## Koichi Hamada, Anil K Kashyap, and David E. Weinstein

Print publication date: 2010

Print ISBN-13: 9780262014892

Published to MIT Press Scholarship Online: August 2013

DOI: 10.7551/mitpress/9780262014892.001.0001

Show Summary Details
Page of

PRINTED FROM MIT PRESS SCHOLARSHIP ONLINE (www.mitpress.universitypressscholarship.com). (c) Copyright The MIT Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a monograph in MITSO for personal use (for details see www.mitpress.universitypressscholarship.com/page/privacy-policy). Subscriber: null; date: 24 April 2019

# Consumption, Land Prices, and the Monetary Transmission Mechanism in Japan

Chapter:
(p.175) 6 Consumption, Land Prices, and the Monetary Transmission Mechanism in Japan
Source:
Japan's Bubble, Deflation, and Long-term Stagnation
Publisher:
The MIT Press
DOI:10.7551/mitpress/9780262014892.003.0006

# Abstract and Keywords

This chapter examines the effect of interest rates on consumption in Japan and explains why monetary transmission via Japan’s household sector is significantly different from that operating in the United States and other industrial countries. It presents an empirical analysis of Japanese consumption and household saving behavior, and explores the role of the household sector in the monetary transmission mechanism in Japan. Using models for aggregate consumption based more on the solved out consumption function approach than the Euler equation approach, the chapter looks at income growth and income uncertainty, along with the role of housing wealth, demography, and residential land prices.

The dotcom stock market bubble collapse of 2001–03 and the global financial crisis of 2008–09 have focused attention on the lessons of Japan’s lost decade. One of the widely accepted key lessons is the need for rapid refinancing of the banking system. However, there are widespread worries that the monetary transmission mechanism in the United States and other industrial economies might be as weak as it appears to be in Japan, with the result that these economies, too, might experience a lost de cade.

This chapter explains why monetary transmission via Japan’s household sector is sharply different from that operating in the United States and other industrial countries, and hence why analogies with Japan should not be taken too far. Indeed, pushing the analogies too far may have contributed to US monetary policy errors in 2002–05.

The research focuses on the empirical analysis of Japanese consumption and household saving behavior and discusses the role of the household sector in the monetary transmission mechanism in Japan. We develop models for aggregate consumption in Japan using the solved out consumption function approach associated with Ando and Modigliani rather than the Euler equation approach. There are three main reasons. First, the Euler approach ignores long-run information, and so is not so useful for understanding historical experience. Second, it is sensitive to the failure of strong assumptions about consumer rationality. Third, the empirical evidence rejects the central prediction of the theory.

Our model incorporates forward-looking income growth expectations and income uncertainty proxies, and the special role of housing wealth, which tend to be neglected in most studies of the solved out form. We also examine the impact of demography, which is an important issue for Japan. The model includes wealth- and interest-rate effects, and investigates the role of residential land prices.

(p.176) A key feature, derived from inter-temporal consumption theory, is to use disposable non-property income rather than total disposable income as the key determinant of consumption in the long run. This is quite important, as property income measured in the national accounts is a poor measure of the income concept that follows from theory. For example, the decline in the saving ratio in the 1990s, despite lower asset prices, is partly the result of lower inflation and the reduction in measured property income in the national accounts—rather than, for example, being necessarily caused mainly by the aging of the Japanese population.

## 0.1 Chapter Outline

The next section provides an overview of the results. section 2 summarizes the Euler equation implied by commonly used constant elasticity of substitution (CES) preferences. It then uses simple inter-temporal consumption theory to explore the likelihood of a positive response of consumption to the real interest rate, and to show that income uncertainty can be an additional factor in generating such a response. The theoretical justification from classical life-cycle theory for a housing wealth effect is then shown to be weak, at least for the conventional national-accounts definition of consumption. However, on a credit channel interpretation, there can be substantial housing-collateral effects, and the role of relaxing down payment requirements for first-time buyers, as well as the access of owners to collateral, are discussed. The implication is that any effect of house prices or housing wealth on consumption is likely to be very dependent on the nature of credit markets and other institutional features, and is therefore likely to differ from country to country.

The section also discusses the role of aggregation and demography for aggregate consumption. It explains the model to be used for empirical work, which incorporates potential credit-channel influences, but encompasses life-cycle models. Some of the relevant literature on consumption and household saving, with special reference to Japan, is reviewed.

section 3 discusses data and measurement issues, and charts the relevant variables. This is followed by a discussion of empirical models for forecasting non-property income, since income growth expectations play an important role in theory models. Our empirical estimates of Japanese aggregate consumption functions are then discussed. Robustness checks with regard to alternative formulations, variable (p.177) definitions, and sample periods are carried out. Co-integration checks are performed and instrumental variables estimates compared.

section 4 examines the factors driving the growth of household debt, looking for any evidence for credit market liberalization.

