Optimal Annuitization with Stochastic Mortality Probabilities Felix Reichling 1 2 Kent Smetters Congressional Budget Oce 2 The Wharton School and NBER 1 May 2013 Disclaimer This research was supported by the U.S. Social Security Administration through grant #5 RRC08098400-04-00 to the National Bureau of Economic Research as part of the SSA Retirement Research Consortium. The ndings, conclusions, and views expressed are solely those of the author(s) and do not represent the views of SSA, any agency of the Federal Government, the CBO, or the NBER. Outline Introduction Yaari Result Overview of Our Results Three-Period Model Model Set-up Yaari Model Stochastic Survival Probabilities Multi-Period Model Model set-up Calibration Results Sensitivity analysis Additional Factors that Reduce Annuity Demand What is a life annuity? Longevity insurance: protects against the risk of outliving assets In exchange for one-time premium, receive a stream of payments with certainty until death Cannot be bequeathed: you risk this principal at death in exchange for a mortality credit for living Examples: Social Security, dened-benet pensions, xed/variable annuities Yaari (1965) Result IF I Individuals have uncertain life spans (deterministic mortality) I Individuals do not have bequest motives I Premiums are approximately fair THEN I Full annuitization is optimal Counterfactual: Annuitization puzzle (Modigliani (1986); many others) Our Modications 1. Allow for the survival probabilities themselves to be stochastic. I Consistent with an investor's health status evolving over her life in a manner that is not a deterministic function of initial health status and subsequent age. I Produces valuation risk where the present value of annuity payments can rise or fall with health shocks 2. Allow health care shocks to produce correlated reductions in income (disability) and increase in costs (long-term care). 3. Explore how rate of time preference (even geometric) becomes important with stochastic mortality probabilities. Contrast with Previous Literature Small previous literature examined the role of stochastic mortality probabilities I Assumed (i) no correlated costs and (ii) fully patient agents I Concluded that Yaari's full annuitization result maintains I Intuition: Agents also want to insure against health reclassication (valuation) risk by buying annuity early I Might explain why this literature never really took o Contrast with Previous Literature Small previous literature examined the role of stochastic mortality probabilities I Assumed (i) no correlated costs and (ii) fully patient agents I Concluded that Yaari's full annuitization result maintains I Intuition: Agents also want to insure against health reclassication (valuation) risk by buying annuity early I Might explain why this literature never really took o In contrast we show that relaxing (i) or (ii) can lead to imperfect annuitization Overview of Our Results Theoretical Results (1) If households are suciently patient and mortality probabilities are the only source of uncertainty, full annuitization is still optimal (same intuition as previous literature) Overview of Our Results Theoretical Results (1) If households are suciently patient and mortality probabilities are the only source of uncertainty, full annuitization is still optimal (same intuition as previous literature) If a bad mortality update is correlated with an additional loss (income or expenses), incomplete annuitization can be optimal, even without ad-hoc liquidity constraints Overview of Our Results Theoretical Results (1) If households are suciently patient and mortality probabilities are the only source of uncertainty, full annuitization is still optimal (same intuition as previous literature) If a bad mortality update is correlated with an additional loss (income or expenses), incomplete annuitization can be optimal, even without ad-hoc liquidity constraints I Intuition: while the annuity can be rebalanced (equivalently, borrowed against), the present value of its income stream falls in states with larger marginal utility I Aside: We discuss asset mix rebalancing below. I These constraints can't bind in the Yaari model. I Introducing them in our model (i.e., asymmetric info with health shocks) only enhances our dierences Overview of Our Results Theoretical Results (2) If households are suciently impatient then incomplete annuitization may be optimal even without any additional loss Overview of Our Results Theoretical