Peer Effects in Retirement Decisions

Peer Effects in Retirement Decisions Mario Meier 1 & Andrea Weber 2 1 University of Mannheim 2 Vienna University of Economics and Business, CEPR, IZA Meier & Weber (2016) Peers in Retirement 1 / 35

Motivation Public retirement insurance systems are most expensive public programs in many countries Often these systems are under great financial pressure Policy often tries to increase labor force participation of the elderly Understanding the incentives that determine when workers retire is important for policy Meier & Weber (2016) Peers in Retirement 2 / 35

Peer Effects in Retirement Decisions Standard theory would attribute retirement age decision to financial incentives only However, finanical incentives are very complex and decision must be made under a fair amount of uncertainty Individuals might refer to the behavior of peers In general, hard to identify endogenous peer effects (Manski, 1993) This paper: Exploits quasi-experimental variation in early retirement ages in Austria Estimates peer effects in co-worker networks Discusses mechanisms and implications of peer effects in retirement decisions Meier & Weber (2016) Peers in Retirement 3 / 35

Overview Findings: For men, we find that a peer retiring one year later induces a co-worker to retire two months later For women, we find only small peer effects Peer effects work mainly through coordination of the retirement age Mechanism: suggestive evidence that information about costs and benefits of early retirement matters Implications: The estimated peer effects imply a substantial social multiplier A fair amount of the bunching response at the early retirement age can be attributed to social interactions Labor supply responses and micro elasticity get amplified by a social multiplier Meier & Weber (2016) Peers in Retirement 4 / 35

Related Literature This paper is connected to several strands of the literature Peer effects literature Peer effects: Beshears et al. (2015), Bursztyn et al. (2014), Brown and Laschever (2012), Dahl et al. (2014), Duflo and Saez (2003, 2002), Manski (1993), Mas and Moretti (2009), Moffitt (2001) Social multiplier: Brock and Durlauf (2001), Glaeser et al. (2003), Glaeser and Scheinkman (2002), Kroft (2008) Retirement insurance literature Brinch et al. (2016), Brown (2013), Carroll et al. (2009), Chalmers et al. (2014), Chetty et al. (2014), Gruber and Wise (2009), Inderbitzin et al. (2016), Madrian and Shea (2001), Manoli and Weber (2011, 2016), Mastrobuoni (2009), Staubli and Zweimüller (2013) Bunching literature Best et al. (2015a, 2015b), Chetty et al. (2011), Einav et al. (2016), Gelber et al. (2016), Kleven et al. (2016), Kleven (2016), Kleven and Waseem (2013), Saez (2010), Seibold (2016) Meier & Weber (2016) Peers in Retirement 5 / 35

Table of Contents 1 Institutional ground & Data 2 Do peers influence their co-workers? 3 Why do peers influence their co-workers? 4 What are the implications of peer effects? Meier & Weber (2016) Peers in Retirement 6 / 35

Pension Reform of 2000 for Men Cohort rules for women Budget constraints for workers Meier & Weber (2016) Peers in Retirement 7 / 35

Data Austrian Social Security Database (Zweimüller et al., 2009) Universe of Austrian private employees Observe exact date of labor force exit and entry into retirement Complete employment histories for every privately employed employee The population of interest consists of 108, 628 men from the cohorts 1939-1944 and 81, 863 women from the cohort 1944-1949 Main outcome of interest: Labor force exit, which we define as the retirement age Descriptive Statistics Meier & Weber (2016) Peers in Retirement 8 / 35

Empirical Strategy Partial population approach for identification of peer effects Focus on 1-to-1 peer networks (Dahl et al., 2014): y 1g = α 1 + β 1 y 2g + γ 1 x 1g + τ 1 x 2g + θ 1 w g + λ 1 p 1g + e 1g (1) y 2g = α 2 + β 2 y 1g + γ 2 x 2g + τ 2 x 1g + θ 2 w g + λ 2 p 2g + e 2g (2) Equation (1) describes the retirement decision y 1g of the peer, where g denotes the peer group Peers face exogenously different costs of retirement p 1g, i.e. a variable that only enters the peer equation and is orthogonal to any unobservable variable affecting the co-worker decision y 2g β 2 is the effect of the peer retirement decision on the co-workers retirement decision (peer effect) Meier & Weber (2016) Peers in Retirement 9 / 35

Research Design Identification Meier & Weber (2016) Peers in Retirement 10 / 35

Results for Men Meier & Weber (2016) Peers in Retirement 11 / 35

Effects of Pension Reform for Men Meier & Weber (2016) Peers in Retirement 12 / 35

