Do immigrants delay retirement and social security claiming?

ABSTRACT

We use data from the Health and Retirement Study (HRS) to examine how immigrants’ retirement and Social Security claiming patterns compare to those of natives. We find that immigrants retire and claim Social Security at older ages than natives, an effect that is similar across immigrant ethnic groups and more pronounced among immigrants who arrived in the U.S. after age 40. We discuss and explore possible mechanisms for these differences. Hazard models suggests that these differences arise because immigrants have lower retirement and claiming hazard rates in their early 60s. We do not find evidence of an increase in the rate at which immigrants move abroad or exit the survey, compared to U.S. natives, in their late 50s through 60s, a finding that is consistent with immigrants retiring in the U.S. rather than abroad. These results are important given the rising share of older immigrants living in the U.S., and they have implications for the impact of immigration on government finances as well as the retirement security of an important subgroup of the population.

KEYWORDS:

• Immigrants
• retirement
• wealth
• social security

JEL CLASSIFICATION:

• J15
• D14
• J26
• H55

Disclosure statement

No potential conflict of interest was reported by the authors.

Notes

¹ Between 1960 and 2000, the share of immigrants aged 65 and older dropped from 33 percent to 11 percent mainly because the 1965 Immigration and Nationality Act led to in an influx of younger, working-age immigrants (Batalova 2012). However, the share of older immigrants has been increasing recently, with 14.4 percent of the immigrant population aged 65 years or older in 2016, due to the ageing of these working-age immigrants as well as an increase in the number of older immigrants sponsored by their adult children (Batalova 2012; Wilmoth 2012; Borjas 2011).

² These differences are more likely among those who immigrated to the U.S. more recently (who have lower wages relative to natives, as documented by Favreault and Nichols 2011) or at older ages (who have lower incomes, Social Security benefits, and Medicare coverage, as documented by O’Neil and Tienda 2014).

³ Aguila and Vega (2017) use data from the Mexican Health and Ageing Study to examine the behaviour and characteristics of older Mexicans who lived in the U.S. in the past.

⁴ From 1993 onwards, individuals could report either a year of entry, an age of entry, or the number of years ago that entry occurred. The 1992 and 1993 surveys simply asked respondents when they arrived in the U.S. In 1998, 2004, and 2010, respondents were asked when they first arrived in the U.S., whether they ever left, and when they subsequently returned. For these waves, we record the date that the respondent first arrived in the U.S.

⁵ The first indicator also includes a category for foreign-born individuals with missing information on age/year of entry.

⁶ Wealth is in nominal dollars at the time of the survey. Wave dummies in the regressions should absorb the effects of inflation.

⁷ In 2015, the foreign-born were twice as likely to live in 5 or more person families than the U.S. born (see http://www.pewhispanic.org/ph_stat-portraits_foreign-born-2015_trends-42b/) and Asians, Hispanics, and immigrants were more likely than whites and U.S. natives to live in multigenerational family households (see http://www.pewresearch.org/fact-tank/2018/04/05/a-record-64-million-americans-live-in-multigenerational-households/and http://www.pewsocialtrends.org/2011/10/03/chapter-3-demographics-of-multi-generational-households/)

⁸ We also drop individuals who report that they receive either SSDI or Supplemental Security Income benefits but don’t know which.

⁹ Whether an individual considers him or herself retired is a separate question from employment status, but it is highly correlated.

¹⁰ We bound individuals’ retirement ages between the last wave in which they report working for pay and the first wave in which they report not working for pay. We use the first observed transition from working to not working. Individuals who are not working report the date at which they last worked. From these reported dates of last work, we can construct an exact retirement age, and we take the minimum of these exact retirement ages across all waves of the survey. If the exact retirement age falls between the upper and lower bounds of the retirement age we derived from the work for pay transitions, we assign this value as the individual’s retirement age. Otherwise, we average the upper and lower bounds to obtain an estimated retirement age.

¹¹ Social Security worker benefits do not grow with delay past age 70. Thus, claiming after 70 is almost never optimal, and anyone who has not claimed by age 70 is very likely to be ineligible. A possible exception is that some individuals (late life immigrants, for example) may not have not accumulated 10 years of covered work by age 70. If these individuals continue to work beyond 70, they could become eligible and claim once they reach the required 10 years. We think such cases are rare as most people do not work beyond age 70. Given that anyone who has not claimed by age 70 is highly likely to be ineligible, this restriction also ensures that unauthorized immigrants are not included in this restricted sample.

¹² Retirement transitions can occur multiple times for a given individual. In the person-level analysis, individuals were assigned a unique retirement age based on their first observed retirement transition. These individuals may appear more than once in the hazard data if they subsequently ‘unretire’ and then retire again. In addition, for a retirement transition to be included in the hazard data, an individual must be observed in two consecutive waves. There may be individuals who were assigned retirement ages in the person-level analysis but are not included in the hazard models because (due to missing data or skipped waves) they are not observed to transition to retirement over two consecutive waves.

¹³ Specifically, we set the attrition indicator to 1 if the RAND HRS variable riwstat is equal to 1 in the previous wave and either 4 or 7 in the current wave. It is equal to 0 if riwstat is equal to 1 in both the previous and the current waves.

¹⁴ Results from OLS regressions are qualitatively similar and available upon request.

¹⁵ For waves 1–12, we use the RAND variable RwWTRESP, which is equal to zero for nursing home residents. For wave 13, that variable is no longer available. Instead, we use the RAND variable RwTRESPE, which reflects new weighting HRS weighting procedures and no longer assigns a zero weight to individuals in nursing homes.

¹⁶ Results available from author upon request.

Additional information

Funding

This study received no funding.