. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
An Evaluation of Optimal Unemployment Insurance Using
Two Natural Experiments
Po-Chun Huang Tzu-Ting Yang
National Chengchi University Academia Sinica
January 2, 2018
Motivation
I Unemployment insurance (UI) benefit protects individuals against the risk of
earnings loss during unemployment
I But UI benefit also distorts incentives to search for jobs
I UI benefit increases unemployment duration through two distinct channels
(Chetty, 2008):
1 Moral hazard effect (welfare cost)
I More generous UI incentivizes people to keep unemployed to get UI benefits
2 Liquidity effect (welfare gain)
I More generous UI assist people with little saving to smooth their consumption
during unemployment
I Distinguishing liquidity effect from moral hazard effects has important
welfare implications
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
This Paper
OverviewI We disentangle the liquidity effect from the moral hazard effect using UI
administrative data and two natural experiments in Taiwan
1 Use discontinuity in eligibility for extended UI benefits to identify the
(total) effects of UI extension:
I Since 2009, UI recipients aged 45 or older at job loss are eligible for 9 months
(270 days) benefits, rather 6 months (180 days) for those under 45
2 Use the effect of reemployment bonus to identify the moral hazard effect
I Since 2003, UI recipients who find a job before exhausting benefits can
receive 50% of remaining benefits
I Reemployment bonus does not change the income stream during
unemployment so it does not have liquidity effect
I But it affects people’s incentive to keep unemployed
I Use variation in bonus offer around the time when bonus was introduced to
estimate the effects of reemployment bonus
Institutional Background
I In Taiwan, job losers aged 15-65 with at least one year of work history in
the three years prior to layoff are eligible for UI benefits
1 Replacement rate
I 60% of recipients’ average monthly earnings during the 6 months prior to
layoff
2 Potential benefit duration
I Since 2009, the potential benefit duration has been 9 months (270 days) for
workers aged 45 or older at job loss, rather than 6 months (180 days) for those under 45
3 Reemployment bonus
I UI recipients can receive bonuses equal to 50% of remaining benefits, if they find jobs before benefit exhaustion and keep the job for at least three months
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Examples
I Claimant 1 is age 44 at job loss and Claimant 2 is age 45 at job loss
Job Search: Static Version
I Consider a static search model in Chetty (2008).
J(s) = max
s (1− s)u(A + b) + su(A + w + r − τ) − g(s), where r = θb
I Optimal search satisfies:
u(ce)− u(cu) = g′(s)
I The effect of an increase in b is a combination of a liquidity effect (∂s
∂A) and
a moral hazard effect (−∂s
∂w). ∂s ∂b = ∂s ∂A− (1 − θ) ∂s ∂w = ∂s ∂A− (1 − θ) ∂s ∂r
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Welfare: Static Version
I Social planner maximizes welfare by choosing b
W(b, θ) = max
b su(A + w + r− τ) + (1 − s)u(A + b) − g(s)
s.t.(1− s)b + sr = sτ;
I The welfare effect of increasing b depends on the relative size of
consumption smoothing benefits and increased unemployment.
dW db /u′(c e) = 1− s s [ u′(cu)− u′(ce) u′(ce) − ϵ1−s,b s ]
I The ratio of the liquidity to moral hazard effect equals consumption
smoothing benefits. u′(cu)− u′(ce) u′(ce) = ∂s/∂A ∂s/∂w = ∂s/∂A ∂s/∂r 7 / 32
Data
I UI claims and earnings records for the population of UI recipients in Taiwan
from 1999 to 2012
I Each observation contains
I date of job loss and date of birth
I insured duration of unemployment and nonemployment duration
I monthly earnings and some demographic information
I Sample for extended benefits: 20,906 UI recipients age 43 to 46 at job loss
between May 2009 and December 2012. (about 10% of all recipients)
I Sample for bonuses: 91,889 UI recipients between January 2002 and July
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
The Effect of UI Extension on Benefit Duration
Figure:Benefit Duration: 2009-2013
Figure:Benefit Duration: 2005-2008
The Effect of UI Extension on Nonemployment Duration
Figure:Nonemployment Duration: 2009-2013. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
The Effect of UI Extension on Monthly Reemployment Hazard
1st-6th month
Figure: Monthly Reemployment Hazard (1st-6th month): 2009-2013
.04 .06 .08 .1 .12 .14
Average Remeployment Hazard in the First Six Months
40 42 44 46 48 50
Age at Job Loss
Figure:Nonemployment Duration: 2005-2008
.04
.06
.08
.1
.12
Average Remeployment Hazard in the First Six Months
40 42 44 46 48 50
Age at Job Loss
Density Covariates Placebo
Estimate the Effect of UI Extension
Research Design: Regression Discontinuity DesignI To estimate the effect of UI extension, we conduct the following RD design:
yi = α + βEBAge45i+ f(ai) + vi
I yi: duration outcomes, reemployment hazard
I ai: ”age at job loss”
I Age45i= 1[ai≥ 45]
I f(ai): a polynomial function of ai interacted with Age45i
I We estimate this regression within specific age range 45− b < ai< 45 + b
I Bandwidth (b): optimal bandwidth proposed by Calonico, Cattaneo, and
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Estimation Results
The Effect of UI Extension
I The Effect of 90-day increase in UI benefits on benefit duration,
nonemployment duration and monthly reemployment hazard
Estimate the Effect of Reemployment Bonus
Description of Bonus TreatmentI The reemployment bonus program reached back to UI recipients who were
receiving benefits when the program took effect in January 1, 2003
I Example: a worker starting UI spell on Nov. 1, 2002 found a job on Jan. 1,
2003 would receive two months of benefits a a bonus.