Results (2) If households are suciently impatient then incomplete annuitization may be optimal even without any additional loss I Intuition: A standard lifetime annuity in the traditional of Yaari and the subsequent literature is not an optimal contract for impatient households. The optimal annuity is actually a designer annuity that is a function of the annuitant's rate of time preference. I In Yaari model, level of patience aects level of savings but not the annuitization choice. Hence, issue largely ignored until now. Overview of Our Results Numerical Results (1) Stochastic lifecycle model to simulate optimal annuitization (extensive and intensive margins). Even with no bequest motives or asset-related transaction costs: Overview of Our Results Numerical Results (1) Stochastic lifecycle model to simulate optimal annuitization (extensive and intensive margins). Even with no bequest motives or asset-related transaction costs: I Most households (63%) should not annuitize any wealth Overview of Our Results Numerical Results (1) Stochastic lifecycle model to simulate optimal annuitization (extensive and intensive margins). Even with no bequest motives or asset-related transaction costs: I Most households (63%) should not annuitize any wealth I Annuitization is mostly found with: I wealthier households (costs correlated with health shocks are small relative to their assets) I households where the expected mortality credit is large relative to the valuation risk (older and sicker households) Overview of Our Results Numerical Results (1) Stochastic lifecycle model to simulate optimal annuitization (extensive and intensive margins). Even with no bequest motives or asset-related transaction costs: I Most households (63%) should not annuitize any wealth I Annuitization is mostly found with: I wealthier households (costs correlated with health shocks are small relative to their assets) I households where the expected mortality credit is large relative to the valuation risk (older and sicker households) I If households are allowed to sell (short) annuities (i.e., positive life insurance) then: I Younger households generally short as a hedge against future negative health correlated costs. I Net aggregate annuitization is negative when summed across the measure of households Overview of Our Results Numerical Results (2) Results intended to are mostly normative, that is, about optimality (i.e., we don't claim to explain the annuity puzzle). But, for kicks, we then throw in various other factors (calibrated bequests, asset management fees, etc) to see if we can reduce the extensive and intensive margins even more (of course, these factors also reduce annuitization in the Yaari model) Results: extensive margin as low as 10%, intensive margin even smaller. Additional realistic modications likely to reduce these amounts even more Overview of Our Results Numerical Results (3) Hence, the rational expectations model is not necessarily fully down and out for explaining annuitization patterns, although various behavioral theories, of course, still likely play an important role. Overview of Our Results Numerical Results (3) Hence, the rational expectations model is not necessarily fully down and out for explaining annuitization patterns, although various behavioral theories, of course, still likely play an important role. Our model is also consistent with another stylized fact from the literature which indicates that households typically view annuities as increasing their risk rather than reducing it. I Brown et al. (2008) interpret this evidence as compatible with narrow framing. I In our rational expectations model herein, the presence of stochastic mortality probabilities implies that annuities deliver a larger expected return (mortality credit) along with more risk (valuation risk). A greater level of risk aversion typically reduces annuitization. Three-Period Model Individuals I Individual can life up to 3 periods, j , I Probability of surviving period depends on health state h j j + 1, to period at age and j +1 j +2 is sj (h) and j I Markov transition probability between health state is 0 with h, h ∈ H , where |H| > 1 P (h0 |h), I Individuals can invest in annuities and bonds I Annuity: Pays $1 per period (say in real terms) conditional on survival I Bonds: WLOG. (Could allow for equities w/ variable annuities) Three-Period Model Annuity premium πj and realized net return ρ(h0 |h) for a survivor sj (h) · 1 sj (h) · ∑h0 P (h0 |h) sj+1 (h0 ) · 1 + (1 + r ) (1 + r )2 sj (h) · 1 ∑h0 P (h0 |h) sj+1 (h0 ) · 1 = · 1+ (1 + r ) (1 + r ) ! sj (h) 0 0 · 1 + ∑ P h |h πj+1 h = (1 + r ) h0 πj (h) = (1) Three-Period Model Annuity premium πj and realized net return ρ(h0 |h) for a survivor sj (h) · 1 sj (h) · ∑h0 P (h0 |h) sj+1 (h0 ) · 1 + (1 + r ) (1 + r )2 sj (h) · 1 ∑h0 P (h0 |h) sj+1 (h0 ) · 1 = · 1+ (1 + r ) (1 + r ) ! sj (h) 0 0 · 1 + ∑ P h |h πj+1 h = (1 + r ) h0 πj (h) = 1 + πj+1 (h0 ) ρj h0 |h = − 1. πj (h) (1) (2) Yaari Model Subcase: Deterministic mortality probabilities ( P h0 |h = h0 = h 0 0, h 6= h 1, (3) Yaari Model Subcase: Deterministic mortality probabilities ( P h0 |h = h0 = h 0 0, h 6= h 1, But standard lifecycle aging eects allowed: sj+1 (h) < sj (h) < 1 (3) Yaari Model Subcase: Deterministic mortality probabilities ( P h0 |h = h0 = h 0 0, h 6= h 1, (3) But standard lifecycle aging eects allowed: sj+1 (h) < sj (h) < 1 πj (h) = sj (h) · (1 + πj+1 (h)) (1 + r ) (4) Yaari Model Subcase: Deterministic mortality probabilities ( P h0 |h = h0 = h 0 0, h 6= h 1, (3) But standard lifecycle aging eects allowed: sj+1 (h) < sj (h) < 1 πj (h) = sj (h) · (1 + πj+1 (h)) (1 + r ) (4) (1 + r ) −1 sj (h) (5) ρj (h) = Yaari (1965) Result Intuition Denition We say that annuities statewise dominate bonds if values of h. ρj (h) > r for all Yaari (1965) Result Intuition Denition We say that annuities statewise dominate bonds if values of ρj (h) > r h. Proposition With deterministic survival probabilities, annuities statewise dominate bonds for any initial health state at age j. Proof. In equilibrium, sj (h) (1 + ρj (h)) = (1 + r ) or ρj (h) = provided that sj (h) < 1 (i.e. people will die). (1 + r ) −1 > r sj (h) for all Yaari (1965) Result Intuition Denition We say that annuities statewise dominate bonds if values of ρj (h) > r for all h. Proposition With deterministic survival probabilities, annuities statewise dominate bonds for any initial health state at age j. Proof. In equilibrium, sj (h) (1 + ρj (h)) = (1 + r ) or ρj (h) = provided (1 + r ) −1 > r sj (h) that sj (h) < 1 (i.e. people will die). Intuition Annuities are investment wrappers that produce a mortality credit. Graphical Representation Annuities statewise dominate bonds Yaari Result is Very Strong (1) Literature has not been able to explain the lack of annuitization observed I Davido et al. 2005 argues that most deviations from Yaari's assumptions don't explain incomplete annuitization. I Brown et al. (2008) conclude: As a whole, however, the literature has failed to nd a suciently general explanation of consumer aversion to annuities. Yaari result is even stronger than commonly appreciated Yaari Result is Very Strong (2) Example: Adverse selection does not undermine full annuitization Yaari Result is Very Strong (3) Robust to other market imperfections (see paper) Social Security only reduces saving, not how it is invested Yaari Result is Very Strong (3) Robust to other market imperfections (see paper) Social Security only reduces saving, not how it is invested Insurance within the marriage, similar Yaari Result is Very Strong (3) Robust to other market imperfections (see paper) Social Security only reduces saving, not how it is invested Insurance within the marriage, similar Moral hazard reduces mortality credit but only if annuitization occurs Yaari Result is Very Strong (3) Robust to other market imperfections (see paper) Social Security only reduces saving, not how it is invested Insurance within the marriage, similar Moral hazard reduces mortality credit but only if annuitization occurs Liquidity Constraints I Really an asset rebalancing constraint I Can't bind in Yaari model since the mortality probability at any age is just a deterministic function of the initial health condition when the annuity was written and current age I Consistently, Sheshinski (2008) writes that no apparent reason seems to justify these constraints. (p. 33). Yaari Result is Very Strong (4) One exception is transaction costs but it's knife-edge in Yaari model With Stochastic Mortality Probabilities As in Yaari (1965), we assume rational expectations: I No ad-hoc liquidity constraints (i.e. we allow full asset mix rebalancing). With Stochastic Mortality Probabilities As in Yaari (1965), we assume rational expectations: I No ad-hoc liquidity constraints (i.e. we allow full asset mix rebalancing). I Realistic? Maybe, maybe not. With Stochastic Mortality Probabilities As in Yaari (1965), we assume rational expectations: I