Genetic Ability, Wealth and Financial Decision-Making

Transcription

1 Genetic Ability, Wealth and Financial Decision-Making Danny Barth Affiliated Faculty, CESR Kevin Thom New York University Nicholas W. Papageorge Johns Hopkins University Econometric Society Summer Meeting 2017 June 15, / 53

2 Introduction Motivation Wealth inequality is large and rising. Saez and Zucman (2016): From: Wealth Inequality in the United States since 1913: Evidence from Capitalized Income Tax Data * Top 0.1% Wealth Share in the United States, Date of download: 2/11/2017 The Author(s) Published by Oxford University Press, on behalf of President and Fellows of Harvard College. All rights reserved. For Permissions, please journals.permissions@oup.com 2 / 53

3 Introduction Motivation Explaining Wealth Inequality. Labor income inequality alone cannot explain dispersion in wealth. Some other explanations: Inheritances. Risk preferences. Entrepreneurship. 3 / 53

4 Introduction Motivation Heterogeneity in returns to wealth appears to be important. One way to match observed distributions: Benhabib, Bisin, and Zhu (2011) Benhabib, Bisin, Luo (2015). Recent empirical work suggests returns are persistent: Fagereng et al. (2016), Bach et al. (2015). Wealthier people enjoy persistently higher returns to wealth. Heterogeneity is captured as fixed effects - sources are not observed. Not clear: mechanisms through which these factors operate. 4 / 53

5 Introduction This Paper. Do genetic factors associated with human capital accumulation also predict wealth independently of labor income? We examine the role of genetic markers linked to education in the Health and Retirement Study (HRS), using a polygenic score (linear index of genetic markers) constructed by Okbay et al. (2016) These genetic endowments are associated with higher wealth through labor income. Are they associated with higher wealth through other mechanisms? 5 / 53

6 Preview of Results. Introduction A higher polygenic score predicts higher wealth. One s.d. higher score is associated with over 9% higher wealth after controlling for education. Gradient not fully explained by income, bequests and business ownership. Stock market participation plays a big role, but why? Risk aversion does not explain key relationships. The polygenic score is related to financial literacy. Higher scores predict objectively correct beliefs. Lower scores predict extreme beliefs. Higher score predicts a longer planning horizon - may be linked to sophistication Defined benefit pensions moderate the gene-wealth gradient. 6 / 53

7 Introduction Contribution. Provide a genetic explanation for differences in returns posited in recent literature on wealth inequality. Also relate our findings to recent work on genes and savings. Cronqvist & Siegel (2016) demonstrate genetic basis for savings. They suggest self-control as a measure because of correlations with smoking and obesity. We show that the education score still predicts wealth after controlling for polygenic scores for smoking, BMI. 7 / 53

8 Introduction Literature. Cognition and Financial Wealth. Christelis, Japelli and Padula (2010) Malmendier and Nagel (2011) Beliefs, Probabilistic Thinking and Economic Decisions. Khaneman and Tversky (1972) Wiswall and Zafar (2014) Beliefs and Financial Decision-Making Dominitz and Manski (2007) Kezdi and Willis (2011, and others) Hector (2016) Genes and Economic Outcomes. Cronqvist and Siegel (2015) Cesarini et al (2010) Papageorge and Thom (2016) 8 / 53

9 Introduction Outline. 1 Brief Background on Genetics and Polygenic Scores. 2 Data. 3 Genes and Wealth. 4 Beliefs and Financial Literacy. 5 Policy: Pensions. 6 Alternate Mechanisms: Using scores for BMI and Smoking. 9 / 53

10 Introduction Genetics Background Human DNA is a sequence of approximately 3 billion nucleotide molecules spread across 23 chromosomes. Each human has two copies of each chromosome: one from each parent. 10 / 53

11 Introduction Genetics Background If we zoom in further, we see that each chromosome contains subsequences of genetic material that are referred to as genes. There are between 20,000-25,000 genes in the human genome. Genes provide instructions for synthesizing proteins that affect body function. 11 / 53

12 Introduction Genetics Background Each gene consists of a sequence of base pairs. Pairs can either be adenine-thymine (AT) pairs,or guanine-cytosine (GC) pairs. So at each address in the human genome, we can either see (AT) or (GC). 12 / 53

13 Introduction Genetics Background At the vast majority of locations in the human genome, there is no variation in the population. All individuals have the same nucleotide pair at such locations. 13 / 53

