Do Borrowing Constraints Matter? An Analysis of Why the ... · is the most important determinant of whether or not a household is borrowing-constrained, (3) that the Euler equation

Do Borrowing Constraints Matter? AnAnalysis of Why the Permanent IncomeHypothesis Does Not Apply in Japan

Miki Kohara and Charles Yuji Horioka

August 2005

Abstract

In this paper, we use micro data on young married householdsfrom the Japanese Panel Survey of Consumers, conducted by the In-stitute of Research on Household Economics, to shed light on (1) theprevalence of borrowing constraints in Japan, (2) what households areborrowing-constrained in Japan, (3) whether the life cycle/permanentincome hypothesis (LCPIH) holds in Japan, and (4) whether the pres-ence of borrowing constraints, the presence of income risk, or the pres-ence of precautionary motives for saving is the reason why the LCPIHdoes not hold in Japan. To summarize our main �ndings, we �nd(1) that about 8 percent of young married Japanese households areborrowing-constrained, (2) that the husband�s educational attainmentis the most important determinant of whether or not a household isborrowing-constrained, (3) that the Euler equation implication is re-jected for both the full sample and for the subsample of unconstrainedhouseholds, (4) that the Euler equation implication is rejected for bothlow income volatility and high income volatility households, and (5)that the Euler equation implication is rejected for households with pre-cautionary motives for saving but cannot be rejected for householdswithout precautionary motives for saving. These results suggest thatthe LCPIH does not apply in Japan and that the presence of precau-tionary motives for saving (not the presence of borrowing constraintsor the presence of income volatility) is the main reason for why it doesnot apply.

1

1 Introduction

If the life cycle/permanent income hypothesis (hereafter LCPIH) holds, changesin consumption should not be sensitive to changes in current income. On theother hand, if this hypothesis does not hold (for example, because house-holds are borrowing-constrained), changes in consumption will be sensitiveto changes in current income. Thus, a commonly used test of the valid-ity of the LCPIH is to estimate an Euler equation to see whether changesin consumption are sensitive to changes in current income. If the LCPIHdoes not hold and the reason is the existence of borrowing constraints, wewould expect changes in consumption to be sensitive to changes in currentincome in the case of borrowing-constrained households but not in the caseof unconstrained households. In this paper, we use micro data on young mar-ried households from the Japanese Panel Survey of Consumers, conductedby the Institute of Research on Household Economics, to shed light on (1)the prevalence of borrowing constraints in Japan, (2) what households areborrowing-constrained in Japan, (3) whether the LCPIH holds in Japan, and(4) whether the presence of borrowing constraints, the presence of incomerisk, or the presence of precautionary motives for saving is the reason whythe LCPIH does not hold in Japan.To summarize our main �ndings, we �nd (1) that about 8 percent of young

married Japanese households are borrowing-constrained, (2) that the hus-band�s educational attainment is the most important determinant of whetheror not a household is borrowing-constrained, (3) that the Euler equation im-plication is rejected for both the full sample and for the subsample of un-constrained households, (4) that the Euler equation implication is rejectedfor both low income volatility and high income volatility households, and(5) that the Euler equation implication is rejected for households with pre-cautionary motives for saving but cannot be rejected for households withoutprecautionary motives for saving. These results suggest that the LCPIHdoes not apply in Japan and that the presence of precautionary motives forsaving (not the presence of borrowing constraints or the presence of incomevolatility) is the main reason for why it does not apply.The paper is organized as follows: in section 2, we present the theoretical

model; in section 3, we describe the data and analyze what households areborrowing-constrained in Japan; in section 4, we present the results of ourEuler equation tests; and section 5 concludes.

2

2 The Model

2.1 Consumption Smoothing

Consumption smoothing behavior is characterized by the Euler equation.Assume that the consumer holds At of total assets at the beginning of periodt and purchases a total of Nt of assets at (the end of) t. These are realexpenditure on total assets1. The consumer earns labor income of wtht, wherew is the exogenous real wage that is constant across individuals and h is workhours, and spends it on the consumption of goods, c, and the purchase ofassets, N . The saving constraint faced by the consumer is described as St =Nt�At = wtht�ct: The consumer also faces an asset accumulation constraintof At+1 = Nt(1+rt+1) where rt+1 is the interest rate at the beginning of periodt + 1. Assume that all agents face the same interest rate. Individuals livefor a �nite lifetime T and leave no bequests at T + 1. Consumers allocateall of their time, say 24 hours a day, between work, ht, and leisure, lt, whichcreates another constraint (a time constraint). Standardizing time to one,ht+ lt = 1: These constraints can be combined into one constraint as followsNt �Nt�1(1 + rt) = wtht � ct:Suppose that the individual�s utility is stationary and additively separable

over time and written as EtnPT

k=t1

(1+�)k�tu(ck)o;where Et is an expectation

operator conditional on information available at t, u is a function that is in-creasing and concave in ct and � is the rate of time preference, which isassumed to be homogeneous over individuals and time. We assume that util-ity is not a function of leisure. The representative consumer�s maximizationproblem can be written as a dynamic programming problem. MaximizingVt = u(ct; lt) +

11+�EtVt+1(At+1; wt+1), we obtain the �rst order condition

(the Euler equation for consumption):

Et

�@ut@ct

� 1 + rt+11 + �

@ut+1@ct+1

�= 0; (1)

implying consumption at t should be chosen so that the expected discountedgain of saving now for the future is equal to marginal utility in this period.In addition, assume, as is often done, that utility is isoelastic, u(cit) =

c1� it =1 � , where is the risk aversion parameter. Marginal utility is con-vex and allows for precautionary saving as a special case. If it is assumed

1We can write the multiple asset case in the same way but using a vector, Nk �(N1;k; :::; NJ;k), where Nj;k is the j th asset the household holds at the end of k.

