Data and Methods
Data for the following analysis come from the Panel Study of Income Dynamics (PSID; Hill 1992) and the German Socio-Economic Panel (SOEP; Wagner, Frick, and Schupp 2007), two of the longest-running household panel studies in the world. SOEP respondents are interviewed annually, as were PSID respondents until 1997, when the study switched to biennial interview intervals. Several key variables, including the income measures, come from the Cross-National Equivalent File (CNEF; Frick et al. 2007), which provides consistent variables for international comparisons. I use PSID data from survey years 1980-2005 and SOEP data from survey years 1984-2010. All income measures are based on retrospective information for the previous calendar year.
Retirement is defined as exit from work and, more technically, as the first prolonged spell without substantial employment after age 50. I restrict the analysis to persons with substantial employment around and/or after age 50 (at least 15 hours per week on average between ages 48 and 50 or during the first three years in the panel if the respondent entered the study at a higher age). A person is treated as having retired in year t if he worked less than 10 hours per week in t and worked less than 10 hours per week two years later (that is, in t+2). Information on work hours in t+1 is ignored because it is not available after the PSID’s move to biennial interviewing in 1997. Some of the analyses differentiate between voluntary and involuntary retirement. A given retirement event is classified as involuntary if the retiree experienced job displacement or a negative health shock at any time between two years before and one year after retirement. See Heisig (2015, chap. 4) for further details.
I characterize income trajectories around retirement in several ways. First, I consider average income losses relative to pre-retirement income, both before accounting for taxes and transfers (pre-government income) and after including them (post-government or disposable income). The underlying income measures are needs-adjusted according to the modified OECD scale. For a man retiring in year t, the relative change (in per cent) from a years before until b years after retirement is calculated as follows:
In principle, this indicator can vary between -100 and +ro. I cap it at +100 to limit the influence of (a very small number of) outliers. Second, in addition to average relative changes, I also look at the proportion of individuals whose disposable income declines by more than a third (‘large loss’) and more than half (‘very large loss’). These indicators are useful for identifying the prevalence of potentially critical declines in disposable income and more generally for studying the inequality of income trajectories. Finally, I also report the proportion of men who enter income poverty after retirement. These entry rates are calculated only on the basis of those who were not poor in the pre-retirement reference year t — a. ‘Poverty’ is defined using a ‘European-style’ threshold of 60 per cent of median needs-adjusted disposable income for the whole adult population.
Results below are based on four-year and six-year income changes, that is, on changes between a pre-retirement reference year and four/six years later. I use t — 1 as the reference year for workers retiring in odd-numbered years and t — 2 for workers retiring in even-numbered years. This approach ensures consistency across countries and over time despite the PSID’s switch to biennial interviews (for details, see Heisig 2015, chap. 4). Four- year changes and six-year changes are similar, and I do not analyze them separately but cluster standard errors on the person-level to correct for serial correlation. Men are included in the analysis if at least one of the changes (four-year or six-year) is available for them (unbalanced panel).
-  PSID/CNEF data from the 2007 wave were available at the time of writing. Unfortunately,exploratory analyses revealed severe problems with a crucial income component—household private pension income—in these data.