CONTRIBUTIONS Blanchett

The Impact of Age Uncertainty on Retirement Outcomes by David M. Blanchett, Ph.D., CFA, CFP®

David M. Blanchett, Ph.D., CFA, CFP®, is head of Executive Summary retirement research at Morningstar Investment Management. He is a recipient of the Journal’s 2007 • This research explores the retire later than expected, and those Financial Frontiers Award and its 2014 and 2015 implications of early retirement targeting retirement at an age after Montgomery-Warschauer Award. on required savings levels and 61 generally retire approximately a variables that can help predict half-year early for each additional A growing body of research notes when someone might retire earlier year targeted past age 61. that retirement age projections are than expected. • Incorporating retirement age inconsistent with actual retirement age • Research suggests people tend to uncertainty into a financial plan decisions—people tend to retire earlier retire earlier than expected. Retiring can have a significant impact on than expected. Retiring early can have early can have a significant (nega- required retirement savings levels. a significantly negative impact on a tive) impact on a retiree’s likelihood • Financial planners should consider retiree’s likelihood of meeting his or her of achieving retirement success. modeling early retirement to retirement income goal by reducing the • A nonlinear relation exists between prepare clients for the (likely) actual and expected retirement possibility that it may occur, time for saving and investing, lowering age, where individuals targeting especially for individuals targeting the potential Social Security benefit, retirement before age 61 tend to retirement past age 65. and potentially extending the retire- ment period (assuming the decision to retire early is independent from any before age 61 tend to retire later than success, assuming no retirement age kind of health shock). expected; individuals targeting a retire- variability. Additional required savings This paper explores the extent to ment age of 61 retire when expected; for equivalent outcomes can exceed 50 which certain variables1 can help and those targeting a retirement age percent, or even 100 percent, for those predict early (or late) retirement; it after 61 generally retire approximately scenarios with a later retirement age also explores the implications of early a half-year early for each additional year (e.g., age 70) or higher probabilities of retirement on required savings levels. of work planned past age 61 (e.g., an success. Few of the variables considered for investor targeting a retirement age of 69 Overall, these findings suggest retire- the analysis have statistically significant would actually retire around age 65). ment age uncertainty is an important coefficients, which suggests there is Incorporating retirement age consideration for financial planners. a significant amount of uncertainty uncertainty into the retirement plan- Financial planners should consider (i.e., randomness) associated with the ning process can significantly impact modeling early retirement to prepare decision to retire early. One variable, required retirement savings levels. clients for the (likely) possibility that it though—expected age of retirement— For example, a 55-year-old targeting a may occur, especially for those individu- drove the majority of the predictive retirement age of 65 and a 4 percent als targeting a retirement past age 65. effect of the model. initial withdrawal rate would need The forecast error associated with approximately 25 percent more savings Retirement Age Trends expected retirement age is nonlinear, (at age 55) to achieve the same income When making a financial plan, the where individuals targeting a retirement level with the same probability of expected retirement age is typically

36 Journal of Financial Planning | September 2018 FPAJournal.org Blanchett CONTRIBUTIONS

Figure 1: RetirementTk Expectations and Reality

Panel A: Actual Versus Expected Retirement Age Panel B: Probability of Retiring When Expected 68 60% 66 50% 64 40%

Age 62 30%

% of Total 20% 60 10% 58 2002 2007 2012 0% 1991 1996 2001 2006 2011 2016 Year Survey Year Average actual retirement age among retirees Earlier About When Later Average expected retirement age among non-retirees

