Motivation Facts Model Estimation MPC Monetary Policy Conclusion MPC Heterogeneity in Europe: Sources and Policy Implications Miguel Ampudia 1 Russell Cooper 2 Julia Le Blanc 3 Guozhung Zhu 4 1 European Central Bank 2 European University Institute and Penn State University 3 Deutsche Bundesbank 4 University of Alberta CEPR-BdF Conference Dec. 6, 2019
55
Embed
MPC Heterogeneity in Europe: Sources and Policy Implications€¦ · MotivationFactsModelEstimation MPC Monetary Policy Conclusion MPC Heterogeneity in Europe: Sources and Policy
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
Motivation Facts Model Estimation MPC Monetary Policy Conclusion
MPC Heterogeneity in Europe:Sources and Policy Implications
Miguel Ampudia1 Russell Cooper2 Julia Le Blanc3
Guozhung Zhu4
1European Central Bank
2European University Institute and Penn State University
3Deutsche Bundesbank
4University of Alberta
CEPR-BdF ConferenceDec. 6, 2019
Motivation Facts Model Estimation MPC Monetary Policy Conclusion
Motivation
Big Theme: Heterogeneity and Non-convexities matter foreconomic policy
Little Theme: What are the implications of heterogeneity in MPCand non-convexities in HH Finance for monetary policy?
no longer business as usual with a simple Euler equation for therepresentative HH linking federal funds rate to consumptionasset market participation and adjustment margins areempirically relevantheterogeneity matters: positive and normative dimensionschallenges: estimation and intervention
Motivation Facts Model Estimation MPC Monetary Policy Conclusion
Motivation
Big Theme: Heterogeneity and Non-convexities matter foreconomic policy
Little Theme: What are the implications of heterogeneity in MPCand non-convexities in HH Finance for monetary policy?
no longer business as usual with a simple Euler equation for therepresentative HH linking federal funds rate to consumptionasset market participation and adjustment margins areempirically relevantheterogeneity matters: positive and normative dimensionschallenges: estimation and intervention
Motivation Facts Model Estimation MPC Monetary Policy Conclusion
Intertemporal Impact of Monetary Policy
dCt+τ
dMPt=
∫s
dc(Y ,Rs ,Rb,Ω)
dYt+τ (Ω)
dYt+τ (Ω)
dMPtdGt+τ (Ω) +∫
s
dc(Y ,Rs ,Rb,Ω)
dRst+τ (Ω)
dRst+τ (Ω)
dMPtdGt+τ (Ω) (1)
Motivation Facts Model Estimation MPC Monetary Policy Conclusion
dC t+τ
dMPt=
∫s
dc(Y ,Rs ,Rb,Ω)
dYt+τ (Ω)
dYt+τ (Ω)
dMPtdGt+τ (Ω) +∫
s
dc(Y ,Rs ,Rb,Ω)
dRst+τ (Ω)
dRst+τ (Ω)
dMPtdGt+τ (Ω) (2)
on consumption
Motivation Facts Model Estimation MPC Monetary Policy Conclusion
dC t+τ
dMPt=
∫s
dc(Y ,Rs ,Rb,Ω)
dYt+τ (Ω)
dYt+τ (Ω)
dMPtdGt+τ (Ω) +∫
s
dc(Y ,Rs ,Rb,Ω)
dRst+τ (Ω)
dRst+τ (Ω)
dMPtdGt+τ (Ω) (3)
on consumption
through income and interest rate channels based on empiricalIRFs
Motivation Facts Model Estimation MPC Monetary Policy Conclusion
dC t+τ
dMPt=
∫s
dc(Y ,Rs ,Rb,Ω)
dYt+τ (Ω)
dYt+τ (Ω)
dMPtdGt+τ (Ω) +∫
s
dc(Y ,Rs ,Rb,Ω)
dRst+τ (Ω)
dRst+τ (Ω)
dMPtdGt+τ (Ω) (4)
on consumption
through income and interest rate channels based on empiricalIRFs
with heterogeneous households and thus heterogenous MPCsobtained from an estimated HH choice model
Motivation Facts Model Estimation MPC Monetary Policy Conclusion
dC t+τ
dMPt=
∫s
dc(Y ,Rs ,Rb,Ω)
dYt+τ (Ω)
dYt+τ (Ω)
dMPtdGt+τ (Ω) +∫
s
dc(Y ,Rs ,Rb,Ω)
dRst+τ (Ω)
dRst+τ (Ω)
dMPtdGt+τ (Ω) (5)
on consumption
through income and interest rate channels based on empiricalIRFs
with heterogeneous households and thus heterogeneous MPCsobtained from an estimated HH choice model
allowing for the evolution of the cross sectional heterogeneity
Motivation Facts Model Estimation MPC Monetary Policy Conclusion
Specifically: What we do
Build a life-cycle model with portfolio choice, participationcosts, credit constraints and bequest motives.