Definition of Peer Network I Consider workplace networks and identify peers and co-workers that work together at the same firm at the onset of the retirement decision of the peer Consider only firm networks with at most 10 workers in the cohorts from 1939 until the end of 1944 Define a male peer as someone who 1 works at least until the age of 60 and at most until 65 2 works from the age of 59 until 60 at a same single firm (no job switches) 3 was born between 1939 and the end of 1942 4 is either exempt or non-exempt from the reform Meier & Weber (2016) Peers in Retirement 13 / 35

Definition of Peer Network II Define a co-worker as someone who 1 works at least until the age of 60 and at most until 65 2 is strictly younger than a peer and is at most 2 years older than the peer 3 was born in between 1939 and the end of 1944 4 worked at the same firm with a peer at the time the peer was aged 59 until age 60 Peer is always the oldest individual; other peers are then set to be co-workers Hence, there is always one peer and possibly multiple co-workers In 1-to-1 networks consider only the first co-worker (younger ones are ignored, due to possible multiplier effects) In total, end up with 9,934 1-to-1 peer networks and 7,066 with a maximum of 10 old workers Meier & Weber (2016) Peers in Retirement 14 / 35

First-Stage Relationship I Meier & Weber (2016) Peers in Retirement 15 / 35

Reduced-form I Meier & Weber (2016) Peers in Retirement 16 / 35

Peer Effects Men I (1) (2) (3) (4) VARIABLES Ret. Age (Co-worker) Ret. Age (Co-worker) Ret. Age (Co-worker) Ret. Age (Co-worker) Retirement Age (Peer) 0.150*** 0.146*** 0.114*** 0.111*** (0.035) (0.036) (0.036) (0.037) Observations 7,066 6,974 6,940 6,880 Region FE - yes - yes Industry FE - - yes yes First Stage F 26.44 25.78 24.85 24.16 Robust standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1 The presence of a peer that retires one year later, implies that the co-worker retires roughly 2 months later. Meier & Weber (2016) Peers in Retirement 17 / 35

First-Stage Relationship II Meier & Weber (2016) Peers in Retirement 18 / 35

Reduced-form II Meier & Weber (2016) Peers in Retirement 19 / 35

Peer Effects Men II (1) (2) (3) (4) VARIABLES Take-up (Co-worker) Take-up (Co-worker) Take-up (Co-worker) Take-up (Co-worker) Take-up (Peer) 0.133*** 0.134*** 0.116*** 0.116*** (0.029) (0.029) (0.030) (0.030) Observations 7,066 6,974 6,940 6,880 Region FE - yes - yes Industry FE - - yes yes First Stage F 44.91 43.47 42.71 41.41 Robust standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1 The presence of a peer that takes up early retirement right at 60, makes it 13% points more likely that the co-worker also retires at 60. Meier & Weber (2016) Peers in Retirement 20 / 35

Mechanisms behind Peer Effects Some Theories: Information transmission or learning about costs and benefits of (early) retirement Co-workers should retire at similar ages than the peer Bunching of the peer should increase probability that co-workers bunch Firm responses If peer retires, firm wants to keep next-old worker longer or shorter Why retirement age coordination? Leisure complementarities between peers and co-workers Peers should coordinate date of labor market exit Peer effect should be similarly strong in both directions Retirement age norms Co-workers derive utility from retiring at the peers age Information about existence of exemptions and early retirement ages Unlikely, exemption and reform were very salient and publicly debated Meier & Weber (2016) Peers in Retirement 21 / 35

Peer Effect by Network Size Regression table Meier & Weber (2016) Peers in Retirement 22 / 35

Peer Effect by Income Difference Regression table Meier & Weber (2016) Peers in Retirement 23 / 35

Reverse Peer Effects (1) (2) (3) (4) VARIABLES RA (Co-worker) RA (Co-worker) RA (Co-worker) RA (Co-worker) Retirement Age (Co-worker) 0.079*** 0.085*** 0.054* 0.060** (0.029) (0.029) (0.029) (0.030) Observations 7,066 6,974 6,940 6,880 Region FE - yes - yes Industry FE - - yes yes First Stage F 23.09 22.68 25.87 25.12 Robust standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1 Meier & Weber (2016) Peers in Retirement 24 / 35