# 1 An Overview of Our Results

Based on our empirical results, we have a good explanation of why lower short-term interest rates do not stimulate total demand in the Japanese economy in the way they do, for example, in the UK (Muellbauer 2007).

First, in the United States and UK, there is an important asset-price channel, which, according to our estimated Japanese consumption function, is not just switched off in Japan, but even works in reverse. Using data going back to 1961, we find that real land prices have a negative effect on consumption in Japan, controlling for income, financial assets and debt, interest rates, and proxies for uncertainty and for income growth expectations. Thus when real land prices rise, young households and other renters have to save more. This dominates the wealth effect for older households, which we believe is small partly because of the inheritance tax advantages in Japan of leaving housing assets to one’s children. However, for shorter sub-samples in which there is less variation in real land prices, this negative land price effect is weaker than for the full period. Nevertheless, for no period can we find a remotely significant positive effect from physical assets or real land prices on consumption.

In the UK and US, in contrast, there is an important house price channel. In these countries, higher housing collateral results in more borrowing and consumer spending. (The Bank of England’s November 2008 Inflation Report appears to have joined the growing consensus, p 19.) The UK mortgage market is dominated by adjustable-rate mortgages, so reductions in short-term rates feed through quickly into borrowing and house prices. First-time buyers in the UK until recently had access to close to 100% mortgages. As a result, saving for a down payment does not have the priority it has in Japan.

In Japan, lack of competition in banking, the dependence of banks on interest income rather than fees and other sources of profits (see Hoshi and Kashyap (2001, p 284–86) for tables showing bank income sources for the US and Japan), and the non-performing loans problem have kept borrowing rates high relative to deposit rates. Japan does not seem to have experienced credit-market liberalization for (p.178) households on the scale seen in the UK from 1980, and in the US over a longer period.

A second reason for the weak, or even perverse, interest-rate transmission mechanism for households in Japan comes from inter-temporal consumption theory. It says households with a high elasticity of inter-temporal substitution and a low asset to income ratio will experience negative effects on consumption from a rise in the real interest rate, while the opposite is likely to be true for households with the opposite characteristics.

Japanese households have among the highest asset to income ratios in the world, particularly for bank deposits. They may also be particularly cautious in the sense of being averse to fluctuations in consumption. Indeed, we find a very significant and robust positive real interest rate effect in our Japanese consumption function. Thus, the fall in short-term rates after 1993 had a negative direct effect on consumption spending in Japan. However, the later rise in real rates because of falling prices supported consumption.

A possible alternative explanation for a positive real interest rate effect on consumption is omitted variable bias: suppose there had been substantial credit-market liberalization, causing a rise in the ratio of consumption to income, and associated with a rise in real interest rates as credit rationing was replaced by market pricing of credit. We examine evidence from models for household debt to see if, between the late 1970s and the 1980s or later, there was any upward shift in debt that cannot be explained by conventional income, interest rate, and asset price or wealth effects. UK and US evidence supports such shifts, but we find no such evidence for Japan.

This does not mean the interest rate channel is missing for the overall Japanese economy. Financial assets have conventional positive effects on household spending of a size consistent with theory and evidence for other countries. Theory predicts lower interest rates have a positive effect on financial asset prices. But this is offset by the negative direct effect of lower real interest rates on consumption, and the negative indirect effect via higher land prices. Thus, the overall interest channel is far weaker than in the UK or the US.

Evidence from GDP and income forecasting models for Japan show that reductions in nominal interest rates do have a positive effect on output at a one-year horizon. This is consistent with investment, and perhaps exports, responding in the conventional way to lower interest rates and the financial asset price changes they induce.

(p.179) It is important to emphasise that our research does not suggest that raising the policy rate will stimulate economic activity.

Our GDP and income forecasting work has important implications on the efficacy of fiscal policy. We find significant negative effects from fiscal deficit to GDP ratios in recent years on future growth of GDP and income. The forecasts from these models are significant in explaining consumption growth, and suggest that there is an important Ricardian (rational) element in the behavior of Japanese households.

The implication is that both fiscal and monetary policy have had severe limitations in Japan in recent years. This is not, of course, a surprising result, but we provide theoretical and econometric evidence to explain the role of households in this fact.

# 2 Theoretical Foundations of the Consumption Function

To interpret empirical results on the direct and indirect effects of interest rates on consumption, it is crucial to be clear on the controls included in the model, and hence on its interpretation using, for example, the Euler equation or solved out consumption function.

## 2.1 The Euler Equation and Inter-temporal Substitution

For a life-cycle utility function additive in each period’s consumption and a constant inter-temporal elasticity of substitution, the period utility function is or log c if ρ → 0.

The first-order condition for optimization (Euler equation) for a consumer facing a linear budget constraint is

$Display mathematics$
(6.1)

where the inter-temporal elasticity of substitution σ = 1/(1 + ρ), ρ 〉 −1, r is the real interest rate, and d is the subjective discount rate.

