Measuring the Impact of Longevity Risk on Pension Systems: The Case of Italy

Abstract

This article estimates the impact of longevity risk on pension systems by combining the prediction based on a Lee-Carter mortality model with the projected pension payments for different cohorts of retirees. We measure longevity risk by the difference between the upper bound of the total old-age pension expense and its mean estimate. This difference is as high as 4% of annual GDP over the period 2040–2050. The impact of longevity risk is sizeably reduced, but not fully eliminated, by the introduction of indexation of retirement age to expected life at retirement. Our evidence speaks in favor of a market for longevity risk and calls for a closer scrutiny of the potential redistributive effects of longevity risk.

Notes

There are exceptions; a notable one is Alho et al. (2008).

denotes the mortality for individuals of age x in year t, where mortality is the probability that a person aged x and alive at the beginning of the year dies within the end of the year. is instead the survivor probability for individuals of age x in year t, which is the probability that an individual will be alive at age x given that he or she has survived up to age . Survivor probabilities are derived recursively for individuals aged 65 and over: If then if then . Frequencies of death for individuals of age x at time t are determined as first differences of survival probabilities: . Finally, life expectancy at 65 is defined as .

http://dati.istat.it/

http://www.mortality.org

See Appendix A for a full description of the adopted identification and estimation strategy.

We are grateful to an anonymous referee for having raised this point.

http://www.bancaditalia.it/statistiche/indcamp/bilfait.

See Appendix B for the analysis of the surveys and their elaboration.

We have also experimented with specifications allowing the long run to be function of the demographic structure of the population. As the results obtained were not stastically strong and unstable over time, we have preferred the simple AR model.

In our computation we have also taken into account the constraint posed by the minum pension and its evolution over time. We obtained the data relative to the minimum monthly pension in Italian liras (from 1994 to 2001) and in euros (from 2002 to 2011) from INPS website. From 2012 onwards, we let the minimum pension of 2011 grow at the constant ECB inflation rate of 2% already discussed when forecasting average salaries, consistent with the assumption that revisions of minimum pensions are made to preserve retirees’ purchasing power. Moreover, future pensions have been revaluated using projected inflation.

In formulas, we have that retirement age in equation 7 is adjusted by a term h reflecting the difference between expected life at 65 from 2012 to 2050 and expected life at 65 in 2009 (which is roughly equal to 20).

Note that the coefficient is always automatically adjusted to the actual “projected” mortality, while the indexation of retirement age depends on the mean expected life in all simulations.

Obtained from the World Bank's website: http://databank.worldbank.org/ddp/home.do.