Default Options and Retirement Saving Dynamics

Overview

Research question. Does automatic enrollment (auto-enrollment) in retirement savings plans increase lifetime wealth accumulation and welfare? The prior literature established large short-run participation effects but had not traced the policy’s consequences over a full working life.

Data. The paper draws on two primary sources. First, a proprietary panel of 401(k) administrative records from nearly 600 U.S. firms, covering roughly 159,216 first-year employees across 86 firms (for the “increasing default” fact) and 6,415 employees across 34 firms (for structural estimation), observed between December 2006 and December 2017. Second, 12 successive waves (2006–2017) of the U.K. Annual Survey of Hours and Earnings (ASHE), a 1% nationally representative panel of approximately 200,000 private-sector employees per year, including 37,120 job-switchers, used to exploit the phased rollout of the U.K. Pension Act of 2008.

Methodology. The paper proceeds in three steps. (1) Three empirical stylized facts are documented using quasi-experimental variation (comparing employees hired before versus after changes in the default contribution rate within the same firm, and exploiting the staggered employer-size-based rollout of U.K. auto-enrollment). (2) A structural lifecycle model is estimated via the Method of Simulated Moments, using three preference parameters—intertemporal discount factor (δ), elasticity of intertemporal substitution (σ), and opt-out cost (k)—identified from the within-firm default variation in 34 U.S. firms. (3) The estimated model is used for out-of-sample validation and counterfactual welfare analysis.

Three stylized facts.

Fact I — Increasing the default reduces participation. Among 159,216 first-year employees in 86 auto-enrollment firms, each percentage-point increase in the default contribution rate reduces 401(k) participation by approximately 1 percentage point and increases contributions strictly below the new default by 1 percentage point. When the default rose from 3% to 6%, workers were 3.2 percentage points more likely to contribute at 1% or 2% of salary. This “drop-out” pattern is consistent with an opt-out cost model but is inconsistent with loss-aversion and psychological-anchoring theories, both of which predict that raising the default should weakly increase low-end contributions.

Fact II — Non-autoenrolled workers catch up within three years. In the estimation sample of 34 U.S. firms offering a 50% match up to 6% and an auto-enrollment default of 3%, median cumulative employee 401(k) contributions of non-autoenrolled workers equal those of autoenrolled workers after three years of tenure. Because non-autoenrolled workers compensate for initial non-participation by contributing more later—earning similar cumulative employer match and tax benefits over the full three-year horizon—a modest opt-out cost suffices to explain the observed inertia. Previous studies (which examined only the first year of tenure and did not allow future contribution adjustment) inferred opt-out costs of $1,000–$2,200 or more; the dynamic model implies a cost of only approximately $250.

Fact III — Prior auto-enrollment reduces saving in the next job. Using the phased U.K. policy rollout, workers who were auto-enrolled in their previous job and then move to a new employer that has not yet implemented auto-enrollment participate 12.8 percentage points less and contribute 0.55% of salary less in the new plan relative to otherwise similar job-switchers from non-auto-enrollment employers. When the new employer also has auto-enrollment, no statistically significant difference is observed. Placebo rollout tests confirm the effect is not a pre-existing selection pattern. This negative spillover contradicts a “savings habit” hypothesis and suggests that auto-enrollment’s short-run boost overstates lifetime savings effects.

Structural estimation results. The estimated quarterly discount factor is δ = 0.987 (approximately 0.949 annually), and the elasticity of intertemporal substitution is σ = 0.435, both standard in lifecycle models. The opt-out cost is estimated at $254 per contribution-rate change (standard error $11). Sensitivity exercises show that combining a short observation window (first year only), sticky contributions (no intra-job adjustment), no income uncertainty, immediate vesting, and penalty-free DC withdrawals yields an opt-out cost of $3,004—broadly matching the range in previous studies. The low baseline estimate is thus driven by the dynamic nature of decisions (ability to compensate later), the illiquidity of retirement accounts (which reduces their perceived value), and income uncertainty (which expands the inaction range).

Long-run wealth effects. Simulating a universal 3% auto-enrollment policy, the model predicts that wealth at retirement changes by less than 2% for the top 7 income deciles. For individuals in the top two deciles, total wealth at age 65 is actually reduced by less than 1% because many high earners who would voluntarily contribute above 3% are pulled down to the default. At the bottom decile, however, auto-enrollment raises total retirement wealth by more than 12%; savings increases are concentrated in the first 20 years of working life and peak around age 45, where bottom-quintile workers hold an additional 20% of average annual lifetime earnings. Even at the bottom, approximately one-third of the early savings gains are offset by lower contributions after age 45, as the wealth effect dominates. Crowd-out of liquid savings is limited: for bottom-quintile individuals, 89% of the increase in retirement savings at age 65 passes through to total wealth; for middle-quintile individuals, 62% passes through.