Take into account rich heterogeneity in income, education,wealth accumulation and portfolio allocation.
Estimate the model using data from the HFCS for France,Germany, Italy and Spain.
Characterize the distribution of MPC across households.
Evaluate effects of monetary policy through income and assetprices from data not a model
Emphasize heterogenous responses on consumption bycountry and household type
Motivation Facts Model Estimation MPC Monetary Policy Conclusion
What we find
Participation and portfolio adjustment costs are present andnecessary to explain the low ownership of risky assets.
Compared to conventional estimates, the discount factor isestimated to be lower, the risk aversion parameter is higher.
The distributions of MPCs are country-specific. Withincountries, the MPC is higher for low income, low educationhouseholds.
Overall, Germany has the largest response to monetaryinnovations through the return channel, Spain has the largestresponse through the income channel of monetary policy.
France and Germany respond more to monetary contractions,Italy and Spain respond more to expansions.
Motivation Facts Model Estimation MPC Monetary Policy Conclusion
Stages
Motivating Facts
Estimation of HH Finance Model
Calculation of MPC distribution from Policy Functions
Use Estimated IRFs from Monetary Policy as Inputs
Simulate Consumption Response
Motivation Facts Model Estimation MPC Monetary Policy Conclusion
Related Literature
Estimating LC Models with Portfolio Choice (Cooper & Zhu(2015), Fagereng et al. (2017), Calvet et al. (2016))
Characterizing marginal propensities to consume acrossheterogeneous households (Carroll et al. (2014), Kaplan et al.(2014))
Distributional Effects of Monetary Policy (Auclert (2017),Kaplan & Violante (2014), Ampudia et al. (2018), Casiraghiet al. (2018))
Motivation Facts Model Estimation MPC Monetary Policy Conclusion
Our Contribution
Estimated Model of HH Choice
Input estimated IRFs into policy analysis
Comparison within and between countries
Missing: Housing!
Motivation Facts Model Estimation MPC Monetary Policy Conclusion
Some data facts
Education is a key determinant of households’ financial behaviour.Between and within country heterogeneity, particularly in participationrates and wealth.
Table: Household Facts by Education across Countries
This table displays the participation rate in stocks (defined in three different ways, row 1: direct, row 3: stocks plusmutual funds invested mainly in stocks and row 5: stocks, mutual funds invested mainly in stocks plus private pensionplans), the share of stocks over total liquid assets (for participants), the median wealth income ratio, with and withouthousing (h) for households in each country by educational attainment, low (no college) and high (college). The momentscome from the HFCS.
Motivation Facts Model Estimation MPC Monetary Policy Conclusion
The model - Main features
Households maximize expected lifetime utilityHeterogeneity in household types: participants andnon-participants, adjustors, non-adjustorsChoice variables: consumption (C ), bond holdings (B), stockholdings (S), asset market participation and stock adjustment.
Idiosyncratic shocks to income and risky financial assetsExogenous income process: deterministic (growth) andstochastic components (persistent and transitory shock).Risky asset return stochastic (Rs), bond return fixed (Rb).
Consumption floor c coming from government transfer.
Ingredients produce precautionary savings and a distributionof MPCs.
Motivation Facts Model Estimation MPC Monetary Policy Conclusion
The model - Income processes
Deterministic income profileEstimated from ECHP, 1994-2001. Labor income net of taxesand transferslog(Yi,t) = const + polynomial(age) + HHComp + TimeEff
Persistent and transitory income shocks
yi ,t = zi ,t + εi ,t
zi ,t = ρzi ,t−1 + ηi ,t
Linear fit for retirement period
Motivation Facts Model Estimation MPC Monetary Policy Conclusion
The model - Income profiles
Source: European Community Household Panel 1994–2001
Motivation Facts Model Estimation MPC Monetary Policy Conclusion
The model - Income processes
Stochastic Processes by education and country
Germany France
ρ σ2η σ2
ε ρ σ2η σ2
ε
No college 0.895*** 0.022*** 0.016*** 0.971*** 0.031*** 0.006*(0.005) (0.001) (0.001) (0.014) (0.006) (0.003)
Motivation Facts Model Estimation MPC Monetary Policy Conclusion
The model: Participants
vt(Ω) = maxvat (Ω), vn
t (Ω), v xt (Ω) Ω = (y ,Ab,As) (6)
Adjust:
vat (Ω) = max
Ab′≥Ab,As′≥0
(1− β)c1−1/θ + β
[(1− νt+1)
(Etvt+1(Ω′)1−γ) 1
1−γ + νt+1
(EtB(A′)1−γ) 1
1−γ]1−1/θ
11−1/θ
(7)
s.t.
