business cycle: a parametric approach · higher-order income risk over the business cycle: a...
TRANSCRIPT
HIGHER-ORDER INCOME RISK OVER THE
BUSINESS CYCLE: A PARAMETRIC APPROACH
Christopher Buscha Alexander Ludwigb
aMOVE-UAB and Barcelona GSE
bSAFE, Goethe University Frankfurt
V Winter Macroeconomics Workshop
Bellaterra, March 2019
Motivation
I Idiosyncratic income risk important for
I Welfare costs of fluctuations
I Benefits from insurance
I Design of tax and transfer policies
I Recent empirical work documents rich deviations from
“Normality”
I Distribution of shocks
I Persistence of shocks
I Question: does it matter from a macroeconomic perspective?
Busch & Ludwig (MOVE-UAB & Goethe U): Higher-Order Income Risk 1 / 38
This paper
1. Characterize idiosyncratic income risk over business cycle
I Counter-cyclical variance; Pro-cyclical skewnessI High (excess) kurtosisI Post government: weaker cyclicality
2. Welfare implications of remaining higher-order income risk?
I Welfare losses (gains) for high (low) risk aversion
3. Does higher-order risk matter for costs of cycles?
I Yes! With high RA, larger welfare losses from fluctuations
4. Does higher-order risk matter for self-insurance?
I Yes! With high RA, better consumption smoothing (moreprecautionary savings)
Busch & Ludwig (MOVE-UAB & Goethe U): Higher-Order Income Risk 2 / 38
This paper
1. Characterize idiosyncratic income risk over business cycleI Counter-cyclical variance; Pro-cyclical skewnessI High (excess) kurtosisI Post government: weaker cyclicality
2. Welfare implications of remaining higher-order income risk?
I Welfare losses (gains) for high (low) risk aversion
3. Does higher-order risk matter for costs of cycles?
I Yes! With high RA, larger welfare losses from fluctuations
4. Does higher-order risk matter for self-insurance?
I Yes! With high RA, better consumption smoothing (moreprecautionary savings)
Busch & Ludwig (MOVE-UAB & Goethe U): Higher-Order Income Risk 2 / 38
This paper
1. Characterize idiosyncratic income risk over business cycleI Counter-cyclical variance; Pro-cyclical skewnessI High (excess) kurtosisI Post government: weaker cyclicality
2. Welfare implications of remaining higher-order income risk?I Welfare losses (gains) for high (low) risk aversion
3. Does higher-order risk matter for costs of cycles?
I Yes! With high RA, larger welfare losses from fluctuations
4. Does higher-order risk matter for self-insurance?
I Yes! With high RA, better consumption smoothing (moreprecautionary savings)
Busch & Ludwig (MOVE-UAB & Goethe U): Higher-Order Income Risk 2 / 38
This paper
1. Characterize idiosyncratic income risk over business cycleI Counter-cyclical variance; Pro-cyclical skewnessI High (excess) kurtosisI Post government: weaker cyclicality
2. Welfare implications of remaining higher-order income risk?I Welfare losses (gains) for high (low) risk aversion
3. Does higher-order risk matter for costs of cycles?I Yes! With high RA, larger welfare losses from fluctuations
4. Does higher-order risk matter for self-insurance?
I Yes! With high RA, better consumption smoothing (moreprecautionary savings)
Busch & Ludwig (MOVE-UAB & Goethe U): Higher-Order Income Risk 2 / 38
This paper
1. Characterize idiosyncratic income risk over business cycleI Counter-cyclical variance; Pro-cyclical skewnessI High (excess) kurtosisI Post government: weaker cyclicality
2. Welfare implications of remaining higher-order income risk?I Welfare losses (gains) for high (low) risk aversion
3. Does higher-order risk matter for costs of cycles?I Yes! With high RA, larger welfare losses from fluctuations
4. Does higher-order risk matter for self-insurance?I Yes! With high RA, better consumption smoothing (more
precautionary savings)
Busch & Ludwig (MOVE-UAB & Goethe U): Higher-Order Income Risk 2 / 38
Approach
1. Individual income risk over the business cycle
I Residual income (pre- and post-government)
I Parametric approach: GMM estimation
2. Evaluate macroeconomic implications
I Tool: Life-cycle model with exogenous income
I First: Analytical illustration in two-period model
3. Intermediate step:
I Transparent shock discretization method
Busch & Ludwig (MOVE-UAB & Goethe U): Higher-Order Income Risk 3 / 38
Literature
I Earnings dynamics over business cycle
I Early Evidence: 2nd Moment Shock (Storesletten, Telmer & Yaron,
JPE, 2004; Bayer & Juessen, EL, 2012)I Newer Evidence: 3rd Moment Shock (Guvenen, Ozkan & Song,
JPE, 2014; Busch, Domeij, Guvenen & Madera, 2018, Huggett & Kaplan,
RED, 2016)
I Rich earnings dynamics
I Arellano, Blundell & Bonhomme, Ecta, 2017; Guvenen, Karahan, Ozkan,
Song, 2018; Holter, Halvorsen, Ozkan & Storesletten, 2018;...