3

that lnci;t+1 and rt+1 have a joint normal distribution, equation (1) becomeslnci;t+1

Et� ln ci;t+1 = �1(Etrt+1 � �) +

1

2 !2i;t: (2)

In the last term, !2i;t is a measure of uncertainty, is equal to the variance of(� ln ci;t+1 � rt+1= ), and shows a precautionary motive for saving.There are at least two ways to test the validity of equation (2). The

�rst way is to test a structural form, estimating utility function parametersusing Generalized Method of Moments (nonlinear instrumental variable) es-timation. This is a direct test using the Euler equation, whose error termshould be orthogonal to information before t. GMM estimation is bene�cialin the sense that we can avoid the approximation of linear marginal utility inconsumption, the assumption of distribution, and the assumption of incomeexogeneity.However, many researchers have for a long time used another way to test

the Euler equation implication. This is a test of the reduced form Eulerequation with additional variables in past information sets. It tests the va-lidity of this additional information (for example, income changes) predictedby previous information. Additional variables should not explain consump-tion changes if the Euler equation holds. For example, consumption changesshould not react to predicted income changes. That is, we test whether � = 0in the equation

� ln ci;t+1 = �1Ft + �2�Xi;t+1 + �31

2 !2i;t + �� ln y

ei;t+1 + "i;t+1; (3)

where � ln yei;t+1 is income predicted by individuals using the informationavailable to them. This is calculated as �tted values from the �rst stageestimation of � ln yi;t+1. Ft is a time-varying variable including �1(Etrt+1��). Preference shifts, described as �Xi;t, could a¤ect the consumption planat any point in time. The third term is the conditional variance of theuncertain components. One of our main focuses is to review past studiesusing proper data on consumption smoothing. Thus, we conduct this reducedform exclusion test. The null hypothesis is that the Euler equation holds andindividuals smooth consumption changes against predicted income changes.That is, � = 0: consumption does not react to predicted income changes.Most past studies drop the term of precautionary motive for savings,

12 !2i;t, simply assuming that it is the same across individuals. In manycases, we cannot avoid assuming this because information is seldom available

4

on precautionary motives. The exception is Jappelli and Pista¤erri (2000),where a nominal (observable) income variance is included as a proxy for theuncertainty term !2i;t into one of the explanatory variables. Jappelli andPista¤erri (2000) emphasize the importance of including this term; if weignore this term and if � ln yei;t+1 is related to uncertainty or precautionarysaving motives, � measures not the sensitivity of consumption to income butthe e¤ect of uncertainty on consumption.This method could, however, be problematic for the following reasons.

First, income uncertainty is only one of many uncertainties individuals faceand is not a su¢ cient indicator of general uncertainty. Second, even if uncer-tainty consists only of income uncertainty, using actual income uncertaintyas a proxy for !2i;t as well as � ln y

ei;t+1 may raise a problem due to the corre-

lation between the two. Fortunately, our data set collects the data needed tocalculate income variances as a proxy of income uncertainty and also collectsdata on the existence of precautionary motives for saving. Therefore, we testthe Euler equation implication using equation (3) dropping the third term of12 !2i;t, but estimating the equation separately for high and low uncertaintygroups2.

2.2 Violation of Consumption Smoothing

If markets are complete and there exist appropriate securities against anyfuture state, each household�s consumption is fully insured against any idio-syncratic shock. Households can smooth consumption changes completely,sharing risks with each other. Previous studies have tested this implicationof consumption full insurance and most have rejected it. Although it is im-portant to �nd a situation (if any) where full insurance holds, the rejectionof the implication is not surprising. A more interesting issue is what can andwhat cannot explain the violation of the implication.For example, the existence of borrowing constraints may cause the Euler

equation implication to be rejected. Many studies have, for a long time, in-quired into the existence of borrowing constraints and the di¤erences in con-sumption behavior between borrowing-constrained and unconstrained house-holds. Households would fail to smooth consumption if they encountered an

2Since our data are panel data on households, we conduct IV estimation controllingfor household-speci�c di¤erences by: (a) taking the �rst di¤erence of all variables in thespeci�cation of the equation (3), (b) applying �xed e¤ects estimation, and (c) applyingrandom e¤ects estimation.

5

unexpected shock and could not borrow to carry out their original plans.Thus, unconstrained households do not react to income shocks, while con-strained households react strongly. Based on this analogy, many researchershave tested the Euler equation such as equation (3) and have interpreted �as the proportion of borrowing-constrained households.There exist many other auxiliary assumptions that are imposed to de-

rive the Euler equation and its test implication. For example, we ignore thepossibility of the existence of precautionary saving (the di¤erence in uncer-tainty among households) on the assumption that it is same across time andindividuals.Once we have rejected the Euler equation implication, we should seek the

reason for it, checking each possible explanation one by one. The followingparts focus on the existence of borrowing constraints, which is the most fre-quently used explanation for the violation of the Euler equation implication.Speci�cally, we identify unconstrained households using unique informationon households�borrowing constraints and test the Euler equation implica-tion using this subsample. If the existence of borrowing constraints is theprimary explanation for the violation of the Euler equation implication, weshould �nd evidence of the Euler equation implication for this sample butnot for the full sample. However, if other explanations for the violation ofthe Euler equation implication matter, we will not �nd support for this im-plication even for the unconstrained sample. We will discuss other possibleexplanations for the violation of the Euler equation implication later afterwe have examined the empirical results for borrowing constraints.

3 The Data

3.1 JPSC Data

This paper uses micro data from the Japanese Panel Survey of Consump-tion, (hereafter the JPSC) (in Japanese, Shouhi Seikatsu ni kansuru PaneruChousa), a panel survey conducted by the Institute for Research on House-hold Economics (in Japanese, Kakei Keizai Kenkyuusho). This survey hassurveyed young married and unmarried women (those between the ages of24 and 34 in 1993) once a year since 1993, and this paper uses the 1993-2004waves from this survey. Because JPSC is a panel survey, we can calculatechanges in consumption from year to year, which is precisely the variable we

6

need to test our theoretical model. We con�ne our analysis to the subsam-ple of married women because most young single Japanese women live withtheir parents and rely on their parents� income but precise information isnot available on their parents�income and consumption. Note that marriedwomen are asked not only about themselves but also about other householdmembers.Borrowing constraints. The �rst and most important variable used in our

analysis is the one pertaining to borrowing constraints. The JPSC asks threeunique questions about borrowing constraints: (1) Have you (or your spouse)ever had a loan application turned down? (2) Have you (or your spouse) everhad the loan amount reduced when you applied for a loan? (3) Have you (oryour spouse) ever decided against applying for a loan because you expectedyour loan application to be turned down? Following Jappelli (1990), we referto households answering "yes" to these questions as "rejected," "reduced,"and "discouraged" borrowers, respectively. Households that replied "yes" toone or more of these questions were regarded as being borrowing-constrained.Unfortunately, this information is available only for the �rst survey. Thus, wehad no choice but to assume that borrowing constraints remain unchangedover the sample period. This is exactly what Jappelli (1990) and Jappelli,Pischke, and Souleles (1998) assume, even though it may be too strong anassumption. We will return to this point in the last part of this section.Consumption. The JPSC collects data on consumption (living expenses)

by all household members during the month of September. In the regressions,we use the growth rate of monthly consumption. The data on monthly con-sumption have at least two advantages: �rst, they include all consumptiongoods and services, unlike in the case of PSID, which collects data only onfood consumption. Thus, we need not make any assumptions about the sep-arability of consumption. Second, using the change in consumption betweentwo non-sequential months has the advantage of avoiding, to some extent,potential serious problems raised by consumption durability and habit for-mation3.Income. The JPSC collects data on several measures of income, includ-

3The change in monthly consumption could be biased if the household engages inpurchases of big-ticket items such as homes and cars. The JPSC asks about spending on�living expenses�during the previous month excluding spending on most big-ticket items.The survey asks separately about purchases �nanced by loans. Thus, we can exclude thepossibility that consumption growth is overestimated as a result of purchases of big-ticketitems.

