What others have done Bernheim, Skinner and Weinberg(2001)use panel data from the psid to estimate euler equations Retirement status is instrumented by taking age specific predicted probabilities conditional on demographics(however cannot explain spikes at ages 62 and 65) Median drop is 14%, but higher for low wealth Sample is split in groups: low wealth-to-income households drop their consumption most 31%of households reduce their consumption by at least 35 percentage points at retirement
• Bernheim, Skinner and Weinberg (2001) use panel data from the PSID to estimate Euler equations. Retirement status is instrumented by taking agespecific predicted probabilities conditional on demographics (however cannot explain spikes at ages 62 and 65). • Median drop is 14%, but higher for low wealth • Sample is split in groups: low wealth-to-income households drop their consumption most. • “31% of households reduce their consumption by at least 35 percentage points at retirement”. What Others Have Done
What others have done Possible explanations and related literature Many workers are surprised by inadequate resources when they retire(not consistent with life-cycle model rational expectations) Work related expenses Home production and/or more efficient shopping (Aguiar and hurst, 2005, Hurd and rohwedder 2006) Miniaci et al(2003)estimate by ols the italian retirement consumption drop at 5.4%
Possible explanations and related literature: • Many workers are surprised by inadequate resources when they retire (not consistent with life-cycle model & rational expectations). • Work related expenses. • Home production and/or more efficient shopping (Aguiar and Hurst, 2005, Hurd and Rohwedder, 2006). • Miniaci et al (2003) estimate by OLS the Italian retirement consumption drop at 5.4%. What Others Have Done
What We do An alternative identification strategy: we estimate the change in consumption at retirement by exploiting the exogenous variability in the retirement decision induced by the eligibility rules of the Italian pension system Information on consumption expenditures, eligibility for retirement and retirement status is obtained from the bank of Italy survey on household Income and Wealth(SHIW). No need of panel data to achieve identification
What We Do • An alternative identification strategy: we estimate the change in consumption at retirement by exploiting the exogenous variability in the retirement decision induced by the eligibility rules of the Italian pension system. • Information on consumption expenditures, eligibility for retirement and retirement status is obtained from the Bank of Italy Survey on Household Income and Wealth (SHIW). No need of panel data to achieve identification
Punchline Key result: household non-durable consumption drops by 9.8% because of male retirement. A larger drop estimated for total food(14.1%) Our strategy provides non-parametric identification only for a subpopulation of those who retire(those who retire at the time they become eligible) We estimate smaller drops for poverty sample Our estimates can be reconciled with utility optimization-in the cross section, drop in work- related expenses and leisure substitutes is large enough to explain changes in consumption
Punch-line • Key result: household non-durable consumption drops by 9.8% because of male retirement. A larger drop estimated for total food (14.1%). • Our strategy provides non-parametric identification only for a subpopulation of those who retire (those who retire at the time they become eligible). • We estimate smaller drops for “poverty sample”. • Our estimates can be reconciled with utility optimization - in the cross section, drop in workrelated expenses and leisure substitutes is large enough to explain changes in consumption
The Causal problem Let s be a variable denoting time to from eligibility for retirement, negative values indicate that the subject is not yet eligible Let r be the retirement status r=l for the retired and r=o otherwise. Since retirement is an option available only to the eligible workers, the probability to retire is zero if S<0(and it is thus discontinuous at S=0) et(Y,Yo be the two potential household consumption expenditures corresponding to the head being retired or not retired, respectively, and let B=YI-Yo et Y=Yo+rB be observed consumption, where Y=Y, for households whose head is retired and y=Yo otherwise
The Causal Problem • Let S * be a variable denoting time to/from eligibility for retirement, negative values indicate that the subject is not yet eligible. • Let R be the retirement status, R=1 for the retired and R=0 otherwise. Since retirement is an option available only to the eligible workers, the probability to retire is zero if S*<0 (and it is thus discontinuous at S*=0 ). • Let (Y1 ,Y0 ) be the two potential household consumption expenditures corresponding to the head being retired or not retired, respectively, and let β=Y1 -Y0 . • Let Y = Y0+Rβ be observed consumption, where Y≡Y1 for households whose head is retired and Y≡Y0 otherwise