Public programmes to promote firms’ exports in developing countries: are there heterogeneous effects by size categories?

Abstract

Several countries have implemented programmes to support their firms’ internationalization efforts. Their impacts are likely to be heterogeneous over firm size categories because these programmes are primarily intended and expected to benefit smaller companies. Whether this is or not the case is still an open question. In this article, we aim at filling this gap in the literature by providing evidence on the effects of trade promotion programmes on the export performance of firms within different size segments using a rich firm level dataset for Argentina over the period 2002 to 2006. We find that these effects are indeed larger for smaller firms.

Acknowledgements

We thank the Foreign Trade Monitoring Unit at the Secretariat for Industry, Trade, and SMEs (UMCE-SCIP by its Spanish acronym); AFIP; and ExportAR who kindly provided us with export, employment and assistance data for Argentine firms. We also thank Alberto Barreix for his unconditional support in helping us build up the dataset used in this article, Oscar Mitnik who generously shared with us his code to implement the nonparametric tests of treatment effect heterogeneity; the editor, an anonymous referee, Juan Blyde, Ana Fernandes, Mauricio Mesquita Moreira and participants at the Workshop on ‘Exporter Dynamics and Productivity’ organized by the Canadian Department of Foreign Affairs and International Trade (Ottawa) for insightful and useful comments; and Markus Frölich, José Martinez and Jeffrey Racine for valuable background suggestions. The views and interpretation in this document are strictly those of the authors and should not be attributed to the Inter-American Development Bank, its executive directors, its member countries; the Secretary for Industry, Trade and SMEs; AFIP; or ExportAR. Other usual disclaimers also apply.

Notes

¹ Some authors argue, in addition, that informational asymmetries provide a rationale for trade policy (see, e.g. Mayer, 1984; Grossman and Horn, 1988; Bagwell and Staiger, 1989).

² Volpe Martincus et al. (2010) estimate quantile treatment effects of trade promotion programmes managed by PROCHILE, thus examining how their impact varies over the distribution of the relevant export outcomes.

³ Section III includes a precise description of our dataset and its coverage.

⁴ An appendix explaining the institutional organization of ExportAR and describing the export promotion programmes managed by this agency is available from the authors upon request.

⁵ Firm location might also play a role in explaining the probability to be assisted. Even though ExportAR has presence in all provinces, most of the offices are not owned by this agency and are not staffed with own personnel and, importantly, they do not always directly provide all the services offered at headquarters. In addition, spatial coverage is very dissimilar across regions. On the other hand, it can be conceivably thought of companies located in Buenos Aires, which is the largest economic centre and the region with better physical and communication infrastructure in the country, as enjoying a locational advantage in terms of access to relevant business information from alternative sources and therefore, as needing less support to trade across borders.

⁶ Other factors that may also play a role are, for example, the ability to cope with other sunk costs of entry, such as those originated in setting up an export department or redesigning products for foreign customers, and differences in access to management capability and financial resources in capital markets.

⁷ Hirsch and Adar (1974) show that large firms can afford to assume more risks than small ones. Further, their risks from foreign operations are less than those of small firms because the large firms benefit from economies of scale in foreign marketing. Hence, the risk premium demanded by large firms from foreign marketing is less than the premium insisted upon by small firms. As a result, the former tend to export a larger fraction of their output.

⁸ Most of these studies show that smaller sized firms seem to benefit from export assistance programmes. However, it should be mentioned that this literature is far away from having reached a clear consensus, as some authors claim that evidence on effectiveness of export promotion activities is limited and inconclusive (see, e.g. Seringhaus, 1986; Kotabe and Czinkota, 1992).

⁹ The use of (natural) logarithm is partially motivated by the scale problem originated in the fact that our binary variable D does not capture the size of the assistance (Lach, 2002). The presentation hereafter focuses on firms’ total exports, but mutatis mutandis also applies to measures of export performance along the extensive margin (number of destination countries and the number of products exported) and the intensive margin (average exports per country, average exports per product and average exports per country and product).