No ad-hoc liquidity constraints (i.e. we allow full asset mix rebalancing). I Realistic? Maybe, maybe not. I But don't confuse models. These constraints make more sense with stochastic mortality probabilities. The dierence between the Yaari model and ours would grow even larger if we further restricted asset mix rebalancing in our model (with one caveat noted below) I I.e., we produce incomplete annuitization even without these additional constraints Stochastic Rankings Dominance of Annuities Proposition With stochastic survival probabilities ( p (h0 |h) > 0), annuities do not generically statewise dominate bonds. Stochastic Rankings Dominance of Annuities Proposition With stochastic survival probabilities ( p (h0 |h) > 0), annuities do not generically statewise dominate bonds. I Intuition: Competitive annuity premium at age j is equal to the present value of the expected annuity payments received at ages j +1 and j + 2, conditional on the health state h 0 But a suciently negative health realization h at age produces a capital depreciation at age the mortality credit received. j +1 at age j. j +1 that is larger than Stochastic Rankings Dominance of Annuities Proposition With stochastic survival probabilities ( p (h0 |h) > 0), annuities do not generically statewise dominate bonds. I Intuition: Competitive annuity premium at age j is equal to the present value of the expected annuity payments received at ages j +1 and j + 2, conditional on the health state h 0 But a suciently negative health realization h at age produces a capital depreciation at age j +1 at age j +1 that is larger than the mortality credit received. I However, lack of statewise dominance does not rule out expected utility maximizers (in fact, it's failure is not surprising) j. Stochastic Rankings Expected return to annuities Proposition The expected return to annuities exceeds bonds if the chance of mortality is positive. I I.e., Annuities will be preferred by risk neutral agents. Stochastic Rankings Expected return to annuities Proposition The expected return to annuities exceeds bonds if the chance of mortality is positive. I I.e., Annuities will be preferred by risk neutral agents. Proposition With stochastic survival probabilities, annuities do not generically second-order stochastically dominate (SOSD) bonds. I I.e., Annuities not necessarily preferred by risk averse agents. Example Simplied mortality to focus on reclassication risk: Sj (h)=1 (1) Example Simplied mortality to focus on reclassication risk: Sj (h)=1 (1) Assuming that r = 0, then for our contract: πj (h) = 1 + 0.5πj+1 (hG ) + 0.5πj+1 (hB ) = 1.5 Example Simplied mortality to focus on reclassication risk: Sj (h)=1 (2) The net return for a survivor in Good health: ρj (hG |h) = 2 1.5 − 1 = 0.33 > r , while the net return for a survivor in Bad health: ρj (hB |h) = 1 1.5 − 1 = −0.33 < r . Example Simplied mortality to focus on reclassication risk: Sj (h)=1 (2) The net return for a survivor in Good health: ρj (hG |h) = 2 1.5 − 1 = 0.33 > r , while the net return for a survivor in Bad health: ρj (hB |h) = 1 1.5 − 1 = −0.33 < r . In other words, statewise dominance fails (but, again, we care about SOSD for EU maximizers) Example (Continued) Failure of SOSD: Assumed Preferences Example (Continued) Failure of SOSD: Assumed Preferences To further simplify, suppose: I Has $1.5 endowment at age I Consumes at ages j +1 j (for sure) and j + 2 (if j +1 survives). I Recall: Good or Bad health revealed at I Allowed at age j to pool health risk by investing $1.5 in an Annuity contract (otherwise, a Bond) Example (Continued) Failure of SOSD: Assumed Preferences To further simplify, suppose: I Has $1.5 endowment at age I Consumes at ages j +1 j (for sure) and j + 2 (if j +1 survives). I Recall: Good or Bad health revealed at I Allowed at age j to pool health risk by investing $1.5 in an Annuity contract (otherwise, a Bond) Time-separable conditional expected utility preferences over consumption: u (cj+1 |hj+1 ) + β · sj+1 (hj+1 ) u (cj+2 |hj+1 ) , (6) Example (Continued) Failure of SOSD: Assumed Preferences To further simplify, suppose: I Has $1.5 endowment at age I Consumes at ages j +1 j (for sure) and j + 2 (if j +1 survives). I Recall: Good or Bad health revealed at I Allowed at age j to pool health risk by investing $1.5 in an Annuity contract (otherwise, a Bond) Time-separable conditional expected utility preferences over consumption: u (cj+1 |hj+1 ) + β · sj+1 (hj+1 ) u (cj+2 |hj+1 ) , The unconditional