Source: Gallup (news.gallup.com/poll/168707/average-retirement- Source: EBRI’s 2017 Retirement Con dence Survey (ebri.org/pdf/briefspdf/ age-rises.aspx) EBRI_IB_431_RCS.21Mar17.pdf) provided by the client. In theory, the ers retiring earlier than expected has 64.2 years in 2012.2 This recent increase client should have a relatively good idea averaged approximately 45 percent in retirement age has corresponded to of when he or she will retire, especially since 1991, versus 50 percent retiring increases in the eligibility for govern- as the client gets closer to retirement. about when expected, and 5 percent ment retirement along with In reality, there is a growing body of retiring later than planned. Greenwald, financial incentives to delay retirement. research demonstrating that investors Copeland, and VanDerhei (2017) noted do not tend to retire when expected, that those who retired early tended to Literature Review primarily retiring earlier. This effect cite health problems or disability (41 Although most workers expect to is noted in Figure 1, where Panel A percent), downsizing or closure (26 retire at a specific age (Settersten and includes data on actual versus expected percent), or having to care for a spouse Hägestad 1996), the academic perspec- retirement age from Gallup surveys, and or another family member (14 percent) tive of retirement has evolved over time. Panel B includes historical data on the as a reason to retire early, while some Researchers now more commonly view probability of retiring when expected reported being able to afford an earlier retirement as a transition rather than a from EBRI’s 2017 Retirement Confi- retirement (24 percent) or wanting to discrete event. A comprehensive model dence Survey (Greenwald, Copeland, do something else (10 percent). of retirement behavior created by Beehr and VanDerhei, 2017). Increasing retirement ages is a rela- (1986) incorporated a variety of factors, Both panels in Figure 1 suggest tively recent phenomenon over the last such as environmental (including work individuals tend to retire earlier than century. Workforce participation of men and non-work factors) and personal expected. For example, in Panel A, the aged 65 and over has decreased from (including health and economic well- average difference between actual and 75 percent in 1880 to approximately being factors). expected retirement has been approxi- 20 percent by 1980, but it’s begun to Fisher, Chaffee, and Sonnega (2016) mately four years (i.e., people are retir- slowly creep back up. Additionally, the provided a detailed review of the ing four years earlier than expected), average retirement age for men in 1962 existing literature on a variety of factors and while the average actual age among was approximately 65, dropping to 62 in related to retirement timing, including retirees has been increasing (from 59 the 1990s, and increasing back to 64 by physical factors (e.g., cognitive health, to 62), so too has the expected age 2013 (Munnell 2015). mental health, and economic status); (from 63 to 66). This is similar to the This recent trend toward later demographic factors (e.g., age, gender, results noted in the EBRI Retirement retirement age is a global phenomenon. education, and race); psychological Confidence Survey in Panel B, where For example, among the 34 countries in factors (preferences and expectations the median expected retirement age the Organisation for Economic Co- regarding retirement and retirement among workers was 65 versus an actual operation and Development (OECD), timing, attitudes toward retirement, age among retirees of 62 (a three-year the average retirement age for men in role identity, and personality charac- difference). 1949 was 64.3 years, decreasing to 62.4 teristics); subjective ; In Panel B, the percentage of work- years in 1999, and then increasing to family factors; work factors; and

FPAJournal.org September 2018 | Journal of Financial Planning 37 CONTRIBUTIONS Blanchett macroeconomic factors. market fluctuations seem to be more though, each of these three key vari- An increasing number of studies important for the retirement plans of ables has some degree of uncertainty. have examined relationships between higher socioeconomic status groups, For example, it’s possible the retirement retirement expectations and subsequent whereas changing rates need could change considerably over behavior at the individual level and tend to have a greater effect among time (e.g., if the client enters a nursing concluded that expectations provide lower socioeconomic status groups home), and it’s impossible to know useful information for projecting (Coile and Levine 2011). Rutledge, exactly how long retirement will last. future retirement outcomes (Anderson, Gillis, and Webb (2015) found that the Just because an input is uncertain, Burkhauser, and Quinn 1986; Bernheim decrease in defined-benefit cov- though, does not necessarily mean it 1987; Dwyer 2001), and the noted roles erage from previous and the decline should be treated as a random vari- of these factors have varied. in retiree health coverage between the able in a financial plan. An important One of the most explored drivers Health and Retirement Study (HRS) consideration when determining which of early retirement is health. A loss of and the Survey of Income and Program variables should be treated as random control of retirement age (i.e., because Participation cohorts each pushed the has to do with the effective control the of a health shock) can negatively affect retirement age up by approximately one individual has over the decision. For physical health, mental health, well- year, all else equal. example, with respect to returns, while being, life satisfaction, and adjustment a client can control the general risk of to retirement (Quine, Wells, De Vaus, The Uncertainty of Retirement Forecasts the portfolio (i.e., the equity alloca- and Kendig 2007). Dwyer and Mitchell Three key variables are required to tion), the client has no control over the (1999) found that health problems influ- estimate the cost of retirement, or underlying components (e.g., the return ence retirement plans more strongly really any financial goal: (1) the need on stocks and bonds). In contrast, the than economic variables, where men in (or annual cash flow,CF ); (2) how long client typically has significant control poor overall health are expected to retire the goal lasts (in years, N); and (3) over the cash flows withdrawn from the one to two years earlier, on average. the expected returns on the portfolio portfolio each year (total control, we