Out-of-sample validation. The U.S.-estimated model is not rejected (at the 10% level) in 8 of 11 response moments in the 86-firm sample where defaults were raised between two positive rates, covering over 85% of workers. Recalibrated to U.K. institutions (using δ and σ from the U.S. and k = £160 via the average USD/GBP exchange rate), the model replicates the roughly 30-percentage-point increase in both participation and contributions at the 1% U.K. default. The model also predicts a 9.6-percentage-point drop in participation when workers move from an auto-enrollment to an opt-in employer, close to the empirical 12.8 percentage points.

Welfare and optimal policy. Under utilitarian preferences (policymaker shares individuals’ discount rate, no redistributive motive), the opt-in regime is always preferred to auto-enrollment regardless of policy incidence, because matching and tax incentives already induce over-saving relative to individuals’ revealed time preferences. Under paternalistic preferences (social discount factor = 1) or inequality-averse preferences (Pareto weights inversely proportional to income, with degree of inequality aversion ν = 1 following Saez 2002), an auto-enrollment default at or near the employer matching threshold (6% of income) maximizes social welfare. A 6% auto-enrollment default improves welfare by 0.3% in lifetime consumption-equivalent for the bottom decile even under a utilitarian policymaker when incidence is on employers. These optimal policy rankings are robust to whether the opt-out cost is treated as fully welfare-relevant (π = 1) or welfare-irrelevant (π = 0), and hold under three incidence scenarios (employer profit reduction, match-rate adjustment, wage adjustment).

Q&A

Q1: What is the core mechanism by which non-autoenrolled workers “catch up” at the median, and why does this reduce the implied opt-out cost relative to prior estimates?

A: Non-autoenrolled workers who do not contribute in their first year are not permanently forgoing employer matching and tax benefits; they can contribute more later in the same job and earn similar cumulative benefits. The paper shows that at the median and 75th percentile, cumulative employee 401(k) contributions among opt-in workers equal those of autoenrolled workers after three years of tenure in 34 U.S. firms offering a 50%-up-to-6% match at a 3% default. This dynamic substitutability means the opportunity cost of initial non-participation is far smaller than one-period back-of-the-envelope calculations suggest. Previous studies, which implicitly or explicitly assumed static contribution decisions or examined only the first year, inferred opt-out costs of $1,000–$2,200; in a fully dynamic model the same inertia requires only ~$254.

Q2: Why does Fact I (higher default reduces participation) specifically rule out loss aversion and anchoring as the primary mechanism, and what does it support instead?

A: Under loss aversion, contributions above the default feel like losses while contributions below the default feel like gains. Raising the default shifts some contributions from the loss domain into the gain domain, making low contributions relatively less attractive. Proposition 2 demonstrates formally that loss-averse preferences predict a weakly lower fraction contributing below the new (higher) default — the opposite of what is observed. Similarly, Proposition 3 shows that psychological anchoring shifts preferences toward the new default, also predicting more participation at low rates when the default rises. Only the opt-out cost model (Proposition 1) predicts that a higher default causes some workers to incur the cost to switch away from the default and end up at lower contribution rates, matching the empirical finding that each 1-percentage-point rise in the default increases contributions strictly below the old default by approximately 1 percentage point.

Q3: What is the quantitative magnitude of the opt-out cost, and what modeling assumptions are responsible for it being much smaller than prior estimates?

A: The baseline estimate is $254 per contribution-rate change (s.e. $11), roughly an order of magnitude smaller than prior estimates of $1,000–$3,000+. Table 4 decomposes the sources of the difference: using only first-year data changes the estimate only slightly (to $226). Assuming contributions cannot be changed within a job (“sticky contributions”) raises the cost to $308 with four years of data or $712 with one year of data. Eliminating income uncertainty raises the estimate to $465. Assuming immediate vesting raises it to $344. Assuming penalty-free DC withdrawals raises it to $609. Combining all these restrictions simultaneously yields $3,004 — closely matching the prior literature. The three key drivers are thus: (1) the ability to adjust contributions over time within a job; (2) the illiquidity of the DC account (early-withdrawal penalties); and (3) income uncertainty widening the inaction range.