expected utility at age EU = 1 2 where recall j is equal to 1 · [u (cj+1 |hG ) + β · u (cj+2 |hG )] + · u (cj+1 |hB ) , 2 sj+1 (hj+1 = hG ) = 1 and (6) sj+1 (hj+1 = hB ) = 0. (7) Example (Continued) Failure of SOSD: High Patience (β = 1) No Correlated Costs (already a known result) I Bond: hG : cj+1 = 0.75, cj+2 = 0.75; hB : cj+1 = 1.5. I Annuity: hG : cj+1 = 1.0, cj+2 = 1.0; hB : cj+1 = 1.0 I I.e., Annuities more eective at smoothing consumption (due to pooling of reclassication risk) Example (Continued) Failure of SOSD: High Patience (β = 1) No Correlated Costs (already a known result) I Bond: hG : cj+1 = 0.75, cj+2 = 0.75; hB : cj+1 = 1.5. I Annuity: hG : cj+1 = 1.0, cj+2 = 1.0; hB : cj+1 = 1.0 I I.e., Annuities more eective at smoothing consumption (due to pooling of reclassication risk) With Correlated Costs: (e.g., $1 lost with Bad health) I Bond: hG : cj+1 = 0.75, cj+2 = 0.75; hB : cj+1 = 0.5. I Annuity: hG : cj+1 = 1.0, cj+2 = 1.0; hB : cj+1 = 0.0. I Under the Inada condition ( 0 ∂ u(c→ ) ∂c → ∞), Bond is chosen I We will refer to this as the correlated cost channel Example (Continued) Failure of SOSD: Low Patience (β → 0) (1) No Correlated Costs I Bond: hG : cj+1 → 1.5, cj+2 → 0; hB : cj+1 = 1.5. I Annuity: hG : cj+1 → 2.0, cj+2 → 0; hB : cj+1 = 1.0. Example (Continued) Failure of SOSD: Low Patience (β → 0) (1) No Correlated Costs I Bond: hG : cj+1 → 1.5, cj+2 → 0; hB : cj+1 = 1.5. I Annuity: hG : cj+1 → 2.0, cj+2 → 0; hB : cj+1 = 1.0. I Annuity increases consumption variation across the two health states at j + 1; the bond investment perfectly smooths it across allocations actually valued by the agent. I Intuitively, the traditional competitive annuity fails to shift consumption across health states in a way that is informed by the agent's time preferences within each health state. I See Appendix for discussion on designer annuities I We'll refer to it as the impatience channel Example (Continued) Failure of SOSD: Low Patience (β → 0) (2) I This channel does not exist in the Yaari model because the health state is xed; the level of patience only aects the desired level of saving and not how to invest it. I Example above is extreme but see the simple illustrative calculation in the paper with CRRA utility Extensions See Appendix if Interested I Shorter Contracts I A Richer Space of Mortality-Linked Contracts I Hybrid and Designer Annuities Stochastic Survival Probabilities A gateway mechanism Multi-Period Model Individuals I enter economy at j = 21 and die with certainty by I work when young and retire at age 1 j = 120 j = 65 2 3 I can be healthy (h ), disabled (h ), and very sick (h ) I can save into a life annuity and a non-contingent risk-free bond (initially, no shorting, which we relax later) I Health transition and mortality probabilities are based on the actuarial model of Robinson (1996) I adjusted estimates for working-age population to better match Social Security disability data Multi-Period Model Wages Wages are product of four factors: I health status h I a predictable age-related productivity ε I an individual random productivity matrix Qkl ; k, l = 1, ..., Ψ, where η with Markov Ψ ∈ {1, ..., 8} transition I the general-equilibrium market wage rate per unit of labor Age-related and individual productivity states and transition probabilities are from Nishiyama and Smetters (2005) w Multi-Period Model Correlated Shocks - Workers 2 3 Disabled (h ) and very sick workers (h ) do not work ⇒ loss of wage income that is not insured by Disability Insurance I We don't model the lack of full insurance for lost wages, rather take it as given Health care loss L faced by a very sick worker is assumed to be fully paid by private insurance Multi-Period Model Correlated Shocks - Retirees Of course, no risk of disability-related wage loss 3 Very sick retirees (h ) face long-term care expenses that are uninsured ⇒ loss L, unless they qualify for Medicaid Long-term care expenses L are equal to 1.2 times the mean wage, consistent with industry surveys (Genworth Financial (2012), MetLife (2010)) I Similarly, we don't model lack of insurance market for long-term care (e.g., Brown-Finkelstein argue that Medicaid reduces the demand) I Historically: Private insurance existed, although quite expensive with low caps. I Today: Many insurers have pulled out (Met, Pru, Unum, Hartford, etc). I We will do sensitivity analysis on this calibration Multi-Period Model Social insurance payments Workers I Social Security Disability