c = y + TR +∑
i=b,s R iAi −∑
i=b,s Ai ′ − F (8)
A′ = RbAb′ + Rs′As′ (9)
TR = max0, c − (y +∑
i=b,s R iAi ). (10)
No Adjust:
vnt (Ω) = max
Ab′≥Ab
(1− β)c1−1/θ + β
[(1− νt+1)
(Etvt+1(Ω′)1−γ) 1
1−γ + νt+1
(EtB(A′)1−γ) 1
1−γ]1−1/θ
11−1/θ
s.t.
c = y + TR + RbAb − Ab′
As′ = RsAs
A′ = RbAb′ + Rs′As′
TR = max0, c − (y +∑
i=b,s R iAi )
Motivation Facts Model Estimation MPC Monetary Policy Conclusion
The model: Non-Participants
wt(Ω) = maxwnt (Ω),wp
t (Ω) (11)
for all Ω.No Entry
wnt (Ω) = max
Ab′≥Ab
(1− β)c1−1/θ + β
[(1− νt+1)
(Etwt+1(Ω′)1−γ) 1
1−γ + νt+1
(EtB(A′)1−γ) 1
1−γ
]1−1/θ 1
1−1/θ
(12)
c = y + TR + RbAb − Ab′
A′ = RbAb′
TR = max0, c − (y + RbAb)
Entry
wpt (Ω) = max
Ab′≥Ab,As′≥0
(1− β)c1−1/θ + β
[(1− νt+1)
(Etvt+1(Ω′)1−γ) 1
1−γ + νt+1
(EtB(A′)1−γ) 1
1−γ]1−1/θ
11−1/θ
(13)
c = y + TR + RbAb − Ab′ − As′ − Γ
A′ = RbAb′ + Rs′As′
TR = max0, c − (y + RbAb).
Motivation Facts Model Estimation MPC Monetary Policy Conclusion
The model - Solution and estimation
Finite dynamic optimization problem solved by backwardinduction
Discretized shocks,initial distribution of assetsValue function iteration
Simulated method of moments estimation
Λ = minΘ(Ms(Θ)−Md)W (Ms(Θ)−Md)′. (14)
Regressions of participation rate, stock share, (liquid)wealth-to-income ratioMoments are age and education coefficients (with home equitycontrols)W is a the inverse of the variance covariance matrix ofmomentsΘ = (β0, β1, γ, Γ,F , L, φ, c , θ,A
b)
Motivation Facts Model Estimation MPC Monetary Policy Conclusion
This table reports parameter estimates and the corresponding standard errors. The last column is model fit from (14) .
Discount factors β0, β1 lower than conventional value (0.95). HH withlow education have even lower β than highly educated HH
High risk aversion coefficients γ (US around 4)
High stock participation costs (highest in Spain, lowest in Germany)estimates are in terms of mean income
Importance of bequests stronger in some countries
Literature: β, γ estimates comparable to Fagereng et al. (2017) forNorway.
Motivation Facts Model Estimation MPC Monetary Policy Conclusion
Example: Germany
20 30 40 50 60 70
age
0.2
0.4
0.6
0.8Participation
20 30 40 50 60 70
age
0.2
0.4
0.6
0.8Stock Share
20 30 40 50 60 70
age
0.5
1
1.5
2Wealth/Income
This figure shows the average profiles of participation, stock share and thewealth to income ratio for Germany. The high education group is indicatedby the broken curves and the low education group by the solid curve.
Motivation Facts Model Estimation MPC Monetary Policy Conclusion
Other Properties of Solution
8-10% of low education HHs hit consumption floor in Italyand Spain.
borrowing constraints rarely bind
local identification through derivative of moments wrtparameters
Few Hand to Mouth Households are present due to portfolioadjustment costs
Motivation Facts Model Estimation MPC Monetary Policy Conclusion
MPCs Distributions by country
given policy functions, simulate income and return shocks.
calculate MPC distributions from responses
heterogeneity across households due to non-linearities
participationadjustmentborrowing constraint
Moderate cross-country heterogeneity
Motivation Facts Model Estimation MPC Monetary Policy Conclusion
MPCs from income shocks by country
Table: MPC Distribution: Income Shock
1% 10%Country All Households Participants All Households Participants
EdInc
low middle high low middle high low middle high low middle high
This table summarizes the distribution of MPC from transitory income shocks. The three columns (low, middle and high) representthree levels of permanent income. The rows, by country, are for low and high educational attainment for all households as well as thoseparticipating in asset markets. The left block is for a 1% shock and the right is for a 10% transitory income shock.