Hansen and Singleton (1983) show that, under the assumption of log normal distributions for consumption and the real interest rate r,

$Display mathematics$
(6.2)

Under rational expectations, εt+1 is a stochastic error unpredictable from information at time t.

The inter-temporal elasticity of substitution ⁡ can, in principle, be estimated from this relationship. One could therefore attempt to compare the average rate of substitution in the preferences of consumers in (p.180) different countries. The intuition for the positive coefficient on the real interest rate and on a measure of consumption uncertainty, is that higher rates at t depress consumption at t, and so raise the planned rate of growth of consumption between t and t + 1.

The news aspect of εt+1 was emphasized by Hall (1978), who popularized the Euler equation approach. Tests of the unpredictability of εt+1 soon began to uncover the “excess sensitivity” puzzle in which log changes in consumption are found empirically to be far too sensitive to predictable log changes in income. (See Campbell and Mankiw (1989, 1991) for comprehensive international evidence.) This casts doubt on the validity of the underlying assumptions, and also on the usefulness of equation (6.2) for comparing inter-temporal preferences across countries. If (6.2) is invalid because of excess sensitivity, estimates of ⁡ will be biased by the correlation of the interest rate with the omitted predicted log change in income.

## 2.2 A Basic Life-cycle Model to Examine Interest Rate Effects

In the standard two-period model of household consumption choices, the inter-temporal budget constraint is given by

$Display mathematics$
(6.3)

where all variables are in real terms, c refers to consumption, y to disposable non-property income, A to end-of-period assets, r to the interest rate, W defines life-cycle wealth, and the e superscript means expected.

Assume the utility function is additive and has the CES form. Then

$Display mathematics$
(6.4)

Maximising (6.4) subject to (6.2) gives a first-order condition of the form (6.1). Combining this with the budget constraint (6.3) results in the solved out consumption function

$Display mathematics$
(6.5)

where

$Display mathematics$
(6.6)

and

For small values of δ and r1.

$Display mathematics$
(6.7)

(p.181) Here the inverse marginal propensity to consume out of assets k1 depends on the weighted average of the subjective discount rate δ and the market rate r1. The responsiveness of consumption to the real interest rate—given A0, y1, and $y2e$—can be examined by differentiating the log of equation (6.5) with respect to rx:

$Display mathematics$
(6.8)

A low value of expected discounted income relative to life-cycle wealth (which corresponds to a high share of assets in life-cycle wealth) and a low value of the elasticity of inter-temporal substitution σ make a positive response more likely. On all counts, Japan appears to qualify.

Japanese households have a high ratio of assets, particularly liquid assets, relative to income. Also, liquid assets exceed debt for the aggregate of households; the opposite has been true in the US and UK since the late 1990s, if not earlier. Households in Japan surely have very moderate income growth expectations, given the aging population and the large size of government deficits. Indeed, official growth forecasts have been low for more than a decade.

Finally, σ measures the elasticity of inter-temporal substitution— the lower σ, the less tolerant households are of such consumption fluctuations, and the more precautionary their saving decisions. Japanese households on average are widely thought to exhibit such cautious tastes.

This simple two-period model can also be used to analyze the effect of income uncertainty on consumption decisions. Muellbauer and Lattimore (1995, 250–51) show that the choice problem under income uncertainty can be reduced to an equivalent problem under certainty in which expected income is replaced by certainty-equivalent income. (Also see Kimball (1990), and Gourinchas and Parker (2001) for convincing micro-evidence on the precautionary motive.) This income is defined by expected income divided by a discount factor.

The uncertainty discount increases with income uncertainty, and is greatest for households with the smallest values of σ—those most averse to consumption instability—other things equal. And while large asset (p.182) holdings reduce the discount, a weak social security system increases it. A larger uncertainty discount is like a lower value of expected income, and so tends to reinforce the arguments for a positive response of consumption to higher real interest rates in Japan.

## 2.3 Housing Wealth Effects

We begin by demonstrating the weakness of the housing wealth effect in classical life-cycle theory.

Let c = non-housing consumption, ph = relative price of housing, H = stock of housing, δ = rate of deterioration of housing, r = real interest rate, yp = permanent real non-property income, and A = real financial wealth. The consumer maximizes life-cycle utility defined on the flows of c, and on the stocks H, in each period.

Suppose expected ph and the real interest rate are constant. Then the multi-period inter-temporal optimization problem is just a two-good problem with budget constraint.

$Display mathematics$
(6.9)

where (r + δ)H = housing services and ph(r + δ) = real user cost. (Note that it is a two-good problem by the Hicks aggregation theorem: goods whose relative prices are fixed can be treated like a single good. Here, expected relative prices for consumption at t, t + 1, t + 2, etc, and, similarly, for housing are being assumed fixed. (See Deaton and Muellbauer 1980, p 121.))