McGarry (2004) noted that subjective each period (ri). For example, Equation could argue). Although withdrawing reports of health are powerful predictors 1 can be used to estimate the cost of less than some target amount may be of an individual’s expected probability retirement (assuming income begins unpleasant (i.e., result in a significant of working at a specified future age, immediately). This approach is similar change in lifestyle), it can generally be and that the effect of poor health is to a concept introduced by Blanchett, accommodated. substantially larger than the effects of Kowara, and Chen (2012), which While many studies have modeled financial variables. Other health-related was used to estimate the “sustainable investment returns—potentially along issues like mortality risk have also been spending rate.” with withdrawal rates and longevity—as noted to affect retirement, because random variables, few have treated those workers who anticipate having retirement age as a random variable, fewer years to live out their retirement and this is the key innovation this study may decide to retire early (Hurd, Smith, [1] offers. and Zissimopoulos 2002). Other things, such as family caregiv- Data ing responsibilities, which are often Data for the analysis was obtained from unanticipated, may also affect actual Portfolio returns are generally the the first 12 survey years, or waves, of the retirement timing, especially for women only variable treated as random in a Health and Retirement Study (HRS), (Zimmerman, Mitchell, Wister, and financial plan (i.e., in a Monte Carlo from 1992 to 2014. The HRS is a panel Gutman 2000). Montalto, Yuh, and simulation). Incorporating uncertainty household survey specifically focused Hanna (2000) noted that planned around future returns is an improve- on the study of retirement and health retirement age increases as people ment over using a single return assump- among individuals over the age of 50 in get older and (to a lesser degree) have tion (e.g., that the portfolio increases the . The HRS is sponsored higher noninvestment income. by 5 percent per year), because the by the National Institute on Aging and Postponement of retirement may randomness enables a financial planner the Social Security Administration occur in periods of economic downturns to provide additional uncertainties and is conducted by the University (Goda, Shoven, and Slavov 2011). Stock associated with investing. In reality, of Michigan. The survey was first

38 Journal of Financial Planning | September 2018 FPAJournal.org Blanchett CONTRIBUTIONS

Figure 1:2: DistributionTk of Planned Retirement Ages by HRS Survey Year (Wave)

85 y = 0.1434x – 222.81 80 R = 0.7053

75

70

65

60 Planned Retirement Age Planned Retirement

55

50 1992 1994 1996 1998 2000 2002 2004 2006 2008 2010 2012 2014 HRS Survey Year

5th 25th 50th 75th 95th Linear (50th)

Source: Health and Retirement Study (these are raw values and do not include weights) conducted in 1992, and respondents are expected retirement age from the year assuming it met the previously noted resurveyed every other year with addi- 2002 to 2014). criteria, was used for the analysis. This tional households added to the survey Various filters were applied to the results in most retirement age estimates to keep it representative of the target available data. First, the analysis was coming from the first wave, 1992. While population. HRS fields are combined limited to respondents coded as working an individual’s retirement estimate is with data from the RAND HRS, which is with a minimum annual of likely to change as he or she approaches a user-friendly version of a subset of the $10,000 (adjusted to 1992 dollars3) for retirement, the earliest available esti- HRS developed at the RAND Center for the respective wave. Second, the respon- mate was used to reflect initial expecta- the Study of Aging with funding from dent must have had at least five years tions for the respective household. This the National Institute on Aging and the before the expected retirement age and also ensured the estimates were ex-ante. Social Security Administration. four years remaining after the expected The HRS has detailed information retirement age to be included (these are Variables about household retirement expecta- each from the respective survey year). The variables considered for the analysis tions and decisions. For example, Third, the respondent had to be coded were: Figure 2 includes information about the as actually becoming fully retired during Actual retirement age: the age when distribution of planned retirement age the period of analysis (i.e., sometime the respondent fully retired. by HRS wave (i.e., survey year) for all between the first and last waves). The Planned retirement age: the age responses available in each wave. second and third filters ensured the when the respondent expected to retire. Planned retirement ages have analysis captured respondents retiring Years to planned retirement age: increased by about three years across before and after their expected retire- the difference between the expected waves (approximately 0.14 years per ment age. These filters mean the last retirement age and the current age. year, across the 22 years). This change is wave that considered expected retire- Years of education: the number of directionally similar but roughly half the ment ages was 2004 (although later years of education completed. annual increase noted in the Gallup poll waves were used to check for the actual Marital status: a dummy variable (Figure 1, Panel A), where the slope, or retirement age). with a value of 1 if the person is mar- year-over-year change, has been 0.29 The first available expected retire- ried, 0 if other. (consistent with a three-year increase in ment age provided by the respondent, Gender: a dummy variable with a

FPAJournal.org September 2018 | Journal of Financial Planning 39 CONTRIBUTIONS Blanchett