7

ing annual (total) income, annual labor income, and monthly labor income.Annual (total) income and annual labor income are inclusive of taxes sowe need to estimate taxes in order to calculate after-tax income. We useafter-tax monthly labor income in the main Euler equation for at least threereasons: �rst, we wanted the period and timing of consumption and incometo match. If we use annual income, there is a danger of underestimating thedegree of consumption smoothing simply because annual income is more sta-ble than monthly income or consumption. Another reason for using monthlyincome is that using annual income would require us to waste the last year ofdata since the survey asks about annual income in the previous year. Finally,the use of monthly labor income helps to reduce the amount of householdheterogeneity because data on monthly labor income are not available for theself-employed. We sum the monthly labor incomes of all household membersand use the growth rate of total monthly labor income in the regressions.Household characteristics. Following the past literature on testing the

LCPIH and the existence of borrowing constraints by estimating an Eulerequation, we take the husband�s age, the household�s consumption needs, asproxied by the number of family members, and year dummies. Althoughwe tried including many other time-variant and time-invariant variables thatmight possibly in�uence consumption, particularly that of young Japanesehouseholds such as those included in our sample, all of the variables wetried including had little e¤ect and their inclusion was not supported statisti-cally4. Time-invariant variables such as residential area dummies are anywaydropped from �xed e¤ects estimation. In estimating expected income change,we use previous year�s income change and husband�s educational attainmentas instruments, adding to the explanatory variables for consumption.The total number of married women (households) was about 1000 in

most years. We kept an observation if it contained enough information forat least a one-year panel. Thus, the sample we used is unbalanced panel.After eliminating observations with missing values for one or more of thevariables included in the regressions, we were left with 1058 households (5674household-years) in the full sample and 980 households (4243 household-years) in the unconstrained sample5.

4For example, neither a dummy variable for those who had their �rst baby during thecurrent year, which could make a big di¤erence in consumption patterns, nor a dummyvariable for those who starting living with their parents during the current year, which isoften observed in Japan as the parents get older, did not change the results below.

5Observations lying outside of the �mean plus or minus two standard deviations�range

8

3.2 Who is Constrained?

Before estimating the Euler equation, we summarize the characteristics ofborrowing-constrained households. Table 1 summarizes the borrowing mo-tives of households that are currently in debt. Although housing and carpurchases are the main reasons for borrowing, a few households do in factborrow to �nance living expenses. More than half of the sample is currentlyin debt, which suggests that borrowing plays an important role in householdplanning.Many past studies have tried to distinguish borrowing-constrained house-

holds from unconstrained households. Since direct data on borrowing con-straints are usually not available, most previous studies have tried to predictwho is borrowing-constrained using a variety of indicators. In our case as wellas in the case of Jappelli (1990), however, direct information is available onwhether or not a given household is borrowing-constrained. Thus, followingJappelli (1990), we analyze what determines whether a given household isborrowing-constrained by regressing a dummy variable that equals one if thehousehold is borrowing-constrained and zero otherwise on various householdcharacteristics using probit estimation. The household characteristics we useinclude assets, income, husband�s age and educational attainment, householdsize, homeownership, debt, city size, and region. We use two measures of as-sets: Asset1, which is de�ned as holdings of bank and postal deposits, bonds,and equities, and Asset2, which is de�ned as Asset1 plus life and non-life in-surance, land, and housing. Debt is de�ned as the amount of outstandingdebt. The other variables are described in the last section.Table 2 shows the characteristics of borrowing-constrained and uncon-

strained households separately. Borrowing-constrained households, espe-cially "reduced" households have lower incomes, assets, husband�s employ-ment rate, and husband�s educational attainment than unconstrained house-holds. "Discouraged" borrowers also have similar characteristics to "denied"and "reduced" households. This �nding underscores the importance of dif-ferentiating "discouraged" households from those completely free from bor-rowing constraints and grouping them together with borrowing-constrainedhouseholds.Who is borrowing-constrained? Table 3 shows the estimation results of

our probit analysis of who is borrowing constrained. The left-hand column ofTable 3 shows the results based on the narrower de�nition of assets (Asset1),

were regarded as outliers and dropped from the sample.

9

while the right-hand column shows the results based on the broader de�n-ition of assets (Asset2). Household assets and income are two variables ofinterest since most past studies have used the ratio of assets to income as anindicator of whether or not a given household is borrowing-constrained, butneither assets nor income are major determinants of whether or not a givenhousehold is borrowing-constrained, regardless of which measure of assets isused. Income and Asset2 are not relevant at all, and Asset1 has very little ef-fect on reducing the probability of being borrowing-constrained6. Our resultsare very di¤erent from the results Jappelli (1990) obtained using U.S. data�that households with less income or assets indeed face a higher probabilityof being borrowing-constrained.By contrast, the most relevant variable in the case of Japan is not income

or assets but the husband�s educational attainment. Educational attainmentcould be an indicator of current as well as future income, and a householdin which the educational attainment of the husband (usually the householdhead and main income earner) is relatively low might be regarded as havinginsu¢ cient ability to repay loans.These results have important implications for analyses of borrowing con-

straints. Although asset holdings may be a signi�cant determinant of theprobability of being borrowing-constrained in the U.S., it is not a signi�cantdeterminant in Japan.We turn now to a check of the accuracy of indicators used by previous

studies to identify borrowing-contrained and unconstrained households. Fol-lowing previous studies, we group the sample into �hypothetically�borrowing-constrained and unconstrained households using various indicators and thencompare these households to �actually� borrowing constrained and uncon-strained households. The results are shown in Table 4.The �rst three indicators, which were originally proposed by Zeldes (1989),

are the most frequently used indicators in many countries and are con-structed by taking the ratio of asset holdings to income. Since householdswho have adequate amounts of assets relative to income can dissave theirassets when necessary and protect their consumption against unexpected in-come shocks, households with a high asset-income ratio are regarded as being

6The self-employed may need to borrow and may face borrowing constraints more fre-quently than others. Also, employment conditions such as tenure and �rm size often a¤ecthousehold decisions in Japan. Thus, we conducted the estimation including variables relat-ing to the husband�s self-employement, tenure, and �rm size, but none of their coe¢ cientswere signi�cant.