Literature:
MPCs highest for low permanent income group, in particular for IT, ES.High education, high income participants have high MPCs due to adjustment costs.Carroll, Slacalek and Tokyoka: Germany =0.26, Spain =0.38 from incomeOther studies using regression analysis: could study in our simulated data too
Motivation Facts Model Estimation MPC Monetary Policy Conclusion
Conditional on participationMPC falls with permanent income level.MPC lower for higher educated HH.
Motivation Facts Model Estimation MPC Monetary Policy Conclusion
Hand-to-Mouth Households
HANK (2017) classification: liquid assets less than halfincome flow
Data
poor have negative illiquid assetsrich have positive illiquid assets
Simulated Data from Estimated Model
both types exist in simulated datalow income HtM consumers generally have higher MPCs
Motivation Facts Model Estimation MPC Monetary Policy Conclusion
0.79
0.13
0.08
0.76
0.20
0.04
0.83
0.14
0.03
0.77
0.21
0.030
.2.4
.6.8
1
DE ES FR IT
Share of HtM households by country
Non HtM Poor HtM Wealthy HtM
Figure: HtM Households
This figure shows the fraction of HtM households in our sample by country. The vertical axis measures the averageyears of schooling within each of the groups.
liquid assets less than half income flowpoor have negative illiquid assetsrich have positive illiquid assets
Motivation Facts Model Estimation MPC Monetary Policy Conclusion
MPCs for HtM Consumers
Table: Hand-to-Mouth Consumers: Income Shock of 1%
This table reports the mean MPC of stock market participants who are hand-to-mouth consumersin response to a return shock that is 1% of the stock value.
Motivation Facts Model Estimation MPC Monetary Policy Conclusion
MPCs for HtM Consumers
Table: Hand-to-Mouth Consumers: Return Shock of 1%
This table reports the mean MPC of stock market participants who are hand-to-mouth consumersin response to a return shock that is 1% of the stock value.
Motivation Facts Model Estimation MPC Monetary Policy Conclusion
Monetary Policy Implications
Impact of monetary policy shocks on consumption throughincome and asset returns.
Combine estimated elasticities of income and asset returns tomonetary policy (100 bp cut) and MPC estimates
Individual response is state dependent and dynamic
cross sectional distribution, Gt(Ω), evolves
Effects on bond returns and fiscal transfers not present.
dCt+τ
dMPt=
Income channel︷ ︸︸ ︷∫s
dc(Y ,Rs ,Rb,Ω)
dYt+τ (Ω)
dYt+τ (Ω)
dMPtdGt+τ (Ω)+∫
s
dc(Y ,Rs ,Rb,Ω)
dRst+τ (Ω)
dRst+τ (Ω)
dMPtdGt+τ (Ω)︸ ︷︷ ︸
Return channel
Motivation Facts Model Estimation MPC Monetary Policy Conclusion
Figure: Aggregate Consumption Response by Country
Motivation Facts Model Estimation MPC Monetary Policy Conclusion
Nonlinear Aggregate Consumption Response
0 5 10 15
-0.02
-0.01
0
0.01
Germany
benchmarkdoubled shock
0 5 10 15-0.01
0
0.01
0.02
Spain
0 5 10 15
year
-0.02
-0.01
0
0.01France
0 5 10 15
year
-0.01
0
0.01
0.02
Italy
Figure: Nonlinear Aggregate Consumption Response by Country
Motivation Facts Model Estimation MPC Monetary Policy Conclusion
Monetary Policy Effect on Income
Table: Monetary Policy Effect on Income by Quintile, year and Country
This table summarizes the consumption response in percentage from a 100 basis point monetary expansionby education and permanent income.
U-shape of consumption response to monetary policy shock.Poor HH have high MPCs and their income is more sensitive to monetary innovations.Rich HH participate in asset markets, their financial income is sensitive to monetary policy.Total effect reflects non-linearities, portfolio adjustments...