I Macroeconomic implications
I Consumption insurance (De Nardi, Fella & Paz-Pardo, JEEA, 2018)I Aggregate savings (Civale, Díez-Catalán & Fazilet, 2016)I Optimal fiscal policy (Golosov, Troshkin & Tsyvinski, AER, 206)I Portfolio choice & equity premium (Catherine, 2019)
Busch & Ludwig (MOVE-UAB & Goethe U): Higher-Order Income Risk 4 / 38
LiteratureI Earnings dynamics over business cycle
I Early Evidence: 2nd Moment Shock (Storesletten, Telmer & Yaron,
JPE, 2004; Bayer & Juessen, EL, 2012)I Newer Evidence: 3rd Moment Shock (Guvenen, Ozkan & Song,
JPE, 2014; Busch, Domeij, Guvenen & Madera, 2018, Huggett & Kaplan,
RED, 2016)
I Rich earnings dynamics
I Arellano, Blundell & Bonhomme, Ecta, 2017; Guvenen, Karahan, Ozkan,
Song, 2018; Holter, Halvorsen, Ozkan & Storesletten, 2018;...
I Macroeconomic implications
I Consumption insurance (De Nardi, Fella & Paz-Pardo, JEEA, 2018)I Aggregate savings (Civale, Díez-Catalán & Fazilet, 2016)I Optimal fiscal policy (Golosov, Troshkin & Tsyvinski, AER, 206)I Portfolio choice & equity premium (Catherine, 2019)
Busch & Ludwig (MOVE-UAB & Goethe U): Higher-Order Income Risk 4 / 38
LiteratureI Earnings dynamics over business cycle
I Early Evidence: 2nd Moment Shock (Storesletten, Telmer & Yaron,
JPE, 2004; Bayer & Juessen, EL, 2012)I Newer Evidence: 3rd Moment Shock (Guvenen, Ozkan & Song,
JPE, 2014; Busch, Domeij, Guvenen & Madera, 2018, Huggett & Kaplan,
RED, 2016)
I Rich earnings dynamics
I Arellano, Blundell & Bonhomme, Ecta, 2017; Guvenen, Karahan, Ozkan,
Song, 2018; Holter, Halvorsen, Ozkan & Storesletten, 2018;...
I Macroeconomic implications
I Consumption insurance (De Nardi, Fella & Paz-Pardo, JEEA, 2018)I Aggregate savings (Civale, Díez-Catalán & Fazilet, 2016)I Optimal fiscal policy (Golosov, Troshkin & Tsyvinski, AER, 206)I Portfolio choice & equity premium (Catherine, 2019)
Busch & Ludwig (MOVE-UAB & Goethe U): Higher-Order Income Risk 4 / 38
LiteratureI Earnings dynamics over business cycle
I Early Evidence: 2nd Moment Shock (Storesletten, Telmer & Yaron,
JPE, 2004; Bayer & Juessen, EL, 2012)I Newer Evidence: 3rd Moment Shock (Guvenen, Ozkan & Song,
JPE, 2014; Busch, Domeij, Guvenen & Madera, 2018, Huggett & Kaplan,
RED, 2016)
I Rich earnings dynamics
I Arellano, Blundell & Bonhomme, Ecta, 2017; Guvenen, Karahan, Ozkan,
Song, 2018; Holter, Halvorsen, Ozkan & Storesletten, 2018;...
I Macroeconomic implications
I Consumption insurance (De Nardi, Fella & Paz-Pardo, JEEA, 2018)I Aggregate savings (Civale, Díez-Catalán & Fazilet, 2016)I Optimal fiscal policy (Golosov, Troshkin & Tsyvinski, AER, 206)I Portfolio choice & equity premium (Catherine, 2019)
Busch & Ludwig (MOVE-UAB & Goethe U): Higher-Order Income Risk 4 / 38
2nd Moment Shock: Symmetric Risk?
yt+1 − yt
-1 -0.5 0 0.5 1
Den
sity
0
0.5
1
1.5
2
Expansion
Busch & Ludwig (MOVE-UAB & Goethe U): Higher-Order Income Risk 5 / 38
2nd Moment Shock: Symmetric Risk?
yt+1 − yt
-1 -0.5 0 0.5 1
Den
sity
0
0.5
1
1.5
2
ExpansionRecession
Busch & Ludwig (MOVE-UAB & Goethe U): Higher-Order Income Risk 5 / 38
3rd Moment Shock: Asymmetric Risk?
yt+1 − yt
-1 -0.5 0 0.5 1
Den
sity
0
0.5
1
1.5
2
Expansion
Busch & Ludwig (MOVE-UAB & Goethe U): Higher-Order Income Risk 6 / 38
3rd Moment Shock: Asymmetric Risk?
