Discussion of \What Determines Annuity Demand at Retirement" by Guiseppe Cappelletti, Giovanni Guazzarotti & Pietro Tommasino Dirk Krueger University of Pennsylvania, CEPR, and NBER ECB Conference June 1, 2010 Background This paper studies empirically the determinants of demand for annuities at retirement. Theory: { Yaari (1965): If price of annuity is actuarially fair, in absence borrowing constraints and bequest motives risk averse households facing longevity risk should fully annuitize their wealth. { Brown, Davido and Diamond (2005) generalize this result. Data: { Very few households annuitize private wealth: 70+ in US AHEAD data: 8% own private annuities { Those that do are far from annuitizing it fully: SCF 2004 for 6585 old households: annuitized income (incl. DB pension, soc. sec)/total income 50%: Well known lack of annuitization puzzle. If ever there were a prediction of economic theory that was blatantly violated by the empirical evidence, it is that of full annuitization. [Je Brown, NBER Reporter 2004] Potential Reasons (Poterba 2008) Demand Side { Precautionary demand for liquid wealth { Bequest motives { Access to public or informal private longevity risk insurance Supply Side: Prices not actuarially fair (adverse selection, limited competition). This paper Uses SHIW data to study annuity demand, for various given prices. Hypothetical questions posed to households about preferences over two income streams, starting at age 65, until age of death T A1 A2W yi = 1000 y65:1 = W + 500 yi = 500 for i = 65:1; : : : ; T for i = 65:2; : : : ; T Question: how many households j state preference A1 A2W for annuity of 500 over lump sum W; and what Xj correlate with this? Main Results Overall stated preferences A1 A260000 80% A1 A280000 68% A1 A2100000 40% Correlation of annuity demand with observables Xj Corr Probit Sex 0 0 Mar. 0 0 Kids 0 0 Health + + Age + + - yd + + Ed. + + a + 0 Lit. + + - Comments (I) Given data on actual purchases of annuities, stated demand seems large. Should we buy stocks of companies that o er annuities? De ne (W; r; T ) = W PJ 500 j=65 (1+r)j 65 : Companies should sell if (W; r; T ) > 0: Risk-neutral, sel sh, non-liquidity constrained households should buy if (W; r; T ) < 0: Istituto Nazionale di Statistica: in 2009, life expectancy in Italy at age 65 was 83.2 years (males) and 86.7 (females) x 10 Profits of Issuer, W=100000 4 Zero Line r=1% r=4% r=7% 8 6 Profits 4 2 0 -2 -4 -6 70 75 80 Lifetime Horizon 85 90 95 x 10 Profits of Issuer, W=80000 4 Zero Line r=1% r=4% r=7% 6 4 Profits 2 0 -2 -4 -6 -8 70 75 80 Lifetime Horizon 85 90 95 x 10 Profits of Issuer, W=60000 4 5 Zero Line r=1% r=4% r=7% Profits 0 -5 -10 70 75 80 Lifetime Horizon 85 90 95 Comments (I) In terms of pure present discounted values, annuities look good for W = 60; 000 and 80; 000 (and not so bad for females and males with higher than average life expectancy). Perhaps not surprising that a large share of households would want to buy (especially with risk aversion). Given the results, why are actual choices so di erent? What is W in the data? Comments (II) SHIW has signi cant panel dimension. Contains additional information about households to help predict annuity demand. Examples: might be able to construct measures of permanent income, income risk, health risk (etc.) that one cannot measure in pure crosssection. Might help to construct household level empirical proxies for a) precautionary demand for liquid wealth, b) bequest motives c) access to informal private longevity risk insurance. Table 1 Response rates in the Survey of Household Income and Wealth Year Contacted Responses Refusals Absent units Ineligible Gross Net response rate response rate families units 1989 22,344 8,274 9,427 3,855 788 37.0 38.4 1991 25,210 8,188 6,962 9,481 579 32.5 33.2 1993 15,759 8,089 3,152 2,761 1,756 51.3 57.8 1995 15,606 8,135 3,653 2,510 1,308 52.1 56.9 1998 16,268 7,147 6,441 2,680 1,400 43.9 48.1 2000 20,882 8,001 10,461 2,420 802 38.3 39.8 2002 23,356 8,011 14,179 1,166 476 34.3 35.0 2004 22,018 8,012 12,991 1,015 549 36.4 37.3 2006 18,510 7,768 6,603 4,139 304 42.0 42.7 Note: Ineligible units include the families of persons unknown, dead or emigrated. The gross response rate is the ratio of responses to contacted families. The net response rate is the ratio of responses to contacted families net of ineligible units. Table A1 Panel households of the SHIW, 1987-2006 Year of first interview 1987 1989 1991 1993 1995 1998 2000 2002 2004 2006 Cross-sectional sample size Percentage of total sample 1987 8027 1989 1206 7068 1991 350 1837 6001 8027 8274 14.6 8188 26.7 22 Year of survey 1993 1995 1998 2000 173 126 85 61 877 701 459 343 2420 1752 1169 832 4619 1066 583 399 4490 373 245 4478 1993 4128 8089 42.9 8135 44.8 7147 37.3 8001 48.4 2002 44 263 613 270 177 1224 1014 4406 2004 33 197 464 199 117 845 667 1082 4408 8011 45.0 8012 45.0 2006 30 159 393 157 101 636 475 672 1334 3811 7768 50.9