Involuntary retirement and the resolution of the retirement-consumption puzzle

Part of the analysis is performed using the subsample of households where the reference person was not retired in wave 1. This subsample of 770 households repre- sents the set of households ‘at risk’ of retirement during the observation period. [...] Given the data requirements of the fixed effects estimator, and the relatively small sample of households observed to make the transition to retirement during the observation window (283 households), the random effects specification is the primary estimator used for the analysis. [...] The inclusion of the Involuntary indicator allows for sepa- ration of the effects of retirement into voluntary and involuntary components, where the latter is equivalent to an expectations error, an unanticipated negative wealth shock, at the time of retirement. [...] The random effects regression estimator in (6) is applied to the composite index of the total number of hardships experienced by the household, and for the time devoted to various household production activities. [...] The parameters of the model are estimated using maximum likelihood methods, and the APE of the retirement indicators on a range of response categories are reported.