yt+1 − yt
-1 -0.5 0 0.5 1
Den
sity
0
0.5
1
1.5
2
ExpansionRecession
Busch & Ludwig (MOVE-UAB & Goethe U): Higher-Order Income Risk 6 / 38
Outline
Intro
Higher-Order Risk in Two-Period Model
Estimation of Idiosyncratic Income Process
Macroeconomic Implications
Conclusion
Two-Period Model: Setup
I EndowmentsI Agents live for two periodsI Non-risky first-period income y0
I Risky second-period income y1 = exp(ε1), ε ∼ Ψ:
a1 = a0 + y0 − c0, a0 given
c1 = a1 + y1
I Preferences: Epstein-Zin-Weil
Busch & Ludwig (MOVE-UAB & Goethe U): Higher-Order Income Risk 7 / 38
Two-Period Model: Preferences
U =
1
1−ρ
(c
1− 1ρ
0 + v (c1, θ,Ψ)1− 1ρ
)for ρ 6= 1
ln (c0) + ln (v (c1, θ,Ψ)) for ρ = 1.
with
v(c1, θ,Ψ) =
(∫
c1(ε)1−θdΨ(ε)) 1
1−θ =(E[c1(ε)1−θ]) 1
1−θ for θ 6= 1
exp(∫
ln(c(ε))dΨ(ε))
= exp (E [ln(c(ε))]) for θ = 1.
I ρ Intertemporal Elasticity of Substitution
I θ Risk Aversion (and “higher risk attitudes”)
Busch & Ludwig (MOVE-UAB & Goethe U): Higher-Order Income Risk 8 / 38
Two-Period Model: Hand-to-Mouth Consumers
I Shut down risk-free savings technology: a1 = 0
I Fourth-Order Taylor approximation around µc1 = E0[c1]:
U ≈c
1− 1ρ
0
1− 1ρ
+
(1
1− 1ρ
− θ
2µc
2 +θ(θ + 1)
6µc
3 −θ(θ + 1)(θ + 2)
24µc
4
)(1)
I Increase in risk: welfare losses for θ > 1(µ
exp(ε)2 ↑, µexp(ε)
3 ↓, µexp(ε)4 ↑
)I Losses ↑ in θ for θ > 1.
I Relative weight ↓ in order of risk
I Equation (1) is for ρ = 1θ ; extends readily to ρ 6= 1
θ
Busch & Ludwig (MOVE-UAB & Goethe U): Higher-Order Income Risk 9 / 38
Two-Period Model: Precautionary Savings Reaction
I Allow for savings. Approximation of First Order Condition:
(y0 − a1)− 1ρ = E0
[(exp(ε) + a1)
−θ]
(2)
≈ (1 + a1)−θ +
θ(1 + θ)
2(1 + a1)
−(2+θ) µexp(ε)2
− θ(1 + θ)(2 + θ)
6(1 + a1)
−(3+θ) µexp(ε)3
+θ(1 + θ)(2 + θ)(3 + θ)
24(1 + a1)
−(4+θ) µexp(ε)4 .
I LHS increasing in a1, RHS decreasing in a1
I Increase in risk→ More savings:(µ
exp(ε)2 ↑, µexp(ε)
3 ↓, µexp(ε)4 ↑
)→ RHS ↑→ a1 ↑
I Reaction depends on risk attitudes (pinned down by θ)
Busch & Ludwig (MOVE-UAB & Goethe U): Higher-Order Income Risk 10 / 38
Note on Logs vs. Levels
I Moments in previous analysis are “in levels”
I Income process characterized “in logs”
I How do moments of logs translate into moments of levels?I In logs: µ3 ↓⇒ In levels: µ2 ↓, µ3 ↓, µ4 ↓
I In logs: µ4 ↑⇒ In levels: µ2 ↑, µ3 ↑, µ4 ↑
⇒ Quantitative question whether change in µ2 & µ4 or change in µ3
dominate
Answer depends on magnitude of change and on risk attitudes
Numerical Example Numerical Illustration
Busch & Ludwig (MOVE-UAB & Goethe U): Higher-Order Income Risk 11 / 38
Outline
Intro
Higher-Order Risk in Two-Period Model
Estimation of Idiosyncratic Income Process
Macroeconomic Implications
Conclusion
Overview of Estimation
1. Log income:
yith = f (Xith,Yt ) + yith
f (Xith,Yt ) = β0t + fh (h) + Ieit =c fEP(EP) + βsize log (hhsizeith)
fh (h) = βage1 h + β
age2 h2 + β
age3 h3
fEP(EP) = βEP0 + βEP
1 t + βEP2 t2
2. Fit stochastic process to residual income yith
3. Inference: Bootstrap
Busch & Ludwig (MOVE-UAB & Goethe U): Higher-Order Income Risk 12 / 38
GMM Estimation Approach
I Residual income process:
yith = χi + zith + εith
zith = ρzit−1h−1 + ηith
I Distributions of stochastic components:
εith ∼ Fε, χi ∼ Fχ, ηith ∼ Fη (s (t))
I Interested in central moments µ2, µ3, and µ4
Busch & Ludwig (MOVE-UAB & Goethe U): Higher-Order Income Risk 13 / 38
GMM Estimation ApproachIdentification
