The Power of Financial Incentives versus the Power of Suggestion for Individual Pension: Are Financial Incentives or Automatic Enrollment Policies More Effective?

Abstract

This study investigates and compares the impacts of two important reforms in retirement savings—matching contribution and automatic enrollment—on participation in the individual pension system (IPS) in Turkey by using a dataset containing information about over 40 million pension contracts for the period of 2004–2021. Unlike the automatic enrollment system (AES), the state matching contribution policy imposes a burden on the state’s budget; understanding the relative effectiveness of the two policies can help protect the public budget from such burdens, especially by taking individual attitudes into account in future policy recommendations. Our estimations indicate that both reforms led to an overall increase in participation. However, the AES is not efficient at encouraging participants to make an active decision on plan characteristics, such as portfolio choices. We also evaluate the effects of the two reforms on IPS participation among different demographic groups. We find that males and married individuals are more responsive to matching contributions, and education levels seem to be closely related with IPS participation. Participants are also sensitive to both the returns of pension funds and alternative investment instruments. All the findings imply that individuals must be encouraged to recognize the need for effective financial preparation in their post-retirement lives. To accomplish this, policymakers could utilize a comprehensive informative campaign and workplace seminars about both the system as a whole and the returns, and they could improve the plan characteristics to be compatible with the needs of individuals before retirement as well as after.

1. Introduction

For decades, the level of retirement savings by households in many countries has been considered inadequate. The World Bank has labeled this as a “crisis” and has developed policies aimed at increasing retirement savings [1]. To spur individuals to save more for retirement and protect their savings, governments have enacted various regulations, such as the Employee Retirement Income Security Act of 1974 in the United States. A variety of policies, including individual retirement accounts, defined benefit plans, defined contribution plans, and personal plans, have been made available by both private companies and governments, and some of these policies have been effective.

As of 2020, the total assets accumulated in these pension plans worldwide exceeded USD 35 trillion. In some countries, such as Denmark, the Netherlands, Iceland, Canada, and the United States, the amount of pension assets exceeds 100% of their GDP, while in Turkey, it accounts for only 3.4% of its GDP [2]. Given the importance of pension savings, this is particularly critical for countries with significant savings deficiencies such as Turkey, which can have both macroeconomic implications, such as low levels of investment and economic growth, and microeconomic implications, such as a lack of financial security during retirement.

In 2013, Turkey introduced a nationwide program designed to enhance retiree well-being, foster employment growth by making long-term funds more accessible, and stimulate economic growth. Under the plan, the government provides matching contributions of up to 25% on funds deposited into individual pension accounts, which is similar to standard programs in many other countries to promote household retirement savings. Since its implementation, the government has been providing these matching contributions to all individual pension account holders, making it one of the most generous matching contribution programs among OECD countries.

In a rational decision-making scenario, it is expected that matching contributions would encourage individuals to save more for their retirement by participating in the individual pension system (IPS). However, it has been noted in the literature that present-biased preferences or behavioral anomalies, such as self-control issues, procrastination, or a preference for the status quo, can prevent some individuals from actively choosing to participate. If individuals are not acting in their best interest despite the implementation of incentivizing policies, it may be necessary to implement intervention policies to address these behavioral biases [3]. In this context, automatic enrollment systems (AESs) have been found to be highly successful in addressing status-quo biases in individuals [4,5,6].

In 2017, Turkey introduced an AES alongside the existing matching contribution policy, which automatically enrolled all newly hired and currently working employees under the age of 45 into an individual pension plan, with an opt-out option. As shown in Appendix

Table A1. Principle Features of IPS and AES Plans. (At the beginning of 2022, new regulations on the AES were implemented in order to increase participation in the government agenda. However, since there are no data available for measuring their effects yet, we do not present the new regulations here).

	IPS with Matching Contribution (1 January 2013)	AES with Matching Contribution (1 January 2017)
Participation in the system	Voluntary	Compulsory
Condition for retirement	Remain in the system for 10 years as of the system entry date and turn 56
Eligibility	Anyone with the capacity to act	Employees under 45 years of age
The party determining the pension company	Participant	Employer
Right of withdrawal in the grace period	Yes (Right of withdrawal within two months after pension contract proposal form signing/approval or after pension company informs employee about participation in the pension plan in AES)
Company’s obligation to bear the possible loss in the accumulations during the right of withdrawal period	No	Yes
Matching contribution amount	30 percent of the paid contributions	30 percent of the paid contributions
		+One-time 1000 TL (1000 TL kick start), in case of remaining in the system after the right of withdrawal period
		+5% percent of the accumulations, in case of opting to receive the right of pension as a 10-year annuity
Conditions for vesting for matching contribution	Of the amounts in the matching contribution account; 15% by staying in the system for at least 3 years, 35% by staying in the system for at least 6 years, 60% by staying in the system for at least 10 years, 100% in the case of death or disability.
Contribution amount	Any amount over the minimum amount stated in the plan	3% of the premium-based earnings (increase option is available)
Additional contribution payment possible anytime	Yes	No
Possible deductions by the pension company
Entry fee	Possible	No
Administrative fees on the contributions paid
Fund management fees	Yes (min.1.09%; max. 2.28% of the annual fund asset value)	Yes (maximum 1.09% of of the fund asset value)
Fund types offered	Equity Fund	Initial Fund
	Bonds and Bills Fund	Standard Fund
	Participation Fund	Other Funds
	Composite Fund	○ Conservative/Cautious
	Money Market Fund	○ Balanced
	Precious Metals Fund	○ Daring/Dynamic/Growth
	Index Fund	○ Aggressive
	Fund Basket Fund
	Variable Fund
	Standard Fund
	Life-Cycle/Target Fund

, the financial plan features of the AES are almost identical. However, differently from the matching contribution policy, individuals are automatically enrolled into the AES and are given the decision to opt out or not, rather than making the decision to opt in or not while enrolling to an IPS. In terms of policy, the AES is unlike the state matching contribution policy which imposes a burden on the state budget; thus, our main aim is to understand the relative effectiveness of the two policies in alleviating such burdens on the public budget, especially by taking into account individuals’ attitudes in future policy recommendations.