Timing of Retirement I Meier & Weber (2016) Peers in Retirement 25 / 35

Timing of Retirement II Dependent Variable: Retirement Choice Co-worker (1) (2) (3) (4) VARIABLES OLS IV IV IV Exit Age Peer 0.038*** 0.125*** 0.109*** 0.109*** (0.004) (0.015) (0.015) (0.015) Cohort-specific ERA 0.315*** 0.310*** 0.305*** 0.306*** (0.009) (0.009) (0.009) (0.009) Cohort-specific ERA x Exempt -0.206*** -0.206*** -0.205*** -0.204*** (0.011) (0.011) (0.010) (0.010) Age 60 Dummy 0.064*** 0.035*** 0.020*** 0.035*** (0.006) (0.007) (0.007) (0.007) Age 60 Dummy x Exempt 0.338*** 0.335*** 0.335*** 0.335*** (0.010) (0.010) (0.010) (0.010) December 0.021*** 0.021*** 0.021*** 0.021*** (0.001) (0.001) (0.001) (0.001) Exit Date Peer 0.038*** 0.041*** 0.032*** 0.031*** (0.009) (0.009) (0.009) (0.009) Observations 438,092 438,092 438,092 438,092 Age Polynomial - - yes - Age Non-parametric - - - yes First Stage F 907.4 862.7 875.3 Clusters 7066 7066 7066 Robust standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1 Meier & Weber (2016) Peers in Retirement 26 / 35

Heterogeneity & Robustness Checks Peer effects in the service industry are considerably larger Industry Heterogeneity Peer effects are larger in Vienna Regional Heterogeneity Changing the considered birth cohorts and estimation window doesn t affect results Alternating the maximal age difference between peer and co-worker does not change the results Age Difference The length of the joint working time at a firm does not change the results Joint Time Placebo tests with fake reform windows generate zero IV estimates Placebo Clustering standard errors on the cohort or firm level does not alter the significance Standard Errors Adding age difference controls between peers and co-workers does not affect results Age Difference Meier & Weber (2016) Peers in Retirement 27 / 35

Bunching and Peer Effects Part of bunching might be due to peer effects Standard theories of social interactions predict herding/bunching behavior in social equilibrium Natural question: How much of the bunching mass is due to a social equilibrium? In a social interactions model bunching might endogenously arise as a social norm/equilibrium Estimated elasticity ε s with respect to benefits is therefore: ε s = mε b Estimated elasticity is the product of a social multiplier and the micro-elasticity Static Social Interactions Model Meier & Weber (2016) Peers in Retirement 28 / 35

The Social Multiplier Glaeser et al. (2003) (with N peers): N S(P) = γ δ d 1 P d d=1 The closest peer gives social utility γ Social utility decays at rate δ with social distance Having a peer that retires early increases the probability that the co-worker retires early As N the model converges to a social equilibrium take-up ratio T = m is the social multiplier 1 δ 1 γ δ T 1 = mt 1 Meier & Weber (2016) Peers in Retirement 29 / 35

Social Multiplier Regression table Meier & Weber (2016) Peers in Retirement 30 / 35

Bunching Decomposition Estimated Parameters: ˆγ = 0.163 and ˆδ = 0.676 = social multiplier: ˆm = 2.01 Roughly 50% of bunching can be attributed to a social equilibrium Calibration of static retirement model to match observed bunching mass of 20%: Meier & Weber (2016) Peers in Retirement 31 / 35

Results for Women Meier & Weber (2016) Peers in Retirement 32 / 35

Peer Effects Women (1) (2) (3) (4) VARIABLES Ret. Age (Co-worker) Ret. Age (Co-worker) Ret. Age (Co-worker) Ret. Age (Co-worker) Retirement Age (Peer) 0.052** 0.051** 0.039 0.038 (0.024) (0.024) (0.025) (0.025) Observations 7,930 7,778 7,639 7,551 Region FE - yes - yes Industry FE - - yes yes First Stage F 75.24 73.26 68.60 67.01 Robust standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1 The presence of a peer that retires one year later, implies a small or no effect on the co-workers retirement age. 1 st stage women Take-up of early retirement results Meier & Weber (2016) Peers in Retirement 33 / 35

Why do women react less to their workplace peers? Many women have not collected sufficient insurance years in order to be eligible for early retirement bunching at 60, otherwise left with nothing Women might refer to different peer groups: husband, friends Husband might play a particularly important role for women Women sample is more selected (labor force participation of older women lower in general) Women work in different occupations than men and earn fewer income on average wealth in retirement depends on husband Meier & Weber (2016) Peers in Retirement 34 / 35

Conclusion We document the importance of peer effects for the retirement decision among men Women show only minor peer effects We find evidence that co-workers time the particular retirement age The implied social multiplier suggests that a large fraction of bunching is due to a social equilibrium Financial incentives are just one margin of the retirement decision Findings have important implications for the understanding of elderly labor supply Understanding the mechanisms behind the peer effects better can have important policy implications Meier & Weber (2016) Peers in Retirement 35 / 35

