funded systems leads to better asset allocation results. This further helps to determine
in which direction reforms should alter a pension system—that is, toward more
individually managed funding (and income provision) or not.
To evaluate investment performance, we use a utility-based welfare benchmark.
Alternative monetary-based benchmarks such as expected wealth (or quantiles of the
distribution) at retirement (e.g., as in Poterba et al., 2007, or in Watson and
Naughton, 2007) are not suitable because we assume heterogeneity in individual
preferences and endowments. For example, for people with different degrees of risk
aversion, it should be completely rational to follow asset allocation strategies with
different types and amounts of risk and thus accept different levels in expected
retirement wealth. Focusing only on expected retirement wealth, reforms may
outweigh high-risk/high-expected wealth strategies.
To calculate our welfare measure, we employ a method similar to Dammon, Spatt,
and Zhang (2004); Cocco, Gomes, and Maenhout (2005); and Yao and Zhang
(2005). As such, we compare the expected lifetime utility an individual receives
following an optimal asset allocation pattern with the expected lifetime utility
received following the investment strategy observed in our data. We achieve this by
taking the following steps: First, we analyze data from two large data sets, the U.S.-
based Survey of Consumer Finances (SCF) and the German Income and Expenditure
Survey (EVS). Using a regression model, empirical asset allocation policy is
estimated as a function of individual characteristics (e.g., age, gender, education) and
endowments (income and wealth). Next, we calculate the optimal—that is,
benchmark—asset allocation policy and the resulting expected utility. For this we
solve the dynamic optimization problem given by a realistically calibrated life-cycle
consumption/saving/asset allocation model with stochastic uninsurable labor income,
asset returns, and life span. Finally, we place the empirical asset allocation policy
functions, instead of the optimal functions, into the expected utility model and
compare the resulting utility with the optimal one.
Our results are highly relevant for policymakers in their deliberations about changes
to public and private pension systems. We are able to identify the population
Security Administration, 2006). In Germany, the government system (Gesetzliche
Rentenversicherung) in 2003 provided around 66% of retirement income (Federal
Ministry of Health and Social Affairs, 2005).