A new perspective on the exporter productivity premium: online trade

ABSTRACT

We use a unique firm-level data set including 9000 companies from 26 European Union countries covering four different sectors to take a close look at the relationship between online exports and productivity. The online exporter productivity premium is estimated using different techniques (ordinary least squares, quantile regressions and robust estimation). Results consistently indicate that the estimated online exporter productivity premium is statistically different from zero, positive and significant from an economic point of view. European online exporters, according to these results, are approximately 2% more productive than non-online exporters. Productivity differences between firms could be related to variables that are not included in the empirical model. More research would be needed to address this issue in the future.

KEYWORDS:

• e-commerce
• exporter productivity premium
• robust estimation

JEL CLASSIFICATION:

• F14
• D24
• L81

I. Introduction

Since its emergence in the mid-nineties, e-commerce has been considered as a trade facilitator due to the significant reduction in information and communication costs it produces. However, online trade has received little attention, mostly due to the lack of appropriate data at the firm level. The growing literature analysing the relationship between exports and productivity (see Wagner 2012 for a recent survey) has consistently found that exporters tend to be more productive than non-exporters. However, the analysis of this relationship for firms operating online has received hardly any attention.

Business-to-Business (B2B) e-commerce can improve productivity, among other reasons, by improved inventory control, by allowing the procurement of more suitable and cheaper inputs and by lowering transaction and information costs. Similarly, Business-to-Consumers (B2C) e-commerce can also boost productivity by reducing time and transaction costs, by reducing handling costs and by facilitating better customer relationships. Although e-commerce in the European Union (EU) has been growing at impressive rates in the past decade, this is happening in domestic markets while cross-border e-commerce seems to be lagging behind. In this note, we look at the relationship between online exports and productivity.

II. Data and methodology

The data used in this report were collected on the basis of a specific questionnaire applied to a sample of 8.705 European firms in 26 Member States in early 2015 (European Commission, 2015). Among these, 3.945 sell online and 1.816 do it across the border. The data are unique; there is no comparable data source covering online trade for European firms. Four sectors can be identified: (i) Manufacturing; (ii) Wholesale and retail trade; (iii) Accommodation and food; and (iv) Information and communication. Productivity is measured as total revenue per employee. More appropriate measures of productivity such as value added per employee (or per hour worked), or total factor productivity, cannot be computed because of the lack of appropriate information in the database.¹ Some companies do not report revenue or the number of employees. Hence, the final sample is composed by 6.933 firms, of which 1.448 are online exporters.

In the literature on international trade and firm heterogeneity, the exporter productivity premium is defined as the relative productivity differential between exporting and non-exporting firms from the same industry and of the same size. Econometrically, it is estimated by regressing the log of productivity on a dummy variable equal to one when the firm is an exporter and zero otherwise (ISGEP 2008). In addition, the number of employees and its squared value and dummy variables for the sectors and countries are included to control for firm size, industry affiliation and country effects.

Empirical firm-level productivity distributions tend to show extreme values, which may condition the results from standard econometric estimators by showing heterogeneous error variances. In fact, both the Breusch-Pagan and White tests reject the null hypothesis of homoscedasticity.² In order to deal with these concerns, and using an identical model specification, we follow two approaches. First, we use the wild bootstrap (Wu 1986) to test the significance level of the main variable. The idea behind this procedure is to resample the dependent variable based on the values of the residuals while keeping the regressors at their sample values. Second, we use robust estimation methods to avoid contamination from observations with large residuals. In this case, we first show the results using quantile regressions to assess the validity of the results in different positions of the distribution of the dependent variable (Powell and Wagner 2014). We run these regressions at the 0.25, 0.5, 0.75 and 0.9 quantiles. Then, we rely on a robust mm-estimator of the exporter productivity premium to take a more precise account of outliers and heteroscedasticity (Verardi and Croux 2009).

In what follows, we apply this methodology to investigate the online exporters’ productivity premium. First, we compare the results from a simple OLS regression with alternative estimations that include corrections for the SEs. Second, we compare methods that control for the presence of outliers and non-constant error terms: quantile regressions and robust estimation (Wagner 2015). Then, we use the robust mm-estimator to check whether the productivity premium for online exporters is still present when we perform several segmentations in the database.

III. Results and discussion

The first set of results is reported in Table 1. The first column of the table shows the standard OLS estimation and indicates that the estimated online export premium is positive, statistically significant and relevant from an economic point of view – the online exporters of our database are on average (ceteris paribus) 2.8% more productive than non-exporters.³ Due to concerns about the validity of the estimates in the presence of heteroscedasticity, we performed several additional estimations. Columns 2–5 show the results of different ways to compute robust SEs. Estimates in column 2 are derived from a Huber/White estimator and columns 3–5 cluster SEs differently. Apart from column 4, the productivity premium for online exporters is consistently statistically different from zero.