14 Introduction Genetics Background A single nucleotide polymorphism (SNP) is a form of genetic variation in which individuals differ in which base pair (e.g. AT or GC) resides at a particular genetic address. We will refer to specific positions in the genetic code by names, such as rs Alleles: The major allele at a position is the more common allele in pop. The minor allele at a position is the less common allele. 14 / 53

15 Introduction Genetics Background Suppose that there is variation at rs , and that the major allele is AT. Then individuals can differ in terms of how many copies of the minor or major allele (AT) they possess (0, 1, or 2 since there are two copies of each chromosome). An individual s genotype at a particular SNP is the number of copies of the reference allele that they possess: rs i {0, 1, 2} 15 / 53

16 Introduction Genetics Background Genome Wide Association Study (GWAS) Basic Procedure: Regress the outcome against individual SNPs, one at a time: y i = µ + β j x ij + ɛ i Collect the GWAS coefficients β j and the associated p-values. Identify a set of SNPs yielding sufficiently small p-values as genome-wide significant. Key to addressing multiple hypothesis testing: apply stringent p-value thresholds (typically ). 16 / 53

17 Introduction The Genetic Score Variable Polygenic Score Variable. Reitveld et al (2013) provide first GWAS evidence of SNPs that predict educational attainment. Okbay et al (2016), build on this work, discover additional associations. Using the GWAS coefficients from the Okbay et al. study, one can form a polygenic score as follows: EA Score i = j β j SNP ij where β j is a transformation of the underlying GWAS coefficients, which accounts for correlation among SNPs. 17 / 53

18 HRS Data, the Polygenic Score and Wealth Sample Construction and Description Genetic Data and the HRS. Longitudinal sample of U.S. over age 50. Surveys begin 1992; occur every two years. Individuals genotyped in three waves (2006, 2008, 2010). The first two are available. Individuals had to survive until at least 2006 to be included. 18 / 53

19 HRS Data, the Polygenic Score and Wealth Sample Construction and Description Analytic Sample. We restrict attention to: Genetic Europeans. Born before Retired in 1996, 1998, Aim: balance sample size and wealth measurement concerns. Resulting analytic sample 4,349 financial respondents. Information provided for an average of 4 waves. 19 / 53

20 HRS Data, the Polygenic Score and Wealth Household Wealth Wealth Data. We begin with the RAND wealth and income files. We need to construct pension data. An important source of wealth near retirement. For many, the only vehicle for stock market participation. Two key types: Defined Benefit: Use reported pension income. Defined Contribution: Reported at current job. Some ambiguity, e.g., dormant plans and stock market participation. We conduct a number of robustness checks using alternative samples. Wealth Data Construction Details 20 / 53

21 HRS Data, the Polygenic Score and Wealth Household Wealth Wealth Distribution p10 p25 p50 p75 p90 Mean St Dev Wealth (Winz) 31, , , ,583 1,293, , ,701 Wealth (No Housing) 13,853 49, , , , ,273 1,393,031 Wealth (No Ret. Wealth) 1,154 65, , ,905 1,119, ,999 1,088,254 Wealth (No H or R) 0 5,408 83, , , , ,615 Notes: Wealth mean and distribution calculated for the full sample of 15,517 household-year observations with non-missing wealth data. 21 / 53

22 Genes, Wealth and Financial Decisions log Real Total Wealth EA Score Notes: The EA Score and Log Wealth using Lowess Regression. 22 / 53

23 Genes, Wealth and Financial Decisions The Polygenic Score, Household Wealth and Earnings Now, we regress wealth on EA Score education. The aim: assess what explains the gradient. Controls in all regressions Principal components (PCs) of the full genetic data matrix Birth year dummy Age dummy Calendar year dummy Male dummy Interactions of birth year and male dummies Interactions of age and male dummies Father education Mother education Flag for problematic observations for 2002 Interaction of EA score and 2002 problem flag We cluster standard errors at the household level. 23 / 53

24 Genes, Wealth and Financial Decisions The Polygenic Score, Household Wealth and Earnings The Polygenic Score and Wealth Panel A: log Wealth (1) (2) (3) EA Score 0.234*** 0.097*** 0.092*** (0.023) (0.022) (0.022) Resp Education No Yes Yes Parental Education No No Yes Obs. 14,766 14,766 14,766 R Panel B: Wealth (level) (1) (2) (3) EA Score 120,453*** 55,830*** 52,509*** (13,256) (12,568) (12,623) Resp Education No Yes Yes Parental Education No No Yes Obs. 15,061 15,061 15,061 R Notes: Relating the genetic score to wealth. 24 / 53