Table 1: Descriptive Statistics for Variables

Percentile Variable 5th 25th 50th 75th 95th Average Actual retirement age 55.00 60.00 62.00 65.00 69.00 61.92 Planned retirement age 56.00 60.00 63.00 66.00 70.00 63.20 Years to planned retirement 5.00 7.00 9.00 12.00 15.00 9.50 Years of education 8.00 12.00 12.00 15.00 17.00 13.00 Married? (1 = true) 0.00 1.00 1.00 1.00 1.00 0.77 Male? (1 = true) 0.00 0.00 0.00 1.00 1.00 0.48 Tenure 0.90 5.60 12.10 22.10 32.70 14.19 ln (household income) 9.75 10.38 10.83 11.21 11.82 10.81 ln (net housing wealth) 4.61 9.91 10.82 11.41 12.16 9.93 ln (non-housing wealth) 4.61 9.31 10.59 11.64 13.12 10.26 ln (Social Security wealth) 4.61 11.42 11.96 12.25 12.39 11.14 Pension? (1 = true) 0.00 1.00 1.00 1.00 1.00 0.76 Probability of working to age 65 0.00 0.00 0.00 40.00 100.00 21.79 Probability of living to age 75 10.00 50.00 70.00 90.00 100.00 67.01 stress level 1.00 2.00 2.00 3.00 3.00 2.11 Physical job level 1.00 2.00 3.00 4.00 4.00 2.82 Health limits work? (1 = true) 0.00 0.00 0.00 0.00 1.00 0.06 Health status 2.00 3.00 4.00 5.00 5.00 3.78 Source: Health and Retirement Study value of 1 if the person is male, 0 if other. Job stress level: whether the respon- most of these variables is somewhat Years of tenure at current job: the self- dent agreed with the statement that his obvious, and many have been consid- reported tenure of current . or her job involves a lot of stress. This is ered in past research exploring retire- Total household income: this technically a categorical variable, with ment ages or decisions. To be included variable is log-transformed to reduce the four possible responses, but was treated in the analysis, data must have been skewness. as continuous as a simplifying assump- available for all variables. There was Net housing wealth/total household tion. Higher values are associated with relatively little correlation among the income: the log-transformed value lower levels of stress. variables; therefore, multicollinearity of the net value of the home minus Job physical effort: the extent to was not a concern for the regressions liabilities (i.e., mortgage) divided by the which the respondent said his or her job (e.g., the variance inflation factors, or log-transformed income. A minimum requires a lot of physical effort. This is VIFs, for all variables were less than 2). value of $100 was assumed. technically a categorical variable, with Table 1 shows descriptive statistics for Non-housing wealth/total house- four possible responses, but was treated the variables. hold income: the log-transformed as continuous. Higher values are associ- value of the non-housing wealth (which ated with lower levels of physical effort. Analysis includes all financial assets) divided by Health limits work: whether an A series of ordinary least squared (OLS) the log-transformed income. A mini- impairment or health problem limited regressions were conducted, and the mum value of $100 was assumed. the kind or amount of paid work for the results are included in Table 2. The Total household Social Security respondent. This is a dummy variable dependent variable for Model 1 was the wealth/total household income: the with a value of 1 if health problems do actual retirement age; the dependent log-transformed total value of Social limit work, 0 if not. variable for Model 2 was the planned Security benefits for the household Health status: the respondent’s retirement age; the dependent variable divided by the log-transformed income. self-reported general health status. This for Model 3 was the difference in the A minimum value of $100 was assumed. is technically a categorical variable, with expected and actual retirement ages Pension: a dummy variable with a five possible responses, but was treated (i.e., does the person tend to retire value of 1 if the person is covered by a as continuous as a simplifying assump- before or after the expected age); and defined-benefit pension, 0 if not. tion. Higher values are associated with the dependent variable for Model 4 was Probability of living to age 75: the lower health status levels. the absolute difference in the actual and self-assessed probability of living to age 75. The economic intuition for including expected retirement age (i.e., the general

