The Analysis of Duration and Panel Data in Economics

Sammanfattning: This thesis is divided into two distinct parts. The first part contains three chapters dealing with the analysis of duration data from an econometric perspective and with application to trade durations. The second part, consisting of the final chapter, focuses on the analysis of panel data and proposes a new test for poolability of the slope coefficients in cointegrated panel regressions. Chapter 2 introduces a novel hazard rate model that is much more flexible with respect to the imposed covariate effects than the conventional cloglog, logit, and probit specifications. In fact, the proposed Pareto hazard model incorporates the most commonly used cloglog and logit specifications as special cases. Using simulated data and data on US unemployment durations, Chapter 2 shows that the Pareto model works very well in practice, and that it also allows for covariate effects that are entirely different from those implied by both the cloglog and the logit specification. Since most durations of economic interest are continuous by nature but artificially grouped into discrete intervals, a further important contribution of Chapter 2 is to show that the proposed discrete-time duration model can be linked to an underlying continuous-time process. More specifically, the choice of hazard specification in the discrete-time framework is motivated by the asymptotic distribution of threshold excesses of a continuous duration variable. Finally, the fact that the Pareto hazard model nests the cloglog as a special case entails an additional advantage. Since the cloglog model is the discrete-time analogue of the Cox model, the Pareto model can be used to test the proportional hazards assumption imposed by the Cox model, even when duration times are coarsely grouped. This particular virtue of the Pareto model is also utilized in Chapter 3. Chapters 3 and 4 (both co-authored with Maria Persson) provide econometrical and empirical contributions to the emerging literature on the duration of trade. Chapter 3 contains an extensive discussion of the ineligibility of the predominantly used continuous-time duration models for the analysis of trade survival, and provides empirical evidence supporting the use of discrete-time models in this context. Based on the findings in Chapter 3, Chapter 4 employs discrete-time methods to analyze the duration of EU imports. As a result, the contributions to the existing literature on trade survival are twofold. First, the empirical analysis in Chapter 4 is based on data that have not previously been used by any of the existing studies on the duration of trade. Second, since there is at this point no clear, commonly accepted theoretical explanation for the ephemerality of trade, Chapter 4 seeks to provide a thorough empirical analysis of the phenomenon, with the intention of thereby facilitating theoretical developments on the subject. While Chapters 2-4 have exploited the rich information contained in panel data by focusing on individual duration times, Chapter 5 (co-authored with Joakim Westerlund) is concerned with a different way of utilizing the information in panel data sets. Specifically, Chapter 5 makes use of the time-series variation in panel data to test whether or not the slope coefficients in panel regressions differ between the cross-sectional units. The chapter proposes a new poolability test based on the Hausman principle, whereby two estimates of the cointegration parameters – one individual and one pooled – are compared. The proposed test allows for dependence among the cross-sectional units, and simulation results suggest that it works very well even in small samples. A further advantage is that the test statistic is based on the maximum difference between the individual and the pooled parameter estimates. This makes it possible to identify the cross-sectional unit that violates the homogeneity assumption in case the null hypothesis of equal parameters is rejected. Thus, subpanels that are suitable for pooling can be identified by applying the Hausman test in an iterative fashion.