The nonpartisan Urban Institute publishes studies, reports, and books on timely topics worthy of public consideration. The views expressed are those of the authors and should not be attributed to the Urban Institute, its trustees, or its funders.
Note: This report is available in its entirety in the Portable Document Format (PDF). (File Size: 4 MB)
This report describes the work the Urban Institute performed to generate MINT4 and MINT4EX version 2. MINT4 includes individuals born between 1926 and 1972 and projects retirement income and assets out to 2039. MINT4EX extends the basic MINT4 projections to 2099 and then uses a statistical matching procedure to generate projections for individuals born between 1973 and 2017. The final MINT4EX file projects retirement income and assets from 1988 to 2099 for individuals born between 1926 and 2017.
MINT4 replicates MINT3 by reproducing the full set of model projections. MINT4 differs from MINT3 in its starting sample, its included set of cohorts, and its projection horizon. MINT4 is based on the 1996 Survey of Income and Program Participation (SIPP) panel matched to Social Security Administration data on benefits, earnings, and mortality, while MINT3 is based on the 1990 to 1993 SIPP panels matched to these same administrative sources. MINT4 includes individuals born between 1926 and 1972, while MINT3 included only individuals born between 1926 and 1965. Finally, MINT4 makes all of the retirement income projections through the year 2039, while MINT3 projects out to 2032. We constructed the MINT4 file so that it is fully compatible with MINT3that is, analysts may use the files separately or combined.
Both MINT3 and MINT4 have been extended to include cohorts born between 1973 and 2017 with projections to 2099. The extended parts of the file (cohorts born between 1973 and 2017) are identical for both MINT3EX and MINT4EX, and the individuals born between 1966 and 1972 included in MINT3EX come directly from the MINT4. This process is documented in chapter 5 of this report.
The MINT4 contract called for no new enhancements or parameter estimates. Instead, it called for the systematic replication of the methods used in MINT3 to generate the demographic, earnings, pension, wealth, retirement, benefit take-up, Social Security and SSI benefit projections, and immigration on the new data files. These methods are detailed in Toder et al (2002).1
These basic projections were modified under a separate MINT contract to address some concerns raised by the expert review panel (Berk et al 2003).2 This enhanced version is the reference file for MINT3. The MINT4 model includes the same model enhancements that we made for MINT3.
While the MINT4 contract asked for a systematic replication of MINT3, under the direction of the Social Security Administration Task Manager, we did make four important changes to the MINT3 methods to improve and extend the model results.
- We included projected earnings up to age 50 in the fixed-effects model used to generate earnings from age 50 to retirement. MINT3 uses only historic earnings from the administrative earnings records up to 1999. The latter change provided better projections for the newly added cohorts (1966 to 1972) who had relatively few years of historic earnings compared to earlier cohorts.
- We adjusted the starting SIPP financial wealth values to match the age-specific wealth distribution found on the 1998 Survey of Consumer Finances (SCF).
- We changed the deferred contribution limits to match legislated limits through 2005. We then price adjust the limits from 2006 on.
- We incorporated the Office of Retirement Policy's (ORP) Social Security benefit calculator into MINT. This calculator had been developed to model numerous Social Security reform options including reforms with personal retirement accounts.
- We used the CSSS method to add individuals born between 1973 and 2017 and to extend the projections to 2099 (Butrica, Cashin, and Uccello 2004).
Our approach was to match as closely as possible the file structure and coding for MINT4 as was used for MINT3. This required careful recoding and remapping of variables that had changed on the 1996 SIPP survey compared to the 1990 to 1993 SIPP surveys. Where possible, we used run time parameters in the code to adjust for the changes in the cohorts and final projection year in MINT4 compared to MINT3. This technique has allowed us to share the bulk of the code in both MINT3 and MINT4, increasing the model's transparency and facilitating future changes to the model.
In the process of generating the MINT4 projections, we found the MINT4 lifetime earnings were lower than the MINT3 lifetime earnings. We performed some sensitivity analysis that revealed two important findings. First, the 1996 SIPP sample earnings are about 5 percent lower than the 1990 to 1993 earnings for the same cohorts and same years. Separate analysis of the 1990 to 1993 panels showed that this variance is within the range of sample variance observed across the older panels. Second, the projected earnings are sensitive to the final year of earnings used in the earning splicing program. Both 2000 and 2001 were recession years. When we include these years as the last years of earnings, the splicing method projects the recession-year earnings to continue into the future. To lessen the impact of this recession, we excluded the 2001 DER earnings from the donor records and projected earnings from 2001 until retirement.
Chapter 2 of this report describes the results of the sensitivity analysis comparing the MINT3 and MINT4 samples that provided the basis for the decision to exclude the 2001 data from the splicing donor record. Chapter 3 describes the wealth adjustments used to align the MINT4 financial assets and retirement account balances to the 1998 SCF distribution. Chapter 4 details the patterns in outcomes generated by the MINT4 data system, focusing on the cohorts as they reach age 62 and age 67, respectively, and on the entire model population still alive in 2020. These patterns are disaggregated along many different dimensions, with specific consideration given to outcomes by sex and cohort. Separate tables examine labor force participation rates, retirement and Social Security take-up ages, earnings, living arrangements, and housing and nonhousing wealth. Summary tables aggregate these sources of income (and SSI) to show total family incomes. We also highlight some of the differences in results between the MINT3 and MINT4 projections. Chapter 5 describes the method for adding the 1973 to 2017 birth cohorts to MINT4EX, and chapter 6 summarizes the results for the full range of projected cohorts out to 2060.
Notes from this section of the report
1. The specific chapters by topic in that volume are as follows: Introduction (Chapter 1), Earnings, mortality, and disability status prior to age 67 (Chapter 2), Health status (Chapter 3), Employment and retirement behavior after age 50 and Social Security benefit take-up (Chapter 4), Pension coverage and benefits (Chapter 5), Non-pension assets (Chapter 6), Living arrangements and eligibility for Supplemental Security Income (SSI) benefits and subsequent participation (Chapter 7), Mortality adjustments (Chapter 8), Summary of model results (Chapter 9), and Projections of poverty in 2020 (Chapter 10).
2. The updated MINT model includes the following enhancements:
- Use the Detailed Earnings Record (DER) instead of the Summary Earnings Record (SER) for historic earnings and parameter estimation. The latter file introduces bias in measured earnings both through capping of earnings above the Social Security taxable maximum and through missing earnings not covered by Social Security. These biases affected projected earnings, retirement, pension wealth, and financial wealth in earlier versions of MINT.
- Separate out self-employment earnings from nonself-employment earnings. Adding this distinction increased the projected heterogeneity in pension types and wealth accumulation for individuals with selfemployment earnings.
Add future immigrants to the MINT sample. Omitted immigrants, who are primarily those entering the US at older ages, systematically have fewer years of Social Security covered earnings and are more susceptible to poverty and SSI eligibility.
Note: This report is available in its entirety in the Portable Document Format (PDF).