We are interested in the effects on a constant-price index of consumption like the one in the national accounts. This includes imputed rent on housing. Holding base prices fixed and differentiating equation (6.9) with respect to ph gives:

$Display mathematics$
(6.10)

But with H ≈ H0, the right side of equation (6.10) is negative, as δ is positive. This point seems to have been overlooked in the classic work by Modigliani and Brumberg (1954), Friedman (1957, 1963), and Ando and Modigliani (1963).

The simple implications of equation (6.10) are liable to be somewhat modified in models with finite lives and transactions costs, and depend on how well imputed rent is measured in the national accounts. Nevertheless, it is hard to place much store on a substantial aggregate housing wealth effect from classical life-cycle permanent-income theory.

## (p.183) 2.4 The Household-Credit Channel

This section discusses how access to mortgage credit interacts with house prices, interest rates, and income growth expectations to influence consumption, and how a change in access to credit changes consumption through two main mechanisms.

The first mechanism concerns the down-payment constraint. Suppliers of mortgage credit set upper limits on loan-to-income and loan-to-value ratios to reduce default risk. This forces young households to save, that is, to consume less than income, the difference depending on the ratio of house prices to income and on the minimum down payment as a fraction of the value of the house. (Note that most potential first-time house buyers saving for a down payment are not credit-constrained in the sense of being unable to smooth consumption. Savings can be run down or increased in anticipation of shorter-term income fluctuations and in response to changes in real interest rates.)

A reduction in credit constraints in the form of a reduction in the minimum down payment as a fraction of the value of the house will raise the consumption of these households relative to income. (See Jappelli and Pagano 1994; Deaton 1999; and micro evidence in Engelhardt 1996.)

Now consider the impact on consumption of higher house prices via the operation of the down-payment constraint. With weak access to credit, potential first-time buyers save more with higher house prices (unless they give up on a house purchase). Increased access to credit will weaken the resulting negative effect on consumption of higher house prices.

Next, consider the second credit channel mechanism, which operates via housing collateral. In a number of countries, the relaxation of rules and spread of competition has made it easier to obtain loans backed by housing equity (Poterba and Manchester 1989). A rise in house prices then makes it possible to increase debt or to refinance other debt at a lower interest rate, given collateral backing. Effectively, the liberalization of credit conditions increases the “spendability” (liquidity) of housing wealth. This gives housing wealth a buffer-stock role.

Combining the down-payment and collateral mechanisms with the life-cycle view relevant for some households, if existing owners have only limited access to home equity loans, the effect of higher house prices on their consumption will be small. Existing owners who are not credit constrained, and whose behavior is governed by the life-cycle model outlined above, taking equation (6.10) literally, will have a small negative response to a real permanent increase in house prices unless they downsize to cheaper accommodation.

(p.184) By life-cycle theory, renters save more with higher house prices, as implied by equation (6.10), when H0 is zero.

Hence, given the above discussion of the down-payment constraint, the aggregate consumption effect of a rise in real house prices is likely to be negative when access to credit is restricted, but switches to positive as access to credit expands.

In countries like the UK where floating rate debt is important, indebted households are subject to short-term shocks to cash flows when nominal interest rates change. (See Jackman and Sutton (1982) for an exposition of the theory.) Their consumption is thus likely to be influenced by changes in their debt-service burden, which can be well-represented by changes in the nominal interest rate weighted by the debt-to-income ratio. Better access to collateral reduces the impact of such changes, as households with positive net equity can more easily refinance to protect cash flows against rises in nominal interest rates. The negative effect of nominal interest rate changes (weighted by the debt-to-income ratio) should thus weaken with credit market liberalization, but increase in a credit crunch.

Greater access to unsecured credit should increase the role of inter-temporal substitution, enhancing the role of income-growth expectations and, on balance, making the real interest-rate effect more negative.

## 2.5 Aggregation Problems and the Incorporation of Demographic Effects

In the stylized solved out multi-period extension of (6.5), where we proxy expected or “permanent” income by current income, micro-level consumption is given by a linear function of assets and non-property income:

$Display mathematics$
(6.11)

where γh, λh vary by age. Hence aggregate or average per capita consumption is

$Display mathematics$
(6.12)

Thus will have non-constant γ*, λ* depending on demography and on the distributions of income and of wealth by demographic groups.

(p.185) Gokhale, Kotlikoff, and Sabelhaus (1996) argue that, in the long run, shifts in γh and Ah by age account for some of the secular decline in the US saving rate. Similar arguments are common in Japan. However, cross-section evidence suggests that ∑h, λh may vary less across households than textbook models imply because of uncertainty about time of death (Bosworth et al 1991; Murata 1999, ch 8). γ*, λ* are likely to evolve slowly over time as the age distribution, distributions of y and A by age, and life expectancies evolve.