Q4: How does the paper validate the structural model out of sample, and what confidence does this provide in the long-run predictions?

A: Two out-of-sample exercises are reported. First, the model estimated on 34 U.S. firms (introduction of auto-enrollment from 0% to 3% default) is used to predict workers’ response when 86 other firms raised the default from one positive rate to a higher rate. The model prediction cannot be rejected at the 10% level in 8 of 11 response-moment cases, covering 71 of 86 firms and more than 85% of workers. Second, the model is re-calibrated to U.K. institutions (keeping U.S. preference estimates, setting k = £160 via exchange rate) and applied to the phased rollout of the U.K. Pension Act of 2008. The model replicates the roughly 30-percentage-point increase in both participation and contributions at the 1% default following the policy, and predicts a 9.6-percentage-point drop in participation when previously autoenrolled workers move to a new opt-in employer — compared with an empirical estimate of 12.8 percentage points (s.e. 5.5 pp).

Q5: What are the distributional implications of a universal 3% auto-enrollment policy for wealth at retirement?

A: The effect is concentrated at the bottom. For the top 7 income deciles, retirement wealth at age 65 changes by less than 2% relative to the opt-in counterfactual. For the top two deciles, total wealth at age 65 is actually reduced by less than 1% because high-earning workers who would voluntarily contribute above 3% are pulled down to the default. For the bottom decile, the policy raises total retirement wealth by more than 12%. Even at the bottom, roughly one-third of the early savings gains are later offset by lower contributions after age 45 as the wealth effect dominates, so even 20-year empirical follow-ups may overstate the policy’s lifetime effect at the bottom.

Q6: How large is crowd-out of liquid savings by auto-enrollment, and what explains the limited degree of substitution?

A: Crowd-out is modest. For bottom-quintile workers, 89% of the increase in retirement savings at age 65 translates into higher total wealth; for middle-quintile workers, 62% passes through. The limited crowd-out arises because liquid assets serve a precautionary motive and DC accounts serve a lifecycle motive — the two assets are not close substitutes. Additionally, as in Kaplan and Violante (2014), the marginal propensity to consume out of liquid assets is high in the model, so autoenrolled workers reduce consumption rather than run down liquid balances. These predictions align with Beshears et al. (2021), who find no significant increase in unsecured debt after four years, and Chetty et al. (2014), who estimate an 80% pass-through to total savings in a different Danish policy.

Q7: Why do previously autoenrolled workers contribute less when they switch to an opt-in employer, and how is this consistent with the model?

A: The most plausible explanation, and the one consistent with the model’s out-of-sample predictions, is a standard wealth effect: workers auto-enrolled early accumulate more retirement wealth and therefore have less incentive to contribute in a new job. The model predicts a 9.6-percentage-point participation drop for AE-to-non-AE movers, close to the empirical 12.8 pp. An alternative explanation — that previously autoenrolled workers rationally expect their new employer to soon adopt auto-enrollment and thus delay active enrollment — is partially ruled out by the finding that the empirical estimate is closer to the model prediction for job-switchers whose new employer is not expected to adopt auto-enrollment in the next 12 months.

Q8: What are the welfare implications of auto-enrollment under utilitarian, paternalistic, and inequality-averse policymakers, and how robust are these to the incidence assumption?

A: Under utilitarian preferences (policymaker shares individuals’ discount factor, no extra redistributive weight), the opt-in regime is always preferred regardless of whether the policy’s cost falls on employer profits, the match rate, or wages. The negative welfare effect is largest when incidence falls on wages (approximately 50% larger than under match-rate reduction). Under paternalistic preferences (social discount factor = 1), a 6% default (equal to the employer matching threshold) is optimal under all three incidence scenarios. Under inequality-averse preferences (ν = 1 Pareto weights), a 6% default is optimal when incidence falls on employers, and a 5% default when incidence falls on workers. These results are identical whether the opt-out cost is treated as fully welfare-relevant (π = 1) or welfare-irrelevant (π = 0). A 6% auto-enrollment default increases welfare by 0.3% in lifetime consumption-equivalent for the bottom income decile even under a utilitarian planner when incidence is on employers.

Q9: How does the paper address heterogeneity in default effects across age and income groups within a parsimonious homogeneous preference model?