Motivation Facts Model Estimation MPC Monetary Policy Conclusion
Distributional Effects of Monetary Policy
Table: Distribution of consumption response to a monetary policy shock
This table summarizes the consumption response in percentage from a 100 basis point monetary expansionby education and permanent income.
U-shape of consumption response to monetary policy shock.Poor HH have high MPCs and their income is more sensitive to monetary innovations.Rich HH participate in asset markets, their financial income is sensitive to monetary policy.Total effect reflects non-linearities, portfolio adjustments...
Motivation Facts Model Estimation MPC Monetary Policy Conclusion
Conclusion
Life-cycle model with portfolio choice, participation costs,credit constraints and bequest motives implies significantdifferences in estimates of deep parameters within and acrosscountries.
Characterize the distribution of MPC across households andcountries. Within countries, the MPC is higher for lowincome, low education households.
Monetary policy effects on consumption through income andasset prices are household specific and nonlinear.
Overall, Germany has the largest response to monetary policyshocks through the return channel, while Spain has the largestresponse through income.
Motivation Facts Model Estimation MPC Monetary Policy Conclusion
Appendix
Motivation Facts Model Estimation MPC Monetary Policy Conclusion
The model: Participants
vt(Ω) = maxvat (Ω), vn
t (Ω), v xt (Ω) Ω = (y ,Ab,As)
Adjust:
vat (Ω) = max
Ab′≥Ab,As′≥0
(1− β)c1−1/θ + β
[(1− νt+1)
(Etvt+1(Ω′)1−γ) 1
1−γ + νt+1
(EtB(A′)1−γ) 1
1−γ]1−1/θ 1
1−1/θ
s.t.
c = y + TR +∑i=b,s
R iAi −∑i=b,s
Ai ′ − F
A′ = RbAb′ + Rs′As′
TR = max0, c − (y +∑i=b,s
R iAi ).
No Adjust:
vnt (Ω) = max
Ab′≥Ab
(1− β)c1−1/θ + β
[(1− νt+1)
(Etvt+1(Ω′)1−γ) 1
1−γ + νt+1
(EtB(A′)1−γ) 1
1−γ]1−1/θ 1
1−1/θ
s.t.
c = y + TR + RbAb − Ab′
As′ = RsAs
A′ = RbAb′ + Rs′As′
TR = max0, c − (y +∑i=b,s
R iAi )
Motivation Facts Model Estimation MPC Monetary Policy Conclusion
The model: Non-Participants
wt(Ω) = maxwnt (Ω),wp
t (Ω)
for all Ω.No Entry:
wnt (Ω) = max
Ab′≥Ab
(1− β)c1−1/θ + β
[(1− νt+1)
(Etwt+1(Ω′)1−γ) 1
1−γ + νt+1
(EtB(A′)1−γ) 1
1−γ]1−1/θ 1
1−1/θ
c = y + TR + RbAb − Ab′
A′ = RbAb′
TR = max0, c − (y + RbAb)
Entry:
wpt (Ω) = max
Ab′≥Ab,As′≥0
(1− β)c1−1/θ + β
[(1− νt+1)
(Etvt+1(Ω′)1−γ) 1
1−γ + νt+1
(EtB(A′)1−γ) 1
1−γ]1−1/θ 1
1−1/θ
c = y + TR + RbAb − Ab′ − As′ − Γ
A′ = RbAb′ + Rs′As′
TR = max0, c − (y + RbAb).
Motivation Facts Model Estimation MPC Monetary Policy Conclusion
This table reports data and model moments. For the wealth-income ratio regression, the regressors include a constant,age, age-squared, college*age, college*age-squared. For all regressions, controls included home equity and homeownershipstatus.
Motivation Facts Model Estimation MPC Monetary Policy Conclusion
Table: MPC Regressions: Income Shock
wealth percentileconst. age age2 income edu 0-50% 50-70% 70-90% 90-95% 95-100%
This table presents regression results of MPCs in response to positive transitory income shocks of 1% and 10%,respectively. The dependent variable is the MPC. The explanatory variables are a constant, age, age-squared, income,education and wealth percentiles.
Motivation Facts Model Estimation MPC Monetary Policy Conclusion
Table: MPC Regressions: Return Shocks
wealth percentileconst. age age2 income edu 5-50% 50-70% 70-90% 90-95% 95-100%
This table presents regression results. The dependent variable is the MPC from a return shock. The explanatoryvariables are a constant, age, age-squared, income, the wealth to income ratio and education.
Motivation Facts Model Estimation MPC Monetary Policy Conclusion
Monetary Policy Effect on Income
Table: Monetary Policy Effect on Income by Quintile, year and Country