10

unconstrained. The �rst indicator is whether or not the household�s holdingsof �nancial assets are more than twice as much as their monthly income,and the second one is whether or not the household�s holdings of total as-sets (�nancial assets plus housing equity) are more than twice their monthlyincome. The third indicator classi�es households with no �nancial assetsas being borrowing-constrained and households whose holdings of �nancialassets are more than twice their monthly income as being unconstrained.The fourth indicator is whether or not the household owns one or more

credit cards. If it owns one ore more credit cards, it can �nance its consump-tion even when it experiences an unexpected income decline. This indicatoris close to the one suggested by Shintani (1994), who classi�es householdsas being unconstrained if they own one or more credit cards or one or morecards with a free-loan feature because they need to pass a credit check inorder to receive one or both kinds of cards. The �fth indicator, proposed byHayashi (1985b), classi�es households as unconstrained if they consume lessthan 85% of their annual disposable income (minus all debt outstanding plus20% of their �nancial assets).Finally, since we found from Table 3 that educational attainment is a sig-

ni�cant indicator of being unconstrained, we propose a new indicator iden-tifying college graduates as being unconstrained. In addition, we constructanother new indicator that is the same as the �rst indicator suggested byZeldes (1989) except that �nancial assets are replaced by a broader conceptof assets�namely, �nancial assets plus housing equity plus life insurance.Table 4 shows the results. The shaded areas indicate the proportion

of households identi�ed properly. The results are summarized in Table 5.As expected, the husband�s educational attainment identi�es unconstrainedhouseholds well, as does Hayashi�s indicator (his consumption-income ratio).By contrast, Zeldes�s asset-income ratio is better at identifying borrowing-constrained households, but even so, about 50% are misclassi�ed.This �nding is similar to Jappelli�s (1990) �nding for U.S. households

that using the asset-income ratio leads to serious misclassi�cation of con-strained and unconstrained households. Moreover, misclassi�cation is evenmore serious in the case of Japanese households. We should identify un-constrained households using information on educational attainment or theconsumption-income ratio rather than using information on the asset-incomeratio. Moreover, it may not be possible to identify borrowing constrainedhouseholds regardless of what indicator is used.

11

4 The Results

4.1 Euler Equation Test

In this section, we present the results of our Euler equation tests, but we�rst present the descriptive statistics for the sample used in the estimationin Table 6.First, we estimated the Euler equation for the full sample using simple

IV estimation. The results are reported in part (a) of Table 7 (see Appendixfor the �rst stage regression results), and as can be seen from this table, thecoe¢ cient of expected income changes is 0.12 and it is signi�cant at the onepercent signi�cance level. We strongly reject the applicability of the Eulerequation implication.Parts (b) and (c) of Table 7 are the results of IV estimations controlling

for individual e¤ects as �xed e¤ects model, or rondom e¤ects model, respec-tively. Although the Wu-Hausman test shows that individual e¤ects are notcorrelated with the explanatory variables so that the random e¤ects model isgood enough to be estimated, the �xed e¤ects model estimator is still consis-tent (but ine¢ cient) and may still be preferred if individual e¤ects relatingto the household�s consumption and income changes is correlated with theother explanatory variables. The coe¢ cient of expected income is about 0.13and 0.12 in the �xed e¤ects and random e¤ects models, respectively. Bothcoe¢ cients are signi�cant at the one percent signi�cance level. Thus, weagain reject the Euler equation implication.The coe¢ cient of expected income changes in (b) might be upward bi-

ased because the process of taking the di¤erence from the mean (within-estimator) to remove �xed e¤ects makes income changes "transitory or sur-prising" rather than "expected or permanent" whereas we want to see thereaction of consumption to the latter. A larger coe¢ cient in the �xed e¤ectsmodel relative to the random e¤ects model may re�ect this possibility.A coe¢ cient of at least 0.12 is broadly consistent with previous studies

for many countries. However, we will show soon that comparing coe¢ cientsin this way is not meaningful. Although there is variation in the magnitudeof the coe¢ cients, the test implications are the same: the Euler equationimplication is rejected. Households do not smooth consumption changesover even expected income changes.

12

4.2 Do Borrowing Constraints Matter?

Most of the past literature attributes the violation of the Euler equationimplication to the existence of liquidity constraints. Using this analogy, thecoe¢ cient of expected income changes, such as the 0.12 value shown in Table7, is sometimes interpreted as the ratio of constrained households. However,this is not correct. As we have already shown, 8% of our sample is actuallyconstrained, while the results in Table 7 suggest that at least 12 percent ofour sample is constrained.If the existence of borrowing constraints are the reason for the violation

of the Euler equation implication, we would expect to �nd that the Eulerequation implication is applicable or close to applicable in the sample ofunconstrained households. Table 8 shows the results for the sample of un-constrained households, and as this table shows, the coe¢ cient of expectedincome changes stays at about the same magnitude and signi�cance level.The di¤erence in the coe¢ cients of expected income changes between thefull and unconstrained samples is quite small and not signi�cant at a onepercent signi�cance level (see the bottom row of Table 8). Thus, the Eulerequation does not hold even for unconstrained households, which suggeststhat the existence of borrowing constraints are not the reason for the viola-tion of the Euler equation implication.Many past studies that identify unconstrained households using the level

of the asset-income ratio make the conclusions even more ambiguous. Inthe previous section, we found that splitting the sample by the asset-incomeratio itself is questionable, especially when we are interested in the behaviorof unconstrained households. In addition to this problem, the sensitivity ofconsumption to expected income changes as measured by the above type ofEuler equation does not show what proportion of households are borrowing-constrained. The results obtained in this paper suggest that the existence ofborrowing constraints are not be a major reason for the violation of Eulerequation implication.

4.3 Other possible explanations

If borrowing constraints are not the explanation, what is the explanation forthe violation of the Euler equation implication? The existence of precau-tionary saving or di¤erences in households�level of uncertainty are possibleexplanations of the violation of the Euler equation implication. Our basic

13

estimation model excludes this possibility, assuming that time and individ-ual �xed e¤ects can control for it, which may be too strong an assumption.Thus, we split the sample into low income volatility and high income volatil-ity households and into households with precautionary motives for savingand households without precautionary motives for saving and estimated theEuler equation for each subsample.When splitting the sample by the amount of income volatility, income

volatility was measured as the variance in the husband�s annual labor in-come during the previous four years. High income volatility households werede�ned as households whose income volatility is higher than the median,and low income volatility households were de�ned as households whose in-come volatility is lower than the median. When splitting the sample by thepresence or absence of precautionary motives for saving, households withprecautionary motives for saving were de�ned as households with a positiveamount of precautionary saving and households without precautionary mo-tives for saving were de�ned as households with no precautionary saving. Thehigh (low) income volatility sample has 728 (775) households and 2269 (2762)household-years, while the sample with (without) precautionary motives forsaving has 908 (727) households and 3651 (2023) household-years.The estimation results are shown in Table 9, and as this table shows,

the coe¢ cient of expected income changes is statistically signi�cant and ofroughly the same magnitude for both low income volatility and high incomevolatility households. It is slightly higher for high income volatility house-holds when IV with �rst di¤erences is used but somewhat higher for lowvolatility households when IV with �xed e¤ects or IV with random e¤ectsare used. Thus, it appears that income volatility is not the explanation forthe violation of the Euler equation implication either.Turning to the impact of precautionary motives for saving, Table 10 shows

that the coe¢ cient of expected income changes is statistically signi�cant andin the 0.15 to 0.16 range in the case of households with precautionary motivesfor saving but statistically insigni�cant and in the 0.05 to 0.09 range in thecase of households without precautionary motives for saving. This suggeststhat precautionary motives for saving are the primary explanation for theviolation of the Euler equation implication.The existence of consumption durability is another possible explanation

for the rejection of the Euler equation implication. If a commodity is durableand expenditure on that commodity is increased in the present period, ex-penditure will be depressed in the next period although people indeed enjoy