The present investigation assesses the ramifications of two recent policies pertaining to retirement savings on individuals’ decisions to participate and examines the financial factors that affect participation in the IPS. We examine whether the matching contribution scheme or automatic enrollment mechanism is more effective at raising the number of participants in the IPS. While the impact of matching contribution is examined through econometric modelling, as is common practice in the literature (e.g., [4,5,6]), the impact of automatic enrollment is mostly assessed by a descriptive analysis. Utilizing a dataset comprising over 40 million individual pension contracts and 400 million data points, the study employs fixed effect estimations for the analysis. The fixed effect estimations suggest that the implementation of a matching contribution plan leads to an increase in participation in a traditional opt-in plan. Considering the number of overall participants in the IPS, the AES proves to be considerably more effective in boosting the number of individuals holding pension saving plans. However, the AES seems to lower participation in a traditional opt-in plan, as expected.

Our results also show that individuals are sensitive to returns of IPS funds and returns of alternative saving instruments. More specifically, a 1% increase in the real returns of IPS funds raises the monthly net new participation to the IPS by 2.2% (1300 participants), while a 1% increase in alternative saving instruments, such as the real returns of gold or the US dollar, decreases monthly new participation by around 1.7% (1000 participants). Additionally, the study analyzes the impact of matching contributions and the automatic enrollment mechanism on participation among diverse demographic groups and finds that married individuals and men exhibit a more favorable response to the matching contribution policy than their counterparts.

The rest of the study is organized as follows: the next section reviews the literature. In Section 3, the methodology and the data are briefly explained. In Section 4 and Section 5, the empirical results and concluding remarks are presented, respectively.

2. Literature Review

Our paper is related to the extensive literature about the saving outcomes of matching contributions and automatic enrollment programs. Many previous studies employed cross-sectional data and showed that matching contributions have a positive impact on participation rates (such as Andrews [7], Bassett, Fleming, and Rogrigues [8], Clark and Schieber [9], Huberman, Iyengar, and Jiang [10]). However, there are a few exceptions, such as Papke [11] and Kusko, Poterba, and Wilcox [12], who found no change in participation rates.

Studies that provide more convincing evidence are those conducted by Duflo et al. [13] and Engelhardt and Kumar [14]. Duflo et al. [13] carried out a randomized field experiment to analyze the effects of different matching rates on middle- and low-income individuals’ willingness to participate in and contribute to individual retirement arrangements. They found that increasing the match rate from 0% to 20% led to a 5% increase in participation and increasing the match rate from 20% to 50% resulted in a 6% increase in participation. Engelhardt and Kumar [14] used pension plan data from the Health and Retirement Study, focusing on older individuals with an average age of 55. They analyzed the effects of variations in match rates on the participation and contribution decisions of individuals. Similarly to Duflo et al. [13], they found that a 25% increase in the match rate leads to a 5% increase in participation.

Recent literature on individual retirement savings has placed significant emphasis on both economic and non-economic factors, such as income, age, job tenure, education, etc., as well as attitudes, such as status-quo bias, procrastination, choice overload, and peer effects [8,15,16,17,18]. Studies show that people’s behaviors towards retirement savings deviate from classical economic models and align with behavioral ones [3,19]. The literature suggests that people tend to procrastinate and do not accumulate enough money for their retirement (e.g., [20,21,22,23]). Beshears et al. [24] argue that the automatic enrollment mechanism is successful at increasing participation rates. Even though there is almost a consensus in the literature that the automatic enrollment mechanism positively affects participation rates (e.g., [4,5,6]), the findings also indicate that individuals either opt-out in a short period of time or stick around the conservative default options. Research indicates that the demographic traits of individuals, as well as features or options provided in various types of retirement savings plans, can have an impact on retirement savings outcomes. These factors include eligibility requirements, the ability for employees to access funds before retirement, the duration of participation [25], and the number of fund options available in 401(k) plans [17]. In general, participation in retirement savings plans tends to increase with age, income, job tenure, and education level.

Our study is closely related to those of Engelhardt and Kumar [26], Chetty et al. [27], and Falk and Karamcheva [28]. Engelhardt and Kumar [26] found that automatic enrollment is more effective in increasing participation in retirement savings plans compared to matching contributions. Chetty et al. [27] examined the impact of retirement savings subsidies and automatic contribution policies on overall savings and wealth accumulation. They calculate that for every dollar the government spends on subsidies, total savings would increase by only one cent. However, policies that raise retirement contributions when individuals do not take action, such as automatic employer contributions to retirement accounts, have a substantial impact on wealth accumulation. They find that automatic contributions are more successful in increasing savings rates compared to subsidies. Falk and Karamcheva [28] examined the participation of individuals in retirement saving plans. The participants had experienced two reforms in retirement savings in the same manner and order as those experienced in Turkey but the time period between the reforms was much longer than that of in Turkey. They found that the introduction of employer matching leads to a higher increase in both participation and contributions rates compared to automatic enrollment.

