BUFFER‐STOCK SAVING AND HOUSEHOLDS' RESPONSE TO INCOME SHOCKS

• Author: Winfried Koeniger,Serafin Frache,Giulio Fella
• Date: 01 August 2020
• Published date: 01 August 2020
• DOI: http://doi.org/10.1111/iere.12459

INTERNATIONAL ECONOMIC REVIEW

Vol. 61, No. 3, August 2020 DOI: 10.1111/iere.12459

BUFFER-STOCK SAVING AND HOUSEHOLDS’ RESPONSE TO INCOME SHOCKS∗

BYGIULIO FELLA,SERAFIN FRACHE,AND WINFRIED KOENIGER1

Queen Mary University of London, London, UK,CFM and IFS; Universidad de Montevideo,

Montevideo, Uruguay; University of St.Gallen, SEW-HSG, St.Gallen, Switzerland,CESifo, CFS

and IZA

We exploit information on the joint dynamics of household labor income, consumption, and wealth in the

Italian Survey of Household Income and Wealth to structurally estimate a buffer-stock saving model. We

compare the degree of consumption smoothing implied by the model to the corresponding empirical estimates

based on the same data set. We estimate that Italian households smooth 12% of permanent income shocks in

the data that is comparable to the model counterpart of 11% . This result contrasts with existing evidence, and

our own findings in this article, of substantially more consumption smoothing in U.S. data.

1. INTRODUCTION

The degree to which self-insurance allows households to decouple consumption from income

shocks determines the scope for tax and social insurance policies and the associated welfare

gains. This article investigates the extent of self-insurance in Italy through the lens of a struc-

turally estimated, buffer-stock saving model.

The article’s contribution is twofold. First, we estimate consumption-insurance coefficients

for permanent and transitory idiosyncratic (labor)2income shocks in the Italian Survey of

Household Income and Wealth (SHIW) for the sample period 1987–2012. These coefficients

measure the degree of consumption smoothing and are defined as the fraction of the shocks

that is not reflected in movements in consumption. The coefficients are identified in the data by

applying the methodology proposed by Blundell et al. (2008) (BPP hereafter). We estimate an

insurance coefficient of 0.12 for permanent shocks and of 0.8 for transitory shocks based on the

SHIW. For comparison, we conduct the same analysis using the U.S. Panel Study for Income

Dynamics (PSID) for the sample period 1999–2015, over which the PSID contains information

on nondurable consumption, income, and wealth as the SHIW. We estimate an insurance

coefficient of 0.29 for permanent shocks and of 0.91 for transitory shocks. These estimates are

in line with the corresponding empirical estimates—respectively, 0.36 and 0.95—by BPP for the

period 1978–1992, using consumption data from the consumer expenditure survey (CEX) to

impute total nondurable consumption in the PSID.3

Second, in the spirit of Kaplan and Violante (2010), we compare the degree of insurance im-

plied by our empirical estimates to that implied by an incomplete-markets model. In particular,

we use indirect inference to estimate a buffer-stock saving model on the same (SHIW or PSID)

∗Manuscript received April 2019; revised February 2020.

1We thank the editor, Hal Cole, two anonymous referees, Dirk Krueger, and participants at various seminars and

conferences for helpful comments. This is a substantially revised version of our work previously circulated as “Financial

Market Imperfections and Households’ Wealth Response to Labor Income Shocks.” Part of this research has been

conducted while Serafin Frache and Winfried Koeniger were at Queen Mary University of London. Please address

correspondence to: Winfried Koeniger, Department of Economics, University of St. Gallen, Varnb¨

uelstrasse 14 -

HSG-SEW, CH 9000, St. Gallen, Switzerland. Phone: +41 71 224 23 09. E-mail: winfried.koeniger@unisg.ch.

2Where it does not engender confusion, we write just “income” instead of “labor income” in the rest of thearticle.

3The PSID contains information only on food consumption instead of total nondurable consumption during that

period.

