Temporary exports and characteristics of destination countries: First evidence from German transaction data

This paper uses information on all export transactions of goods by German firms with countries outside the European Union from 2009 to 2014 to document for the first time the patterns of export participation at the firm-good-destination level over time and to investigate the link between the duration of export patterns and characteristics of destination countries. It turns out that only 6.5 percent of all combinations were recorded in each year, while more than half of all patterns are only observed once. In line with theoretical hypotheses, the likelihood of permanent trade patterns increases within a firm with proximity and market size of destination countries. JEL F14


Motivation
A growing literature that is based on data for exports at the transaction level which include information on which goods of which value and which weight are exported by which firms to which destination countries in a year 1 documents that export relationships tend to be highly dynamic in the short run. Using data for Chile, Álvarez, Faruq and López (2010) report that an important fraction of firms start to export new products to new markets each year. Previous experience in exporting a certain product, or exporting to a certain market, increases the probability to export these products to new markets, or new products to the same markets. Again for Chile, Blum, Claro and Horstmann (2013) find that one third of exporters enter into and exit from exporting multiple times, and that most continuing exporters enter and exit specific export destinations multiple times. Rahu (2015) report that in Estonia adding and dropping new products in exports is rife, about half of all firms change their export portfolio annually. Similarly, Buono and Fadinger (2012) find that export relationships are highly dynamic in France, where a large fraction is created and concluded each year. For Hungary, Békés and Muraközy (2012) report that about one third of firm-destination and about one half of firm-product-destination export spells are temporary only. Amador and Opromolla (2010) document frequent switching of products and destinations by firms. Similarly, Damijan, Konings and Polanec (2014) report that in Slovenia the average firm changes about one-quarter of imported and exported product-markets every year. For Spain, Esteve-Pérez, Requena-Silvente and Pallardó-Lopez (2013) find that, while the firm export status is highly persistent, firms' destination portfolio is very dynamic with a median duration of firm-country exporting relationship of two years, but the risk of exiting sharply falls afterwards. Geishecker et al. (2017) report that in Denmark one third of all firmproduct-destination export spells are isolated single-month one-off export transactions that are observed only once in a 49-month time window.
This high degree of short-lived export spells at the firm-good-destination level comes as a surprise because export activities incur sunk costs (e.g., for market research, adoption of the product to local conditions, or finding partners to trade with) that a firm has to pay for each good exported to each market at the start of an export relationship. "As this sunk cost is an investment that can only be recovered from a stable stream of revenues, firms are expected to export a given product to a given destination over a long period of time." (Békés and Muraközy 2012, 232) Evidence cited above point out that, contrary to this, firms often do not export a given product to a given destination over a long period of time, at least not in the countries looked at hitherto. This paper contributes to the literature by adding evidence for Germany, the third largest actor on the world market for exports of goods -keeping in mind that "the credibility of a new finding that is based on carefully analyzing two data sets is far more than twice that of a result based only on one" (Hamermesh 2000, p. 376 The rest of the paper is organized as follows: Section 2 introduces the transaction level data for exports of goods in Germany. Section 3 presents descriptive evidence on the frequency of patterns of exports over the years 2009 to 2014. Section 4 reports results from an econometric investigation of the hypothesis put forward by Békés and Muraközy (2012) that the likelihood of a long trade spell increases with proximity and market size of destination countries.

Transaction level data for exports of goods in Germany
In Germany information on the goods traded internationally and on the countries with which these goods are traded is available from the statistic on foreign trade (Außenhandelsstatistik). This statistic is based on two sources. One source is the reports by German firms on transactions with firms from countries that are members of the European Union (EU); these reports are used to compile the so-called Intrahandelsstatistik on intra-EU trade. The other source is transaction-level data collected by the customs on trade with countries outside the EU (the so-called Extrahandelsstatistik). 2 The raw data that are used to build the statistic on foreign trade are transaction level data, i.e. they relate to one transaction of a German firm with a firm located outside Germany at a time. Published data from this statistic report exports or imports aggregated at the level of goods traded and by country of destination or origin.
The data used in this paper are based on the raw data at the transaction level.
The unit of observation in these data is a single transaction between economic agents located in two countries, e.g. the export of X kilogram of good A with a value  combinations of these firms this may lead to an incorrect classification of a combination as non-existent in the respective year. Therefore, in this paper we will only investigate export transactions with non-EU countries. Here all transactions that exceed 1,000 Euro (or have a weight that exceeds 1,000 kilogram) are registered and, therefore, the problem of "false zeros" does not vanish completely but is much less severe.