Research on the factors that influence individuals’ decision to opt-in or opt-out of an IPS and the impact of matching contributions and automatic enrollment systems in Turkey is scarce. Sener and Akın [29] argue that demographic characteristics, such as income and education level, play a significant role in individuals’ decision to enroll in an IPS, and that IPSs are not well-suited to low-income groups in Turkey who are in most need of retirement savings. Ozel and Yalcın [30] suggest that pension reform programs in Turkey may have limited success in increasing savings due to their reliance on voluntary participation and high fund management fees. Ertugrul, Gebesoglu, and Atasoy [31] found that although matching contributions to individual pension accounts initially increased participation, its impact decreased over time. Eren and Ileri [32] found a stronger effect for the matching contribution policy compared to that of proposed tax benefits on IPS participation in Turkey. They also recommended the elimination of fees for increasing IPS returns. Aydın and Selcuk [33] reported that people are more likely to participate in an IPS as they get older and earn more. They also show that continuation rates among AES participants were almost three times higher for individuals who are already enrolled in an IPS. Rudolph [34] conducted a recent study on pension plans and the AES in Turkey and highlighted the successes and failures of the system. Studies by Sataf and Yildirim [35] and Meral and Sener [36] suggested that financial constraints among participants and lack of confidence in the pension system are major weaknesses of IPSs in Turkey. While several studies on IPSs in Turkey exist, only Aydın and Selcuk [33], who used a limited sample, have investigated the effects of automatic enrollment on participation rates, and no study has ever compared the effects of matching contribution and automatic enrollment on participation rates in retirement savings plans in Turkey.

This study examines the effects of two retirement saving policies on a comprehensive population at the country level, using more than 40 million individual observations over a long period of time, which is unique from other studies which generally use company- or plan-level data for a limited period of time. Additionally, it compares the effects of matching contribution and automatic enrollment policies on participation rates in retirement saving plans in Turkey. Furthermore, the study examines the impact of the AES on the private pension system both directly and indirectly, thanks to the fact that records are kept separately for the AES and the individual pension system in Turkey. Indirect effects are also observable on the participation behavior of individuals over 45 who are not eligible for the AES.

3. Methodology and Data

To compare the effects of matching contributions and the automatic enrollment system on participation in the individual pension system in Turkey, we utilize the following empirical model, similar to that in Ertugrul et al. [31] (One might be tempted to think that interest rate must be included in the model. We chose not to include the real monthly interest rate in our model because more than three-fourths of net asset values of IPS funds (and even more than 90% before 2019) are deposited in interest-bearing IPS fund groups. Thus, when included in the model, the coefficient of the real monthly interest rate is found positive, contrary to expectations, and causes insignificant coefficients of return on IPS funds):

$$Y_{it}={\beta }_1+{\beta }_{2}IPSFund{s}_t+{\beta }_{3}Gol{d}_t+{\beta }_{4}Stoc{k}_t+{\beta }_{5}Dolla{r}_t+{\beta }_6{D}_{MC t}+{\beta }_7{D}_{AE t}+u_{it}$$

(1)

where Y_it (in order to minimize the coefficients of independent variables, we divide the Y values by 10,000) is the number of monthly net new participants to the IPS (not through automatic enrollment mechanism) in a pension company i in time t; IPSFunds_t is the monthly index measuring average real return on all IPS funds; Gold_t is the monthly real return on gold prices; Stock_t is the monthly real return on the Istanbul Stock Exchange Index (BIST 100); Dollar_t is the monthly real return on price of the US dollar in terms of Turkish lira; D_MC is the matching contributions dummy: 1 for the years the state provides subsidies for IPS participation (after 1 January 2013), 0 for other years; D_AE is the dummy for the AES: 1 for the years government implements the AE system (after 1 January 2017), 0 for other years; and u_it is the identically and independently normally distributed error term with zero mean and constant variance. The model also includes time dummies, but the results on time dummies are not reported. The model is estimated for nineteen pension companies for the all-sample period: 2004M01–2021M07.

All the investment instruments are hypothesized to affect individual pension system participation negatively since individual retirement accounts can be an alternative for private saving accounts [37], while averages of monthly real return on all IPS funds are hypothesized to have positive effects. It is worthwhile to note that households in Turkey have idiosyncratic saving attitudes. They allocate a large portion of their savings in gold as physical coins or jewelry, which is followed by foreign currencies [38,39]. In terms of foreign currency, the US dollar has been the most preferred currency followed by the Euro in Turkey. We chose to employ the dollar returns in the estimates because the correlation between the dollar and euro series is almost at the level of 0.80. (We also estimated our regressions with the Euro returns instead of the US dollar returns and obtained very similar results. These estimates are available upon request.) The expected effect of matching contribution dummy is positive, while the expected effect of automatic enrollment dummy on individual pension system is negative.

Given that financial return data are characterized by significant fluctuations, to mitigate these variations, all variables are employed in the form of a 12-month moving average. The study employs the fixed effect within-group estimation model, as the dataset exhausts the population, rather than being a sample drawn from the population. The fixed effect estimator, by controlling for selection characteristics, proves to be more resilient to issues of selection bias compared to the random effects estimator. The fixed effect within-group estimation approach allows for more robust results, particularly through the elimination of heterogeneity among cross-sections (by utilizing demeaned variables) and by addressing the problem of omitted variable bias that is unobservable across cross-sections.