40 Journal of Financial Planning | September 2018 FPAJournal.org Blanchett CONTRIBUTIONS

Table 2: OLS Regressions

Actual Planned Actual Planned Actual Planned Dependent Variable Retirement Age Retirement Age Retirement Age Retirement Age Model Number 1 2 3 4 Coe t-stat Coe t-stat Coe t-stat Coe t-stat Intercept 11.204 34.845 2.124 –0.069 Planned retirement age 25.712 Years to planned retirement –3.911 25.074 –12.902 15.578 Education years 0.193 –0.253 0.268 –0.517 Married? 0.762 –3.630 1.958 –0.531 Male? 0.492 7.822 –2.197 –0.843 Job tenure –2.133 2.307 –2.797 0.197 ln (household income) –2.121 –4.016 –0.634 –0.020 ln (net hhld housing wealth) / ln (hhld income) –3.030 1.904 –3.502 1.015 ln (non-housing hhld wealth) / ln (hhld income) –0.537 3.397 –1.665 –0.222 ln (Social Security wealth) / ln (hhld income) –0.139 –4.031 1.244 0.199 Covered by pension? 0.809 –2.239 1.525 –0.643 Probability working to age 65 5.115 9.281 1.773 0.633 Probability living to age 75 0.975 0.554 0.727 0.430 Job stress –0.041 3.137 –1.109 0.515 Physical job 1.368 –0.682 1.519 1.786 Health limits work? –0.786 0.288 –0.837 1.712 Health status 1.975 –1.577 2.396 –1.481 R 40.52% 36.29% 11.53% 14.43% Adjusted R 39.88% 35.64% 10.63% 13.56% Observations 1,596 1,596 1,596 1,596 Notes: **signi cant at 1% level; *signi cant at 5% level accuracy of the prediction). The inde- were not covered by a pension, had a forecast retirement age, the difference pendent variables changed slightly based higher self-assessed probability of work- (i.e., forecast error) was largest for those on the model. The regressions include ing to age 65, and more job stress. The who were further from retirement. weights from the respective HRS wave. years to planned retirement age variable Certain variables (e.g., years of educa- Table 2 offers important takeaways, was obviously somewhat endogenous to tion, income, or financial assets) were explored next by model number. planned retirement age, therefore the not predictive, suggesting the retire- Model 1. With respect to actual high t-statistic is not surprising. ment age is effectively random (beyond retirement age, respondents who tended Model 3. With respect to the planned retirement age). to retire at older ages planned to retire difference in the actual and planned Figure 3 includes information about later, were closer to retirement, had retirement age, among respondents with the distribution of differences in actual shorter job tenures, lower household positive differences (i.e., tended to retire and planned retirement ages by age incomes, lower levels of housing wealth, later than expected), the difference (Panel A) and the values that result had a higher self-assessed probability was largest for those who were further using the regression models introduced of working to age 65, and tended to from retirement, had more years until in Appendix 1 (Panel B). be in worse health. The significance retirement, longer job tenures, higher Figure 3 demonstrates that the error of the retirement age coefficient (the incomes, more housing wealth, and in actual and planned retirement ages t-statistic is 25.712) suggests individuals had a lower self-assessed probability of is nonlinear. Individuals targeting a have some general idea of when they are working to age 65. This is perhaps the retirement before age 61 tend to retire going to retire. most important model among the four, later (e.g., someone targeting to retire Model 2. With respect to planned because it drives the primary objective at age 57 will actually retire around age retirement age, respondents who of the analysis. 58). Those targeting a retirement age of planned to retire later tended to be fur- Model 4. With respect to the absolute 61 tend to retire approximately at that ther from retirement, were single, male, difference in the actual and planned age, and those targeting retirement after had lower incomes, less financial assets retirement age, which provides some age 61 generally retire approximately a saved, lower Social Security benefits, perspective on the ability to correctly half a year early for each additional year

FPAJournal.org September 2018 | Journal of Financial Planning 41 CONTRIBUTIONS Blanchett

Figure 1:3: TkDierence in Actual Retirement Age and Planned Retirement Age

Panel A: Actual Percentiles Panel B: Regression Models

10 3

2 5 1

0 0 –1 –5 –2

–3 –10

and Planned Retirement Age and Planned Retirement –4 and Planned Retirement Age and Planned Retirement

Di erence in Actual Retirement Age Age in Actual Retirement Di erence –15 Di erence in Actual Retirement Age Age in Actual Retirement Di erence –5 55 60 65 70 55 60 65 70 Planned Retirement Age Planned Retirement Age

