What Causes Workers to Retire Before They Plan? Alicia H. Munnell, Geoffrey T. Sanzenbacher, and Matthew S. Rutledge Center for Retirement Research at Boston College 17th Annual Meeting of the Retirement Research Consortium Washington, DC August 6, 2015 Introduction • Workers are internalizing the message to work longer. o Between 1991 and 2014, one survey found the number of workers planning to work past age 65 tripled from 11 to 33 percent. • But, an analysis of the Health and Retirement Study (HRS) finds that 41 percent of workers retired earlier than planned. • What are most important factors driving earlier-than-planned retirements? 1 What we know • The literature points to several correlates of earlier-thanplanned retirement: o Deterioration in health (e.g., Őrestig, Strandh, and Stattin 2013, Dwyer and Hu 2000, and Disney and Tanner 1999); o Marital changes (Disney and Tanner 1999); o Changes in finances (Andersen 1985); and o Changes in employer (Munnell, Jivan, and Triest 2004). 2 What we don’t know • Most studies focus on a subset of the following “shocks”: o Health and its interaction with retiree health insurance (RHI); o Employer change and job loss; o Spousal retirement and other familial changes; and o Changes in finances. • This project will address this gap by including all these shocks in a model of earlier-than-planned retirement. o Allows analysis of relative importance of shocks. 3 The data • To estimate a model of earlier-than-planned retirement, data from the 1992-2012 waves of the original HRS cohort is used. • Sample consists of individuals who are working at the interview closest to their 58th birthday. • HRS poses questions related to planned retirement for all workers. 4 The dependent variable • Dependent variable equals one if worker retires earlier-thanplanned. • The planned retirement year is obtained in one of two ways: 1. A person’s answer to when they “plan” or “think” they are going to retire as of their age-58 interview. 2. If a person says “never” at their age-58 interview, the answer to those same questions at the next closest interview is used. • The individual’s actual retirement year is the first year the individual claims to be “fully” retired. 5 Most of the sample has a planned retirement age near their age-58 interview. Distribution of Reporting Ages of Expected Retirement Age and Percent Retiring Early Age reported 55 56 57 58 59 60 61 62+ Share of sample 3.0% 11.2 22.4 23.4 15.0 11.6 7.8 5.7 Avg. planned retirement age 63.7 63.3 63.3 63.3 63.8 64.2 64.6 68.1 Percent retiring early 52.3% 45.5 42.4 39.7 39.4 39.3 40.1 42.0 Source: Authors’ calculations from the Health and Retirement Study 1992-2012 waves. 6 Independent variables: health shocks • Paper uses an objective health index that sums 13 indicators for whether an individual has certain health conditions. o These include conditions like “activity limiting lung disease,” “heart condition,” “arthritis with medication.” o And limitations like “needs help walking.” • The max change in health between when plans are made and the planned retirement year (or retirement) is the “health shock.” o This change is also interacted with RHI to see if having outside health insurance increases effect of shocks. 7 Independent variables: other shocks • Other shocks also represent changes between the time the plan is made and the planned retirement year (or retirement), including: o A change in employer; o Job loss through layoff or business closing; o Spouse’s retirement; o Spouse’s deteriorating health; o A child leaving home; o A divorce or new marriage; and o Large changes in financial wealth. 8 Health shocks are common, job changes and familial changes are less common. Frequency of Shocks, Original HRS Cohort Shock Increase health index Switch employer Laid off/business close Share experiencing shock 38.2% 10.7 12.3 Share retiring early 44.8% 35.0 54.9 Spouse retires Spouse health worsens 19.5 7.8 46.1 44.5 Resident child leaves Marital change Wealth declines 50% Average regardless of shock: 16.8 7.7 32.4 41.3 44.6 43.5 41.5 Source: Authors’ calculations from the Health and Retirement Study 1992-2012 waves. 9 Independent variables: inclusion of demographic and initial characteristics • Individuals implicitly take into account their own characteristics when making plans – so shouldn’t impact early retirement. • At the same time, some characteristics may be correlated with planning ability and prevalence of shocks. o For example, the less educated may overestimate working life and underestimate likelihood of worsening health. o This could cause an overstatement of the impact of shocks. 10 Worsening health, job loss, spousal retirement associated with early retirement. Marginal Effect on Probability of Retiring before Plan Health Effect (%) Employment Effect (%) Increase in index of 1 2.9** Change employer, no job loss -2.5 Increase with RHI 2.1 Change employer, with job loss -7.9 Initial index 2.9*** Job loss with no new employer 27.2*** Familial Effect (%) Wealth Effect (%) Spouse retires 5.0* Decrease wealth 50% 1.2 Spouse worse health 4.8 Increase wealth 50% 0.1 Marital change 0.0 Resident child leaves 1.4 Note: Coefficients are significant at the 10-percent (*), 5-percent (**), or 1-percent (***) level. Regression includes full set of demographic controls and initial conditions for all shocks. Source: Authors’ calculations from the Health and Retirement Study 1992-2012 waves. 11 Interpreting the regression results • For effect size, losing a job through layoff or business closing is most important, followed by spousal retirement. • Worsening health and poor initial health also have statistically significant results. o Interestingly, retiree health insurance does not significantly increase the likelihood of early retirement. • But the importance of these variables depends on two things: 1) the size of the effect; and 2) the frequency with which shocks occur. 12 A counterfactual exercise • We can use the regression results to perform “counterfactuals” seeing how many people would retire early if: o No one’s health got worse; o Everyone was healthy (Index = 0) to start with; o No one lost their job; o No one’s spouse retired; and o Nothing “bad” happened (i.e., no one had anything happen to them that increases the probability of early retirement). 13 In counterfactuals, health is the most important factor, but much is unexplained. 45% 0.9% 40% 1.7% 1.9% 2.8% 4.5% 6.3% 35% 41.3% 40.4% 39.6% 39.4% 38.5% 36.8% 35.0% 30% Actual No spouse retires No health shocks No job loss No initial health problems No health shocks/all initially healthy No "bad" effects Source: Authors’ calculations from the Health and Retirement Study 1992-2012 waves. 14 Conclusion • Health is the most important driver of early retirement – both changes in health and poor initial health. o RHI does not have a significant effect on early retirement in response to a health shock. • Job loss and then spousal retirement are next biggest causes of early retirement. • Much remains unexplained – even eliminating all “bad” shocks in the model and assuming all are healthy, 35 percent retire early. 15 Next steps • Investigating why much remains unexplained: o Is there a way to proxy planning accuracy that we are missing? o Are shocks being measured in the right way? o Any major shocks we are missing? • Exploring the option of making the analysis a “survivor” type regression. o An advantageous approach is that it more accurately models the timing of shocks. 16