calculated using the U.S. legal bend point formula I We assume no waiting periods and all claims accepted (in reality, most are rejected) Retirees I OASI: Social Security retirement benets according the bend point formula I Medicaid: Very sick retirees can receive a larger medical payment if they have minimal assets Multi-Period Model Household preferences U= " J J ∑ β j u(cj ) = ∑ 1 j= 1 j= cj1−σ 1−σ # + ξ Dj Aj+1 Multi-Period Model Household optimization Given the state variables (A, η, h, j ) and prices (w ,r ,ρ ), household solve: Vj (Aj , ηj , hj , j) = max {u(cj ) + β s(hj , j)E [Vj+1 (Aj+1 , ηj+1 , hj+1 , j + 1)]} c,α subject to: Aj+1 = R(αj , hj , hj+1 )(Aj + Xj − cj ) α ≤1 0 where ≤ cj ≤ Aj + Xj Xj = I (h = h1 )(1 − T )εj ηj w + Bj + Trj − Lj R(αj , hj , hj+1 ) = αj ρj (hj , hj+1 ) + (1 − αj )r Multi-Period Model Production and payroll taxes I Output Y is generated by I Utility weight β Y = θ K λ L1−λ chosen to produce 2.8 capital-output ratio I Social Security, Disability and Medicare transfers are nanced through a labor income tax, assuming a balanced budget. Multi-Period Model General equilibrium A general equilibrium, therefore, is fairly standard and so a formal denition will be skipped. In particular: 1. Household Optimization: Households optimize utility, taking as given the set of factor prices and policy parameters; 2. Asset Market Clearing: The factor prices are derived from the production technology, with the aggregate levels of saving and labor properly integrated across the measure of households; 3. Policy Balance: The policy parameters are consistent with balanced budget constraints (i.e., tax revenue equals spending); and, 4. Bequest Clearing: Bequests left equal bequests received. Calibration Health transition probabilitiesthe healthy Calibration Health transition probabilitiesthe disabled Calibration Health transition probabilitiesthe very sick Calibration Survival probabilities Calibration Annuity returns by age and health state transitionthe healthy Calibration Annuity returns by age and health state transitionthe disabled Calibration Annuity returns by age and health state transitionthe very sick Calibration Disability rates: data vs. model Source: Social Security Administration and authors' calculations. Calibration Age structure of the population: data vs. model Source: U.S. Census Bureau (2011) and authors' calculations. Fractions are based on the population 21 years and older. Calibration Wealth distribution: data vs. model (1) Sources: (1) Nishiyama (2002), (2) U.S. Census Bureau (2012a). Calibration Wealth distribution: data vs. model (2) I The poverty rate among all workers between the ages of 18 to 64 in our model is 4.2%, which is a bit below the Census value of 7.2% (U.S. Census Bureau 2012b). Our underestimate is likely due to the U.S. Census having a broader denition of working relative to our model's wage data. Calibration Wealth distribution: data vs. model (2) I The poverty rate among all workers between the ages of 18 to 64 in our model is 4.2%, which is a bit below the Census value of 7.2% (U.S. Census Bureau 2012b). Our underestimate is likely due to the U.S. Census having a broader denition of working relative to our model's wage data. I The poverty rate among all disabled people between the ages of 18 to 64 is 33.5% in our model, which is pretty close to the empirical counterpart of 28.8%, as estimated by U.S. Census Bureau (2012b), although lower than the value of 50% estimated by Congressional Budget Oce (2012). Results Annuitized fraction of wealth at Age 65 for healthy person (Black Lines) and population distribution (Gray) (1) Results Annuitized fraction of wealth at Age 65 for healthy person (Black Lines) and population distribution (Gray) (2) I At CRRA = 2, households with wealth at age 65 below 6 times the national average wage don't annuitize Results Annuitized fraction of wealth at Age 65 for healthy person (Black Lines) and population distribution (Gray) (2) I At CRRA = 2, households with wealth at age 65 below 6 times the national average wage don't annuitize I At CRRA = 5, that threshold increases to 8 times the national average Results Annuitized fraction of wealth at Age 65 for healthy person (Black Lines) and population distribution (Gray) (2) I At CRRA = 2, households with wealth at age 65 below 6 times the national average wage don't annuitize I At CRRA = 5, that threshold increases to 8 times the national average I At CRRA = 3, 53% of age-65 retirees