Berlin: Technische Universität, 2012, XIV, 334 S.
The overhaul of the German system of old-age provision raises concerns about the future distribution of pension rights. Therefore, it is critical to gain a thorough understanding of the role pension wealth plays in the individual's total wealth holdings, but also what individual-level and institutional factors facilitate or impede successful pension building that allow individuals to retire with a high enough pension on their own. These analyses require the use of complex micro data that are not readily available in Germany. Despite significant improvements of the data infrastructure, none of the existing sources provides a satisfying basis for the research questions of interest. While survey data, in particular the population representative German Socio-Economic Panel, stands out for the large sample size, the longitudinal design, and extensive contextual information at the individual and household level, it lacks complete life cycle earnings and reliable information on the individual's social security wealth. Administrative pension records, in turn, provide these data but fall short of other standard income and wealth categories as well as household level information. Ideally, survey dataand administrative pension records could be linked over a unique identifier, so-called record linkage. However, record linkage is infeasible because of the absence of a common identifier, but also because it requires the written consent of respondents. A viable alternative to record linkage is statistical matching. The elaboration of a statistical matching procedure to augment augment population representative survey data from the German Socio-Economic Panel (SOEP) with administrative pension records coming from the Sample of Active Pension Accounts (SAPA) is one central contribution of this doctoral thesis. The second contribution was to use these new and unique matched in two empirical applications. The dissertation provides a blueprint for the preparation of a statistical matching. The preparatorysteps included: First, a systematic account of strengths and weaknesses of administrative pension records and survey data; second, the identification of potential matching variables and the correct specification of matching populations in both datasets; and third, a comparison of the distributions and the underlying relationships between the matching variables of interest in both datasets and if necessary a further alignment of these variables. The actual implementation of the statistical matching was based on the population representative Socio-Economic Panel (SOEP) and the Sample of Active Pension Accounts (SAPA) that is representative of all individuals holding a pension account. In order to assess which data fusion technique performs best for the data at hand, we compare a total of four statistical matching and imputation techniques. The unique properties of the linked data allowed for a straight control of the quality of matches and the performance of each technique for the group of retirees. The comparison of the four approaches identified Mahalanobis distance matching to be the best performing technique. Given that the conditional independence assumption holds, social security wealth (SSW) was included in a wealth inequality analysis. The key findings of this first application of the matched data were: The inclusion of SSW almost doubles the level of individual net worth. Inequality (Gini coefficient) decreases by almost 25 percent. Moreover, the results reveal striking differences along occupational lines. Civil servants benefit most from the inclusion of SSW. However, the self-employed stay on top of the wealth distribution, not because of their high social security wealth but rather because of their investments in life insurance policies and property. The second application analyzed the relationship between marital trajectories and retirement outcomes as well as the process of pension building of women.
Keywords: statistical matching, social security wealth, inequality, wealth distribution, divorce
Frei zugängliche Version: (pi)