¹⁰ We will use interchangeably assistance, support, treatment and participation throughout this article.

¹¹ This is the potential outcomes framework due to, among others, Fisher (1935), Roy (1951) and Rubin (1974).

¹² In this exercise, we ignore general equilibrium effects so that outcomes for each firm do not depend on the overall level of participation in the activities performed by the agency (Heckman et al., 1998). Further, we do not consider information spillovers either. It is well known that firms may learn about export opportunities from other firms through employee circulation, customs documents, customer lists and other referrals (Rauch, 1996). Evidence on spillovers has been presented in several papers, e.g. Aitken et al. (1997), Greenaway et al. (2004), Mañez et al. (2004), Álvarez et al. (2007) and Koenig et al. (2010). If these spillovers would be associated with participation in export promotion activities, i.e. untreated firms obtain business information from treated firms, then the treatment effects, as estimated here, would be underestimated.

¹³ If there were time-varying firm-specific factors leading to improved export performance that are not observable to us and these were overrepresented among assisted firms, then our procedure would overestimate the causal effects of trade promotion. Regrettably, we cannot rule out this possibility.

¹⁴ Given that support primarily involves a subset of actions that, at least in the short run, are more likely to result in foreign sales (as opposed to other promotion initiatives such as, for example, the provision of generic information), estimated effects reported below should be more properly interpreted as an upper bound on the true impact of export promotion.

¹⁵ Unfortunately, data on these assistances are not consistently available over the sample period.

¹⁶ These data can then be seen as a census of formal Argentine employment. There is of course some risk of misreporting, which would generate measurement errors. As long as these are systematic across firms, they will be eliminated by the time differentiation implemented in the estimation methods used in this article.

¹⁷ This is the standard classification used in the literature (see, e.g. Álvarez, 2004; Hollenstein, 2005; Observatorio PyME, 2008).

¹⁸ This adds to the evidence reported in the empirical international trade literature suggesting that larger firms are more likely to export (see, e.g. Roberts and Tybout, 1997; Bernard and Jensen, 2004), tend to export more (see, e.g. Görg and Strobl, 2007) and have a higher export intensity (see, e.g. Barrios et al., 2003).

¹⁹ The adjusted R ² values of these regressions range between 0.825 and 0.894, with an average of 0.857.

²⁰ Thus, for instance, Sinani and Hobdari (2008) and Lawless (2009) find that foreign ownership is associated with a higher likelihood to export.

²¹ In general, it can be expected that, over time, growth in the number of total destinations (products) will be associated with introduction of new trade partners (products). In particular, this is indeed the case in our sample.

²² Volpe Martincus et al. (2010) present consistent evidence based on data at the country level.

²³ While the original sample corresponds to the period 2002 to 2006 and has 41 224 observations, these restricted samples only cover the period 2003 to 2006 and have 39 286 and 37 217 observations, respectively.

²⁴ Importantly, given the potential existence of lagged effects of export promotion, the latter sample is likely to produce the cleanest estimates.

²⁵ The adjusted R ² values are similar to those reported for our benchmark estimations.

²⁶ It is well known that the conventional difference-in-differences estimator is based on the assumption that, in absence of the treatment, the average outcomes for firms participating in export promotion programmes and firms not participating in these programmes would have followed parallel paths over time, i.e. both average outcomes would have experience the same variation over time (Abadie, 2005). This can be informally assessed by performing a so-called ‘placebo test’. If we are accurately identifying the impact of these programmes, we should see no difference between the average export outcomes of the treated and control groups in the pre-intervention period. We, therefore, compare the rate of change of each export indicator for firms that have been assisted in at least one sample year with those of nonassisted firms over periods in which the formers have not received yet their first assistance. More specifically, we carry out t-tests for differences in means for the logarithmic differences of the variables in question. Reassuringly, the relevant test statistics suggest that these differences are not significant, i.e. supported and never-supported firms seem to behave similarly when no participation in export promotion programmes takes place. A table with these test statistics is available from the authors upon request.