Patterns of exports by firm-good-destination over time
In a first step it is documented how many firm-good-destination transactions were  Permanent export in the sense of exports of one good by one firm to one destination country in each year, therefore, is rare. On the other hand, 54.17% or more than half of all patterns are only observed once -one-time exports by a firm of a good to a destination, therefore, are quite common. Perforated pattern that include zeros between ones (like 111001, or 101001) tend to be rare, while patterns with some ones in a row and zeros otherwise (like 111100, or 000011, or 000111) are more common. The big picture reported here is in line with results from similar investigations reported for other countries that are summarized in the introductory section.
[ Table 1 near here] Why do patterns of export by firm-good-destination differ? Why do we only rarely observe permanent export on the one hand and why are one-time exports quite common on the other hand? Obviously, characteristics of the exported goods will play a role here. You will not expect a shipyard to export submarines to a certain destination country each year (leading to a pattern 111111), and you will not be surprised to learn that such an export deal did only happen once over a period of six years (with a pattern like 000100, or 010000). On the other hand, you might expect that, for example, Volkswagen exports cars from a given HS6-category to several destinations each year (leading to a number of patterns 111111).
For confidentiality reasons it is not possible to look at the patterns for different goods separately. However, some evidence on the role of the characteristics of the exported goods for the patterns of export by firm-good-destination might be revealed by an investigation that distinguishes between goods from the three so-called Basic Classes of Goods -Capital goods, Intermediate goods, and Consumption goods. 4 [ Table 2, Table 3 and Table 4 near here] The big picture is indeed somewhat different for the three types of goods. The frequency of permanent export is smallest (5.87%) and the frequency of one-time exports is highest (58.58%) for capital goods, while permanent export is more common (6.12%) and one-time export is less often observed (55.95%) for consumption goods and permanent export is most often observed among intermediate goods (7.93%) where the share of one-time exports is the smallest (52.21%). These inter-class differences, however, are not of an order of magnitude that deserves a closer inspection. It seems that the three Basic Classes of Goods are much too broadly defined to help to understand differences in pattern of export at the firm-good-destination level over time.

Export patterns and characteristics of destination countries
Békés and Muraközy (2012)  from the data at hand. One such hypothesis is that the likelihood of permanent trade (defined here as an export pattern represented by the six-digit string 111111) rises with proximity and market size of destination countries. 5 To test this hypothesis empirical models are estimated. The sample is made of export activities observed at the firm-good-destination level, where each observation is classified either as permanent (with an export pattern represented by the six-digit string 111111) or not. For each observation we measure the proximity of the destination country to Germany and the economic size of the destination market.
Proximity is measured by the distance between Germany and the destination country of exports taken from the CEPII's GeoDist database (Mayer and Zignago 2011). The "distw" -measure is used that calculates the distance between two countries based on bilateral distances between the biggest cities of those two countries, those intercity distances being weighted by the share of the city in the overall country's population (see Mayer and Zignago (2011, p. 11) (2012), like productivity and capital costs). The estimated regression coefficients, therefore, refer to the within-firm variation of stability of export patterns over time due to variation in proximity and market size of the destination country.
The empirical models are estimated by Ordinary Least Squares, i.e. a Linear Probability Model is used. 6 Results are reported in Table 5. In line with theoretical hypotheses the likelihood of permanent trade patterns increases within a firm with proximity and market size of destination countries -for all goods, and for goods from each of the three basic classes of goods. Pattern | Frequency Percent ----------+--------------------