For the evaluation of pension reforms and reform options it is necessary to understand their effect on retirement behavior. A common method used for estimating retirement behavior is the option value model of retirement. The probability to retire at a given age is estimated by using a probit model with an option value reflecting the financial incentives as explanatory variable. Specifically, the option value describes the utility gains from retiring now versus in the future.
Using administrative data from the German public pension provider (Versichertenkontenstichprobe) this project has two major objectives. First, the effects of the option value on both the labor market exit and the pension claiming decision are estimated. Thus, we are the first to extend the option value model by separately determining the optimal pension claiming and the optimal labor market exit age. Second, in addition to the Stock and Wise utility function which is the standard utility function used in this literature we use the Cobb-Douglas utility function following the argumentation of Börsch-Supan (MEA-DP 27-2014) that the Stock and Wise utility function is degenerated.
Additionally, we control for individual’s eligibility for specific pathways into retirement directly. We deal therefore with the strong correlation observable between the early/normal entitlement ages and actual labor market exit and pension claiming behavior. Former studies, which used age (group) dummies instead, linked pronounced labor market exit/pension claiming ages to the respective ages.
We find a strong relation between the option value of retirement and a person’s labor market exit and pension claiming behavior. We find furthermore that the labor market exit behavior reacts stronger to changes in the financial incentive than the pension claiming behavior. In contrast, the pension claiming behavior is adapted more to changes in the early entitlement ages.
Unfortunately the simultaneous estimation of all preference parameters fails for both utility functions. Overall, the Cobb-Douglas utility function provides better convergence properties and the parameter estimation succeeds if additional assumptions are made. The findings so far are therefore not entirely satisfying and leave room for future research. There are several possibilities to proceed. First, we could reconsider the design of our utility function by e.g. using age-specific leisure preferences. Second, we could use a data set (e.g. SHARE_RV) which includes additional information on individual’s household context and wealth other than social security wealth. Another route would be to develop a full dynamic optimization model.
The current findings are summarized in a chapter which is part of the dissertation of Johannes Rausch.