I Persistent shocks accumulate
⇒ Cross-sectional moments identify distribution of shocks:Fη (s (t)) varies with aggregate state s(t): cross-sectionaldistribution at age h differs across cohorts t − h
I Identifies F s(t)=EXPη and F s(t)=CON
η
I Extend Storesletten, Telmer & Yaron (2004) to 3rd central
moment
Busch & Ludwig (MOVE-UAB & Goethe U): Higher-Order Income Risk 14 / 38
Variance as Share of Recessions
I More recessions: higher varianceBusch & Ludwig (MOVE-UAB & Goethe U): Higher-Order Income Risk 15 / 38
Skewness as Share of Recessions
I More recessions: stronger left skewness
Busch & Ludwig (MOVE-UAB & Goethe U): Higher-Order Income Risk 16 / 38
Approach: Theoretical Moments
Conditional on history St = [s0, . . . , st ]:
µ2 (yith; θ) = E[(yith − E [yith])2 | St
]µ11 (yith, yit+1h+1; θ) = E
[(yith − E [yith]) (yit+1h+1 − E [yit+1h+1]) | St]
µ3 (yith; θ) = E[(yith − E [yith])3 | St
]µ21 (yith, yit+1h+1; θ) = E
[(yith − E [yith])2 (yit+1h+1 − E [yit+1h+1]) | St
]µ4 (yith; θ) = E
[(yith − E [yith])4 | St
]µ31 (yith, yit+1h+1; θ) = E
[(yith − E [yith])3 (yit+1h+1 − E [yit+1h+1]) | St
]
Busch & Ludwig (MOVE-UAB & Goethe U): Higher-Order Income Risk 17 / 38
Approach: Theoretical Moments
µ2 (yith; θ) = µχ2 + µε2 +∑h−1
j=0ρ2jµη2 (s (t − j))
µ11 (yith, yit+1h+1; θ) = µχ2 + ρ∑h−1
j=0ρ2jµη2 (s (t − j))
µ3 (yith; θ) = µχ3 + µε3 +∑h−1
j=0ρ3jµη3 (s (t − j))
µ21 (yith, yit+1h+1; θ) = µχ3 + ρ∑h−1
j=0ρ3jµη3 (s (t − j))
µ4 (yith; θ) = µχ4 + µε4 +∑h−1
j=0ρ4jµη4 (s (t − j)) + g1 (ρ,−→µ 2)
µ31 (yith, yit+1h+1; θ)) = µχ4 + ρ2∑h−1
j=0ρ4jµη4 (s (t − j)) + g2 (ρ,−→µ 2)
More details on 4th moment
Busch & Ludwig (MOVE-UAB & Goethe U): Higher-Order Income Risk 18 / 38
Moment Conditions
I Parameters to estimate:
θ =(ρ, µχ2 , µ
ε2, µ
η,E2 , µη,C2 , µχ3 , µ
ε3, µ
η,E3 , µη,C3 , µχ4 , µ
ε4, µ
η,E4 , µη,C4
)
I Stack moment conditions into F (θ) and minimize (weighted with
#obs):
minθF (θ)′WF (θ)
Busch & Ludwig (MOVE-UAB & Goethe U): Higher-Order Income Risk 19 / 38
Data and Sample Selection
I Germany (GSOEP) and US (PSID)
I GSOEP: about 5,000 households in 1984 / 10,500 in 2013
I PSID: about 2,000 households each wave (annual 1968 –1997,
biannual afterwards)
I Sample Selection:I 2 adults present in householdI Head of household age 25-60I Annual household income ≥ 520 hours at half minimum wage
I Pre-sample aggregate information: expansions, contractions
Busch & Ludwig (MOVE-UAB & Goethe U): Higher-Order Income Risk 20 / 38
Estimation Results PSID
HH Pre HH Post
ρ .9873 .9815µχ2 .2150 .1926µε2 .1038 .0936µη,C2 .0132 .015µη,E2 .0076 .0067µχ3 −.1248 −.0966µε3 −.1489 −.1219µη,C3 −.0121 −.0145µη,E3 −.003 −.0004µχ4 .1386 .1113µε4 .4393 .3125µη,C4 .0413 .0584µη,E4 .0137 .0117
Busch & Ludwig (MOVE-UAB & Goethe U): Higher-Order Income Risk 21 / 38
Estimation Results PSID
HH Pre HH Post
ρ .9873 .9815µχ2 .2150 .1926µε2 .1038 .0936µη,C2 .0132 .015µη,E2 .0076 .0067µχ3 −.1248 −.0966µε3 −.1489 −.1219µη,C3 −.0121 −.0145µη,E3 −.003 −.0004µχ4 .1386 .1113µε4 .4393 .3125µη,C4 .0413 .0584µη,E4 .0137 .0117
Busch & Ludwig (MOVE-UAB & Goethe U): Higher-Order Income Risk 21 / 38
Estimation Results PSID
HH Pre HH Post
ρ .9873 .9815µχ2 .2150 .1926µε2 .1038 .0936µη,C2 .0132 .015µη,E2 .0076 .0067µχ3 −.1248 −.0966µε3 −.1489 −.1219µη,C3 −.0121 −.0145µη,E3 −.003 −.0004µχ4 .1386 .1113µε4 .4393 .3125µη,C4 .0413 .0584µη,E4 .0137 .0117
Busch & Ludwig (MOVE-UAB & Goethe U): Higher-Order Income Risk 21 / 38