As demographic characteristics have significant impacts on individuals’ saving behavior and aggregate data analysis of participation rates fails to capture the substantial variation among participants with different demographic characteristics, this study estimates the model by taking various demographic features of participants, such as gender, age, marital status, and educational attainment, into account. The database is segmented by demographic characteristics and the number of monthly net new participants is calculated for each group, based on gender, age groups, and education groups. The study employs four age groups and four educational groups. Participants’ age is considered at the entry date into the system. The age groupings are determined in accordance with those used by the Pension Monitoring Center, and only participants over 45 years of age are grouped in light of AES provisions, as previously stated. Educational attainment level indicates the last graduated level of education of participants at the time of entry into the system. Participants are grouped by educational attainment levels, including illiterate, primary, secondary, and tertiary. It is worth noting that the sample size of the demographically grouped participants is smaller than that of the aggregate data due to missing demographic information of participants in the dataset.

In this study, we obtained the data about participants and IPS fund indices from the Pension Monitoring Center in Turkey. Our dataset includes more than 40 million observations and covers both panel data at the company level in the private pension system from January 2004 to July 2021. We also obtained data for the returns of alternative investment instruments from The Central Bank of the Republic of Turkey. We present summary statistics for the 12-month moving average of monthly net new participant numbers at the company level in

Table 1. Summary statistics for 12-month moving average of monthly net new participant number to the IPS in Turkey.

	Number of Participants in Individual Pension System
Identity	Obs. (Number of Months)	Mean	Std. Dev.	Min.	Max.
1	170	615	413	8	1449
2	111	821	1284	1	4292
3	211	9274	3491	1511	16,980
4	211	9607	4434	2442	17,893
5	183	545	563	1	1662
6	115	526	267	9	962
7	211	648	482	60	1831
8	140	2348	765	87	3826
9	153	1323	978	8	2893
10	211	1842	594	585	2944
11	211	10,774	4809	2071	21,432
12	91	5341	2150	4	9067
13	211	1033	811	123	3296
14	85	3738	1010	3	5155
15	211	2934	865	1107	4626
16	199	4976	3613	461	18,027
17	101	5911	2657	33	9224
18	211	6090	2483	2296	11,766
19	12	14,471	1866	11,880	17,614
Total	3073	58,560	25,371	14,455	101,236

. To avoid confusion, we represent pension companies by their identity numbers, not by their names.

Table 2. Summary statistics for real monthly returns in Turkey (12-month moving averages, 2004–2021, %).

Variable	Obs.	Mean	Std. Dev.	Min	Max
IPS Funds	3073	0.23	0.60	−2.13	3.25
Stock	3073	0.55	2.30	−17.66	15.95
Gold	3073	0.82	1.50	−6.72	18.87
Dollar	3073	0.19	1.27	−6.42	12.23

shows the descriptive statistics of the investment instruments used in the study. It is worth noting that, on average, IPS funds yield relatively lower levels of real returns compared to other investment instruments. None of the maximum levels of returns are as attractive as matching contributions though. Therefore, we expect that matching contributions raise net new participation in the IPS.

4. Results

As illustrated in Figure 1, the number of participants in the AES has increased over time. The effectiveness of the AES in promoting participation is evident when compared to the matching contributions mechanism. Within the first five years of the AES’s implementation, the number of participants surpassed the number of IPS participants in the 20 years prior to the AES’s introduction. Given that the plan characteristics of both policies are largely similar, the significant difference in the growth rate of IPS and AES participation cannot be solely attributed to economic factors [4,5,40].

However, the AES policy has been less successful in augmenting contribution amounts. The average contribution amount of IPS participants consistently exceeds that of AES participants—for example, in December 2021, the average AES contribution amount was approximately 130 TL, while that of the IPS was approximately 450 TL. The low level of contributions may be attributed to the potential influence of the 3% default contribution rate as a focal point or anchor. Additionally, it is possible that the AES policy draws in a more diverse income demographic.

The opt-out rate among automatically enrolled employees is also substantial. As presented in

Table 3. Duration of participants or certificates in the pension system according to enrollment period by enrollment types (Due to the non-availability of data, we could not present the same indicator (certificates in the IPS, participants in the AES) for the duration period in either system. Since the share of participants with only one pension contract in the IPS is about 85%, to us, the difference between the indicators does not undermine the importance of the subject).

Participation or Enrollment Year	Voluntary—IPS (by 31 December 2021)				Compulsory—AES (by 30 June 2021)
	Terminated Pension Contracts	Contracts in Effect	Share of Contracts in Effect (%)	Number of Years in Effect	Opted-Out Participants	Participants in the System	Share of Participants in the System (%)
2003	19,518	2568	11	19
2004	299,937	46,410	13	18
2005	406,067	73,839	15	17
2006	487,489	103,707	17	16
2007	515,378	141,098	21	15
2008	510,976	146,979	22	14
2009	501,845	146,752	22	13
2010	512,086	156,557	23	12
2011	591,938	209,195	26	11
2012	695,486	325,245	31	10
2013	1,123,132	528,690	32	9
2014	1,125,516	603,896	34	8
2015	1,238,188	745,619	37	7
2016	1,052,174	790,579	42	6
2017	759,758	763,002	50	5	6,838,400	1,309,315	16
2018	613,308	769,042	55	4	4,848,607	1,597,785	25
2019	455,984	885,821	66	3	3,236,534	1,546,659	32
2020	279,608	854,873	75	2	1,236,794	968,310	44
2021	143,234	1,267,583	89	0–1	403,648	484,091	54

, the current status of all AES participants included in the system as of the report date indicates a high average opt-out rate of 70%. This is notably higher than the opt-out rates observed in developed countries [4,5,16,34]. The high opt-out rate in Turkey may be attributed to financial constraints and a lack of trust in the pension system [35,36].