A: The model has only three estimated preference parameters (δ, σ, k), yet it endogenously replicates empirical heterogeneity. Conditional on participating, workers in their 20s are approximately 20 percentage points more likely to stay at the 3% default than workers in their late 50s and early 60s; the model attributes this to the option value of waiting: young workers can compensate for current non-saving by contributing more later, so the cost of opting out is effectively smaller for them. The lowest-income workers are approximately 40 percentage points more likely to remain at the default than the highest-paid; the model explains this primarily because the fixed opt-out cost of $254 represents a larger share of earnings for low-income individuals (and secondarily because high-income workers have more to gain from active contribution decisions due to higher marginal tax rates and a lower Social Security replacement rate). All model-predicted coefficients fall within the 95% confidence intervals of the empirical estimates.

Q10: What does the paper conclude about the broader relevance of the “dynamic opt-out cost” framework beyond retirement saving?

A: The paper argues that wherever individuals can compensate for present inaction with future actions — as in retirement saving — the observed inertia at a default understates the freedom of choice preserved by the nudge, and short-run effects overstate long-term consequences. In contrast, in domains such as healthcare plan choice or school selection, future actions cannot easily offset present inertia; opt-out costs are likely to remain large; and the distinction between a nudge and a hard mandate collapses. The paper therefore argues that the appeal of “libertarian paternalism” (Thaler and Sunstein 2003) is domain-specific and is strongest precisely where intertemporal adjustment is possible.

Key Concepts

Opt-out cost (k). In this paper, a utility cost — estimated at $254 per contribution-rate change — that individuals must pay every time they choose a retirement contribution rate different from the current default. The cost is modeled as a consumption reduction and captures both real transaction costs (form-filling, adviser fees) and behavioral costs (cognitive cost of attention and optimal-choice search). It is fixed and homogeneous across individuals, and applies symmetrically in any direction of deviation from the default.

Auto-enrollment default contribution rate. The positive contribution rate at which new hires are automatically enrolled in a defined-contribution plan, with the option to opt out by incurring the opt-out cost. In the paper’s estimation sample, this is 3% of salary. The default is exogenous at the start of each new job but endogenous thereafter: once established, the default for subsequent periods equals the worker’s contribution rate in the previous period.

Default effect. The empirically observed tendency of workers to remain at the default contribution rate rather than actively choosing a different rate. In this paper, the default effect is explained by opt-out costs rather than loss aversion or psychological anchoring — a distinction identified through the novel prediction that raising the default from a positive rate to a higher positive rate reduces overall participation (the “drop-out” effect), a pattern consistent only with opt-out costs.

Drop-out effect. The paper’s term (following Caplin and Martin 2017) for the empirical finding that increasing the auto-enrollment default contribution rate causes some workers to stop contributing altogether or to contribute at rates strictly below the initial default. This effect is used as a discriminating test between competing theories of the default effect.

Dynamic opt-out cost framework. The paper’s core modeling insight: that opt-out costs must be estimated in a fully dynamic lifecycle model that allows workers to adjust contributions over time, to hold liquid assets and unsecured debt, and to face labor market risk. In a static or short-horizon model, the opportunity cost of initial non-participation appears large (because the worker permanently forgoes match and tax benefits), requiring large opt-out costs. In the dynamic model, the ability to compensate later shrinks the implied opportunity cost and hence the opt-out cost required to rationalize observed inertia.

Crowd-out of liquid savings. The extent to which higher DC retirement contributions induced by auto-enrollment reduce liquid asset holdings (or increase unsecured borrowing), rather than increasing total wealth. The paper estimates limited crowd-out (89% pass-through to total wealth for bottom-quintile workers, 62% for middle-quintile workers), attributable to the different roles of liquid assets (precautionary motive) and DC accounts (lifecycle motive) in the model.

Policy incidence. The channel through which employers balance their budget in response to higher matching costs created by auto-enrollment. The paper considers three scenarios: employers absorb costs through reduced profits; employers reduce the match rate; employers reduce wages. Optimal policy rankings and welfare magnitudes differ across these scenarios, but the qualitative conclusions — utilitarian policymaker prefers opt-in; paternalistic or inequality-averse policymaker prefers AE at 6% — are robust across incidence assumptions.

Consumption-equivalent variation (γ). The welfare metric used in the paper: the proportional increase in consumption in every period and every state of the world that would make the policymaker indifferent between an auto-enrollment policy at default d and the opt-in regime. A 6% default increases welfare by 0.3% in consumption-equivalent for the bottom income decile under a utilitarian policymaker when incidence is on employers.