Estimation Results PSID
HH Pre HH Post
ρ .9873 .9815µχ2 .2150 .1926µε2 .1038 .0936µη,C2 .0132 .015µη,E2 .0076 .0067µχ3 −.1248 −.0966µε3 −.1489 −.1219µη,C3 −.0121 −.0145µη,E3 −.003 −.0004µχ4 .1386 .1113µε4 .4393 .3125µη,C4 .0413 .0584µη,E4 .0137 .0117
Busch & Ludwig (MOVE-UAB & Goethe U): Higher-Order Income Risk 21 / 38
Implied Standardized Moments
HH Pre HH Post
αχ3 −1.25 −1.14
αε3 −4.45 −4.26
αη,C3 −7.96 −7.90
αη,E3 −4.58 −0.82
αχ4 3 3
αε4 40.76 35.69
αη4 237.31 260.03
αk = µk/µk/22 : standardized k th moment
I Ratio of Std. Dev. Contraction/Expansion = 1.5
Busch & Ludwig (MOVE-UAB & Goethe U): Higher-Order Income Risk 22 / 38
Implied Age Profiles: Post Government
I Age profiles, data and model, post government
Pre Government
Busch & Ludwig (MOVE-UAB & Goethe U): Higher-Order Income Risk 23 / 38
Outline
Intro
Higher-Order Risk in Two-Period Model
Estimation of Idiosyncratic Income Process
Macroeconomic Implications
Conclusion
Questions & Approach
I 3 Questions:
1. What’s the welfare cost of higher-order income risk?
2. Does it matter for welfare costs of business cycles?
3. Does it matter for consumption self-insurance?
I Tool: Life-cycle model with exogenous income
I No analysis of
I nature of realized income changes
I direct effect of aggregate risk
Busch & Ludwig (MOVE-UAB & Goethe U): Higher-Order Income Risk 24 / 38
Welfare Criterion
I Ex-ante welfare: Into which world would you prefer to be born
into as the average individual?
I Welfare measure: CEV gc in (a) to make households indifferent
to (b)
I gc < 0: welfare losses fromI higher order riskI aggregate risk
I Decomposition into mean and distribution effects
Busch & Ludwig (MOVE-UAB & Goethe U): Higher-Order Income Risk 25 / 38
Insurance Coefficients
I Model-version of Blundell, Pistaferri & Preston (2008) a la
Kaplan & Violante, 2010
I Share of variance of shock x ∈ {ε, η} which does not translate
into consumption growth
I By age h and aggregate state s ∈ {C,E}:
φxh(s) = 1−
cov(∆ ln ch(s), xh(s))
var(xh(s))
Busch & Ludwig (MOVE-UAB & Goethe U): Higher-Order Income Risk 26 / 38
Quantitative Model: Overview
I H-period lived households
I Working and retirement phase, retirement at age hr
I Endowments:
I exogenous
I discretization of estimated income process
I zero borrowing constraint
I Preferences: Epstein-Zin-Weil: RA θ and IES ρ
More details
Busch & Ludwig (MOVE-UAB & Goethe U): Higher-Order Income Risk 27 / 38
Quantitative Model: Shock Distributions
I Model shock distributions Fε, Fη: discrete
I Fit to estimated central moments: continuous Fε, Fη (s (t))
I Approach: Flexible Generalized Lambda Distribtion (FGLD):
1. Target estimated µ2 − µ4
2. Discretize FGLD distribution on grids Gε,Gη,
I Lowest income level unchanged over scenarios
Busch & Ludwig (MOVE-UAB & Goethe U): Higher-Order Income Risk 28 / 38
Quantitative Model: The FGLD
I Quantile function of FGLD (Freimer et al. 1988):
Q(p) = λ1 +1λ2
(pλ3 − 1λ3
− (1− p)λ4 − 1λ4
)
I Closed form expressions for moments µ1 − µ4
I Numerical solution for λ3, λ4 to match µ3, µ4
I Analytical solution for λ1, λ2 to match µ1, µ2
I Then discretization
More details
Busch & Ludwig (MOVE-UAB & Goethe U): Higher-Order Income Risk 29 / 38
Quantitative Model: Fit of Discretized Distribution
Moment µ2 µ3 µ4
Transitory Shock
estimate 0.094 -0.122 0.312fitted 0.094 -0.122 0.313discrete 0.091 -0.112 0.297
Persistent Shock