Using calibrations broadly consistent with micro data from the Family Saving Survey, Murata (1999, ch 5) finds that aggregate consumption models in which γ*, λ* are constant have very similar implications, and fit as models where they evolve according to sample survey data (discussed later). Furthermore, as households make long-run portfolio decisions, the level and composition of assets is likely to reflect the demographic evolution, implying a smaller direct impact on consumption of shifts in γ*, λ* due to demographic change.

## 2.6 A Solved Out Consumption Function

The Friedman-Ando-Modigliani consumption function requires an income forecasting model to generate permanent non-property income. Unlike the Euler equation, it does not ignore long-run information on income and assets. The solved out consumption function has advantages for policy modeling and forecasting. This basic aggregate life-cycle, permanent-income consumption function has the form:

$Display mathematics$
(6.13)

where c is real per capita consumption, yp is permanent real per capita non-property income, and A is the real per capita level of net wealth.

This equation also has a basic robustness feature missing in the Euler equation. Euler equations require well-informed households continuously trading off efficiently between consuming now and consuming next period. An extension of (6.13) in which current income potentially also plays a role is also consistent with a fairly rudimentary comprehension of life-cycle budget constraints. Any household with some notion of wanting to sustain consumption will realize that not all of assets can be spent now without damaging future consumption, and that future income has a bearing on sustainable consumption. Practical applications of extensions of (6.13) capture these basic ideas.

(p.186) Dividing (6.13) by yt and a little manipulation gives:

$Display mathematics$
(6.14)

The right side of (6.14) has the form 1 + x, where x is usually a fairly small number. This is because one thinks of λ* as being around 1, γ* on the order of 0.03 or 0.04, and asset-to-income ratios on average perhaps 4. The proportionate deviation of permanent from current income on average would be expected not to exceed 20% given historical data on fluctuations in real income and plausible discount rates.

We can then take logs, using the fact that log (1 + x) ≈ x and ≈ $log⁡(ytp/yt)(ytp−yt)/yt$. We then see that

$Display mathematics$
(6.15)

where γ = γ*/λ* and α0 = log λ*. Thus, α0 embodies the evolving distribution of income and demography, while ∑ embodies the evolving relative influences of the asset- and income- distributions and demography. One might attempt to proxy a0 by inclusion of demographic variables such as the population proportions in different age groups. In micro applications where data are more variable, the second-order approximation log (1 + x) ≈ x − 0.5x2 is preferable.

The log ratio of permanent-to-current income reflects expectations of income growth and, in practice, can be proxied by functions of forecast income growth rates. The log formulation is very convenient with exponentially trending macro data, because residuals are likely to be homoscedastic. (Note that the widely used alternative log-linearization of (6.13) using log A cannot cope with negative net assets. It also suffers from large approximation errors for small values of assets, so that splitting log assets into logs of components is not advisable.)

Adding further realistic features, such as habits, roles for variable interest rates and income uncertainty, splitting assets into different types, and introducing a role for the credit channel gives rise to a modern empirical version of the Friedman-Ando-Modigliani consumption function that encompasses the basic life-cycle model given by (6.15). Habits or adjustment costs result in a partial adjustment version of (6.15), where β is the speed of adjustment (Muellbauer 1988):

$Display mathematics$
(6.16)

Economic theory suggests a role for a variable real interest rate r and for income uncertainty 0 because of precautionary behavior, and we include these as linear terms in the extension of (6.16) to (6.17) below. (p.187) This extension also splits total net worth into three components and adds two additional effects which could reflect credit constraints or rule-of-thumb behavior by some households, and the possibility of time variation in some of the parameters induced by shifts in credit availability, discussed further below.

$Display mathematics$
(6.17)

measures income growth expectations. NLA/y is the ratio of liquid assets minus debt to non-property income, IFA/y is the ratio of illiquid financial assets to non-property income, and HA/y is the ratio of housing wealth to non-property income. The cash-flow impact on borrowers of changes in nominal rates is measured by $Δnrt(DBt−1/yt)$ where nr is the nominal interest rate on debt and DB is debt. The speed of adjustment is β, and the γ parameters measure the marginal propensities to consume for each of the three types of assets. The term in the log change of income can be rationalized by an aggregation argument over credit-constrained and unconstrained households (Muellbauer and Lattimore 1995).

Note that β = 1, α1t = 0, α2t = 0, γ1 = γ2 = γ3t = γ, β1t = 0, β2t = 0 and α3t = 1 are the restrictions that result in the basic life-cycle, permanent-income model (equation (6.15)).

The credit channel features through the different marginal propensities to consume (MPC) for net liquid assets (Otsuka 2006) and for housing; through the cash-flow effect for borrowers; and by the possibility of parameter shifts with credit market liberalization.