²⁷ Average effects on export outcomes of firms within these two size categories are not statistically significantly different from each other.

²⁸ See, e.g. Wagner (1995), Argentine Law 24.476/1995 (reformed), Burdisso et al. (2001), OECD (2005) and Gallup (2007).

²⁹ We have also performed estimations based on alternative definitions that only change one of the limits, namely, (i) large firms are those whose number of employees exceeds 250; (ii) small firms are those whose number of employees does not exceed 40; (iii) larger firms are those whose number of employees exceeds 150; and (iv) small firms are those whose number of employees does not exceed 60. The estimation results are similar to those reported here and are available from the authors upon request.

³⁰ In the labour market literature, this is known as Ashenfelter's dip (Ashenfelter, 1978).

³¹ Estimators of treatment effects that weight on functions of the probability of treatment are based on the statistic proposed by Horvitz and Thompson (1952) (Abadie, 2005).

³² Note that, if adding a new destination country (product) requires incurring specific sunk costs of entry, then trading with a larger number of countries (a larger number of products) will reflect higher productivity (Bernard et al., 2006). Thus, by including those export indicators, we are also implicitly partially accounting for productivity differences across (groups of) firms. More generally, we estimate a probit model where the dependent variable is the firms’ assistance status and the explanatory variables are constant, lagged (natural logarithm of) total exports, lagged (natural logarithm of) number of destination countries, lagged (natural logarithm of) number of products exported, lagged binary variables capturing size categories in terms of employment (large is the omitted category), a dummy variable for location (City of Buenos Aires = 1 and 0 otherwise) and year fixed effects. The estimated coefficients and the marginal effects of the aforementioned variables suggest that the probability of participating in export promotion activities organized by ExportAR is significantly higher for small- and medium-size firms and lower for those located in the City of Buenos Aires. In addition, it increases with the number of countries to which the companies export and decreases with their total exports and the number of goods they sell abroad. Detailed tables with these estimates are available from the authors upon request.

³³ More precisely, combining regression with weighting can lead to additional robustness by both removing the correlation between omitted variables and by reducing the correlation between omitted and included variables (Imbens and Wooldridge, 2009).

³⁴ Despite the fact that we are including lagged values controlling for previous export performance, these estimates are also based on the period 2002 to 2006 because we are using export data from 2001 as firms’ export outcomes antecedents in 2002.

³⁵ These procedures also rely for identification on the assumption that there are no time-varying unobserved effects influencing selection into trade promotion programmes and exports.

³⁶ We use here a result from Rosenbaum and Rubin (1983), according to which matching can be performed on the propensity score instead of on whole set of observable characteristics. This allows significantly reducing the dimensionality problem associated with comparison of multiple characteristics. Notice, however, that the propensity score is in fact based on fitting a parameter structure (probit or logit). It is, therefore, necessary to test whether the estimated propensity score is successful in balancing the values of covariates between matched treatment and comparison groups. We assess the matching quality using five alternative tests: the stratification test; the standardized differences test; the t-test for equality of means in the matched sample; the test for joint equality of means in the matched sample or Hotelling's test; and the pseudo R ² and the joint insignificance test of all regressors included in the propensity score specification (see, e.g. Smith and Todd, 2005; Girma and Görg, 2007; Caliendo and Kopeinig, 2008). These tests are reported in an appendix available from the authors upon request.

³⁷ Export promotion activities are likely to have different effects on export performance over firms exporting good bundles with different degrees of differentiation and thus facing varying levels of information incompleteness (Volpe Martincus and Carballo, 2010b).

³⁸ Properly shaping the marketing strategy to meet these markets’ requirements is an information-intensive activity. For instance, firms need to learn and understand the preferences of foreign consumers; the nature of competition in foreign markets; the structure of distribution networks and the requirements, incentives and constraints of the distributors (see, e.g. Artopoulos et al., 2007).

³⁹ Detailed tables reporting these estimation results are available from the authors upon request.