Contraction
estimate 0.015 -0.015 0.058fitted 0.015 -0.014 0.059discrete 0.015 -0.014 0.058
Expansion
estimate 0.007 0 0.012fitted 0.007 0 0.012discrete 0.007 0 0.012
I For US post government earnings
Busch & Ludwig (MOVE-UAB & Goethe U): Higher-Order Income Risk 30 / 38
Quantitative Model: Persistent Income ShocksIncome Profiles
(a) NORM (b) LK
(c) LKSW
Busch & Ludwig (MOVE-UAB & Goethe U): Higher-Order Income Risk 31 / 38
Quantitative Model: Other Parameters
Working period 25 (h = 0) to 60
Maximum age 80
IES ρ = 1
RA θ ∈ {1,2,3,4}Discount factor (2nd stage) β ∈ {0.971,0.969,0.967,0.964}Interest rate r = 0.042
Borrowing constraint a = 0
Pension contribution rate τ = 0.124
Aggregate shocks π(s′ = c | s = c) = 0.43
π(s′ = e | s = e) = 0.79
Initial ass. / perm. inc. 1.54
I estimate β to match assets / permanent income
Busch & Ludwig (MOVE-UAB & Goethe U): Higher-Order Income Risk 32 / 38
Welfare Costs of Higher-Order Risk, IES ρ = 1CEV gc gmean
c gdistrc g lcd
c gcsdc
Risk Aversion, θ = 1LK -0.99 1.141 -2.131 -2.107 -0.024LKSW 0.609 0.66 -0.052 -0.064 0.012Risk Aversion, θ = 2LK -1.754 1.267 -3.021 -3.006 -0.016LKSW -0.097 0.661 -0.757 -0.793 0.036Risk Aversion, θ = 3LK -4.867 1.62 -6.487 -6.456 -0.031LKSW -4.247 1.137 -5.383 -5.379 -0.005Risk Aversion, θ = 4LK -11.097 2.286 -13.383 -13.264 -0.119LKSW -11.98 2.019 -13.999 -13.857 -0.142
I Positive mean effects
I Substantial welfare losses from higher-order risk in distribution effects
I Losses increase in measure of risk aversion
Busch & Ludwig (MOVE-UAB & Goethe U): Higher-Order Income Risk 33 / 38
Welfare Costs of Aggregate Risk, IES ρ = 1CEV gc gmean
c gdistrc g lcd
c gcsdc %-change
Risk Aversion, θ = 1NORM -1.334 0.329 -1.663 -1.621 -0.042 0LK -1.7 0.389 -2.089 -2.044 -0.046 27.42LKSW -0.984 0.277 -1.261 -1.227 -0.034 -26.259Risk Aversion, θ = 2NORM -2.512 0.638 -3.149 -3.049 -0.101 0LK -2.794 0.601 -3.395 -3.305 -0.09 11.234LKSW -2.686 0.61 -3.296 -3.211 -0.085 6.933Risk Aversion, θ = 3NORM -3.553 0.903 -4.455 -4.298 -0.157 0LK -4.641 0.851 -5.493 -5.355 -0.138 30.645LKSW -5.961 1.109 -7.07 -6.899 -0.171 67.785Risk Aversion, θ = 4NORM -4.442 1.138 -5.58 -5.374 -0.205 0LK -7.252 1.13 -8.382 -8.188 -0.194 63.276LKSW -10.236 1.63 -11.866 -11.587 -0.28 130.443
Busch & Ludwig (MOVE-UAB & Goethe U): Higher-Order Income Risk 34 / 38
Life-Cycle Consumption1st and 2nd Moment (Log), ρ = 1, θ = 4
(a) Mean (b) Variance
I Higher-order income risk: lower consumption when young
I LK: higher variance, LKSW: lower variance
I Precautionary savings reaction
Busch & Ludwig (MOVE-UAB & Goethe U): Higher-Order Income Risk 35 / 38
Insurance CoefficientsTransitory & Persistent Shock, ρ = 1, θ = 4
(a) Transitory Shock (b) Persistent Shock
I LK: less insurance against transitory shocks
I Total effect: more insurance b/c of higher precautionary savings
3rd and 4th
Busch & Ludwig (MOVE-UAB & Goethe U): Higher-Order Income Risk 36 / 38
Outline
Intro
Higher-Order Risk in Two-Period Model
Estimation of Idiosyncratic Income Process
Macroeconomic Implications
Conclusion
SummaryEmpirical Analysis
I GMM estimator of 2nd–4th moments of transitory and (cyclical)
persistent component of income process
I German (SOEP) and US (PSID) households:
I Variance of persistent shocks countercyclical
I Skewness of persistent shocks procyclical
I Highly leptokurtic shocks
I Role of government taxes and transfers?