Credit market liberalization should raise the intercept α0, as saving for a down payment falls, implying a higher level of log(c/y). It should make the real interest rate coefficient α1 more negative, and raise the impact of expected income growth α3, because more opportunities for inter-temporal substitution arise with easier access to credit; and raise the MPC for housing collateral γ3 as access to home equity loans increases. It should also lower the current-income growth effect β1 because this could reflect the presence of credit-constrained households, and the cash-flow impact of the change in the nominal rate β2 because, with easier access to credit, refinancing in the face of a rise in nominal rates is more likely to be possible. In our work on the UK, (Aron et al 2008), we handle these shifts by writing each of these time-varying parameters as a linear function of an index of credit-supply conditions (p.188) (CCI) so that CCI enters the model as an intercept shift and in interaction with several economic variables.

For Japan, empirical versions of (6.17) reduce to a far simpler and more parsimonious model, in part because we can find no significant effects of credit market liberalization and floating rate debt is relatively unimportant.

## 2.7 Review of Literature on Japan

Several studies have examined asset effects on consumption in Japan. Surveys on consumption or saving behavior up to the middle of the 1990s include Hayashi (1997) and Horioka (1993). Horioka (2004) discusses reasons for Japan’s past high saving rate and recent decline.

Most analyses have found the MPC out of assets is around 0.05. Ando et al (1986) used micro data from the National Survey of Family Income and Expenditure and found an MPC out of assets below 0.05. Ogawa et al (1996) examined the asset effects on consumption using prefecture-level data from the same survey, confirming an MPC of around 0.05 for liquid assets. Using the Japanese Panel Survey of Consumers, covering young and middle-aged households, Hori and Shimizutani (2003) found the MPC out of assets to be about 0.05–0.10, with similar effects for liquid assets and shares. Using national accounts data, Horioka (1996) estimated a consumption function based on specifications in Modigliani and Brumberg (1954) and Ando and Modigliani (1963). He obtained an MPC for net worth of 0.02–0.04 at the sample mean.

The effects of financial imperfections or down-payment constraints have also been examined. Hayashi et al (1988) present a life-cycle simulation analysis, comparing saving rates in Japan and the United States, suggesting that while it was likely that Japanese households saved more early in the life cycle to meet the higher down-payment requirement, the contribution of the early saving appears too small to explain a large differential in the aggregate household saving rate between Japan and the US. However, the study may have understated differences in the down-payment constraint between the two countries. Moriizumi (2003) finds micro evidence for a large effect on saving by young households in Japan linked to the constraint.

Horioka (1988) focused on saving by motive, concluding it was likely that housing-related saving, including down-payments, was considerable. However, this was approximately offset by dissaving in the form of depreciation of the housing stock.

(p.189) Concerning interest-rate effects on consumption, in a leading study on the permanent-income hypothesis as applied to Japan, Hayashi (1985) estimated Euler equations using national accounts data and found the coefficient on the interest rate was not statistically significant (except for durables). Nakagawa (1999) estimated an equation explaining the saving rate in terms of the real interest rate, income risk, and income-growth expectations using aggregate data for income quintile groups and obtained a positive and significant interest rate effect on consumption for higher-income households (negative effect on the saving rate). Nakagawa and Oshima (2000) estimated consumption functions following the consumption-CAPM, using national accounts data for the US, UK, France, and Japan. They found that the coefficients on the real interest rate were negative and significant for the US and UK, negative but insignificant for France, and positive but insignificant for Japan.

Horioka (2006) has argued that the stagnation of Japan’s consumption during the 1990s is attributable to the stagnation of household disposable income, a decline in household wealth, and increased uncertainty about the future. However, household income is itself endogenous. Unfortunately, he does not conduct an empirical analysis in his paper. There has been little empirical investigation of the role of increased uncertainty in the stagnation of household consumption during this period. Exceptions are Murata (2003) and Saito and Shiratsuka (2003).

On monetary policy transmission, Ito and Mishkin (2004) and Hamada (2004), for example, implicitly accept the conventional view that monetary policy via households in Japan works much as in the US and UK. There has not been a rigorous examination of whether this is correct. Horioka (2004) suggests consumer behavior in Japan is similar to Continental Europe but different from the US and UK. However, he does not discuss the implications for monetary policy transmission.

Drawing on joint research reported in Aron et al (2008), Muellbauer (2007) applied the consumption function outlined in section 2.5 to UK and US data. He finds an MPC out of net liquid assets of around 0.1, and around 0.02 for illiquid financial assets. He also found an MPC increasing with a measure of credit availability for housing wealth, reaching a maximum of around 0.03 for the UK and even higher for the US. In both countries there is evidence for a negative real interest-rate effect on consumption. For the UK, where floating rate mortgages dominate, there is an important negative time-varying effect from the debt-weighted change in nominal borrowing rates. This is consistent with the discussion in (p.190) section 2.6. In both countries, forecast income-growth rates are quite significant in explaining consumption.