We also document the new IPS participants with their past participation histories; in

Table 4. IPS–AES participants’ interaction on the basis of new certificates of IPS participants.

	Share of New Participants with an IPS Certificate in Effect (%)	Share of New Participants with a Terminated IPS Certificate (%)	Share of New Participants with an AES Certificate in Effect (%)	Share of New Participants with a Terminated AES Certificate (%)	Share of New Participants Never Have an IPS Certificate before, and Entered IPS for the First Time (%)
1st half of 2017	31.04	20.70	3.23	3.11	58.63
2nd half of 2017	32.86	21.66	7.24	9.23	56.81
1st half of 2018	32.83	23.33	10.24	14.79	55.82
2nd half of 2018	34.74	25.72	11.49	18.10	52.99
1st half of 2019	35.24	28.77	12.29	23.48	50.51
2nd half of 2019	33.50	28.14	13.39	28.67	52.18
1st half of 2020	34.44	30.32	14.69	33.64	49.87
2nd half of 2020	31.88	28.15	14.28	32.46	53.54
1st half of 2021	31.64	28.64	13.94	31.67	53.58
2nd half of 2021	21.81	20.14	10.99	24.03	67.46
Average	32.00	25.56	11.18	21.92	55.14

, it can be seen that on average, 22% of the opted-out AES participants participated in the IPS. On the one hand, it propounds that the AES could have an indirect positive effect on IPS participation even among active decision makers. On the other hand, it corroborates Thaler [41], who claimed once individuals start saving for retirement, they are probably going to continue to save.

The data presented in

Table 5. Regression Results for Monthly Net New Participation in Individual Pension System, 2004–2021.

Variables	1	2	3	4	5	6	7	8	9	10	11	12	13
	All	Female	Male	Age 0–24	Age 25–34	Age 35–44	Age 45+	Single	Married	Illiterate	Primary	Secondary	Tertiary
Matching Contribution Dummy	1.701 ***	0.734 ***	1.018 ***	0.357 ***	0.544 ***	0.443 ***	0.435 ***	0.634 ***	1.046 ***	0.133 *	0.244 ***	0.375 **	0.224 ***
	(9.490)	(10.09)	(8.949)	(10.04)	(8.718)	(8.392)	(8.437)	(9.090)	(8.099)	(1.805)	(3.904)	(2.204)	(2.806)
Automatic Enrollment Dummy	−0.311 ***	−0.110 ***	−0.203 ***	−0.028 **	−0.124 ***	−0.108 ***	−0.055 **	−0.096 ***	−0.230 ***	−0.014	−0.044 **	−0.130 **	−0.084 ***
	(−4.100)	(−3.761)	(−4.217)	(−2.034)	(−4.929)	(−4.856)	(−2.521)	(−3.226)	(−4.153)	(−0.744)	(−2.460)	(−2.468)	(−3.157)
Real Return on IPS Funds	0.129 ***	0.060 ***	0.081 ***	0.027 ***	0.037 **	0.033 **	0.042 ***	0.071 ***	0.046	0.043 *	0.022	0.009	0.019
	(2.677)	(3.163)	(2.636)	(2.952)	(2.280)	(2.331)	(3.009)	(3.790)	(1.335)	(1.658)	(1.112)	(0.163)	(0.737)
Stock Market Real Return	−0.070 ***	−0.035 ***	−0.041 ***	−0.017 ***	−0.022 ***	−0.018 ***	−0.020 ***	−0.028 ***	−0.041 ***	−0.012	−0.007	0.001	−0.005
	(−5.533)	(−6.716)	(−5.039)	(−6.561)	(−5.036)	(−4.651)	(−5.408)	(−5.634)	(−4.431)	(−1.614)	(−1.236)	(0.037)	(−0.662)
Real Return on Gold	−0.095 ***	−0.049 ***	−0.052 ***	−0.023 ***	−0.026 ***	−0.022 ***	−0.030 ***	−0.036 ***	−0.074 ***	−0.020 **	−0.003	−0.005	−0.004
	(−5.455)	(−6.908)	(−4.714)	(−5.808)	(−4.250)	(−4.283)	(−5.856)	(−5.109)	(−5.722)	(−2.287)	(−0.558)	(−0.317)	(−0.590)
US Dollar Real Return	−0.106 ***	−0.045 ***	−0.063 ***	−0.024 ***	−0.035 ***	−0.027 ***	−0.026 ***	−0.037 ***	−0.058 ***	−0.010	−0.016 ***	−0.013	−0.008
	(−6.187)	(−6.720)	(−5.826)	(−6.489)	(−5.980)	(−5.290)	(−5.240)	(−5.698)	(−4.730)	(−1.523)	(−2.808)	(−0.847)	(−1.053)
Constant	−0.971 ***	−0.447 ***	−0.570 ***	−0.246 ***	−0.317 ***	−0.230 ***	−0.246 ***	−0.371 ***	−0.576 ***	−0.103	−0.140 **	−0.105	−0.06
	(−5.972)	(−6.767)	(−5.525)	(−7.641)	(−5.592)	(−4.805)	(−5.254)	(−5.867)	(−4.923)	(−1.447)	(−2.415)	(−0.664)	(−0.812)
Observations	3048	3045	3039	3005	3046	3027	3048	2880	2887	1424	2615	2442	2482
R-squared	0.788	0.806	0.774	0.700	0.798	0.785	0.744	0.739	0.786	0.578	0.735	0.726	0.768