hold no annuities and only 11% fully annuitize Results Annuitized fraction of wealth at Age 65 for healthy person (Black Lines) and population distribution (Gray) (2) I At CRRA = 2, households with wealth at age 65 below 6 times the national average wage don't annuitize I At CRRA = 5, that threshold increases to 8 times the national average I At CRRA = 3, 53% of age-65 retirees hold no annuities and only 11% fully annuitize I Annuitization occurs at larger wealth levels because retiree has enough assets to pay for any potentially correlated long-term care cost from the annuity stream itself Results Annuitized fraction of wealth at Age 65 for healthy person (Black Lines) and population distribution (Gray) (2) I At CRRA = 2, households with wealth at age 65 below 6 times the national average wage don't annuitize I At CRRA = 5, that threshold increases to 8 times the national average I At CRRA = 3, 53% of age-65 retirees hold no annuities and only 11% fully annuitize I Annuitization occurs at larger wealth levels because retiree has enough assets to pay for any potentially correlated long-term care cost from the annuity stream itself I Recall: Results not driven by ad-hoc liquidity constraints. Adding those would further decrease annuitization. Results Some quick checks Results Some quick checks I We ran a Yaari version of our model where health shocks are turned o and the mortality rate by age is set equal to the average mortality rate weighted by the share of households in each health state at that age. I 100% of wealth annuitized, as we would expect Results Some quick checks I We ran a Yaari version of our model where health shocks are turned o and the mortality rate by age is set equal to the average mortality rate weighted by the share of households in each health state at that age. I 100% of wealth annuitized, as we would expect I We also ran our model but assumed that long-term care costs were fully insured (but with health shocks) I Not the same thing as Yaari model due to presence of valuation risk I Still, 100% of wealth at retirement fully annuitized I Hence, the impatience channel is not driving these results either. I Results being driven by the correlated cost channel. Results Annuitized fraction of all wealth, by age (intensive margin). Population density in gray. (1) Results Annuitized fraction of all wealth, by age (intensive margin). Population density in gray. (2) I Not monotonic in age due to rate of return to annuities being a function of the mortality credit (which is monotonic in age) and valuation risk (and, such heterogeneity across health groups, the shares of which vary by age). Results Fraction of households with annuities, by age (extensive margin) (1) Results Fraction of households with annuities, by age (extensive margin) (2) I At CRRA=3, 37% of all households in the economy hold any positive level of annuities; 24% of households fully annuitize. But 57% of wealth is annuitized (due to skewed wealth distribution) Sensitivity Analysis Reducing long-term care costs Sensitivity Analysis Allowing short-sales (intensive margin) Sensitivity Analysis Allowing short-sales (extensive margin) Sensitivity Analysis When healthy, young savers short annuities I Equivalent to purchasing life insurance I In other words, they pay (instead of earn) the mortality credit. I But the mortality credit is cheap when young, and they young still face a lot of health reclassication risk (and, hence, correlated expenses) Then, after a negative health shock, buy (go long in) annuities I Current life policy is very valuable; annuity is cheap I Life policy could be sold (life settlement) or borrowed against to nance annuity (a fully collaterialized swap) This short-long trade produces a windfall that can be used to pay for correlated expenses Additional Factors Management fees and bequests (1) How far can our rational model be pushed to produce a low level of annuitization? Additional Factors Management fees and bequests (1) How far can our rational model be pushed to produce a low level of annuitization? Management fees and bequests also reduce annuitization in the Yaari model. So we typically run both models. Additional Factors Management fees and bequests (2) Management fees: Management fees for annuities are about 1.0% larger than for bonds (see paper), not including surrender charges Additional Factors Management fees and bequests (2) Management fees: Management fees for annuities are about 1.0% larger than for