# 3 Empirical Results for Consumption

This section discusses data and measurement issues, models income growth expectations, and presents results for Euler equations and aggregate consumption.

## 3.1 Data and Measurement Issues

### 3.1.1 Consumption Data

Our consumption expenditure series is defined as “actual final consumption” which consists of “final consumption expenditure” plus “individual consumption by the government,” such as medical expenses paid by pension funds, textbooks at school, and the like, plus consumption by non-profit institutions (which is negligible).

### 3.1.2 Non-property Income

Inter-temporal consumer theory uses a concept of non-property income. The national accounts define personal disposable income (PDI) as the sum of labor, transfer, and property income, after taxes and subsidies, and operating surplus (after tax). To measure non-property income (NPDI) we subtract after-tax property income and part of operating surplus from PDI, broadly following Blinder and Deaton (1985) as explained in annex 6.1.

Figure 6.1 shows the log ratio of consumer expenditure to non-property income. This is a preferable measure to the log ratio of consumer expenditure to PDI, which is approximately 1 − s, where s is the household saving ratio. The reason is that the property income component of PDI is distorted by inflation. A large element of property income is interest on deposits, which depends on the nominal interest rate. With the real rate of interest constant, a decline in inflation reduces measured PDI even when the real budget constraint is unchanged.

The figure also reveals a negative correlation between the log real land price and the other two ratios.

The changing demographic structure of Japan has often been linked with the decline in Japan’s household saving rate and the positive trend seen in figure 6.1. (See Horioka (1997) for the most striking claims.) Data revisions in 2006 resulted in a substantial upward revision of the consumption-to-income ratio (downward revision of the household saving ratio) (Masubuchi 2006).

(p.191)

Figure 6.1 Ratios of Consumption to Total Disposable Income and to Non-property income, and

Real Land Price, 1955–2006.

Note: All variables are in logs.

Source: Cabinet Office; Japan Real Estate Institute.

Figure 6.2 shows the falling share of the population aged 14 or less, the rising proportion aged 65 or more, as well as shares of the population aged 45 to 59 and 40 to 64. Drawing robust conclusions about the demographic shift’s role in explaining the consumption-to-income ratio is difficult. (This is because the population shares are I(2) (integrated of order two so that the series need to be differenced twice to make them stationary) variables, while the consumption-to-income ratio is I(1), so needing to be differenced only once to achieve stationarity.)

Prima facie, it is not obvious whether there is any net effect: while the consumption needs of the young may be less than those of the elderly, their proportionate decline is also greater.

Moreover, life-expectancy in Japan has risen steadily in this period (figure 6.3). An aging population is supposed to raise the average saving rate according to the Modigliani life-cycle hypothesis, as the proportion of elderly dissavers rises. However, rising life expectancy with (p.192)

Figure 6.2 Shares of Population, 1955–2006.

Source: Ministry of Internal Affairs and Communications.

Figure 6.3 Life Expectancies in Japan.

Note: Simple averages of males and females.

Source: Ministry of Health, Labour and Welfare.

(p.193) which it is positively correlated, has the opposite effect since it increases the need of the working population to save, see Murata (1999 ch 8) for some evidence on the effect of rising life-expectancy on the consumption-to-income ratio.

Data from other countries suggest that those in the decade or two before retirement tend to have the highest saving rates. The peaking in these proportions in the 1990s could be associated with the slight dip or flattening of the consumption-to-income ratio in that decade, and its later rise. But that is pure speculation at this juncture; robust conclusions may not be available, even with a more complete model.

### 3.1.3 Wealth and Debt Data

Japan has produced flow of funds accounts and balance sheet data since the 1950s. These are available quarterly from 1964. End-of-year balance sheet data are published as part of the extended national income and expenditure accounts. The latter include estimates of physical assets, and can differ somewhat in other dimensions, including estimates of shares in companies which are not publicly traded, among other differences. As is the usual case, the household sector includes unincorporated businesses.

Data have been published on two bases: the 1968 SNA and the 1993 SNA. On a 1968 basis, rather less detail is available, particularly on the debt side. For example, mortgage and non-mortgage debt are not reported separately. On a 1993 basis, not only are mortgages separated, there is information on unincorporated business debt back to 1979.

This poses the issue of what definition of debt to adopt. One is total household liabilities. The data back to 1979 indicate that around 30% to 40% of these are debt of unincorporated businesses. A second definition of household debt excludes these loans. To generate the data before 1979, we subtract a fixed 41% of debt, the proportion in 1979.

Assets can be divided several different ways. We follow the division set out in section 2.5. Liquid financial assets are currency and deposits, and illiquid financial assets are the sum of shares, pension funds, and other financial assets, including bonds. This means two definitions of net liquid assets: liquid assets minus the two definitions of debt.