5th 25th 50th 75th 95th Nonlinear Linear

planned to work past age 61 (e.g., a client benefit at full retirement age (FRA), or cap equity portfolio and a U.S. aggregate targeting a retirement age of 69 would age 67, was assumed to be $30,000. All bond portfolio. The assumed real annual actually retire around age 65). These values are in real terms (i.e., adjusted for arithmetic return of the portfolio was 2 differences are an important distinc- inflation) and taxes were ignored for the percent with a standard deviation of 8 tion to consider, instead of assuming analysis. Note, the savings and Social percent. These return assumptions were all individuals retire early, because the Security values were estimated assum- similar to JP Morgan Asset Management’s implications of early retirement predomi- ing annual of $100,000. 2017 capital market assumptions5 as well nately affect individuals with older target Social Security retirement benefits as Morningstar Investment Management’s retirement ages (e.g., past age 65, since were assumed to be claimed upon retire- 2017 20-year capital market assumptions.6 they are retirees who are actually retiring ment (although no earlier than age 62). The returns implicitly assumed an invest- before the expected retirement age). The benefit amount at full retirement age ment management fee of 0.5 percent (i.e., (assumed to be $30,000) was adjusted the assumed return would be 2.5 percent The Impact of Retirement Age Uncertainty based on early or late retirement based if a fee was not included). Retirement was on Required Retirement Savings on the adjustments noted by the Social assumed to last until age 95 regardless of Early retirement, especially the Security Administration.4 For example, when it commenced. uncertainty surrounding it, can if claimed at age 62, the annual benefit Three approaches were considered. potentially have significant implications would be $21,000 (70 percent of the FRA For the first approach, the retirement on required retirement savings. This benefit); and if claimed at age 70, the age was assumed to happen with cer- section quantifies the potential cost annual benefit would be $37,209 (124 tainty. This is the traditional approach associated with this uncertainty. percent of FRA benefits). The average taken by planners when generating a For this analysis, a 5,000-run Monte indexed monthly earnings (AIME) was financial plan. Carlo simulation was conducted. not assumed to change based on early For the second approach, the The individual’s age for the analysis or late retirement (i.e., Social Security nonlinear regression model introduced was assumed to be 55. Retirement benefits were not assumed to increase in the Appendix was used to determine ages between 60 and 70, in one-year even if the person worked longer). the actual retirement age based on the increments, were considered. Total The portfolio was assumed to be projected retirement age (e.g., for a retirement savings were assumed to invested in 40 percent equities and 60 target retirement age of 69, the actual be $500,000 (at age 55), with annual percent fixed income, with respective retirement age was assumed to be 65). savings of $10,000. The Social Security components generally being a U.S. large- For this model, there was no assumed

42 Journal of Financial Planning | September 2018 FPAJournal.org Blanchett CONTRIBUTIONS

Figure 1:4: TkImpact of Retirement Age Uncertainty on Retirement Success

100% 90% 80% 70% 60% 50% 40% 30%

Probability of Success Probability 20% 10% 0% 60 61 62 63 64 65 66 67 68 69 70 Planned Retirement Age Retire When Planned (No Randomness) Nonlinear Empirical Model, No Uncertainty Nonlinear Empirical Model, With Uncertainty randomness associated with the retire- was not assumed to change based on for individuals targeting a younger ment age decision (you were assumed to retirement age. The amount represents retirement age (i.e., those targeting a potentially retire at an age that differed a median withdrawal of 3.7 percent from retirement age at 60) because these from your target, but that retirement the portfolio in the first year of retire- retirees tended to retire later than age was constant across runs). ment at age 65. expected, but resulted in lower success The third approach built on the rates for those targeting an older retire- second approach but incorporated ment age. potential randomness around the actual Early retirement can Third, incorporating uncertainty retirement age. The randomness was potentially have significant (beyond simply using the nonlinear assumed to follow a normal distribution model) did not result in significantly with a mean of zero and a standard implications on required different outcomes than the base non- deviation of 3 (consistent with the linear model (i.e., being aware of when error associated with the nonlinear retirement savings. someone was likely to retire given their model in the Appendix). The maximum target age appeared to capture the clear positive change in retirement age was Figure 4 contains a variety of informa- majority of the impact of the model). two standard deviations (i.e., six years), tion. First, delaying retirement can have Next, an analysis was conducted although Social Security benefits were a significant, positive impact on the to determine the additional savings always assumed to commence no later probability of an investor achieving a required to achieve the same probability than age 70 and never before age 62. retirement goal. Within the context of of success for a given withdrawal rate The maximum negative change in this analysis, delaying retirement gives when incorporating retirement age retirement age was also two standard the investor more time to save, more uncertainty versus more traditional deviations (six years) or the current time for the portfolio to grow, a larger retirement modeling (i.e., assuming age of the investor (e.g., a 55-year-old Social Security retirement benefit, and the individual always retired when targeting a retirement age of 60 could a shorter retirement period (retirement expected). This was determined through retire no earlier than age 55). is assumed to end at age 95 regardless of stochastic simulation (i.e., different Figure 4 includes the results from retirement age). current savings levels were tested simulations where the base retirement Second, incorporating average retiree until the one that resulted in the same need was assumed to be $50,000 (in behaviors (i.e., the nonlinear error probability of success was determined). constant real dollars). The amount model) resulted in higher success rates For example, assuming a retiree is