I Similar transitory shocks
I Dampen cyclicality of persistent shocks
Busch & Ludwig (MOVE-UAB & Goethe U): Higher-Order Income Risk 37 / 38
SummaryMacroeconomic Implications
I Household life-cycle model
I (Cyclical) higher-order risk welfare implications:
I θ = 1: welfare gain of approx. .6%
I θ = 4: welfare losses of approx. 12.0%
I Quantitatively relevant for welfare costs of business cycles:
I For θ = 1, welfare cost of fluctuations lower by 26%
I For θ = 4, welfare cost of fluctuations larger by 130%
I Better consumption insurance against shocksBusch & Ludwig (MOVE-UAB & Goethe U): Higher-Order Income Risk 38 / 38
Fourth Central Moment of Income Processµ4 (yith ; θ) = E
[(yith − E [yith ])4
]= µ
χ4 + µ
ε4 +
∑h−1j=0
ρ4jµη4 (s (t − j)) + 6
(µχ2 µ
ε2 + . . .(
µχ2 + µ
ε2
)∑h−1j=0
ρ2jµη2 (s (t − j)) + . . .
∑h−2j=0
∑h−1k=j+1
ρ2jρ
2kµη2 (s (t − j))µ
η2 (s (t − k))
)
µ31(yith, yit+1h+1) = E[
(yith − E [yith ])3 (yit+1h+1 − E[yit+1h+1
])|St]
= E[µχ4 + ρ
h−1∑j=0
ρ4jµη4 (s(t − j)) + 3
(µχ2 µ
ε2 + µ
χ2
h−1∑j=0
ρ2jµη2 (s(t − j)) + . . .
(µχ2 + µ
ε2
)ρ
h−1∑j=0
ρ2jµη2 (s(t − j)) + . . .
ρ
h−1∑j=0
ρ2jµη2 (s(t − j))
h−1∑k 6=j
ρ2kµη2 (s(t − k))
︸ ︷︷ ︸∑j−1k=0 ρ
2kµη2 (s(t−k))+∑h−1
k=j+1 ρ2kµη2 (s(t−k))
)]
Back
Busch & Ludwig (MOVE-UAB & Goethe U): Higher-Order Income Risk 39 / 38
Implied Age Profiles: Pre Government
I Age profiles, data and model, pre government
Go Back
Busch & Ludwig (MOVE-UAB & Goethe U): Higher-Order Income Risk 40 / 38
Note on Logs vs. Levels: Numerical Example
I Numerical example with 3-point distribution Details
Moments of Innovation in Logs, ε
µε1 µε2 µε3 µε4
NORM –.25 0.5 0 0.75
LK –.57 0.5 0 7.5
LKSW –.11 0.5 -1.7678 7.5
Moments of Innovation in Levels, exp(ε)
µexp(ε)1 µ
exp(ε)2 µ
exp(ε)3 µ
exp(ε)4
NORM 1 0.5868 0.6691 1.3045
LK 1 11.6316 299.3406 7842.5727
LKSW 1 0.1039 0.0190 0.0523
I Welfare gain of LKSW with log-utility (hand-to mouth)Back
Busch & Ludwig (MOVE-UAB & Goethe U): Higher-Order Income Risk 41 / 38
Two-Period Model: Illustration
I Effects of higher-order risk
I Take 3-point distribution for yt Details
I Three distributions: “normal” (NORM), leptokurtic (LK),
leptokurtic and skewed, (LKSW)
I Two normalizations: E [ε] = 0 (LN), E [exp ε] = 1 (EN)
I Two institutional setups: No capital markets (NCM), natural debt
limit (NDL)
I Two degrees of “risk aversion”: low (θ = 1), high (θ = 4)
Busch & Ludwig (MOVE-UAB & Goethe U): Higher-Order Income Risk 42 / 38
Simple Model: Numerical Example
CEV relative to NORM:
(a) LN (b) EN
Scen / CEV LK LKSW LK LKSW
θ = 1
NCM 0 0 -0.1859 0.1096
θ = 4
NCM -0.7888 -0.8554 -0.8281 -0.8396
CM -0.6978 -0.8004 -0.7540 -0.7785
I (N)CM: (No) capital markets
I Important to distinguish mean and distribution effects
I Distribution effects dominate with increasing risk aversion
Back
Busch & Ludwig (MOVE-UAB & Goethe U): Higher-Order Income Risk 43 / 38
Two Period Model: DiscretizationLet x ∈ {xl , x0, xh}, drawn with respectiveprobabilities {(1− p) · q,p, (1− p) · (1− q)}. Then,
1. if and only if α(3) = 0 or2. if α(3) 6= 0 and α(4) ≥ 1 + α(3)2
we match E [exp(x)] = 1 and µ(2), α(3), α(4) by determining
q =12
+12
√1−
4 α(4)
α(3)2−4
4 α(4)
α(3)2−3if α(3) > 0
−12
√1−
4 α(4)
α(3)2−4
4 α(4)
α(3)2−3if α(3) < 0
0.5 if α(3) = 0
p =
1− (2q−1)2
q(1−q)α(3)2 if α(3) 6= 0
1− 1α(4) if α(3) = 0
Busch & Ludwig (MOVE-UAB & Goethe U): Higher-Order Income Risk 44 / 38
Two Period Model: Discretization (cont’d)
∆x =
√µ(2)α(3)
2q−1 if α(3) 6= 0
2õ(2)
√α(4) if α(3) = 0,
and
xl = − ln [p exp ((1− q)∆x ) + (1− p) (q + (1− q) exp(∆x ))]
x0 = xl + (1− q)∆x
xh = xl + ∆x .