1359

C

(2020) by the Economics Department of the University of Pennsylvania and the Osaka University Institute of Social

and Economic Research Association

1360 FELLA,FRACHE,AND KOENIGER

data set used for our empirical estimates. Instead of basing estimation on a set of unconditional

moments, indirect inference targets the parameters of an auxiliary statistical model that pro-

vides a reduced form for the structural model. We choose as auxiliary model the reduced-form

regressions of consumption and wealth changes over income changes at different time hori-

zons. As argued by Krueger and Perri (2011), the long-run wealth response to income shocks is

potentially informative about the degree of partial insurance against permanent income shocks.

We use the estimated structural model to simulate a panel of individual histories for consump-

tion and permanent and transitory income shocks and estimate insurance coefficients applying

the BPP methodology on the simulated data. We find that, for Italy, the amount of insurance

implied by the model is in line with the empirical estimates. Through the lens of the model,

this suggests that households in Italy do not have access to more consumption insurance than

implied by self-insurance through a single, noncontingent asset in the buffer-stock model. This

contrasts with our findings for the United States that, in line with Kaplan and Violante (2010),

confirm that the empirical estimates imply a substantially higher degree of insurance than the

self-insurance predicted by the model.

A large literature has tried to estimate the amount of insurance available to households by

analyzing the response of consumption to income shocks. Two polar benchmark models have

provided the theoretical framework for this effort. On the one hand, the complete-markets

model assumes that agents can perfectly insure ex ante against all idiosyncratic contingencies.

Tests of this hypothesis (Cochrane, 1991; Mace, 1991) do not need to distinguish between

permanent and transitory income shocks since consumption should not respond to any kind

of idiosyncratic income shock. The predictions of the complete-markets model are typically

strongly rejected by the data (Attanasio and Davis, 1996).

On the other hand, the textbook permanent income hypothesis (PIH) assumes that noncon-

tingent borrowing and lending is the only way to (self-)insure against income shocks and implies

that consumption should respond fully to permanent shocks but only marginally to temporary

ones. Since the consumption response depends on the persistence of income shocks, testing

the PIH requires identifying shocks of different persistence when only total income changes

are observed. For this reason, some authors focus on the correlation between consumption

and total income changes as a way to restrict the set of models consistent with it (Altonji and

Siow, 1987; Krueger and Perri, 2005, 2006, 2011), while others use proxies for transitory and

permanent income changes such as changes in hours or changes in wages and involuntary job

loss, respectively (Cochrane, 1991; Dynarski and Gruber, 1997). Finally, some authors exploit

panel data and the cross-equation restrictions implied by the linear or linearized consumption

function to separately identify the consumption response to permanent and transitory income

shocks (Hall and Mishkin, 1982; BPP).

BPP’s estimate of 0.36 for the consumption insurance coefficient for permanent shocks4

suggests a substantial degree of excess insurance compared to the theoretical prediction of zero

under the PIH, or of a value close to zero for the linear approximation of the consumption

function. Although this finding constitutes prima facie evidence against the textbook PIH

and, more generally, the linearized consumption function, it is not necessarily at odds with the

nonlinearized incomplete-markets model. Carroll (2009) shows that a buffer-stock saving model

with impatient consumers, constant-relative-risk-aversion (CRRA) preferences, and a single,

noncontingent asset implies an insurance coefficient for permanent shocks ranging between

0.08 and 0.25, for plausible degrees of patience and risk aversion. For this reason, Kaplan and

Violante (2010) suggest that a better way to assess the degree of excess insurance is to compare

the empirical BPP estimates to their counterparts estimated on data simulated from a calibrated

incomplete-markets model. This is the route we take in this article with the only difference that,

instead of calibrating our model as in Kaplan and Violante (2010), we structurally estimate by

indirect inference. This last aspect is shared with Guvenen and Smith (2014) with the important

4Hall and Mishkin (1982) impose zero insurance of permanent shocks and estimate only the response to transitory-

ones.