Table 1. Online exporters’ productivity premium, basic estimates.

		SEs
			Clustered at
	OLS	Robust	Country	Sector	Country–sector
	(1)	(2)	(3)	(4)	(5)
Online exporter	0.0279**	0.0279**	0.0279*	0.0279	0.0279**
	(0.0137)	(0.0137)	(0.0146)	(0.0151)	(0.0135)
Size	−5.42e-05	−5.42e-05***	−5.42e-05**	−5.42e-05***	−5.42e-05**
	(3.32e-05)	(1.62e-05)	(2.56e-05)	(4.99e-06)	(2.14e-05)
Size^2	3.81e-09	3.81e-09*	3.81e-09	3.81e-09*	3.81e-09
	(4.56e-09)	(2.11e-09)	(2.96e-09)	(1.58e-09)	(2.75e-09)
Constant	0.404***	0.404***	0.404***	0.404***	0.404***
	(0.0310)	(0.0429)	(0.00918)	(0.0421)	(0.0374)
Observations	6933	6933	6933	6933	6933
R-squared	0.056	0.056	0.056	0.056	0.056

Notes: All regressions include sector and country dummies. SEs in parentheses.

***p < 0.01, **p < 0.05, *p < 0.1.

To confirm the validity of these results, we employ the wild bootstrap method to test whether the main variable of interest, the online exporter productivity premium, is still statistically different from zero when explicitly taking heteroscedasticity into consideration. Table 2 shows that the null of a zero productivity premium for online exporters is rejected at the 5% level using several specifications for the residual transformations implied by the wild bootstrap and also with different number of replications.

Table 2. Significance levels using the wild bootstrap.

	Method for weighting residuals
Replications	Normal	Rademacher	Mammen	Webb
1,000	0.0320	0.0460	0.0450	0.05
5,000	0.0388	0.0422	0.0462	0.0432
10,000	0.0386	0.0402	0.0432	0.046

Note: The null is that the coefficient of online exporter is equal to zero.

The second set of results is presented in Table 3. Columns 1–4 show the estimated coefficients for the different quantiles. When we move up in the productivity distribution, the differences widen up as well, passing from a productivity premium of 0.6% in the first quantile (0.25) to a 5.9% premium in the last quantile (0.9). The last column reports the coefficients from the robust estimator that avoids contamination from outliers and accounts for heteroscedastic disturbances. The estimated online exporter productivity premium is again statistically highly significant and large from an economic point of view. In this case, the point estimate is smaller but still significant in economic terms. The online exporter productivity premium is 0.8%. These results are in line with previous findings concerning the exporter productivity premium in different countries and industries (Wagner 2012).

Table 3. Online exporters’ productivity premium, robust estimates.

	Quantile
	(0.25)	(0.5)	(0.75)	(0.9)	Robust
	(1)	(2)	(3)	(4)	(5)
Online exporter	0.00591***	0.0108***	0.0297***	0.0570**	0.00819***
	(0.00205)	(0.00362)	(0.00866)	(0.0253)	(0.00209)
Size	1.06e-05**	2.32e-05**	4.84e-05**	−7.01e-05	0.000108***
	(5.14e-06)	(9.01e-06)	(2.06e-05)	(4.79e-05)	(1.65e-05)
Size^2	−2.13e-09***	−4.05e-09***	−6.95e-09***	2.51e-09	−4.75e-08***
	(5.86e-10)	(1.02e-09)	(2.38e-09)	(5.16e-09)	(8.71e-09)
Constant	0.0806***	0.207***	0.403***	0.743***	0.165***
	(0.00247)	(0.00570)	(0.0227)	(0.0759)	(0.00848)
Observations	6933	6933	6933	6933	6933
R-squared	0.254	0.273	0.164	0.059

Notes: All regressions include sector and country dummies. SEs in parentheses.

***p < 0.01, **p < 0.05, *p < 0.1.

Next, we perform a series of robustness checks to analyse whether this online exporter productivity premium is consistent along different segmentation variables. In all these cases, for simplicity and accuracy, we focus only on the results obtained when using the robust estimator. The first case, reported in Table 4, refers to sector differences. The table shows that the online exporter productivity premium is positive and statistically significant in the four sectors considered, ranging from 0.3% in the case of Accommodation and food to 1.1% in the case of Wholesale and retail trade.

Table 4. Online exporters’ productivity premium, by sector.