25 Genes, Wealth and Financial Decisions The Polygenic Score, Household Wealth and Earnings Explanatory Power. Incremental R 2. After the usual set of controls (not education): 2.2% After conditioning on education: 0.3% For comparison: Education accounts for 10% of wealth variation. The EA Score incremental R 2 is 6.6% for education. 25% of variation in portfolio risk is genetic (Cesarini, 2010). For saving: 35% (Siegel and Cronqvist, 2015). 25 / 53

26 Genes, Wealth and Financial Decisions The Polygenic Score, Household Wealth and Earnings Factors Explaining the Gene-Wealth Gradient. Next, we regress wealth on EA Score and other factors. These include: Household income. Inheritances. 44% Self Employment 36% Stock market participation. 47% We continue to control for education. Sample: full set of explanatory variables. 26 / 53

27 Genes, Wealth and Financial Decisions The Polygenic Score, Household Wealth and Earnings Total Wealth, Income Flows, and Financial Decisions Dep. Var: log Tot. Wealth (1) (2) (3) (4) (5) EA Score 0.129*** 0.120*** 0.107*** 0.099*** 0.061** (0.029) (0.027) (0.027) (0.027) (0.024) Avg log HH Inc 0.398*** 0.384*** 0.388*** 0.310*** (0.037) (0.037) (0.037) (0.031) log Sum Inher *** 0.022*** 0.014* (0.008) (0.008) (0.007) Ever Rec Inher (0.095) (0.095) (0.081) Ever Self Emp *** 0.202*** (0.054) (0.048) Owns Stocks 1.043*** (0.043) Obs. 7,151 7,151 7,151 7,151 7,151 R Notes: The dependent variable in all specifications is the log of total wealth. 27 / 53

28 Genes, Wealth and Financial Decisions The Role of Stock Market Participation Polygenic Score and Stock Ownership Dep. Var: Owns Stocks (1) (2) (3) EA Score 0.039*** 0.029*** 0.029*** (0.009) (0.010) (0.010) Lag of log Wealth 0.152*** 0.149*** (0.008) (0.009) Avg log HH Inc (0.010) Obs. 8,035 5,047 5,047 R Notes: Relating the genetic score to stock ownership. 28 / 53

29 Genes, Wealth and Financial Decisions Risk Preferences Risk Aversion. Do endowments operate through risk preferences? We use survey items for choices between gambles. 1 Guaranteed income for life chance of double lifetime income or cut by X% Dummy for always choosing the guaranteed option. Indicates highest degree of risk aversion we can detect. About 32% of the sample takes a value of / 53

30 Genes, Wealth and Financial Decisions Risk Preferences Risk Aversion, Wealth, and Stock Ownership (1) (2) (3) (4) (5) Dep Var: Risk Averse Tot. Wealth Tot. Wealth Owns Stocks Owns Stocks EA Score * 0.110*** 0.108*** 0.030*** 0.030*** (0.006) (0.031) (0.031) (0.010) (0.010) Risk Averse ** *** (0.064) (0.020) Obs. 5,346 6,752 6,752 7,139 7,139 R Notes: The EA Score, risk aversion, wealth and stock ownership. 30 / 53

31 Genes, Wealth and Financial Decisions Risk Preferences Summary. There is a substantial gene-wealth gradient. Not explained fully by education, income, etc. Stock market participation is important. However, this is not operating through risk aversion. Additional results: robust to consideration of non-financial respondent. 31 / 53

32 Genes, Financial Literacy and Expectations Does the EA Score explain differences in how people think about financial decisions? We investigate the relationship between: 1 The EA Score and financial literacy. 2 The EA Score and subjective expectations. 3 Subjective expectations, investment behavior and wealth 4 Financial planning horizon 5 Relationship to self control (via variants tied to BMI and smoking). 32 / 53