bonds (see paper), not including surrender charges I Yaari model: I Intensive margin: 73% of all wealth annuitized; 100% of retiree wealth; 65% of non-retiree wealth I Extensive margin: 43% of households annuitize; 90% of retirees (10% have no wealth); 33% non-retirees I But knife-edged: most households either 100% or 0%. Fraction of households that fully annuitize also equal to 43%. Additional Factors Management fees and bequests (2) Management fees: Management fees for annuities are about 1.0% larger than for bonds (see paper), not including surrender charges I Yaari model: I Intensive margin: 73% of all wealth annuitized; 100% of retiree wealth; 65% of non-retiree wealth I Extensive margin: 43% of households annuitize; 90% of retirees (10% have no wealth); 33% non-retirees I But knife-edged: most households either 100% or 0%. Fraction of households that fully annuitize also equal to 43%. I Our model: I Intensive margin: 27% of all weath annuitized; 42% of retiree; 20% of non-retiree I Extensive margin: 22% of all households annuitize; 49% of retirees; 17% non-retirees I Not knife edge: Only 7% fully annuitize Additional Factors Management fees and bequests (3) With positive bequests management fees) (ξ > 0) and a 2.5% bequest-GPD ratio (no Additional Factors Management fees and bequests (3) With positive bequests (ξ > 0) and a 2.5% bequest-GPD ratio (no management fees) I Yaari model: 67% of wealth annuitized and 90% of households hold a positive level of annuity I Our model: 38% of wealth annuitized and 35% of households hold a positive level of annuity Additional Factors Management fees and bequests (3) With positive bequests (ξ > 0) and a 2.5% bequest-GPD ratio (no management fees) I Yaari model: 67% of wealth annuitized and 90% of households hold a positive level of annuity I Our model: 38% of wealth annuitized and 35% of households hold a positive level of annuity Add back in the management fees: 11% of wealth annuitized; 14% with any annuities in our model Additional Factors Management fees and bequests (3) With positive bequests (ξ > 0) and a 2.5% bequest-GPD ratio (no management fees) I Yaari model: 67% of wealth annuitized and 90% of households hold a positive level of annuity I Our model: 38% of wealth annuitized and 35% of households hold a positive level of annuity Add back in the management fees: 11% of wealth annuitized; 14% with any annuities in our model Uneven bequests (top 40%): Our model: 8% of wealth annuitized; 10% hold a positive level in our model Possible Future Extensions Dierential transaction costs More Worker Risk Works Cited in Talk I Brown, Jerey R., Jerey R. Kling, Sendhil Mullainathan, and Marian V. Wrobel, Why Don't People Insure Late-Life Consumption? A Framing Explanation of the Under-Annuitization Puzzle, American Economic Review, May 2008, 98 (2), 30409. Congressional Budget Oce, Policy Options for the Social Security Disability Insurance Program, July 2012. Davido, Thomas, Jerey R. Brown, and Peter A. Diamond, Annuities and Individual Welfare, American Economic Review, December 2005, 95 (5), 15731590. Genworth Financial, Genworth 2012 Cost of Care SurveyExecutive Summary, 2012. MetLife, The 2010 MetLife Market Survey of Nursing Home, Assisted Living, Adult Day Services, and Home Care Costs, October 2010. Works Cited in Talk II Modigliani, Franco, Life Cycle, Individual Thrift, and the Wealth of Nations, The American Economic Review, June 1986, 76 (3), 297313. Nishiyama, Shinichi, Bequests, Inter Vivos Transfers, and Wealth Distribution, Review of Economic Dynamics, October 2002, 5 (4), 892931. and Kent Smetters, Consumption Taxes and Economic Eciency with Idiosyncratic Wage Shocks, Journal of Political Economy, October 2005, 113 (5), 10881115. Robinson, Jim, A long-term care status transition model, in The Old-Age Crisis: Actuarial Opportunities The 1996 Bowles Symposium Georgia State University, Atlanta 1996, pp. 7279. Chapter 8. Sheshinski, Eytan, The economic theory of annuities, Princeton University Press, 2008. Works Cited in Talk III U.S. Census Bureau, Statistical Abstract of the United States: 2012, Washington, DC, September 2011. , Household Income Inequality Within U.S. Counties: 2006-2010, Washington, DC, February 2012. , Income, Poverty, and Health Insurance Coverage in the United States: 2011, Washington, DC, September 2012. Yaari, Menahem E., Uncertain Lifetime, Life Insurance, and the Theory of the Consumer, The Review of Economic Studies, April 1965, 32 (2), 137150.