Harmonization of the ifo Business Survey’s Micro Data

1 Introduction

The micro data of the ifo Business Survey (IBS) provide unique opportunities to study firms’ economic behavior and expectation formation along a multitude of dimensions. Its advantages are manifold: first, the survey receives approximately 9000 monthly responses from a set of firms that is representative for the German economy.^[1] Second, the number of observations per firm is very large (on average, firms are observed more than 70 times) as response rates are uncommonly high and attrition is low compared to other business surveys. Third, the set of questions is extensive. Every month, firms are asked about, inter alia, current trends and expectations for the development of their general business situation as well as their volume of production, demand situation, and changes in employment and prices. In addition, the survey covers industry-specific questions that, e. g. refer to constraints to business activity, credit supply by banks, or subjective uncertainty. At times, the IBS is supplemented by questions about recent events such as the integration of refugees into the German labor market, the introduction of a statutory minimum wage in 2015, and their perceptions with respect to Brexit.

However, existing studies based on the micro data of the IBS could not tab its full potential because the survey is divided into four separate industry surveys covering the main sectors of the economy (manufacturing, services, retail/wholesale, construction) that could not have been jointly analyzed so far. This is mainly due to the fact that the industry-specific surveys are heterogeneous in the level of observational units (product vs. firm) as well as in the industry classification systems used in the micro data. Hence, researchers usually restricted their analyses to the subset of manufacturing firms (e. g. Pesaran/Timmermann 2009; Bachmann et al. 2013; Carstensen et al. 2013; Strasser 2013; Fidrmuc/Hainz 2013; Bachmann/Elstner 2015; Bachmann et al. 2018; Massenot/Pettinicchi 2018; Huber 2018; Bachmann et al. 2019; and Enders et al. 2019a; Enders et al. 2019b).

The goal of this paper is to harmonize the micro data of the most important industry-specific surveys of the IBS in a way that allows for a joint econometric analysis of the firms. In order to ensure a high level of generality, I use the three main surveys: (a) IBS-IND (2017) covering manufacturing firms since 1980, (b) IBS-TRA (2017) including retailers and wholesalers since 1990, and (c) IBS-SERV (2017) containing service companies since 2004.^[2] For this purpose, I standardize the level of observational units across surveys by aggregating to the firm level. Thereby, I adjust for a methodological break in the retail and wholesale survey in 2006. Based on this procedure, the full panel of data for the retail and wholesale industries can be analyzed for the first time.^[3]

Moreover, I harmonize the industry classification in the micro data of the IBS that are originally coded according to heterogeneous classification systems within and across the different industry surveys. The respective codes are transferred to the German standard classification systems of 2003 and 2008 (henceforth, WZ 03 and WZ 08), which largely correspond to the European “NACE Rev. 1.1” and “NACE Rev. 2” classification systems. Codifying all firms to a unique classification system expands the scope of empirical analyses based on the IBS due to at least two reasons: First, fixed effects can be applied along the same level of industry aggregation in order to flexibly control for industry-specific variation of any kind. Second, the IBS data can be merged to administrative data at disaggregate industry levels.

The remainder of this paper is organized as follows: Section 2 provides information on how the micro data of the IBS and the codes of the harmonization procedure can be accessed. Section 3 presents the harmonization of the IBS micro data and Section 4 provides descriptive statistics of the harmonized IBS micro data set. Section 5 concludes.

2 Data access

The micro data of the IBS can be accessed under strict non-disclosure agreements on-site the ifo-LMU Economics & Business Data Center (EBDC) in Munich, Germany. In addition to the IBS micro data, the EBDC also provides access to the micro data of other surveys conducted by the ifo Institute such as the “ifo Investment Test,” the “ifo Innovation Test,” and the “World Economic Survey” as well as combined panel data sets of the surveys with balance sheet data. The website of the EBDC (https://www.ifo.de/EBDC) provides a comprehensive overview and description of the data sets that are hosted at the ifo Institute and explains the application procedure for data access in detail. The codes of the data set harmonization presented in this paper are also available at the EBDC and can be jointly used with the IBS micro data.