Back to Illustration Back to Numerical Example
Busch & Ludwig (MOVE-UAB & Goethe U): Higher-Order Income Risk 45 / 38
Quantitative ModelEndowments
I 2 aggregate states of economy s ∈ c,eI Transition: time-invariant Markov
I Households:I live for H periodsI 1 unit of labor at h ∈ 0, ...,hr − 1; retired hr , . . . ,HI Household i ’s (post-government) log income:
yith = eh + zith + εith
zith = ρzit−1h−1 + ηith
I eh : fh and fEP from first stageI zit+1h+1 = ρzith + ηit+1h+1
I εith ∼ Fε, ηih ∼ Fη (s (t))
Busch & Ludwig (MOVE-UAB & Goethe U): Higher-Order Income Risk 46 / 38
Quantitative ModelEndowments (cont’d)
I Contributions to pension system: τp
I Retirement income: risk-free, US bend-point formula
I Savings: risk-free 1-period bonds, rate of return r
I Dynamic budget constraint with borrowing constraint:
ait+1h+1 = aith (1 + r) + (1− τp)yith − cith ≥ −a
I Only idiosyncratic effects of aggregate risk→ lower bound of
welfare losses
Busch & Ludwig (MOVE-UAB & Goethe U): Higher-Order Income Risk 47 / 38
Quantitative ModelPreferences
I Epstein-Zin-Weil recursive utility: separate IES ρ from RA θ
Uh (x , z; s) =
((1− β) c
1− 1ρ
h + β(
ce (Uh+1 (·))1− 1ρ
))1/(
1− 1ρ
)if ρ 6= 1
exp {(1− β) ln ch + β ln (ce (Uh+1 (·)))} otherwise
with
ce (Uh+1 (·)) =
(
Eh
[Uh+1 (x ′, z′; s′)
1−θ])1/(1−θ)
if ρ 6= 1
exp (Eh [lnUh+1 (x ′, z′; s′)]) otherwise
I x = a(1 + r) + y : cash on hand
Go Back
Busch & Ludwig (MOVE-UAB & Goethe U): Higher-Order Income Risk 48 / 38
Details on FGLD
I 1st moment:
µ1 = E [X ] = a + bE [Y ] = a + bν1
= λ1 −1
λ2λ3+
1λ2λ4
+1λ2ν1
I Moments of Y by binomial expansion:
E[(Y − E [Y ])k
]= E
k∑j=0
(kj
)(−1)j (Y )k−j ν(1)j
=
k∑j=0
(kj
)(−1)j E
[(Y )k−j
]ν(1)j
Busch & Ludwig (MOVE-UAB & Goethe U): Higher-Order Income Risk 49 / 38
Details on FGLD (cont’d)
I Higher moments of X :
µ2 =1λ2
2(ν2 − ν2
1)
µ3 =1λ3
2
(ν3 − 3ν1ν2 + 2ν3
1
]µ4 =
1λ4
2
(ν4 − 4ν1ν3 + 6ν2
1ν2 − 3ν41
).
I Jointly estimate λ3, λ4 with more weight on skewness:
min{λi∈(− 1
4 ,λi ]}4i=3
4∑i=3
(pi
(µi(~λ)− µi
)2)
Back
Busch & Ludwig (MOVE-UAB & Goethe U): Higher-Order Income Risk 50 / 38
Life-Cycle Earnings1st Moment and 2nd Moment of Log
Back
Busch & Ludwig (MOVE-UAB & Goethe U): Higher-Order Income Risk 51 / 38
Life-Cycle Earnings3rd and 4th Moment (of Logs)
Back
Busch & Ludwig (MOVE-UAB & Goethe U): Higher-Order Income Risk 52 / 38
Life-Cycle Consumption3rd and 4th Moment (of Logs), θ = 4
Go Back
Busch & Ludwig (MOVE-UAB & Goethe U): Higher-Order Income Risk 53 / 38