merge FAME into the ARD; nevertheless, these have been largely unsuccessful (see Harris and Li, 2007, ch. 2, for more details).

17. Where there are fewer than 10 enterprises in any subgroup, these data are not available to avoid disclosure of confidential information in these ONS data. This results in a loss of some 4% of the total turnover available in the ARD.

18. Note, we do not weight the FAME data for 34 industries because the FAME data have better coverage in terms of total turnover than the ARD. These 34 industries (out of 215 in total) account for just 2.9% of total FAME turnover. Note also, the ARD does not contain data for Northern Ireland, but since this region is rather small, it will not have much of an effect on the weights used.

19. We have also undertaken a further check of the usefulness of the weighted data on exports by comparing it to information from the 2004 Community Innovation Survey (CIS4) that contains information on which establishments exported in 2004. Our findings suggest that while there are differences across industries, the relative magnitudes of the estimates of the percentage of firms that export for the two datasets are very similar.