3 Preparing the IBS micro data for panel data research

The goal of this section is to harmonize the micro data of the IBS such that firms in the most important industry-specific surveys can be analyzed jointly. Panel A of Table 1 summarizes the two challenges this task includes: First, in the original micro data, the level of observational units is heterogeneous between the industry surveys and has been changed over time in the retail and wholesale survey. Second, the original micro data provides information with respect to each firm’s main sector of business based on different classification systems across and within the industry-specific surveys. In principle, the harmonization procedure applies to all survey questions in the IBS. For the ease of exposition, the description is exemplified using the questions regarding current and expected business conditions^[4]:

Q1“Current situation: We evaluate our current business condition as [1] good, [0] satisfactory (typical for the season), [- 1] bad.”

Q2“Expectations for the next 6 months: After elimination of purely seasonal fluctuations, the development of our business will be [1] more favorable, [0] about the same, [- 1] more unfavorable.”

3.1 Standardization of observational units to the firm level

In the original IBS micro data, the questions either refer to the firm as a whole or to specific products of the firm. Aiming at the construction of a homogeneous data set, I define the firm as the unit of observation and aggregate the micro data to this level. As the anonymized micro data of the IBS do not allow to sharply discriminate between subsidiaries of larger conglomerates in different locations and stand-alone firms, the term “firm” refers to both types of entities interchangeably. As the identification of manufacturing firms is not unambiguous in many cases during the 1980s, I restrict the dataset to firms that responded to the IBS for more than one time between 1990 and 2017.^[5]

To begin with, I have to cope with a methodological change in the retail/wholesale survey: Since February 2006, firms have been asked to answer each survey question only once while referring to the firm as a whole. Before this date, firms responded to the same set of questions for different products. Answers for all products had to be given on the same questionnaire. As summarized in Column (3) Panel B of Table 1, almost half of the firms (117930/239115 = 49 percent) provided assessments of expected business conditions with respect to more than one product per point in time prior to the break. This fraction decreased to less than 1 percent after February 2006, see Column (4).^[6]

Figure 1:

Correction of the Methodological Break in the Retail/Wholesale Survey. Notes: The figures plot the cross-sectional means (upper graphs) and standard deviations (lower graphs) of current business conditions (left) and expected business conditions (right) of retail and wholesale firms over time. Before January 2006, the reports to the retail and wholesale survey referred to different products. This data is aggregated to the firm level by taking means and rounding to the next integer. Cross-sectional means and standard deviations of the aggregated data are plotted using a dashed line. Accordingly, the solid lines display the cross-sectional means and standard deviations of reports firms after February 2006 that referred to each firm as a whole.

To use the entire universe of the micro data in the retail/wholesale survey, I aggregate the data before January 2006 to the firm level by taking arithmetic means of the answers.^[7] This is feasible as the answers before the methodological break were almost perfectly correlated within firms at a given point in time. If retail or wholesale firms provided assessments with respect to multiple products at the same date, they reported exactly the same expected business conditions for all products in more than 82 percent of cases, see Panel B of Table 1. Hence, it is reasonable to assume that the current and expected business conditions of each firm are well captured by the average reports with respect to different products. Figure 1 plots cross-sectional means and standard deviations of aggregated reported business conditions and expectations of retail and wholesale firms over time. The methodological break between January and February 2006 did not generate differing patterns in the data.^[8] Hence, the data before January 2006 can be aggregated to the firm level without major concerns.

The aggregation is more straightforward for the manufacturing and services surveys of the IBS. While the questions in the services survey refer to the firm as a whole, firms in the manufacturing survey are asked for an assessment regarding a specific product. As the latter usually refers to the main product of the firm and firms are mostly observed only once per month in the survey, the responses are interpreted as referring to the current and expected business conditions of the firm as a whole. In turn, the micro data in the manufacturing (services) survey contain multiple reports per firm and date in 0.8 percent (0.3 percent) of cases only, see Table 1. As these simultaneous reports are usually identical, the respective observations are again aggregated to the firm level by taking means and rounding to the next integer.