Appendix Meier & Weber (2016) Peers in Retirement 36 / 35

Pension Reform of 2000 for Women Meier & Weber (2016) Peers in Retirement 37 / 35

Financial Incentives of Workers Meier & Weber (2016) Peers in Retirement 38 / 35

Descriptives Men Women Variables N Mean s.d. N Mean s.d. Long Insurance Years 108,628 0.543 (0.498) 81,863 0.264 (0.441) Retirement Age 108,628 58.55 (2.656) 81,863 56.69 (2.244) Claim Age 108,628 59.40 (2.467) 81,863 57.36 (2.028) Take-up of Early Retirement 108,628 0.726 (0.446) 81,863 0.285 (0.451) Manufacturing 108,158 0.573 (0.495) 81,658 0.268 (0.443) Vienna 107,850 0.304 (0.460) 81,459 0.328 (0.469) Notes: This table shows summary statistics for men born in cohorts 1939-1944 and women born in cohorts 1944-1949. Meier & Weber (2016) Peers in Retirement 39 / 35

Identification We aim to estimate an endogenous peer effect in the fashion of Manski (1993) Instrument vector: p 1g = (q c,1g, ν 1g, q c,1g ν 1g ) where q c,1g is a set of peer quarter-of-birth dummies and ν 1g is a dummy for the potential exemption from the reform Potentially exempt are all men who have contributed at least 38 years to the retirement system at age 53 Full interaction captures institutional environment and incentives for peers and co-workers completely Identification Assumptions Meier & Weber (2016) Peers in Retirement 40 / 35

Identification Assumptions Conditional Independence Conditional on X, the cohort of the peer and the exemption status of the peer is independent of any unobservable characteristic (e 2g ) affecting the co-workers retirement decision Exclusion Restriction The birth cohort of the peer and the exemption status of the peer does not directly enter the retirement decision of the co-worker First Stage The reform significantly alters the decision to retire for the peer Monotonicity The reform does not induce any peer to retire earlier than in the counterfactual with no reform Meier & Weber (2016) Peers in Retirement 41 / 35

Peer Effect by Network Size (1) (2) (3) VARIABLES Size: less or equal 3 Size: between 4 and 9 Size: 10 or more Retirement Age (Peer) 0.199*** 0.098** 0.010 (0.059) (0.045) (0.046) Observations 2,706 4,002 3,226 First Stage F 10.44 16.83 15.24 Robust standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1 The larger the firm (number of workers within considered birth cohorts), the smaller the peer effect and vice versa could be that the influence of a single peer depends on the number of other potential peers Meier & Weber (2016) Peers in Retirement 42 / 35

Peer Effect by Income Difference (1) (2) (3) VARIABLES Large Income Difference Pooled Sample Small Income Difference Retirement Age (Peer) 0.084* 0.150*** 0.199*** (0.044) (0.034) (0.052) Observations 3,732 7,066 3,334 First Stage F 16.89 26.96 11.20 Standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1 Meier & Weber (2016) Peers in Retirement 43 / 35

Peer Effects by Industry (1) (2) (3) (4) VARIABLES RA Manufact RA Not Manufact Take-up Manufact Take-up Not Manufact Retirement Age (Peer) 0.117** 0.340*** (0.052) (0.055) Take-up (Peer) 0.112*** 0.269*** (0.043) (0.044) Observations 3,602 3,463 3,602 3,463 First Stage F 13.91 13.70 25.93 20.94 Robust standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1 Peer Effects in the service industry are considerably larger than in manufacturing. Meier & Weber (2016) Peers in Retirement 44 / 35

Peer Effects by Region (1) (2) (3) (4) VARIABLES RA Vienna RA Not Vienna Take-up Vienna Take-up Not Vienna Retirement Age (Peer) 0.204*** 0.116** (0.051) (0.046) Take-up (Peer) 0.220*** 0.188*** (0.050) (0.038) Observations 2,247 4,808 2,247 4,808 First Stage F 11.93 16.35 17.32 30.73 Robust standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1 Peer Effects in the service industry are considerably larger than in manufacturing. Meier & Weber (2016) Peers in Retirement 45 / 35

Peer Effects by Age Difference (1) (2) (3) VARIABLES Age diff 1 Age diff 2 Age diff 3 Retirement Age (Peer) 0.147*** 0.160*** 0.134*** (0.043) (0.035) (0.033) Observations 4,794 6,937 8,046 First Stage F 18.74 26.60 30.69 Robust standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1 Meier & Weber (2016) Peers in Retirement 46 / 35