14

consuming that commodity. Households can derive bene�ts from consumingnow rather than later, showing excess sensitivity of consumption7. In thiscase, the error terms in the equation (3) contains the e¤ects past consumptionand are correlated with the explanatory variables (Mankiw (1982), Hayashi(1985b, 1999)). This is, however, less serious in our results, since the surveywe used measures consumption at separate time periods one year apart).Furthermore, data problems are also a possible explanation for the re-

jection of the Euler equation implication. If we do not have data on totalconsumption but only on consumption of a certain good, we need to assumeseparability between goods. If we cannot obtain the appropriate micro datato test the Euler equation implication, we have to assume that aggregationis possible. If we cannot �nd valid instruments in the limited informationset, there exists misspeci�cation of the stochastic structure of income. Thelast problem is related to informational constraints on households. However,these problems are not critical in the present analysis. Our consumptiondata is total consumption expenditure, and moreover, our data set containsdata on a large number and variety of household attributes, making it easierto �nd appropriate instruments.Thus, the only explanation for the violation of the Euler equation im-

plication that is potentially relevant in the case of the present analysis andthat we have not yet investigated is the assumption of separability betweenconsumption and leisure.

5 Conclusion

In this paper, we used micro data on young married households from theJapanese Panel Survey of Consumers, conducted by the Institute of Re-search on Household Economics, to shed light on (1) the prevalence of bor-rowing constraints in Japan, (2) what households are borrowing-constrainedin Japan, (3) whether the life cycle/permanent income hypothesis (LCPIH)holds in Japan, and (4) whether the presence of borrowing constraints, the

7The misspeci�cation of the assumption that consumption and leisure are separableis yet another possible explanation for the violation of the Euler equation implication.Unless this assumption is imposed, we cannot derive the above type of Euler equationtest, where the consumption decision is made independently of the leisure decision.Habit formation is also an example of nonseparable consumption over time. In this

case, consumption must increase over time and households try to save now for futureconsumption, showing excess smoothness of consumption.

15

presence of income risk, or the presence of precautionary motives for savingis the reason why the LCPIH does not hold in Japan. To summarize ourmain �ndings, we found (1) that about 8 percent of young married Japanesehouseholds are borrowing-constrained, (2) that the husband�s educational at-tainment is the most important determinant of whether or not a household isborrowing-constrained, (3) that the Euler equation implication is rejected forboth the full sample and for the subsample of unconstrained households, (4)that the Euler equation implication is rejected for both low income volatilityand high income volatility households, and (5) that the Euler equation impli-cation is rejected for households with precautionary motives for saving butcannot be rejected for households without precautionary motives for saving.These results suggest that the LCPIH does not apply in Japan and that thepresence of precautionary motives for saving (not the presence of borrowingconstraints or the presence of income volatility) is the main reason for whyit does not apply.Turning to the policy implications of our �ndings, our �nding that the

presence of precautionary motives for saving is the main reason why theLCPIH does not apply in Japan implies that the best way to make Japaneseconsumers better o¤and to enable them to optimize their consumption pathsis to reduce uncertainties about the future so that households do not feel theneed to do saving for precautionary motives. Possible ways to achieve thiswould be to improve the safety net for those who experience job loss, incomeloss, illness, accidents, natural disasters, and other unforeseen emergencies,to relax borrowing constraints so that those experiencing unforeseen emer-gencies can borrow as much as they need to tide them over, and to revampthe public old-age pension, health insurance, and long-term care insuranceprograms so that worry about life during retirement is alleviated.

16

References

[1] Arellano, M., and Bond, S., (1991), �Some Tests of Speci�cation forPanel Data: Monte Carlo Evidence and an Application to EmploymentEquations,�Review of Economic Studies, 58, 277-297.

[2] Campbell, J. H. and Mankiw, G. N. (1991), �The Response of Consump-tion to Income: A Cross-Country Investigation,�European EconomicsReview, 35, 723-756.

[3] Hayashi, F. (1985a), �The Permanent Income Hypothesis and Consump-tion Durability: Analysis Based on Japanese Panel Data,� QuarterlyJournal of Economics, 100, 1083-1113.

[4] Hayashi, F. (1986), �An Extension of the Permanent Income Hypothesisand Its Test�(in Japanese) Keizai Bunseki (Economic Analysis), 101,1-23, Institute of Economic Reserach, Economic Planning Agency of theJapanese Government.

[5] Hayashi, F. (1997), Understanding Saving (Cambridge, Massachusetts:The MIT Press).

[6] Institute for Research on Household Economics, Japanese Panel Surveyof Consumers, 1993 through 2004.

[7] Jappelli, T. (1990), �Who Is Credit Constrained in the U.S. Economy?�Quarterly Journal of Economics, 105, 219-234.

[8] Jappelli. T., and Pistaferri, L., (2000), �Using Subjective Income Ex-pectations to Test for Excess Sensitivity of Consumption to PredictedIncome Growth,�European Economic Review, 44, 337-358.

[9] Jappelli, T., Pischke, J., and Souleles, N. S. (1998), �Testing for Liquid-ity Constraints in Euler Equations with Complementary Data Sources,�Review of Economics and Satistics, 80, 251-62.

[10] Mankiw, N. G., (1982), �Hall�s Consumption Hypothesis and DurableGoods,�Journal of Monetary Economics, 10, 417-25.

[11] Ogawa, K. (1990), �Cyclical Variations in Liquidity-Constrained Con-sumers: Evidence from Macro Data in Japan,�Journal of the Japaneseand International Ecnomy, 4, 173-193.

17

[12] Runkle, D. E., (1991), �Liquidity Constraints and the Permanent-income Hypothesis,�Journal of Monetary Economics, 27, 73-98.

[13] Shibata, A. and Shintani, M. (1998), �Capital Mobility in the WorldEconomy: An Alternative Test,� Journal of International Money andFinance, 17, 741-756.

