The Micro Data Linking-Panel: A Combined Firm Dataset

3.1 Basic information

Column 3 of Table 2 shows a total of 246.556 surveyed firms that are available in the MDL-Panel for reference year 2013. For example in NACE section B and C, that is mining/quarrying and manufacturing, three structural business surveys were combined to a total of 22.517 firms. This population consists of 16.625 firms with at least 20 employed persons originating from a complete survey with a cut-off limit regarding a firm’s investments and an associated sample survey regarding other structural variables. The sample survey with a reduced questionnaire for 5.892 firms with less than 20 employed persons was added to complement the population. This allows, among others, for an analysis of smaller firms compared to larger firms, which is at the moment not possible with the available individual statistics of the RDC.

Table 3 shows that in the time dimension there are 47.106 firms which were surveyed continuously in each reference year from 2008 to 2013. The share of these firms, compared to all surveyed firms, is roughly one fifth per reference year of the MDL-Panel because most sample surveys rotate parts of their covered population after a fixed or variable amount of time. This is done in order to smooth the response burden among the reporting firms. Note that sample survey designs are commonly optimized with respect to sampling variance, which gives a very large firm a higher probability to be observed in each reference year compared to a very small firm.

3.2 Data integration and validation

As in Germany the structural business statistics are organized domain-specific with distinct laws for each target population, linking the related surveys to create the “backbone” of the MDL-Panel was not without issues. Each survey is conducted based on a specific law which imposes restrictions on the contents of the survey. It is obvious that some structural variables relevant in a certain industry like manufacturing are not necessarily relevant for the service sector as well and hence the respective laws restrict the scope of the survey to the variables most meaningful for covering the target population. This is also a reason for the limited number of variables available in MDL across different industries. Also we faced the problem that due to time inconsistencies it was not always possible to link the firms originating from the structural business statistics surveys to the other business statistics, which resulted in the loss of a few firms compared to the actual number of surveyed firms in the individual statistic. Even though the collected survey data is cleaned by means of automatic and manual plausibility checks within the scope of each statistical domain, we conducted further validation checks considering the time dimension like attrition, changing industry affiliation or demographic events to ensure the validity of the data.

3.3 International trade in goods data at the level of the firm

A key feature of the MDL-Panel is the availability of a firm’s import and/or export values, which are derived from the international trade in goods statistics. However, in Germany the firm is not necessarily the reporting unit in this statistic and certain assumptions have to be made to make this data applicable for a firm level analysis at all. First, the system of data collection varies between goods traded with members of the EU and goods traded with third countries outside the EU. While in the latter case the data is compiled via customs declaration, reporting obligations for exports and/or imports within the EU arise for an entrepreneur by being subject to value-added tax in Germany. ^[5] Both systems, however, share the registration of all trade in goods transactions via the firm’s tax id. Now, this means that tax schemes like group taxation (“Organschaft”) become an issue for Official Statistics as they comprise multiple firms in order to form a single tax entity (“VAT group”). In the following, the head of the VAT group (“Organträger”) is referred to as the parent and any other firm within the VAT group is referred to as a subsidiary (“Organgesellschaft”).

In the statistical domain, the parent’s tax id appears by customs declaration for traded goods with third countries outside the EU. Regarding the international trade in goods with EU members, the parent usually obeys the reporting obligations itself. As a VAT group consists of at least two entities, we do not know whether the parent itself caused a goods movement within its firm or any other subsidiaries with their firms as they all share the tax id of the parent. Most parents are located in NACE group M 70.1 (Activities of head offices) while frequently having subsidiaries in manufacturing or wholesale and retail trade. We assume that firms in the service sector are not predominantly involved in trading goods but rather trading services. If we do not adjust the firm level data to this assumption, a large amount of the actual traded goods would be attributed to these firms. ^[6] To evade this data collection issue, we reallocate the exports within a VAT group from a parent in the service sector to its subsidiaries in NACE sections B to G, if there are any. After this reallocation only 4 % of the total exported goods volume remains in the service sector (Kaus/Leppert 2017). Regarding imports, the redistribution within a VAT group is also carried out but regardless of the industry affiliation of the parent’s subsidiaries. The redistribution is done via a key variable available from the business register, which represents the estimated turnover for each member of the VAT group (Wagner 2004). Hence, the firm’s export and import values are interconnected with turnover, which is, given the available data source, the best measure to provide trade in goods data at the firm level. Our redistribution is conceptually comparable with the official approach carried out by the trade by enterprise characteristics statistics (TEC) (Allafi 2012). We used the business register to link the international trade data to the structural business statistics surveys via the tax id.

Note that the problem to attribute exports and imports within a VAT group to the respective firms extends to the available breakdown by country of destination (exports) and country of origin (imports). Given the data, it is not clear which firms within a VAT group do trade with the countries reported by the parent’s tax id and we have to assume that all eligible subsidiaries trade with each partner country according to their relative share of the VAT group’s turnover. The reallocation approach described above remains unaffected and tends to lead to a higher number of firms engaged in foreign trade and a higher number of listed partner countries respectively.

There is a possibility to improve the allocation of exported goods with EU members within a VAT group by using VIES data. ^[7] This data source allows to breakdown the export shares of the parent and each subsidiary within the VAT group and has already been used for crafts statistics with promising results (Feuerhake/Giebenhain 2017). Not only does this warrant more exact export figures at the firm level but also provides the actual countries of destination for each subsidiary. However, for goods traded with third countries outside the EU there is currently no remedy to circumvent the described allocation issue.

3.4 Extrapolation of linked survey data

As the MDL-Panel includes the sampling weights of each respective sample survey, the firms can in general be extrapolated to the survey’s target population. However, related figures published by the Federal Statistical Office cannot be exactly reproduced with these sampling weights. For example, a complete and sample survey with the same target population is extrapolated such that the sampling weights of the sample survey are tied to a certain variable population total of the complete survey. Hence, it is not possible to produce the exact figures with the sampling weights of the sample survey that we linked to the complete survey. Moreover, the sampling weights are designed only for extrapolating variables from the structural business statistics surveys and not for variables originating from the other linked business statistics (see Table 1). For example the population total value of imports and exports, originating from the international trade in goods statistics, cannot be reproduced with the sampling weights originating from the structural business statistics surveys. In the first project we tried to solve this issue with adjustment factors to improve the consistency of MDL figures with published figures like IFATS (Jung/Käuser 2016). Due to the variety of statistical domains and different survey methodologies this approach proved unfeasible. In the second project we considered the General Regression Estimator (Deville/Särndal 1992) which can ensure consistent extrapolated figures across different target populations. However, this approach yielded only reliable results for highly aggregate figures. On the other hand, the trend of the total extrapolated import and export values from 2009 to 2013 can be captured with the sampling weights and is also in line with figures published by the Federal Statistical Office, although underestimating the total import and export values by 2 % and 4 % respectively (Kaus and Leppert 2017).

3.5 Selected applications of the MDL-Panel

Kaus and Leppert (2017) use the MDL-Panel to assess the contribution of firms which are involved in international trade regarding structural variables like value added, wages and salaries or gross investment in tangible goods. Furthermore, they demonstrate the heterogeneous nature of firms which are involved in international trade by characterizing them with respect to structural variables as well as firm characteristics like type of trader, group affiliation or the number of countries traded with. The relationship between a firm’s productivity and its type of international trade activity (i. e. being an importer, exporter or both) and number of countries traded with is examined as well.