33 Genes, Financial Literacy and Expectations The Polygenic Score and Financial Literacy The financial literacy module includes 4 questions. 1 Compounding Interest: First, suppose you had $100 in a savings account and the interest rate was 2% per year. After 5 years, how much do you think you would have in the account if you left the money to grow more than $102, exactly $102, or less than $102? 2 Real Interest Rate: Imagine that the interest rate on your savings account was 1% per year and inflation was 2% per year. After 1 year, would you be able to buy more than today, exactly the same as today, or less than today with the money in this account? 3 Diversify Stocks: Do you think that the following statement is true or false: buying a single company stock usually provides a safer return than a stock mutual fund? 4 Inflation and Lending: Are creditors or debtors helped by inflation? Note: only for a small sample of HRS respondents. This highlights the advantage of observed genetic variants. 33 / 53

34 Genes, Financial Literacy and Expectations The Polygenic Score and Financial Literacy EA Score and Financial Literacy (1) (2) (3) (4) (5) Dep Var: Compound Real Diversify All Correct Inflation Interest Interest (1)-(3) and Lending EA Score * * 0.080*** (0.018) (0.016) (0.021) (0.021) (0.020) Obs R Notes: Relating genetic ability to financial literacy. 34 / 53

35 Genes, Financial Literacy and Expectations The Polygenic Score and Financial Expectations Expectations. Economists have long recognized limits to probabilistic thinking (Khaneman and Tversky, 1972). More recent work examines (possibly biased) subjective beliefs. Relative earnings and college major choice. Teacher expectations and student outcomes. Beliefs about the stock market and investments. We examine the EA Score and expectations. Idea: genetic endowments may relate to information processing. This could affect both educational and wealth (through financial decisions). 35 / 53

36 Genes, Financial Literacy and Expectations The Polygenic Score and Financial Expectations Beliefs Patterns. We focus three belief questions: Probability the stock market has + return over the next year. Probability of U.S. economic recession. Probability of double digit inflation. We show: Low genetic score predicts heaping on extreme beliefs (0, 100). These focal beliefs predict behavior. We are not the first to examine these beliefs patterns. We are the first to demonstrate a genetic basis for them. 36 / 53

37 Genes, Financial Literacy and Expectations The Polygenic Score and Financial Expectations Density Subjective Probability Stock Market Up Notes: Histogram of the probability the stock market appreciates in value 37 / 53

38 Genes, Financial Literacy and Expectations The Polygenic Score and Financial Expectations Density Subjective Probability Major Recession Notes: Histogram of the probability of a major U.S. recession 38 / 53

39 Genes, Financial Literacy and Expectations The Polygenic Score and Financial Expectations Density Subjective Probability Double Digit Inflation Notes: Histogram of the probability of double digit inflation 39 / 53

40 Genes, Financial Literacy and Expectations The Polygenic Score and Financial Expectations EA Score and Beliefs (1) (2) (3) (4) Dev. from 0% Prob 50% Prob 100% Prob Objective Panel A: Market Up EA Score *** *** ** (0.141) (0.001) (0.003) (0.001) Obs. 39,743 39,743 39,743 39,743 R Panel B: U.S. Depression EA Score *** *** ** (0.125) (0.002) (0.003) (0.002) Obs. 33,048 33,048 33,048 33,048 R Panel C: Double Digit Inf EA Score *** *** *** (0.184) (0.002) (0.004) (0.002) Obs. 19,551 19,551 19,551 19,551 R Notes: Relating genetic ability to beliefs. 40 / 53

41 Genes, Financial Literacy and Expectations The Polygenic Score and Financial Expectations Magnitude and Interpretation. A one s.d. rise in the EA Score predicts 0.5 percent reduction in the probability that individuals believe there zero chance of a major depression in the next 10 years. 6.7% of individuals respond with 0 for this item. Do extreme beliefs reflect respondent confusion? If so, it is consistent with lower ability. Less helpful in explaining EA Score and financial decisions. Do beliefs predict behavior and wealth? 41 / 53

42 Genes, Financial Literacy and Expectations Linking Beliefs to Behavior Beliefs, Behavior and Household Wealth (1) (2) (3) (4) log Wealth log Wealth Owns Stocks Owns Stocks Ever Pr Mrkt Up 0% *** * *** *** (0.054) (0.063) (0.017) (0.021) Ever Pr Mrkt Up 100% 0.261*** 0.261*** 0.101*** 0.099*** (0.063) (0.063) (0.021) (0.021) Ever Pr Rec 0% (0.055) (0.056) (0.018) (0.018) Ever Pr Rec 100% *** *** (0.059) (0.079) (0.018) (0.025) Ever DD Inf 0% (0.067) (0.066) (0.021) (0.021) Ever DD Inf 100% *** *** ** (0.069) (0.093) (0.020) (0.029) Max Dev Mrkt. Up ** *** (0.001) (0.000) Max Dev Rec *** *** (0.002) (0.001) Max Dev DD Inf (0.002) (0.001) Obs. 13,655 13,655 14,093 14,093 R Notes: Relating heaping on focal beliefs to wealth and stock ownership. 42 / 53