FPAJournal.org September 2018 | Journal of Financial Planning 43 CONTRIBUTIONS Blanchett

Figure 1:5: TkAdditional Current Savings Required to Achieve the Same Probability of Success When Treating Retirement Age as Uncertain

120% 117% 103% 100% 93% 89% 80% 81% 80% 69% 66% 67% 55% 60% 52% 52% 41% 40% 40% 37% 31% 28% 27% 21% 17% 20% 13% 15% 8% 8% 4% 2% 5% Additional Savings Required Savings Additional 0% –2% –2% –6% –10% –8% –20% –15% 60 61 62 63 64 65 66 67 68 69 70

Target Retirement Age High Success Target Moderate Success Target Low Success Target

targeting a retirement age of 65 and a 4 for individuals targeting later retire- early retirement to potentially get them percent median initial withdrawal rate ment ages (e.g., age 70) and with a to save more than they would using a (across simulations) the total retirement higher target probability of success (i.e., more traditional approach where retire- need (at age 65) would be $52,076.7 The a retiree who would take a 3 percent ment age is treated as certain. probability of success of achieving this initial withdrawal). If the “average income level (to age 95) was 56 percent. investor” is assumed to be seeking a 4 Conclusion To achieve the same probability of suc- percent initial withdrawal rate (i.e., a A financial plan is only as good as its cess (56 percent) for the same income moderate success target) and wants to assumptions. While financial planners level ($52,076), the current balance at retire at age 65, the impact of retire- are increasingly treating returns as age 55 would need to increase by 28 ment age uncertainty would require a random variable in financial plans percent to $641,450 (from the assumed approximately 25 percent more savings (through Monte Carlo simulations), $500,000). than suggested by traditional models there is obviously a good deal of This analysis was conducted for retire- (28.29 percent to be exact). uncertainty associated with other key ment ages from 60 to 70, in one-year assumptions. This paper explored the increments, and for initial withdrawal Incorporating average implications of retirement age uncer- rates (at the target retirement age) of tainty and found that many individuals 3 percent, 4 percent, and 5 percent. To retiree behaviors resulted are likely to retire at earlier ages than make the results more intuitive, simply in higher success rates for expected, and that this can potentially note the 3 percent initial withdrawal have a significant impact on required target as a high probability of success, those targeting a younger savings (or retirement success). the 4 percent initial withdrawal target Therefore, it is essential that financial as a moderate probability of success, and retirement age. planners understand the implications the 5 percent initial withdrawal target as These results suggest ignoring retire- associated with the expected retirement a low probability of success. The results ment age uncertainty can potentially age assumption. are shown in Figure 5. have a significant (negative) impact on Incorporating retirement age potential retirement outcomes. There- Endnotes uncertainty can significantly increase fore, financial planners should consider 1. The variables tested were: target retirement the required savings levels, especially showing clients the implications of an age, years until planned retirement, years

