Pension Projection Models


This is a brief description of just one of the PSG models, a unit of three that work seamlessly together as described on the homepage.

PENSIM is a dynamic microsimulation model for analysis of the retirement income implications of government policies affecting employer-sponsored pensions, employer offerings of pensions, and employee behavior with respect to pensions.

PENSIM development has been funded between 1997 and 2014 by the Office of Policy and Research at the Employee Benefits Security Administration of the U.S. Department of Labor. Subsequent development has been supported by the Policy Simulation Group and by organizations licensing the PSG models.

PENSIM uses discrete event simulation methods to generate a cohort sample of life histories. These same microsimulation methods are used by Statistics Canada in their LifePaths model. Consult the Thumbnail Sketch of PENSIM section of the PENSIM Overview documentation for more details.

Detailed pension characteristics are imputed using a model estimated with 1996-98 establishment data from Bureau of Labor Statistics Employee Benefit Survey, which is now known as the National Compensation Survey. Consult the documentation for more details.

The main trends in employer pension offerings from 1975 through 2005 are represented in the imputation model, but because 1975 offerings are simply projected backward in time, lifetime pension accumulation for cohorts born before 1955 may not be realistic.

The current version of PENSIM produces a cohort sample of 100,000 life histories, plus life histories for each individual's spouses, in just a couple of minutes on a fast personal computer (see fast-running).

The simulated life histories have been subjected to a wide range of validation tests, the results of which suggest that samples of simulated life histories are realistic (see extensively-validated).

PENSIM has been used in a number of interesting pension analysis projects by well-known users.

This page was last revised on February 17, 2016.