State Auto-IRA policies and firm behavior: Lessons from administrative tax data

Highlights

• We study the effects of state “auto-IRA” policies on firms using tax data.
• 17 percent of non-offering firms start a retirement plan in response to state policy.
• These policies do not induce firms to stop offering retirement plans (crowd-out).
• Firms induced to start retirement plans are largely similar to those that do not.
• We explore and rule out some explanations for the observed behavior.

Abstract

Several states have recently attempted to boost retirement saving by adopting “auto-IRA” policies that require employers not currently offering an employer-sponsored retirement plan (ESRP) to either (1) establish an ESRP or (2) enroll employees in state-facilitated Individual Retirement Accounts (IRAs). We identify the effect of these state policies on firm decisions to offer ESRPs, exploiting the phased rollout of these policy treatments across states and employer size categories. Using U.S. tax microdata, we estimate that about 17% of treated firms have been induced to offer an ESRP by these policies, although there is substantial heterogeneity in these effects across firm and worker characteristics. This effect is large considering that, for employers, establishing and maintaining an ESRP is more costly than utilizing the state-facilitated IRAs. We explore both rational and behavioral explanations for why firms might choose the higher-cost option to comply with state auto-IRA policies.

Introduction

Fringe benefits – largely in the form of employer-sponsored retirement plans (ESRPs) and health insurance – have grown dramatically as a share of employee compensation since the middle of the 20th century (Chen, 1981, Wiatrowski, 1999, U.S. Bureau of Labor Statistics, 2024). A large literature has established that workers place substantial value on nonwage compensation, and that benefits like health insurance and defined benefit pensions affect labor supply and job transitions (e.g., Gruber and Madrian, 1994, Garthwaite et al., 2014, Ni and Podgursky, 2016, Nyce et al., 2013, Kim, 2020). There has also been considerable analysis regarding how fringe benefits and other nonwage job attributes affect measurement of income growth and inequality (e.g., Pierce, 2001, Burkhauser et al., 2012, Guvenen et al., 2022, Maestas et al., 2023). By contrast, there has been comparatively little research on how employers determine whether to offer fringe benefits to workers. In this paper, we use recent, quasi-experimental variation in state policies to shed light on why employers choose to establish ESRPs.

Most private retirement saving in the U.S. takes place through ESRPs.¹ Over the last fifty years, defined contribution (DC) plans, in which workers contribute to a personal retirement savings account established by their employer, have gradually replaced traditional defined benefit (DB) plans, which pay a retirement benefit according to a formula (U.S. Department of Labor 2023). Federal policy has also incentivized employers to automatically enroll workers in their DC plans (requiring workers to opt out if they do not wish to participate), leading to large increases in participation rates. However, a significant proportion of the American workforce works for employers that do not offer ESRPs.²Those who do not have access to an ESRP are disproportionately likely to be young, Black or Hispanic, to have less education and lower incomes, and to be employees of smaller firms (Sabelhaus 2022). While individuals who do not have access to an ESRP may contribute to non-employment based Individual Retirement Accounts (IRAs) – which also provide access to tax-preferred retirement saving – most do not (see, e.g., Investment Company Institute 2022a). In recent years, many states have attempted to boost retirement savings by mandating that all employers offer a way for employees to save for retirement through payroll deduction. Employers can comply with the policy mandate by either offering an ESRP or automatically enrolling employees in an IRA that is facilitated by the state and managed by professional retirement service providers. We refer to these state-facilitated IRAs as state auto-IRA programs (or just ‘state programs’), and to a combination of an employer mandate and state auto-IRA program as an auto-IRA policy. Employees may opt out of either of these employer mediated savings options.

Starting in 2017, there has been a staggered rollout of these auto-IRA policies across states and firm size categories. In each of the policy expansions that we study, we find clear and substantial increases in the share of affected firms establishing an ESRP immediately upon implementation. We refer to this induced increase in ESRP offerings as the “crowd-in” effect of the policy. We do not find evidence of any offsetting “crowd-out”, i.e. firms terminating existing ESRPs in favor of utilizing the state auto-IRA program. We estimate that approximately one sixth of firms in the affected employer size ranges were induced to offer an ESRP by the policies. These effects are also substantial relative to the number of firms participating in the state auto-IRA programs directly; we find that ESRP crowd-in accounts for 30 to 45 percent of the total increase in employer coverage, depending on the state. Using a treatment effects framework, we then estimate mean characteristics of “complier” firms induced to offer ESRPs by the policy, as well as their employees and owners. We find that complier firms are more similar to firms that never offer an ESRP than they are to firms that already offer ESRPs.

We analyze our findings in the context of a neoclassical framework where rational employers weigh the costs and benefits of offering ESRPs. ESRPs have substantial costs, as they are typically administered by third parties for a fee and generally require compliance with nondiscrimination rules. Employers do not pay a fee to participate in state auto-IRA programs; however, they do face an administrative burden of signing up for the state’s program and facilitating contributions by employees. On the benefit side, workers value ESRPs because they offer access to tax-advantaged saving. Owners may capture this benefit both directly in their role as an owner-employee (i.e., by participating in the ESRP themselves) and indirectly because their employees value ESRPs. While fully rational workers gain no benefit from the state auto-IRA program – since workers have long been able to contribute to IRAs on their own – “behavioral” workers may value the convenience of automatic contributions, as well as the commitment mechanism they provide. The large crowd-in that we observe empirically requires some combination of high perceived administrative cost and low perceived benefits of state auto-IRA programs relative to ESRPs. By contrast, if workers value auto-IRAs and ESRPs similarly and the administrative cost of state auto-IRA programs is low, then employers may have an incentive to drop their ESRPs and utilize the less costly state auto-IRA program instead (“crowd out”).³

