CAED Research and Data Agenda
(contributed by John Earle)
Executive and
Planning Committees
History of CAED
Previous conferences
2008 - Budapest
2006 - Chicago
2005 - Cardiff
2003 - London
About CAED
Ten years ago, an economist leafing through the major professional journals would have been hard-pressed to find many articles using firm-level data. Particularly unusual were studies using comprehensive panel data on all enterprises in a single economy, and still rarer, practically unknown, were analyses of such data for multiple countries. One of the most important developments in economic research over the last decade is the growing analysis of such databases. The new data provide the opportunity for revisiting many of the classic empirical questions in economics, this time with data at the appropriate level of aggregation – the business that is the decision-making unit. The data also permit and stimulate the analysis of many new questions that economists could not even dream of addressing with previous data resources. Together, the data and accompanying research agendas are transforming much of economics and public policy analysis.
New Types of Enterprise Data
While firm-level data sets have been studied by economists for a long time, the quantity, quality, and availability of the data have all vastly expanded in recent years. Earlier data sets on businesses tended to be small sample surveys focused on specialized topics and containing only cross-sectional information. Individual researchers frequently assembled these data on their own, or they organized the collection for the purpose of a particular research project. Limited funding generally resulted in small-scale data sets, little standardization of variables, and little sharing of data among researchers. Moreover, despite the usefulness, indeed the necessity, of such data for answering a variety of questions, the tendency of the economics profession to award little credit for data collection meant that most economists felt only weak incentives to expend effort in this area. It was much easier to work with standard, existing databases on households or industry, regional, and economy-wide aggregates.
The new data sets on businesses tend to rely on governmental sources, and as a consequence they are more systematic and much larger in scale. Both the national statistical offices and the agencies administering government program have regularly collected data on firms and establishments in order to monitor the macroeconomy, collect taxes, and evaluate policies. But researchers were unable to obtain access to the business-level information. A number of recent developments – growing openness of governmental agencies, increasing pressure from the research community, improving technologies to process data and protect confidentiality, and mounting emphasis on empirical research, particularly at the micro level – have led to accelerating access and analysis of the microdata.
The new data sets have several important advantages. Numbers of observations are much larger, permitting stronger conclusions from a given analysis. In some cases, the data are “universal,” covering the entire population of businesses in a country or a particular sector (all of manufacturing industry, for example). In other cases, these databases make use of universal sampling frames, which solves one of the biggest problems confronting researchers carrying out their own firm surveys: constructing a representative, random sample.
The new data are also usually longitudinal, containing multiple observations over time for a given business. This panel dimension of the data facilitates statistical techniques to control for unobserved differences across firms and it permits researchers to study dynamics – changes in firm behavior and responses to shifts in the firm’s operating environment. When combined with universal coverage, a dynamic analysis can also address questions involving entry and exit of businesses from the market. The dynamics of firm turnover are not only of great potential consequence for economic growth and worker welfare, but they may also be important statistical factors to control for, as analyses based on only continuing firms are likely to be biased if exit or entry is nonrandom.
A significant drawback of the new data sets is the limitation to a relatively small set of variables used for constructing aggregates and evaluating programs. Moreover, while the data have become much more readily available for researchers than they were in the past, large obstacles to access persist in many countries. Cross-country comparisons are facilitated by some degree of standardization in the collection of data for national income accounting, but idiosyncrasies in definitions of variables and in the rules for inclusion of observations (in the sample or universe) remain. Thus, individual surveys focused on particular topics will continue to play an important role, as will the collaboration of researchers with knowledge of local idiosyncrasies in data, policies, and institutions with the data providers to link, harmonize, improve, and make available many types of data.
One important sub-class of enterprise data sets deserves particular mention: linked employer-employee data (LEED). Such data contain information on the composition of employment within firms, including the characteristics and, usually, the wages of workers. The information is useful for controlling for differences across firms in the workforce and for studying many questions involving the internal organization and compensation structures of firms. When the data contain longitudinal information on both employers and employees, it is also possible to control for unobserved firm attributes in analyzing worker outcomes and to study job mobility of workers across firms. In essence, the data permit analysis of both the demand and the supply sides of the labor market.
New CAED Research
The new data permit many of the fundamental questions in empirical economics to be studied at the level of the enterprise, the decision-maker for many questions underlying economic growth and welfare. In traditional economics, for example, the entire economy (or an entire industry) is modelled as if it were a single firm with a single production function it uses to transform inputs into output. With the aggregate data corresponding to such a model, estimation of basic parameters is at best difficult, because sample sizes are too small to permit reliable inference. More importantly, the assumption of a common technology across diverse industries is untenable. To take one example, in projecting the impact of a rise in oil prices on employment levels, the researcher needs to estimate cross-elasticities of input demands, which depend on technological ease of substitution and market factors that vary across industries. Using industries as observations to estimate these relationships for the economy as a whole fails because the industry and economy-wide aggregate relationships are in general different, so no inferences are possible. Only by moving to the firm-level, using the data corresponding to the decision-maker and estimating separately by groups operating in relatively homogeneous markets and with homogeneous technology, can these problems be avoided.
Many important economic questions cannot even be empirically posed in the absence of appropriate enterprise data. One of the earliest lessons from firm-level research is that firms display enormous heterogeneity in their performance and behavior, even within narrowly defined categories and industries. The diversity of outcomes contradicts standard theoretical models of competitive industries and frictionless environments as well as empirical analyses based on aggregate (sectoral or regional) data. The factors leading some firms to be more productive than others are fundamental determinants of economic growth, and they are fundamental puzzles for economists, but they can only be satisfactorily investigated with firm-level panel data. Among the factors that the new literature is addressing are technology and R&D, ownership and corporate governance, and government policies and institutions.
Another set of questions that require enterprise data, ideally with universal coverage, concern industry dynamics. Stretching back to Schumpeter, there had been much casual discussion of the potentially important role played by the creative destruction process in capitalist economies. But the data required to investigate the nature of exit and entry have only recently become available. The important research issues concern the pace and the determinants of the firm turnover process, as well as its consequences for growth: the shares of entrants and exiters, their relative productivity levels, and the magnitudes of costs of entry, exit, and remaining in the market. Closely related policy questions involving these costs include regulatory barriers to entry, financial constraints on growth, competition from entrants and international trade, provision of complementary institutions, and softness of budget constraints.
A final set of issues involves the consequences of the firm-level restructuring and reallocation processes for workers. Do employees gain when their employer’s productivity improves? Or does the improvement more often come at their expense? These questions can be addressed in the context of any of the factors or policies affecting firm performance and industry dynamics. Using firm-level data, the outcomes for levels of employment and average wages of the firm may be estimated. Using LEED, it is possible to estimate heterogeneous outcomes for different types of workers and, in some cases, to trace the mobility patterns and long-term consequences for displaced workers.
Ultimately, the analysis of firm-level data promises new insights into the causes of differences in the “wealth of nations.” Most firm-level research has been carried out for single countries, but many of the important decisions underlying international differences in productivity and incomes reside within firms. International comparative research on enterprise data can exploit variation in policies and institutions and their effects on enterprise behavior, thus revealing the reasons for cross-country differences in firm performance and industry dynamics.
< Back to top >