4 Descriptive statistics of harmonized IBS micro data

The harmonized sample of the IBS comprises of on average approximately 5500 firms per month between 1990 and 2017 if the data is restricted to firms that responded to the survey in at least two months. As displayed in Figure 2, the number of reporting firms per period is largest in the manufacturing survey (on average approximately 3000 firms), but declining over time. In contrast, the number of firms in the retail and wholesale survey has been relatively constant (on average roughly 1350 firms). Moreover, an average number of 2300 firms has responded to the services survey since its introduction in October 2004. Hence, researchers can build their analyses on more than three times as many firms in the recent waves of the harmonized IBS panel compared to only using the subset of manufacturing firms.

Figure 2:

Number of Firms in the Different Industry-Specific Surveys of the IBS. Notes: This figure plots the number of firms in the respective industry-specific surveys of the IBS over time conditional on responding in at least two months.

Conditional on responding to the survey more than once, attrition is very low and firms are observed for 7.5 years on average.^[10] While attrition of firms is not found to be random, dropout of firms does not appear to be problematic for most research questions. As documented in Column (1) of Table 2, firms reported current business conditions—coded as { - 1,0,1}—that were significantly worse by - 0.14 on average during the last month in the sample compared to the mean of reported business conditions in all other observations.^[11] However, dropout of firms does not appear to be driven by industry-specific shocks because the estimated average condition in the month prior to firms’ dropout is largely unaffected by the inclusion of date fixed effects at the level of four-digit industries, see Column (2). In contrast, the coefficient drops by roughly two thirds relative to its size in Column (1) after controlling for firm fixed effects in Column (3). Hence, firms that drop out of the sample generally performed worse during all other periods compared to the remaining firms. Furthermore, the pattern of attrition does not appear to have changed substantially over time. As can be inferred from Table 3, the survival rates are rather constant across different cohorts of firms.

Moreover, the response rates to the IBS are remarkably high despite of the fact that participation is voluntary. On average, firms answer in 80.5 percent of months and response rates do not differ substantially between firms in the industry surveys on manufacturing (82.3 percent), retail and wholesale (79.3 percent), and services (77.5 percent).^[12]

5 Conclusion

This paper describes how two major obstacles that have prevented panel data research based on the universe of firms in the IBS can be removed: (1) heterogeneity across different industry-specific surveys of the IBS in the level of the observational units (product vs. firm) as well as (2) heterogeneity in the industry classification systems used in the micro data. For this purpose, I aggregate the survey responses of the most important industry-specific surveys covering manufacturing, retail/wholesale, and services to the firm level and transfer the sector identifiers contained in the micro data to the official German industry classification systems WZ 03 and WZ 08.

The harmonized firm level dataset expands the scope for economic research based on the micro data of the IBS to more firms in a broader set of sectors. While existing studies that use the IBS micro data usually restricted the analysis to the subset of manufacturing firms due to the inconsistencies between industry-specific surveys, the harmonized dataset contains of more than three times as many firms in the recent waves of the IBS. The value added of the harmonization procedure is largest for the micro data in the retail and wholesale survey as it adjusts for a break in the methodology in the survey in 2006. This adjustment permits to jointly analyze firms’ survey responses in the time periods before and after the break. For example, Buchheim and Link (2017) make use of this opportunity in order to study the formation of firms’ expectations regarding their own future business conditions across firms in all main sectors of the economy. Moreover, they are able to examine anticipation effects of retail firms to an increase in German value added taxes in 2007 based on data before and after the methodological break.

In addition, the transfer of the sector identification variables to the standard industry classification systemsWZ 03 and WZ 08 is useful for at least two reasons. First, fixed effects can be applied along the same level of industry aggregation in order to flexibly control for industry-specific variation of any kind. Second, the harmonized IBS data can be merged to industry level data from other sources which are usually coded according to the classification systems WZ 03 and WZ 08. Based on this, Link (2019) matches firms in the IBS to data on the distribution of wages in their industry and location in order to identify the degree to which firms are affected by the introduction of a statutory minimum wage in Germany in 2015.