We then test whether this framework can explain our empirical findings. On the benefit side, we do not find substantial evidence that the crowd-in is driven by perceived low benefits of state auto-IRAs relative to ESRPs. For example, auto-IRAs may have lower benefits than ESRPs because IRAs have lower contribution limits. However, we estimate that the auto-IRA contribution limits typically do not bind, either for employees or owners. On the cost side, while we do not have direct evidence on administrative burden, we find that take-up rates for a federal tax credit for establishing an ESRP – a task that may require several hours of compliance efforts – are very low, suggesting that that paperwork and compliance burdens may weigh heavily on this population.

Finally, we consider how behavioral factors at the firm level might cause a state auto-IRA policy to lead to ESRP crowd-in. In the presence of inertia, for instance, removing the default option of offering no plan may induce employers to revisit their ESRP decision and choose to offer a plan. Furthermore, employers may be responding to marketing that ESRP administrators have undertaken in response to the auto-IRA policies.

There exists a small, recent literature studying state auto-IRA policies. Several papers have studied their direct effect – i.e., the boost in IRA participation brought about by the state auto-IRA programs (Dao, 2024, Quinby et al., 2020, Chalmers et al., 2022). Willingham (2024) finds that employees at adopting firms increase their retirement savings and tend not to make withdrawals even at job separation. Closer to our paper, previous work by Bloomfield et al. (2024) used individual-level data from the Current Population Survey (CPS) and firm-level ESRP filings and found that auto-IRA policies induced ESRP offers.⁴ We add to this literature and provide three substantial contributions beyond Bloomfield et al. (2024). First, our paper confirms and refines the effects identified by Bloomfield et al. (2024). While statistically significant, their CPS-based results are potentially confounded by survey response biases, as household surveys are known to underreport retirement coverage. Thus, these findings may primarily (or entirely) reflect increased worker awareness of existing benefits rather than new ESRP formation. Bloomfield et al. (2024) attempt to address this limitation by examining plan-level disclosure filings, but these data provide limited visibility into firm-level behaviors.⁵ Second, our paper uses administrative tax data that allows for precise treatment assignment (i.e., by firm size and time) and detailed analysis of heterogeneity across firms, workers, and owners. In contrast, the previous study lacked precision in assigning treatment at the firm level and offered limited granularity with regard to firm and worker characteristics. The data precision and granularity in the current study is crucial for understanding the heterogeneous effects of these policies. Third, and most importantly, the comprehensive data in our paper enables us to explore and rule out potential mechanisms driving the employer behavior that we observe.

Our study contributes to several additional streams of research. First, a large body of research has shown that – contrary to the predictions of the rational model – automatic enrollment substantially boosts employee participation in retirement saving accounts in the short run (Madrian and Shea, 2001, Choi et al., 2004), although the impact on long-term saving is less clear (Choukmane, 2024, Beshears et al., 2022, Derby et al., 2023, Beshears et al., 2024). However, this literature primarily analyzes worker decisions to participate in ESRPs rather than employer decisions to establish them. We focus on firms’ decisions to establish ESRPs and show that frictions may affect these choices as well.

Second, we contribute to the empirical literature on why firms offer fringe benefits to workers. Previous papers in this literature have considered the role of unions (Freeman 1981), tax policy (Long and Scott, 1982, Long and Scott, 1984, Turner, 1987a, Turner, 1987b), worker characteristics (Rhine 1987), economies of scale in benefit provision and search costs (Oyer 2008), and the desire of employers to attract female employees (Liu et al. 2023). Third, our study contributes to the literature on how firms respond to government mandates concerning compensation level and structure. For example, recent papers in the minimum wage literature have examined how minimum wage laws, which regulate monetary compensation, affect nonwage compensation such as health insurance (Clemens et al., 2018, Clemens, 2021, Meiselbach and Abraham, 2023). Some studies have also examined the impact of the mandates in the Affordable Care Act (ACA), as well as an earlier Massachusetts law, which required some employers to offer health insurance (Kolstad and Kowalski, 2016, Lyons, 2017, Abraham et al., 2019). We contribute to both literatures by investigating how state policies affect firms’ decisions to offer DC ESRPs, which have become the dominant retirement savings vehicle in the U.S.

Finally, we contribute to a small but growing literature on the role of behavioral economics in explaining firm decision-making (a recent paper is Gertler et al. 2025; for reviews, see Heidhues and Koszegi 2018; Malmendier 2018). Our results suggest that neoclassical factors cannot fully explain the large crowd-in that we find, leaving behavioral factors such as inertia, salience, and marketing as plausible channels. Thus, these findings are consistent with the hypothesis that owners of small- and medium-sized businesses are susceptible to many of the same behavioral biases that are well-studied in the context of consumers, savers, and employees.

The policy implications of this study are particularly pertinent. As states and the federal government (through the SECURE Act and SECURE 2.0, as well as prospective federal laws under consideration) continue to implement measures designed to boost retirement saving, understanding the impact of state auto-IRA policies on firm and worker behavior is paramount. More broadly, employer mandates are a widespread policy tool that governments use to affect societal goals, so it is important to assess their impact on labor markets. This paper seeks to contribute to a more robust knowledge base that can guide policymakers and employers in their decisions with regard to retirement savings options.