Robust t-statistics in parentheses, and *** p < 0.01, ** p < 0.05, * p < 0.1.

utilize a panel analysis of all available information. The first column displays the results of the basic model for the entire sample for the period between 2004 and 2021. The study finds that matching contributions increase monthly net new participation in the IPS by 17,000 individuals, or 29% of the mean 12-month moving average of total IPS participation, as seen in Table 1. Conversely, the AES decreases participation by 3110 individuals, or 5%. This suggests that the AES may have a crowding-out effect on IPS participation. It should be noted that if AES participation is not tracked through a separate account, the coefficient on the AES dummy may have a positive effect. (When we consider monthly net new participation for all participation (IPS + AES) as the dependent variable for the robustness check for the whole period (2004–2021), the estimation results are mostly in line with our main results. We found the positive but insignificant coefficients for the dummy of the AES in most cases, which could stem from the modelling of mandatory participation. The results are presented in Appendix

Table A4. Regression Results for Monthly Net New Total Participation in IPS and AES, 2004–2021.

	1	2	3	4	5	6	7	8	9	10	11	12	13
Variables	All	Women	Men	Age 0–24	Age 25–34	Age 35–44	Age 45+	Single	Married	Illiterate	Primary	Secondary	Tertiary
State Subsidy Dummy	3.806 ***	1.593 ***	2.302 ***	1.231 ***	1.247 ***	0.797 ***	0.703 ***	1.885 ***	2.001 ***	−0.972	0.144	−2.135 *	−0.079
	(4.443)	(5.394)	(3.862)	(5.094)	(3.556)	(2.734)	(10.29)	(3.971)	(3.776)	(−0.660)	(1.028)	(−1.800)	(−0.450)
Automatic Enrollment Dummy	0.543	0.127	0.410	0.363 ***	0.147	0.057	−0.034	0.530 ***	0.036	0.136	−0.018	0.363	0.082
	(1.441)	(0.990)	(1.621)	(3.612)	(0.995)	(0.461)	(−1.139)	(2.593)	(0.159)	(0.365)	(−0.426)	(1.317)	(1.283)
Real Return on IPS Funds	0.315	0.109	0.229	0.125 **	0.116	0.071	0.046 **	0.224 *	0.099	−0.254	0.002	−0.173	−0.033
	(1.365)	(1.375)	(1.424)	(2.009)	(1.220)	(0.899)	(2.482)	(1.753)	(0.694)	(−0.486)	(0.035)	(−0.600)	(−0.848)
Stock Market Real Return	−0.199 ***	−0.083 ***	−0.124 ***	−0.069 ***	−0.072 ***	−0.044 **	−0.035 ***	−0.103 ***	−0.100 ***	0.112	0.003	0.135	0.01
	(−3.275)	(−3.945)	(−2.936)	(−4.027)	(−2.837)	(−2.100)	(−7.255)	(−3.027)	(−2.673)	(0.760)	(0.236)	(1.510)	(0.924)
Real Return on Gold	−0.331 ***	−0.144 ***	−0.200 ***	−0.126 ***	−0.121 ***	−0.088 ***	−0.020 ***	−0.182 ***	−0.153 ***	0.001	0.000	0.016	0.083 ***
	(−4.347)	(−5.382)	(−3.430)	(−6.020)	(−3.427)	(−3.099)	(−3.262)	(−3.823)	(−2.980)	(0.003)	(0.006)	(0.185)	(5.055)
US Dollar Real Return	−0.248 ***	−0.107 ***	−0.149 ***	−0.092 ***	−0.079 **	−0.045	−0.042 ***	−0.129 ***	−0.119 **	0.078	−0.006	0.245 **	0.040 **
	(−3.247)	(−4.071)	(−2.618)	(−4.354)	(−2.361)	(−1.622)	(−6.805)	(−2.883)	(−2.373)	(0.570)	(−0.477)	(2.108)	(2.276)
Constant	−2.632 ***	−1.124 ***	−1.587 ***	−0.975 ***	−0.865 ***	−0.488 *	−0.460 ***	−1.385 ***	−1.338 ***	1.008	−0.036	2.215 **	0.211
	(−3.388)	(−4.199)	(−2.936)	(−4.443)	(−2.722)	(−1.844)	(−7.424)	(−3.219)	(−2.786)	(0.712)	(−0.276)	(2.040)	(1.345)
Observations	3073	3072	3071	3071	3070	3071	3073	2927	2939	1494	2726	2604	2620
R-squared	0.667	0.672	0.659	0.651	0.648	0.620	0.769	0.652	0.627	0.356	0.599	0.484	0.667

Robust t-statistics in parentheses, and *** p < 0.01, ** p < 0.05, * p < 0.1.

. However, when considering the period 2014–2021, we found positive and statistically significant effects of the AES on all participation, as can be seen in Appendix

Table A5. Regression Results for Monthly Net New Total Participation in IPS and AES, 2014–2021.