	Manufacturing	Wholesale and retail	Accommodation and food	Information and communication
Online exporter	0.00830**	0.0113**	0.00345*	0.0104**
	(0.00374)	(0.00489)	(0.00200)	(0.00433)
Size	0.000106***	0.000163***	5.54e-05***	1.57e-05
	(2.17e-05)	(4.28e-05)	(1.63e-05)	(2.67e-05)
Size^2	−3.41e-08***	−7.18e-08**	−4.40e-08***	2.41e-09
	(5.97e-09)	(2.93e-08)	(1.00e-08)	(6.64e-09)
Constant	0.161***	0.236***	0.104***	0.162***
	(0.0112)	(0.0261)	(0.00636)	(0.0164)
Observations	1877	3099	810	1147

Notes: All regressions include sector and country dummies. SEs in parentheses.

***p < 0.01, **p < 0.05, *p < 0.1.

A second-segmentation exercise is to perform the same analysis separating the firms by age. The database identifies the firms that have been created before and after 2012. In our case, we define the latter as young firms and the former as old firms and re-run the estimations taking these groups into account. Results are reported in Table 5 and show that although older firms that are exporting online show a positive and statistically significant productivity premium of 1.1% over old firms that are not exporting online, this is not happening in the case of young firms. This result is consistent with existing evidence from the offline world where exporter firms tend to be large, and size is normally reached after some years of operations (Mayer and Ottaviano 2007).

Table 5. Online exporters’ productivity premium, by age.

	Young	Old
Online exporter	−4.22e-05	0.0106***
	(0.00416)	(0.00228)
Size	−2.77e-05	8.04e-05***
	(0.000122)	(1.58e-05)
Size^2	1.53e-07	−2.65e-08***
	(2.12e-07)	(7.07e-09)
Constant	0.139***	0.170***
	(0.0171)	(0.00982)
Observations	1026	5894

Notes: All regressions include sector and country dummies. SEs in parentheses.

***p < 0.01, **p < 0.05, *p < 0.1.

An additional dimension to control for is the type of product the firm is selling. In this case, it is possible to distinguish firms that are selling goods from those firms that are selling services (these categories are not mutually exclusive). Table 6 reports the results when we separate the firms in these two categories. Again, the estimated productivity premium is positive and statistically significant for online exporters of goods (0.7%) and of services (0.9%). The difference in the estimated premium among these groups could be explained by the fact that service firms include firms selling not only services that are provided physically, but particularly firms that are selling digital goods, downloaded directly from the clients into their computers and which do not require physical distribution (digital media and software, basically).

Table 6. Online exporters’ productivity premium, by type of product.

	Goods	Services
Online exporter	0.00741***	0.00931***
	(0.00219)	(0.00347)
Size	8.90e-05***	4.43e-05***
	(1.65e-05)	(1.57e-05)
Size^2	−2.70e-08***	−7.78e-09***
	(7.12e-09)	(2.39e-09)
Constant	0.168***	0.171***
	(0.00941)	(0.0114)
Observations	6482	3522

Notes: All regressions include sector and country dummies. SEs in parentheses.

***p < 0.01, **p < 0.05, *p < 0.1.

IV. Conclusions

In this article, we offer some first insights into EU companies’ online internationalization. Although an important limitation of the analysis is not being able to control for individual firm effects, the findings described in this note are important and novel. The productivity premium for online exporters is relevant: online exporters, according to our results, tend to be approximately 2% more productive than non-online exporters. Although this figure seems plausible, productivity differences between firms could be related to alternative variables not included in the empirical model, either because information is missing or because they are unobservable to the researcher. More research would be needed to address this issue in the future, in particular to explain the model behind the data to detect the main differences between traditional and online trade.

Acknowledgements

We thank Martin Falk for many constructive comments as well as participants at the 17th ETSG annual conference in Paris for helpful suggestions. We also thank an anonymous referee who made relevant suggestions for improvement. However, remaining errors are our sole responsibility. The views and opinions expressed in this note are the authors’ and do not necessarily reflect those of the JRC or the European Commission. This research did not receive any specific grant from funding agencies in the public, commercial or not-for-profit sectors.

Disclosure statement

No potential conflict of interest was reported by the author.

Notes

¹ Controlling for firm fixed effects, however, could absorb much of the differences derived from firms’ unobserved characteristics. Unfortunately, we only have a cross-section of firms.

² Available from the authors upon request.

³ The estimated coefficient for the online exporter dummy has been transformed by 100(exp(ß)-1). The transformation shows the average percentage difference in labour productivity (ceteris paribus) between online exporters and non-online exporters.