[14] Shintani, M. (1994),�Nippon no Shouhisha to Ryuudousei Seiyaku,�(Japanese Consumers and Liquidity Constraints: A Test Based onCredit Information), Osaka Economic Papers, 44, 41-56.

[15] Wakabayashi, M., and Horioka, C. Y. (2005), "Borrowing Constraintsand Consumption Behavior in Japan," Discussion Paper No. 640, Insti-tute of Social and Economic Research, Osaka University.

[16] Zeldes, S. P. (1989), �Consumption and Liquidity Constraints: An Em-pirical Investigation,�Journal of Political Economy, 97, 305-346.

18

Table 1. Borrowing MotivesAll 784 households

Borrowing? No / Yes 47.59% / 52.41%If yes, housing 45.9%to finance: car 39.1%

durables 10.1%clothing 8.7%leisure 4.6%

children' s education 3.6%children's marriage 1.4%

payment for accidents 5.6%living expenses 5.3%

repayment of debt 5.1%new business 1.2%

others 4.3%

Note.1. Households answering "yes" to "Borrowing?" are households that are currently in debt, and the figures in the second line and below showthe proportion of households currently in debt that are borrowing for each motive. Since some households are borrowing for more than one motive, the figures sum to more than 100.2. These data are from the first wave of the Japanese Panel Survey of Consumers, 1993.

Table 2. Characteristics of Constrained and Unconstrained HouseholdsAll Sample Unconstrained Constrained541 households 92.23% 7.76% Denied (59.52%) Reduced (11.90%) Gave up (71.43%)

Mean Std.Dev. Mean Std.Dev. Mean Std.Dev. Mean Std.Dev. Mean Std.Dev. Mean Std.Dev.HusAge 33.0 4.8 32.94 4.77 33.31 5.67 34.00 5.82 31.20 5.89 32.97 6.25HusCollege % 36.8% 48.3% 39.7% 49.0% 2.4% 15.4% 4.0% 20.0% 0.0% 0.0% 3.3% 18.3%HusEmployed % 99.6% 6.1% 99.8% 4.5% 97.6% 15.4% 96.0% 20.0% 100.0% 0.0% 96.7% 18.3%NumFamily 4.17 1.41 4.17 1.45 4.14 0.95 4.16 1.11 4.60 0.89 4.00 0.83Homeown % 23.7% 42.6% 24.0% 42.8% 21.4% 41.5% 20.0% 40.8% 0.0% 0.0% 23.3% 43.0%Asset1 (\10 thousand) 323.3 435.6 342.0 445.4 101.8 186.8 140.2 229.7 32.4 65.9 55.4 74.0Asset2 (\10 thousand) 1212.0 1861.6 1265.9 1907.3 550.6 955.1 450.2 717.5 77.5 55.0 628.2 1073.6Debt (\10 thousand) 424.9 945.6 419.2 958.6 491.6 776.5 436.1 718.5 316.6 204.9 557.4 869.9Income (\10 thousand) 517.3 229.7 520.8 229.1 475.9 235.4 492.7 226.5 567.1 411.4 413.7 160.2Consumption (\ thousand) 204.7 87.0 203.5 84.2 218.8 115.4 225.7 128.9 244.6 161.5 235.1 124.2

Notes.1. The data are from the first wave of the Japanese Panel Survey of Consumers, 1993.2. "Constrained" households include 'rejected, 'reduced' and 'discouraged' households. Since some households experienced two or more of the three events, the totals sum to more than 100%. 3. HusAge, HusCollege and HusEmployed denote the husband's age, college graduation, and employment status, respectively. 4. NumFamily denotes the number of family (household) members.5.. Asset1 denotes the household's total holdings of bank and postal deposits, bonds, and equities. Asset2 includes, in addition to Asset1, the paid-in value of life and non-life insurance and the value of land and housing they own. In units of 10,000 yen. 6. Income is the annual disposable income of all household members (in units of 10,000 yen). 7. Consumption is monthly expenditure (in units of 1,000 yen). ]

Table 3. Who Is Constrained?Dependent Variable: "Constrained" =1 if either rejected, reduced, or discouraged.

Coefficient CoefficientHusAge 0.0018 -0.0028

(0.0046) (0.0069)HusAge*HusAge 0.0000 0.0001

(0.0001) (0.0001)Income 0.0001 0.0002

(0.0001) (0.0003)Income*Income 0.0000 0.0000

(0.0000) (0.0000)Income*HusAge 0.0000 0.0000

(0.0000) (0.0000)Asset1 -0.0003 **

(0.0002)Asset1*Asset1 0.0000

(0.0000)Asset1*Income 0.0000

(0.0000)Asset1*HusAge 0.0000

(0.0000)Asset2 0.0000

(0.0000)Asset2*Asset2 0.0000

(0.0000)Asset2*Income 0.0000

(0.0000)Asset2*HusAge 0.0000

(0.0000)HusCollege -0.0367 *** -0.0523 ***

(0.0165) (0.0202)HusEmployed -0.1703

(0.2376)NumFamily 0.0002 0.0021

(0.0023) (0.0041)Homeown -0.0062 0.0481

(0.0074) (0.0572)Debt 0.0000 0.0000 *

(0.0000) (0.0000)AreaScale_med -0.0173 ** -0.0218 *

(0.0115) (0.0156)AreaScale_small -0.0122 ** -0.0175 *

(0.0075) (0.0104)Number of obs (households) 541 450Log likelihood -111.51 -91.05Likelihood Ratio 72.32 53.86

Notes:1. The coefficients show marginal effects.2. For the definitions of the variables, see the text and the notes to Table 1. 3. AreaScale_med and AreaScale_small are dummy variables indicating that the household lives in medium-sized and and small-sized cities, respectively, rather than in a large metropolitan area. Metropolitan areas denote the thirteen ordinance-designated cities. 4. The equation also includes eight regional dummies.