	1	2	3	4	5	6	7	8	9	10	11	12	13
Varıables	All	Women	Men	Age 0–24	Age 25–34	Age 35–44	Age 45+	Single	Married	Illiterate	Primary	Secondary	Tertiary
Automatic Enrollment Dummy	2.089 ***	0.758 ***	1.331 ***	1.013 ***	0.567 **	0.343 *	0.165 ***	1.501 ***	0.692 **	0.291	−0.011	0.483	−0.013
	(3.697)	(3.832)	(3.532)	(6.888)	(2.413)	(1.704)	(5.502)	(5.105)	(1.996)	(0.590)	(−0.210)	(1.093)	(−0.152)
Real Return on All IPS and AES Funds	1.646 **	0.542 *	1.104 **	0.602 ***	0.440	0.235	0.369 ***	0.901 **	0.681	−0.152	0.037	−0.082	0.033
	(2.072)	(1.949)	(2.083)	(2.911)	(1.330)	(0.831)	(8.725)	(2.213)	(1.395)	(−0.247)	(0.508)	(−0.149)	(0.304)
US Dollar Real Return	−0.628 ***	−0.234 ***	−0.394 ***	−0.254 ***	−0.160 *	−0.076	−0.138 ***	−0.363 ***	−0.241 *	0.049	−0.016	0.054	−0.021
	(−2.917)	(−3.104)	(−2.746)	(−4.534)	(−1.788)	(−0.988)	(−12.05)	(−3.289)	(−1.822)	(0.317)	(−0.875)	(0.327)	(−0.662)
Real Return on Gold	−0.432 *	−0.188 **	−0.244	−0.264 ***	−0.098	−0.046	−0.024 *	−0.336 ***	−0.096	−0.034	−0.005	−0.103	−0.005
	(−1.825)	(−2.274)	(−1.543)	(−4.275)	(−0.995)	(−0.547)	(−1.937)	(−2.772)	(−0.661)	(−0.172)	(−0.225)	(−0.543)	(−0.150)
Stock Market Real Return	−0.736 ***	−0.250 ***	−0.486 ***	−0.234 ***	−0.211 **	−0.124	−0.166 ***	−0.380 ***	−0.325 **	0.074	−0.017	0.120	−0.041
	(−3.254)	(−3.159)	(−3.221)	(−3.981)	(−2.246)	(−1.536)	(−13.83)	(−3.280)	(−2.342)	(0.442)	(−0.820)	(0.514)	(−0.905)
Constant	0.185	0.022	0.163	−0.258	0.122	0.115	0.205 ***	−0.269	0.303	−0.146	0.098	−0.137	0.203*
	(0.268)	(−0.09)	(0.355)	(−1.433)	(0.425)	(0.468)	(5.598)	(−0.759)	(0.716)	(−0.275)	(1.596)	(−0.258)	(1.898)
Observations	1548	1548	1548	1548	1548	1548	1548	1496	1496	712	1380	1350	1342
R-squared	0.766	0.757	0.764	0.770	0.728	0.694	0.922	0.765	0.724	0.584	0.801	0.667	0.808

Robust t-statistics in parentheses, and *** p < 0.01, ** p < 0.05, * p < 0.1.

). Additionally, the study finds that increases in real returns for IPS funds lead to increased participation, while increases in alternative investment instruments lead to decreased participation, indicating that investment returns are a significant factor for IPS participants. Specifically, a 1% increase in real returns for IPS funds leads to a 2.2% increase in monthly net new participation in the IPS, or 1300 individuals, while a 1% increase in alternative saving instruments, such as real returns on gold or the US dollar, leads to a 1.7% decrease in monthly new participation in the IPS, or 1000 individuals.

The second and third columns of Table 5 present the estimations for female and male participants, respectively. Consistent with the results from the full sample, both male and female participants exhibit responsiveness to returns on alternative investment instruments, matching contributions, and returns on IPS funds. However, the relatively higher estimated coefficients for all dependent variables for males compared to those for females suggest that female participants are relatively less sensitive to the returns of their investments. This difference in sensitivity is also reflected in the trading frequency of female participants on the Pension Fund Trading Platform (BEFAS), with females between the ages of 20 and 50 displaying an average trading frequency three times lower than that of their male counterparts.

Research by Chetty et al. [27] indicates a positive correlation between income/wealth level and the sensitivity of individuals to saving/price subsidies. Given that risk awareness or lower risk taking might be an income effect and female employees in Turkey earn less than their male counterparts at every education and occupational level [42], this may explain why women are relatively less sensitive to matching contributions [43]. The same may be true for married participants, as discussed below. Furthermore, a significant proportion of women in Turkey are not participating in the labor market and do not possess an independent income source. According to the World Bank’s labor force statistics, the female labor force participation rate in 2012 stood at 30.3%, which is significantly lower than the male labor force participation rate of 70.9%. As a result, a significant number of women may not possess the financial means to contribute to the IPS. This may also account for the lower sensitivity of women to matching contributions policies and investment returns in their decisions to participate in the IPS. Furthermore, the small differences in our estimations regarding gender may also be attributed to the well-established finding of finance/financial behavior literature that women tend to have higher levels of financial illiteracy and risk aversion.

Before examining the regression results for different age groups, it would be useful to see the age distribution of IPS participations. Figure 2a,b show the trend for monthly net new IPS participation of different age groups. Regarding the shares of each age group in total monthly net new participation, the decreasing trend of the 25–34 age group and the increasing trend of the eldest group were notable. In line with findings in the related literature, the eldest people have the highest participation ratio (e.g., [4,44]).