44 Journal of Financial Planning | September 2018 FPAJournal.org Blanchett CONTRIBUTIONS

of education, marital status, gender, job Appendix: Actual Minus Planned Retirement tenure, household income, net housing wealth, Age OLS Regressions non-housing wealth, Social Security wealth, Appendix: Actual Minus Planned Retirement Age OLS Regressions whether the individual is covered by a pension, Dependent Variable Actual Planned Retirement Age self-assessed probability of working to age 65, Model Number 1 2 self-assessed probability of living to age 75, job Coe t stat Coe t stat stress, physical nature of job, whether health Intercept 22.305** 17.456 –100.607 –9.126 limits the individual’s ability to work, and Planned retirement age –0.374** –18.426 3.592* 10.145 self-assessed heath status. Planned retirement age2 –0.032** –11.220 2. See “OECD Pensions at a Glance” at dx.doi. R 17.56% 23.60% Adjusted R 17.51% 23.50% org/10.1787/pension_glance-2013-en. Observations 1,961 1,961 3. 1992 was used as the base year because it was the Notes: ** signi cant at 1% level; * signi cant at 5% level most common retirement age expectation wave. 4. See ssa.gov/OACT/ProgData/ar_drc.html. 5. See am.jpmorgan.com/lu/institutional/our- The Accuracy of Expected Retirement Dates Montalto, Catherine Phillips, Yoonkyung Yuh, thinking/ltcma-2017. and the Role of Health Shocks.” Center for and Sherman Hanna. 2000. “Determinants of 6. Not publicly available; contact the author for Retirement Research Working Paper No. 2001-8. Planned Retirement Age.” Financial Services more information. Available at crr.bc.edu/working-papers/planning- Review 9 (1): 1–15. 7. Note that $26,000 of this would be covered for-retirement-the-accuracy-of-expected- Munnell Alicia H. 2015. “The Average Retirement from Social Security retirement benefits, so the retirement-dates-and-the-role-of-health-shocks. Age—An Update.” Center for Retirement remainder ($26,076) would need to be funded Dwyer, Debra Sabatini, and Olivia S. Mitchell. Research Working Paper No. 15-4. Available at by the portfolio, which had a median value of 1999. “Health Problems as Determinants crr.bc.edu/briefs/the-average-retirement-age-an- $652,147 at retirement, before withdrawals of Retirement: Are Self-Rated Measures update. began. This resulted in a 4.0 percent initial Endogenous?” Journal of Health Economics 18 (2): Quine, Susan, Yvonne Wells, David De Vaus, and withdrawal rate ($26,076/$652,147 = 4.0 173–193. Hal Kendig. 2007. “When Choice in Retire- percent). Fisher, Gwenith G., Dorey S. Chaffee, and ment Decisions Is Missing: Qualitative and Amanda Sonnega. 2016. “Retirement Timing: Quantitative Findings of Impact on Well-being.” References A Review and Recommendations for Future Australasian Journal of Ageing 26 (4): 173–179. Anderson, Kathryn H., Richard V. Burkhauser, Research.” Work, Aging and Retirement 2 (2): Rutledge, Matthew, Christopher Gillis, and Anthony and Joseph F. Quinn. 1986. “Do Retirement 230–261. Webb. 2015. “Will the Average Retirement Age Dreams Come True? The Effect of Unanticipated Goda, Gopi Shah, John B. Shoven, and Sita Nataraj Continue to Increase?” Center for Retirement Events on Retirement Plans.” ILR Review 39 (4): Slavov. 2011. “What Explains Changes in Retire- Research Working Paper No. 2015-16. Available 518–526. ment Plans During the Great Recession?” The at crr.bc.edu/wp-content/uploads/2015/07/ Beehr, Terry A. 1986. “The Process of Retirement: American Economic Review 101 (3): 29–34. wp_2015-16.pdf. A Review and Recommendations for Future Greenwald, Lisa, Craig Copeland, and Jack Settersten Jr., Richard A., and Gunhild O. Hägestad. Investigation.” Personnel Psychology 39 (1): 31–55. VanDerhei. 2017. “The 2017 Retirement 1996. “What’s the Latest? Cultural Age Deadlines Bernheim, B. Douglas. 1987. “The Timing of Confidence Survey—Many Workers Lack for Family Transitions.” The Gerontologist 36 (2): Retirement: A Comparison of Expectations and Retirement Confidence and Feel Stressed About 178–188. Realizations.” NBER Working Paper No. 2291. Retirement Preparations.” EBRI Issue Brief Zimmerman, Lillian, Barbara Mitchell, Andrew Available at nber.org/papers/w2291. No. 431. Available at ebri.org/pdf/briefspdf/ Wister, and Gloria Gutman. 2000. “Unantici- Blanchett, David, Maciej Kowara, and Peng Chen. EBRI_IB_431_RCS.21Mar17.pdf. pated Consequences: A Comparison of Expected 2012. “Optimal Withdrawal Strategy for Retire- Hurd, Michael D., James P. Smith, and Julie M. and Actual Retirement Timing Among Older ment Income Portfolios.” Retirement Management Zissimopoulos. 2002. “The Effects of Subjective Women.” Journal of Women & Aging 12 (1-2): Journal 2 (3): 7–18. Survival on Retirement and Social Security 109–128. Coile, Courtney C., and Phillip B. Levine. 2011. Claiming.” NBER Working Paper No. w9140. “The Market Crash and Mass Layoffs: How the Available at nber.org/papers/w9140. Citation Current Economic Crisis May Affect Retire- McGarry, Kathleen. 2004. “Health and Retire- Blanchett, David M. 2018. “The Impact of ment.” The B.E. Journal of Economic Analysis & ment: Do Changes in Health Affect Retirement Retirement Age Uncertainty on Retirement Policy 11 (1), article 22. Expectations?” The Journal of Human Resources Outcomes.” Journal of Financial Planning 31 (9): Dwyer, Debra S. 2001. “Planning for Retirement: 39 (3): 624–648. 36–45.

FPAJournal.org September 2018 | Journal of Financial Planning 45