Table 4. Indicators of Constrained Households used in the Past Literature PanelA- appropriateness of the past indicators PanelB- other possible indicators①Asset (Basic　Split)* ⑥Basic Split by widely defined assets**701sample unconstrained constrained 535sample unconstrained constrainedThe indicator unconstrained 65.76 3.00 The indicator unconstrained 82.99 5.23

predicts: (71.81) (35.59) predicts: (90.06) (66.67)constrained 25.82 5.42 constrained 9.16 2.62

(28.19) (64.41) (9.94) (33.33)

②Asset (Total Wealth Split)462sample unconstrained constrained ⑦College Graduates (husband)The indicator unconstrained 75.54 2.81 810sample unconstrained constrained

predicts: (79.50) (56.52) The indicator unconstrained 88.89 7.65constrained 19.48 2.16 predicts: (96.90) (92.54)

(20.50) (43.48) constrained 2.84 0.62(3.10) (7.46)

③Asset (Extreme Split)582sample unconstrained constrainedThe indicator unconstrained 79.21 3.61 *Saving accounts and time deposits

predicts: (85.69) (47.73) **Saving accounts and time deposits, life and non-life insurance, and landconstrained 13.23 3.95 and house values

(14.31) (52.27)

④Credit Card Holders816sample unconstrained constrainedThe indicator unconstrained 83.70 7.60

predicts: (91.07) (93.94)constrained 8.21 0.49

(8.93) (6.06)

⑤Consumption-Income Ratio581sample unconstrained constrainedThe indicator unconstrained 84.85 6.88

predicts: (92.50) (83.33)constrained 6.88 1.38

(7.50) (16.67)

Actually

ActuallyActually

Actually

Actually

Actually

Actually

Table5. Which indicator should be used?

pick up pick up the unconstrained properly pick up the constrained properly

the constrained & the unconstrained properly

1 ⑦College Graduates (husband) 89.51 ⑦College Graduates (husband) 96.90 ①Asset (Basic Split) 64.412 ⑤Consumption-Income Ratio 86.23 ⑤Consumption-Income Ratio 92.50 ③Asset (Extreme Split) 52.273 ⑥Basic Split by widely defined assets* 85.61 ④Credit Card Holders 91.07 ②Asset (Total Wealth Split) 43.484 ④Credit Card Holders 84.19 ⑥Basic Split by widely defined assets* 90.06 ⑥Basic Split by widely defined assets* 33.335 ③Asset (Extreme Split) 83.16 ③Asset (Extreme Split) 85.69 ⑤Consumption-Income Ratio 16.676 ②Asset (Total Wealth Split) 77.70 ②Asset (Total Wealth Split) 79.50 ⑦College Graduates (husband) 7.467 ①Asset (Basic Split) 71.18 ①Asset (Basic Split) 71.81 ④Credit Card Holders 6.06

Table 6. Descriptive Statistics for the Euler Equation Estimation

Mean Std. Dev. Min Max Mean Std. Dev. Min MaxConsumption change rates 0.0307 0.4038 -3.0681 2.8622 0.0338 0.4003 -3.0681 2.8622Income change rates 38.5129 5.4427 24 65 39.0384 5.4031 25 61Income Change rates at one

year before 4.5192 1.3854 1 11 4.5150 1.4062 1 11Income Change rates at two

years before 14.3780 19.0988 0 61 15.0575 19.5221 0 61HusAge 0.0261 0.3152 -3.2189 2.8622 0.0290 0.3216 -3.2189 2.6231family size change rates 0.0341 0.3062 -3.2189 2.8622 0.0334 0.3145 -3.2189 2.8622Husage*HusCollege 0.0329 0.3113 -3.2189 3.4012 0.0259 0.3011 -3.2189 2.8622







Note 1. HusCollege and Husage are same definitions in table 2. Husage*HusCollege is intersection term between Husage and HusCollege. Family size is the number of faily members. Consumption and income changes are changes in monthly values. Changes are all taken as growth rates.

The sample with precautionary motives (number of theobservations (households) = 3651 (908) )

The sample without precautionary motives (number of theobservations (households) = 2023 (727) )

The entire sample (number of the observations (households) =5674 (1058) )

The unconstrained sample (number of the observations(households) = 4243 (980) )

The sample with high income volatility (number of theobservations (households) = 2269 (728) )

The sample with low income volatility (number of the observations(households) = 2762 (775) )

Table 7. Euler Equation Estimation Dependent Variable: change in log consumptionFor Constrained and Unconstrained Sample

Expected Income change* 0.1215 *** 0.1255 *** 0.1247 ***(0.0267) (0.0372) (0.0402)

Husage 0.0572 -0.0288 0.0000(0.2052) (0.0229) (0.0011)

family size change 0.0282 0.0053 0.0043(0.0235) (0.0139) (0.0039)

Husage*HusCollege 0.0214 0.0076 0.0004(0.0209) (0.0053) (0.0003)

constant -0.0875 1.0895 0.0072(0.2077) (0.9541) (0.0448)

Number of observations (households) 5674 (1058) 5674 (1058) 5674 (1058)Test for indiv.effects: Ui=0 --- 0.40 ---Test for all coef.=0 53.19 *** 64.49 *** 2.84σu --- --- 0.000σv 0.680 0.425 0.425R-squared (for within-estimator) 0.002 0.016 ---Wu-Hausman Test for RE ---1st stage: R-squared 0.53 0.26 --- Test for indiv.effects: Ui=0 --- 1.10 ** --- Test for all coef.=0 427.70 *** 137.20 *** 1234.00 ***

Note 1. For the 1st stage estimations to calculate Expected Income Change, see Appendix Table 1. 2. Squared of σu and σv are vaiances of ui and vit where error terms are written as eit=ui+vit.

3. Wu-Hausman test statistic, 4.57, suggests that individual effects are not correlated to explanatory variables and random effect model is good enough to be estimated. However, the fixed effect model result is reported additionaly, since that is consistent (but not efficient) even in this case. 4.The fixed effect model here takes a difference from the mean of time for each individual and adds the mean of total number of the observations (both over individuals and time). Since the last term is backed into a "usual" mean difference, a constant is remained in the estimation. 5. Time dummies are included in all the estimations. 6. Also see the footnote in table 6.

Table 8. Do Constraints Matter? For Unconstrained SampleDependent Variable: change in log consumption

Expected Income change* 0.1113 *** 0.1184 *** 0.1081 **(0.0303) (0.0410) (0.0452)

Husage 0.1483 -0.0402 * -0.0003(0.2119) (0.0235) (0.0012)

family size change 0.0404 0.0161 0.0081 *(0.0270) (0.0171) (0.0045)

Husage*HusCollege 0.0033 0.0090 0.0005(0.0245) (0.0078) (0.0003)

constant -0.1820 1.4799 -0.0004(0.2144) (0.9674) (0.0502)

Number of observations (households) 4243 (980) 4243 (980) 4243 (980)Test for indiv.effects: Ui=0 --- 0.48 ---Test for all coef.=0 36.05 *** 53.22 *** 26.36 ***σu --- --- 0.006σv 0.670 0.424 0.424R-squared (for within-estimator) 0.001 0.015 ---Wu-Hausman Test for RE ---1st stage: R-squared 0.54 0.28 --- Test for indiv.effects: Ui=0 --- 1.05 --- Test for all coef.=0 369.63 *** 128.47 *** 920.00 ***Difference in Consumption reaction to Expected Income between full sample & the unconstrained

Difference (standard error)

Note 1. Since Constrained Sample is rather small, we do not estimate for them, and compare the unconstrained results in this table to the full sample results in the previous table7. 2. There are no significant difference in consumption reaction to Expected Income between full sample and the unconstrained. 3. Also see the footnote in table 7.