Robustness: Length of Shared Time at Same Firm (1) (2) (3) VARIABLES shared time 1 shared time 3 shared time 5 Retirement Age (Peer) 0.160*** 0.162*** 0.135*** (0.035) (0.038) (0.039) Observations 6,937 6,291 5,726 First Stage F 26.60 23.20 21.92 Robust standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1 Meier & Weber (2016) Peers in Retirement 47 / 35

Robustness: Placebo Test with Fake Reform Window Meier & Weber (2016) Peers in Retirement 48 / 35

Robustness: Clustered Standard Errors (1) (2) (3) VARIABLES cohort cluster peer cohort cluster firm cluster Retirement Age (Peer) 0.150*** 0.150*** 0.150*** (0.034) (0.029) (0.036) Observations 7,066 7,066 7,066 First Stage F 82.50 396.4 26.42 Robust standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1 Meier & Weber (2016) Peers in Retirement 49 / 35

Robustness: Age Difference Controls (1) (2) (3) (4) VARIABLES Ret. Age (Co-worker) Ret. Age (Co-worker) Take-up (Co-worker) Take-up (Co-worker) Retirement Age (Peer) 0.158*** 0.118*** (0.036) (0.037) Take-up (Peer) 0.138*** 0.117*** (0.030) (0.031) Observations 7,066 6,880 7,066 6,880 Age Difference yes yes yes yes Region FE - yes - yes Industry FE - yes - yes First Stage F 25.93 23.64 39.89 36.92 Robust standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1 Meier & Weber (2016) Peers in Retirement 50 / 35

Evolution of Bunching through Peer Network Meier & Weber (2016) Peers in Retirement 51 / 35

A Static Social Interactions Model Individual decides whether to take-up early retirement or not x i = Utility features social interactions { 1 if take-up 0 otherwise max U i = max x i (u(b) + S(P)) + (1 x i ) (u(w) θ i ) x i x i θ i : heterogeneous, unobserved cost of work S(P): social interactions function P: vector of take-up decisions of peers Individual retires early iff: u(b) + S(P) u(w) θ i Meier & Weber (2016) Peers in Retirement 52 / 35

Social Multiplier II (1) (2) (3) VARIABLES 1 st Co-Worker 2 nd Co-Worker 3 rd Co-Worker Peer bunching dummy 0.163*** 0.136* 0.108 (0.058) (0.080) (0.091) Observations 7,275 3,863 2,130 First Stage F 9.991 5.068 3.995 Robust standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1 The presence of a bunching peer increases the probability that the first co-worker bunches by 16 percentage points second co-worker bunches by 14 percentage points third co-worker bunches by 11 percentage points (not significant) Meier & Weber (2016) Peers in Retirement 53 / 35

Social Multiplier III Sequential Ordering: y 1g = α 1 + λp 1g y 2g = α 2 + γy 1g y 3g = α 3 + γy 2g + γδy 1g y 4g = α 4 + γy 3g + γδy 2g + γδ 2 y 1g Identification: Reduced-form coefficients π identify linear combination of γ and δ Two coefficients sufficient for identification (3 rd co-worker delivers overidentification restriction) π 1 = γ π 2 = γ 2 + γδ π 3 = γ 3 + 2γ 2 δ + γδ 2 Meier & Weber (2016) Peers in Retirement 54 / 35

Retirement Age of Non-exempt Women by Cohort Meier & Weber (2016) Peers in Retirement 55 / 35

Retirement Age of Exempt Women by Cohort Meier & Weber (2016) Peers in Retirement 56 / 35

Effects of Pension Reform for Women Meier & Weber (2016) Peers in Retirement 57 / 35

First-Stage Relationship Women I Meier & Weber (2016) Peers in Retirement 58 / 35

Reduced-form Women I Meier & Weber (2016) Peers in Retirement 59 / 35

Effects of Pension Reform for Women Meier & Weber (2016) Peers in Retirement 60 / 35

First-Stage Relationship Women II Meier & Weber (2016) Peers in Retirement 61 / 35

Reduced-form Women II Meier & Weber (2016) Peers in Retirement 62 / 35

Peer Effects Women II (1) (2) (3) (4) VARIABLES Take-up (Co-worker) Take-up (Co-worker) Take-up (Co-worker) Take-up (Co-worker) Take-up (Peer) 0.053** 0.048* 0.041 0.036 (0.027) (0.027) (0.027) (0.028) Observations 7,930 7,778 7,639 7,551 Region FE - yes - yes Industry FE - - yes yes First Stage F 34.84 34.48 33.54 32.93 Robust standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1 The presence of a peer that takes up early retirement right at 60, has a minor or no effect on the co-workers take-up of early retirement right at 60. Meier & Weber (2016) Peers in Retirement 63 / 35