1 Cohorts starting UI spell before July 5th, 2002
I They were not eligible for reemployment bonus
2 Cohorts starting UI spells between July 5th, 2002 and December 31st, 2002
I They were partially exposed to the bonus program due to the reach back
provision
3 Cohorts starting UI spells after January 2003
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Estimate the Effect of Reemployment Bonus
Description of Bonus Treatment
0
90
180
Potential Reemployment Bonus
-360 -180 0 180 360
Number of Days from Jan. 1, 2003
Average Reemployment Hazard and UI Starting Date
Figure:Monthly Reemployment Hazard from Jan. 2002 to July 2003
.07
.08
.09
.1
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Average Reemployment Hazard and UI Starting Date (Age 35-50)
Figure:Monthly Reemployment Hazard from Jan. 2002 to July 2003
.04 .05 .06 .07 .08 .09
Monthly Reemployment Rate in the First Six Months
−360 −300 −240 −180 −120 −60 0 60 120 180 Number of Days from Jan. 1, 2003
Figure:Monthly Reemployment Hazard from Jan. 2000 to July 2001
.04 .05 .06 .07 .08 .09
Monthly Reemployment Rate in the First Six Months
−360 −300 −240 −180 −120 −60 0 60 120 180 Number of Days from Jan. 1, 2001
Density Covariates
Estimate the Effect of Reemployment Bonus
Research Design: Regression Kink DesignI We estimate the following hazard model:
him = α + γ(ti− c) + β(ti− c) · D
I him: the reemployment probability in month m + 1 given worker i was not
employed in month m
I t: the first date of benefits receipt
I c: the cutoff date
I D = 1[ti− c ≥ 0]
I Using kink 1, 180· β identifies the effect of a 90-day increase in benefits as
a bonus.
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Estimation Results
The Effect of Reemployment Bonus
I The effect of a reemployment bonus equivalent to 90 days of UI benefits on
monthly reemployment hazard (1st-6th month) for workers aged 35-50 at job loss
Liquidity Effect and Moral Hazard Effect
I To incorporate dynamics, we need to discount the moral hazard by
multiplying S(P) ∂st ∂P |{z} Total Effect = b∂st ∂At | {z } Liquidity Effect −(1 − θ)St+1(P) b ∂st ∂rt |{z}
Moral Hazard Effect
I Plug in the estimated effects of extended benefits and reemployment
bonuses
−0.017 = b∂s0
∂A0 − 0.5 · 0.6 · 0.020
I The liquidity effect explains 65% of the total effect of UI extension
I b ∂s0 ∂AP b∂s0 ∂P = 0.011/0.017 = 0.65
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Conclusion
I We disentangle the liquidity effect from the moral hazard effect and
estimate the welfare effect of a UI extension
I Use UI administrative data and two natural experiments in Taiwan
I A 90-day increase in potential benefit duration
I increases benefit duration and nonemployment duration by 57 days (40%) and
41 days (15%)
I reduces reemployment hazard by 1.7 percentage points in the first six months.
I Eligibility for the reemployment bonus (90-day of benefits)
I increases reemployment hazard by about 2 percentage points for middle-aged
workers.
I We estimate that the liquidity effect accounts for 65% of the effect of
extended UI benefits.
I Our results suggest a marginal increase in potential benefit duration
improves welfare.
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Sufficient Statistic Approach - Chetty (2009)
Sample 1
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Sample 2
Estimate the Effect of Reemployment Bonus
RD—Density Test
back 250 300 350 400 450 500 550 600 Number of UI Spells. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
RD—Smoothness of Observables
back 27 / 32RD—Predicted Nonemployment Duration
back 285 290 295 300 305 310. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
RD—Smoothness of Observables
back 29 / 32Average Reemployment Hazard and UI Starting Date
After UI Benefits ExhaustionFigure:Monthly Reemployment Hazard After Reform: 2002-2003
.04 .05 .06 .07 .08 .09
Monthly Reemployment Rate in the First Six Months
−360 −300 −240 −180 −120 −60 0 60 120 180 Number of Days from Jan. 1, 2003
Figure:Monthly Reemployment Hazard Before Reform: 2001-2002
.05
.06
.07
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
RK—Density Test
back 2000 4000 6000 8000 Number of UI Spells −400 −200 0 200 400Number of Days from Jan. 1, 2003