Publications “Worker and Spousal Responses to Automatic Enrollment" Journal of Public Economics Volume 223, July 2023, 104910 with Kathleen Mackie and Jake Mortenson This paper estimates the saving effects of automatically enrolling employees in retirement plans, examining a large set of firms and incorporating savings responses beyond employer-sponsored plans. We construct an original data set − using tax returns, payroll filings, and retirement distributions from information returns − for employees at 751 US-based firms that adopted automatic enrollment between 2010 and 2016. We use these data estimate the effects of the policy on retirement plan contributions, withdrawals, and net retirement savings − for both employees and their spouses − by comparing workers hired in the years before and after each firm adopted automatic enrollment. We estimate that in the first year after hiring, automatic enrollment increases plan participation by approximately 80 percent (36 percentage points) and increases retirement savings contributions (as a percent of wages) by about 50 percent (1.2 percentage points). Spouses of employees at these firms do not alter their saving behavior in response to the policy, and employees do not alter IRA contributions. However, automatic enrollment also increases the likelihood that an employee will take a withdrawal from their retirement account by 35 percent (4 percentage points). This effect is driven by employees who take withdrawals when separating from their employer. We find that savings effects are increasing in wages − with the lowest wage quintile increasing savings at roughly one-quarter the level of the top quintile − consistent with the “percent of compensation” structure of default contributions. In the medium run (after three years) the effects dissipate but remain discernible from the control group.
“Changes in Retirement Savings During the COVID Pandemic” University of Pennsylvania, Wharton Research Council, August, 2022 with Lucas Goodman, Kathleen Mackie, and Jake Mortenson This paper documents changes in retirement saving patterns at the onset of the COVID-19 pandemic. We construct a large panel of U.S. tax data, including tens of millions of person-year observations, and measure retirement savings contributions and withdrawals. We use these data to document several important changes in retirement savings patterns during the pandemic relative to the years preceding the pandemic or the Great Recession. First, unlike during the Great Recession, contributions to retirement savings vehicles did not meaningfully decline. Second, driven by the suspension of required minimum distribution rules, IRA withdrawals substantially declined in 2020 for those older than age 72. Third, potentially driven by partial suspension of the early withdrawal penalty, employer-plan withdrawals increased for those under age 60.
Working Papers “Does Growing Up in Tax-Subsidized Housing Lead to Higher Earnings and Educational Attainment?” This paper investigates the effects of the Low-Income Housing Tax Credit (LIHTC) on residents of buildings qualifying for the credit. Specifically, it analyzes whether individuals who grow up in LIHTC housing are more likely to enroll in post-secondary education programs and have higher earnings as adults. Using administrative tax records, I find that each additional year spent in LIHTC housing as a child is associated with an average 4.3 percent increase in the likelihood of attending a higher education program for four years or more, and a 5.7 percent increase in future earnings. Furthermore, I find that there are heterogeneous effects when comparing individuals who live in LIHTC housing located in neighborhoods with different characteristics, and among families that have varying income levels and varying levels of housing security prior to moving into a LIHTC building. Based on this analysis, it is likely that the housing subsidy provides some families with a more stable living situation and with more disposable income. "Assessing the Accuracy of using BISG to Estimate Tax Disparities" with Connor Dowd, and Jake Mortenson Bayesian Improved Surname Geocoding (BISG) is a widely used method for inferring race and ethnicity in administrative data when this information is missing. Estimated probabilities of race and ethnicity are often used to investigate disparities in outcomes among different racial and ethnic groups. It is well-known that the assumptions underlying BISG can fail, but the effect of these failures on analyses of disparities is not well understood. In this paper, we use individual-level Low Income Housing Tax Credit (LIHTC) data – which contains race and ethnicity – combined with individual income tax data to estimate the accuracy of BISG in estimating disparities in the earned income tax credit among LIHTC residents. Our analysis reveals that BISG tends to overestimate the probability that non-white LIHTC residents are white. When using these estimated probabilities to estimate disparities in EITC take-up and receipt, the estimated BISG probabilities tend to overstate the take-up rates and EITC dollars claimed by white LIHTC residents and understate take-up rates and dollars claimed among non-white LIHTC residents. This leads to errors in estimated disparities, in one case flipping the direction of the disparity.