43 Genes, Financial Literacy and Expectations Linking Beliefs to Behavior Financial Planning Horizon. Respondents asked In planning your (family s) saving and spending, which of the following time periods is most important to you:. THE NEXT FEW MONTHS THE NEXT YEAR THE NEXT FEW YEARS THE NEXT 5-10 YEARS LONGER THAN 10 YEARS Possible this reflects time preference. However, this may also reflect ability in forecasting future events. Gabaix and Laibson (2017) develop a model where households with more precise beliefs about future events will behave as if they are impatient even with the same rate of time preference. 43 / 53

44 Genes, Financial Literacy and Expectations Linking Beliefs to Behavior Financial Planning Horizon, Wealth, and the EA Score. Planning Horizon and Household Wealth (1) (2) (3) log Wealth Owns Stocks Horizon > Next Year Plan. Horizon: Next Year 0.375*** 0.095*** (0.054) (0.017) Next Few Yrs 0.633*** 0.155*** (0.049) (0.016) Next 5-10 Yrs 0.821*** 0.230*** (0.069) (0.024) > 10 Yrs 1.018*** 0.179*** (0.153) (0.052) EA Score 0.078*** 0.024*** 0.017*** (0.021) (0.007) (0.005) Obs. 14,702 15,341 8,526 R / 53

45 Genes, Financial Literacy and Expectations Linking Beliefs to Behavior Financial Planning Horizon and Discount Factors. Individuals asked a series of questions - would you rather accept $100 today or $XXX one year from now. Can use these questions to construct an annual discount factor, β. Mean discount factor by horizon: Avg. β Next Few Months Next Year Next Few Yrs Next 5-10 Yrs > 10 Yrs Inconsistent with time preference driving the association between planning horizon and wealth. 45 / 53

46 Genes, Financial Literacy and Expectations Linking Beliefs to Behavior Relationship Between the EA Score and Wealth (1) (2) (3) (4) Log Wealth Log Wealth Log Wealth Log Wealth EA Score 0.090*** 0.087*** 0.074*** 0.072*** (0.027) (0.026) (0.026) (0.025) Standard Controls Yes Yes Yes Yes Belief Controls No Yes No Yes Planning Horizon Controls No No Yes Yes Obs R / 53

47 Genes, Financial Literacy and Expectations Pensions Defined-Benefit Pensions Suppose: Genetic endowments predict wealth. Part of the explanation is financial decision-making. Then: Genetic Endowments would matter less if individuals had less financial decision-making autonomy. We compare individuals with and without pensions. Caveat: pensions are not randomly assigned. 47 / 53

48 Genes, Financial Literacy and Expectations Pensions Pensions and Household Wealth (1) (2) (3) (4) (5) Has Pension log Pen. Wealth log Wealth log Wealth log Wealth EA Score *** 0.135*** 0.115*** (0.008) (0.022) (0.022) (0.035) (0.035) DB Pension 0.459*** 2.376** (0.039) (1.129) (1.109) EA Score x DB ** * (0.038) (0.038) Max Dev Mrkt. Up *** (0.002) Max Dev Rec *** (0.002) Max Dev DD Inf ** (0.002) (Max Dev Mrkt. Up) x DB 0.004** (0.002) (Max Dev Rec.) x DB 0.006** (0.003) (Max Dev DD Inf) x DB 0.005** (0.002) Obs. 13,655 7,727 13,655 13,655 13,655 R Notes: Relating pensions to the genetic ability-wealth gradient. 48 / 53

49 Genes, Financial Literacy and Expectations Alternative Mechanisms Relating to Alternative Mechanisms. We have stressed beliefs and information processing. Cronqvist and Siegel (2016) demonstrate a genetic basis for savings related to smoking and BMI. They do this using massive amounts of Swedish twins data. Indeed, this is why they can say something about observables. They suggest: genes governing self-control also relate to savings. Is the EA Score we study capturing the same genes? Alternatively, are these distinct pathways? 49 / 53