4.57

(c) IV with randomeffects

(b) IV with fixedeffects

4.80

0.010 (0.040) 0.007 (0.055) 0.017 (0.060)

(a) IV taking firstdifferences



(c) IV with randomeffects

Table 9. Does Income Risk Matter? Dependent Variable: change in log consumption

For the Households with High-Income-Volatility For the Households with Low-Income-Volatility

Expected Income change* 0.1273 *** 0.1206 ** 0.1099 ** 0.0931 *** 0.1751 *** 0.1853 ***(0.0410) (0.0509) (0.0536) (0.0507) (0.0609) (0.0671)

Husage -0.0584 -0.0373 0.0012 -0.3917 -0.0503 -0.0003(0.6814) (0.0405) (0.0016) (0.4625) (0.0327) (0.0015)

family size change -0.0057 -0.0108 0.0062 0.0462 -0.0073 0.0025(0.0447) (0.0271) (0.0062) (0.0372) (0.0229) (0.0054)

Husage*HusCollege 0.0396 0.0152 0.0007 0.0029 0.0038 0.0003(0.0378) (0.0101) (0.0004) (0.0321) (0.0086) (0.0004)

constant -0.0068 1.3714 -0.0704 0.3448 2.1008 0.0446(0.6831) (1.6984) (0.0696) (0.4654) (1.3539) (0.0631)

Number of observations 2269 (728) 2269 (728) 2269 (728) 2762 (775) 2762 (775) 2762 (775)Test for indiv.effects: Ui=0 --- 0.57 --- --- 0.54 ---Test for all coef.=0 16.59 23.16 ** 17.47 * 38.40 *** 59.81 *** 33.21σu --- --- 0.000 --- --- 0.000σv 0.680 0.433 0.433 0.652 0.410 0.410R-squared (for within-estimator) 0.007 0.029 --- 0.001 0.020 ---Wu-Hausman Test for RE --- ---1st stage: R-squared 0.58 0.33 --- 0.49 0.29 --- Test for indiv.effects: Ui=0 --- 0.88 --- --- 1.36 *** --- Test for all coef.=0 152.50 *** 62.45 *** 609.00 *** 142.56 *** 66.15 *** 518.00 ***Difference in Consumption reaction to Expected Income between High & Low financial assets groups


Note 1. Income volatility is measured by the variance of husband's income among the past four years. High income volatility means the variance is higher than the median, and low income volatility measn the variance is lower than the median. 3. Time dummies are included. 4. Also see the footnote in table 7.

Table10. Precautionary Behavior Matters? Dependent Variable: change in log consumption

For the Households with Precautionary Motives For the Households without Precautionary Motives

Expected Income change* 0.1529 *** 0.1550 *** 0.1614 *** 0.0876 0.0827 0.0485(0.0350) (0.0461) (0.0474) (0.0542) (0.0660) (0.0721)

Husage 0.1740 -0.0300 0.0009 -0.0447 -0.0016(0.2389) (0.0285) (0.0013) (0.0450) (0.0018)

family size change 0.0487 0.0012 0.0033 0.0437 0.0235 0.0068(0.0331) (0.0185) (0.0047) (0.0477) (0.0279) (0.0072)

Husage*HusCollege 0.0237 0.0045 0.0001 0.0213 0.0155 0.0009(0.0282) (0.0069) (0.0003) (0.0485) (0.0119) (0.0005)

constant -0.1983 1.2175 -0.0129 -0.0001 1.5694 0.0383(0.2429) (1.1928) (0.0563) (0.0570) (1.8642) (0.0767)

Number of observations 3651 (908) 3651 (908) 3651 (908) 2023 (727) 2023 (727) 2023 (727)Test for indiv.effects: Ui=0 --- 0.55 --- --- 0.70 ---Test for all coef.=0 34.60 *** 49.42 *** 24.83 *** 19.02 ** 22.74 ** 21.14 **σu --- --- 0.000 --- --- 0.056σv 0.675 0.421 0.421 0.654 0.433 0.433R-squared (for within-estimator) 0.000 0.016 --- 0.011 0.019 ---Wu-Hausman Test for RE --- ---1st stage: R-squared 0.57 0.29 --- 0.49 0.29 --- Test for indiv.effects: Ui=0 --- 0.97 --- --- 1.23 *** --- Test for all coef.=0 255.07 *** 92.47 *** 946.00 *** 86.01 *** 43.11 *** 362.00 ***Difference in Consumption reaction to Expected Income between High & Low financial assets groups


Note 1. Households with precautionary motives are those who have precautionary savings. 2. In the estimation of IV taking the first difference for households without precautionary motives (NPa), Age is dropped by perfect collinearity. 3. Time dummies are included. 4. Also see the footnote in table 7.

1.46

6.71

(Ha) IV taking thefirst difference

(La) IV taking thefirst difference

(Hb) IV with fixedeffects

(Hc) IV with randomeffects

(Pa) IV taking thefirst difference

(NPa) IV takingthe first difference

4.32

0.034 (0.65) -0.055 (0.079) -0.075 (0.086)(Ha) <--> (La) (Hb) <--> (Lb) (Hc) <--> (Lc)

(Lb) IV withfixed effects

(Lc) IV withrandom effects

(Pb) IV with fixedeffects

(Pc) IV with randomeffects

(NPb) IV withfixed effects

(NPc) IV withrandom effects

0.065 (0.065) 0.072 (0.080) 0.113 (0.086)

4.79

(Pa) <--> (NPa) (Pb) <--> (NPb) (Pc) <--> (NPc)

Appendix. Prediction of Income Changes for Table 7Dependent Variable: Changes in Log Income

Income Change(t-1) -0.9412 *** -0.5742 *** -0.4585 ***(0.0133) (0.0143) (0.0132)

Income Change(t-2) -0.4744 *** -0.2953 *** -0.2092 ****(0.0134) (0.0143) (0.0130)

Husage 0.0404 -0.0143 -0.0019 ***(0.1073) (0.0152) (0.0007)

family size change 0.0073 0.0251 *** 0.0094 ***(0.0123) (0.0092) (0.0028)

Husage*HusCollege 0.0057 -0.0049 0.0005 **(0.0109) (0.0036) (0.0002)

constant -0.0573 0.6271 0.0912 ***(0.1086) (0.6354) (0.0319)

Number of observations 1613 689 1489Test for indiv.effects: Ui=0 --- 0.40 ---Test for all coefficients=0 53.19 *** 64.49 *** 2.84σu --- --- 0.000σv 0.680 0.425 0.425R-squared (for within-groupestimator) 0.080 0.074

Note 1. See the footnote for Table 7.



(c) IV withrandom effects

Documents

Do Borrowing Constraints Matter? An Analysis of Why the ... · is the most important determinant of whether or not a household is borrowing-constrained, (3) that the Euler equation