Many studies refer to the life-cycle model and behavioral economics theory in their explanations about how participants’ age affects participation in retirement saving plans [8,18,25]. According to the life-cycle model, individuals in their 20s or 30s, who could also be in the beginning of their careers with relatively low earnings, might save for home ownership or prepare for parenthood, neglecting retirement savings [18]. (Our results on age groups could also reflect the inclusive structure of age in terms of tenure and earnings since we omit these variables due to the nonavailability of data. While the effect of earnings is clear, tenure has an important influence on participation; not only are more-tenured people older, but also they have more financial experience (obtained via workplace financial education) and social interaction on the job [45,46]. The life-cycle theory along with retirement plan provisions could also explain the decreasing share of participation for young to middle-aged adults (between 25 and 34 years old). For the sake of clarity, individuals obtain the right of retirement, which also means the acquisition of full benefits of matching contributions, by satisfying two principal conditions: the first is remaining in the system for ten years and the second is turning 56 years old. Indeed, these two provisions are important because they affect a crucial factor of participation—the length of an individual’s saving horizon [25].

The results presented in Table 5 demonstrate that, similar to the full sample, participants in all age groups exhibit similar behavior with regard to participation in the IPS in terms of returns. The coefficients estimated for matching contributions reveal minimal variations across age groups, while those for the AES reveal significant variations, likely due to the inclusion criteria for the AES, which exclude non-workers and employees under the age of 45. As a result, the negative impact of the AES on IPS participation is smallest in the 0–24 age group and the group aged 45 and above. Furthermore, the results for the youngest group may be attributed to their financial constraints and higher levels of educational attainment, which make them less sensitive to retirement savings considerations.

The minor differences in the results across age groups comply with the life-cycle theory and also with the provisions/requirements of the pension system itself. Younger individuals seem to be less sensitive to matching contributions because they are in the early stages of retirement and they may seek more liquid investments with short-term returns in order to prepare for marriage or parenthood, whereas the opposite is the case for elderly individuals. However, it should be kept in mind that the effect of age might contain the effects of tenure or income, which we cannot observe.

Columns 8 and 9 of Table 5 present the results for the single and married participants. We found that married participants are more sensitive to matching contributions than their single counterparts. The reason for this may be that married participants may be more risk averse than their single counterparts and they evaluate matching contributions as a saving instrument with guaranteed yields [47].

The final four columns of Table 5 present the fixed effect estimations for education groups. Our dataset includes education data for 65% of all the participants in the sample. Of these, approximately 50% have a secondary education, 27% have a tertiary education, and 20% have a primary education. The percentage of participants who are illiterate is negligibly small at around 3%. The estimates for education groups reveal mixed and intriguing results. Although all coefficients have the expected signs, only the matching and automatic enrollment dummies are statistically significant. As expected in the literature, the effect of matching contribution on IPS participation is increasing with the education level. Although one might expect that the higher the education level is, the higher the responsiveness to matching contributions would be, the negligible difference in our estimates may stem from the fact that they are disproportionately included in the IPS. The majority of the coefficients on returns are not statistically significant. Given the mostly insignificant coefficients, it can be argued that the returns of financial investment instruments do not have a significant effect on participation in the IPS among different education levels. As education might not be a proper proxy for financial literacy, our examination of the effect of education could be improved by taking into account the participants’ financial literacy level since it is also an important determinant of financial behaviors.

When all of the results are taken into consideration, matching contributions promotes participation in the IPS, and the AES has a negative impact on participation. Male participants are more responsive to matching contributions than females, younger individuals are less responsive than other age groups, and education level does not appear to have a significant effect on individuals’ attitudes towards IPS participation. Overall, participants are sensitive to both the returns of IPS funds and returns of alternative investment instruments. Although returns of all alternative investment instruments have similar effects on participation, gold and dollar returns have a higher relative impact. As discussed above, gold is considered a reliable store of wealth in Turkey, particularly among older generations, and gold and foreign exchange are the most commonly preferred investment instruments in Turkey.

5. Conclusions

This study examines the impact of two significant reforms related to retirement savings in Turkey on individuals’ participation in the Individual Pension System (IPS) between 2004 and 2021, using fixed effect estimations. We find that matching contributions led to an increase in participation in a traditional opt-in plan, while the AES had a crowding-out effect on IPS participation. However, both mechanisms ultimately led to an overall increase in participation. Our results support the hypothesis that the AES is more effective than the matching contribution mechanism in increasing total participation, despite the high opt-out rates relative to those in the USA, the United Kingdom, and New Zealand, probably because it does not effectively encourage individuals to make active decisions regarding contribution rates, portfolio choices, or asset classes, as participants tend to mostly rely on default options.

This study also looks at the effects of matching contributions and the AES on participation among different demographic groups and finds that married and male individuals respond more strongly to matching contributions. Our results also indicate that education levels moderately affect the IPS participation levels. This study finds that IPS fund returns raise IPS participation while the returns on alternative investment instruments decrease it.

As for policy implications, based on our estimates, our recommendations for increasing participation include implementing regulations to increase returns on IPS funds and providing information on returns through various channels, as well as allowing individuals to access their assets before retirement without penalty. As a developing country with a financially constrained population (often lacking income), our estimation results of the comparative effects of state matching contributions and subsequent automatic enrollment policies in Turkey serve a good benchmark for other middle-income countries in the world.