50 Genes, Financial Literacy and Expectations Alternative Mechanisms Additional GWAS. There are polygenic scores from smoking and BMI GWAS. We incorporate them and ask: 1 Are these scores related to the EA Score? 2 How do the BMI and smoking scores related to wealth? 3 Do all three independently predict wealth? 50 / 53

51 Genes, Financial Literacy and Expectations Alternative Mechanisms Household Wealth, Expectations, and Other Scores Panel A: Correlations between Polygenic Scores: EA Score BMI Score Cigs. Score EA Score 1.00 BMI Score Cigs. Score Notes: Correlations among polygenic scores. 51 / 53

52 Genes, Financial Literacy and Expectations Alternative Mechanisms Household Wealth, Expectations, and Other Scores Panel B: Other Scores, Wealth, and Expectations: log Wealth Dev. from Dev. from Dev. from Objective: Objective: Objective: Market Up Depression Double Digit Inf. EA Score 0.080*** *** ** *** (0.022) (0.144) (0.128) (0.187) BMI Score *** (0.021) (0.140) (0.120) (0.183) CPD Score ** (0.021) (0.138) (0.120) (0.177) Obs. 14,575 39,288 32,686 19,335 R Notes: EA Score, polygenic scores, beliefs and wealth. 52 / 53

53 Genes, Financial Literacy and Expectations Alternative Mechanisms Summary. Our findings validate previous twins studies work. The EA Score independently predicts wealth. Savings not related to education (Siegel and Cronqvist, 2015). Smoking and BMI scores are not related to beliefs issues. Conclusion: there are distinct genetic pathways to wealth. Self control, BMI, smoking, savings and wealth. Information processing, education, beliefs, financial decisions. 53 / 53

54 Appendix Notes on Wealth Data Wealth Data. Two key aims: 1 Construct a complete picture of wealth. 2 Include correct information on allocation, e.g., stocks. We begin with the RAND wealth and income files. Problem: employer-sponsored pensions are excluded. IRAs and other non-employer plans are included. Excluded plans include 401(k) or 403(b) plans. They are substantial portion of wealth near retirement. May also be the only vehicle for investment in stocks. Go Back 54 / 53

55 Appendix Notes on Wealth Data Pension Data. Employer-sponsored pension data is not straightforward. There are two types of plans: 1 Defined-Benefit Plans, e.g., traditional pensions. 2 Defined-Contribution Plans, e.g., 401(k). Go Back 55 / 53

56 Appendix Notes on Wealth Data Defined-Benefit Plans. Identified from retired individuals reporting pension income. The income stream is guaranteed. Thus, not counted as stock market participation. Go Back 56 / 53

57 Appendix Notes on Wealth Data Including Retirement Income in Wealth. We follow Yogo (2016). We include DB-Pensions, SS income and annuities. Calculate a p.d.v of this income based on a 1.5% risk-free interest rate. T t 1 s u=1 P t = Y p t+u t R where Y t is total retirement income. s=1 p t is the recipient s survival probability in period t. This is a function of gender, cohort and age. R= Go Back 57 / 53

58 Appendix Notes on Wealth Data Defined-Contribution Plans. Creates its own set of problems. There are two ways these plans are measured. Non-dormant Plans: Plans for current; last employer if retired. We observe stock market participation. Dormant Plans: Previous employer plans, not rolled into IRA. Asked about in 1996, 1998, Stock allocation of not measured. Possible hack: use earlier waves; construct a dummy. Go Back 58 / 53

59 Defined-Contribution Plans. Appendix Notes on Wealth Data Do dormant plan measurement issues drive results? To fix ideas, the following would be a potential problem: 1 Individual s stock market participation (SMP) limited to a dormant plan. 2 That is, they do not participate anywhere else. 3 SMP for this group must also have a different relationship with the EA Score. 4 There would need to be many of these people. Note, we focus on extensive, not intensive margin, thus if they participate elsewhere, we are fine. Also: dormant plans affect SMP, not total wealth. Results do not change if we assign them a SMP value of 1. Go Back 59 / 53

60 Appendix Notes on Wealth Data Wealth Data: Bottom Line. Alternative samples are used for robustness. For example: Exclude housing wealth. Only look at RAND files (without pensions). Include years without dormant plan data. Include workers with dummy for expected DB. Results are robust across samples. Go Back 60 / 53