Physical assets are dominated by land. We take the data back to 1955, splicing with the 1968 SNA data. The data on shares pose some problems. Appendix 6.1 explains how we created a market-value series.

Figure 6.4 shows that assets substantially exceed debt, and this is true even with the broadest definition of household debt. Further, the net liquid asset ratio does not show the strong negative trends seen in (p.194)

Figure 6.4 Ratio to Non-property Disposable Income of Selected Balance Sheet Items.

Debt excludes unincorporated business debt (as explained in the text).

Source: Cabinet Office; Bank of Japan.

the US and UK equivalents illustrated in Muellbauer (2007). On the face of it, there seems little sign here of the household credit market liberalization seen in many other countries.

The ratios to non-property income of illiquid financial assets and physical assets are shown in figure 6.5. The latter is clearly correlated with the real land price index, also shown. The peak of physical assets relative to income in 1992 substantially exceeded even the 2007 peak of the UK equivalent, but illiquid financial assets relative to income have been lower than in the UK, and especially lower than in the US with its substantial direct and indirect ownership of equities.

### 3.1.4 Other Data

For the representative short-term interest rate we take the overnight call rate (Rc), which is available for the full sample. Rc is measured as an annual percentage rate divided by 100. The real tax-adjusted interest rate (RCR) is defined by Rc(1 − tax rate) − Δlog PC. Here the tax rate is the property tax rate and PC is the consumer expenditure deflator.

One indicator of income insecurity is taken to be the change in the unemployment rate (DUR), which is highly relevant in explaining UK (p.195)

Figure 6.5 Ratios to Non-property Income of Assets, the Log Real Price of Land and the Real Interest Rate.

Note: The real interest rate is tax adjusted. The real land price is in logs.

Source: Cabinet Office; Bank of Japan; Japan Real Estate Institute.

and US consumption. The other indicator of income insecurity or volatility, SY, is defined to be the absolute value of the deviation between the current growth rate of real per capita non-property income, Δlog y, and its average growth rate over the previous 5 years. Figure 6.6 displays the growth rate of income and the two indicators of income insecurity.

## 3.2 Modeling Income Growth Expectations

Income-growth expectations are a central feature of inter-temporal consumption models. Indeed, it has sometimes been claimed that housing wealth or collateral effects on consumption are an illusion, being merely a proxy for omitted income growth expectations. (See King (1990) commenting on Muellbauer and Murphy (1990), Attanasio and Weber (1994), and Attanasio et al (2009). However, the last two fail to control for current income, perhaps the most obvious driver of consumption.)

Poterba (2000) similarly makes the point that the size of the stock market wealth effect on consumption depends on the source of the asset-price (p.196)

Figure 6.6 Income Growth, Income Volatility, and the Change in the Unemployment Rate.

Note: Income is real per capita non-property income.

Source: Cabinet Office; Ministry of International Affairs and Communications.

shock, as some shocks may signal a shift in income expectations. This makes it important to control for income growth expectations.

We follow Muellbauer (1996), which forecasts US income growth using a general-to-specific methodology in paring down a very general model to a parsimonious form. The general model includes a trend, a split trend from 1973 for the slowdown in Japanese growth which occurred then, and the level of log real per capita income to capture trend reversion. Other variables include log US GDP, the log real exchange rate, log real oil prices, log real asset prices, the real interest rate, change in the nominal interest rate, and the government surplus- and debt-to-GDP ratios. Table 6.1 reports the parsimonious specifications found after testing down.

There is strong evidence for reversion to the split-growth trend and the 3-year moving average of government balance to GDP has a positive coefficient. The table also shows an alternative specification in which lags in the ratio of government debt to GDP replace government balance to GDP. The lagged government debt to GDP ratios have highly significant coefficients, negative in the long run, as shown in the last column.

(p.197)

Table 6.1 Estimates of the Forecasting Equation for Change in log y+1

Dependent variable: Change in log y+1

1959–2005

1959–1992

1975–2005

1959–2005

(1)

(2)

(3)

(4)

(5)

Intercept

−3.814 ***

−3.547 ***

−4.121 ***

−3.144 ***

−4.440 ***

(0.693)

(0.624)

(0.726)

(0.640)

(0.669)

Trend

0.028 **

0.024 ***

0.025 **

0.026 **

(0.010)

(0.008)

(0.011)

(0.012)

Split trend at 1973

−0.024 ***

−0.022 ***

−0.023 ***

−0.024 ***

(0.006)

(0.005)

(0.007)

(0.007)

log y

−0.458 ***

−0.404 ***

−0.446 ***

−0.356 ***

−0.473

(0.100)

(0.068)

(0.097)

(0.068)

(0.111)

3-year change in nominal call rate

−0.199 ***

−0.212 ***

−0.203 **

−0.159 **

−0.235 ***

(0.064)

(0.062)

(0.076)

(0.057)

(0.061)

log US GDP−1

0.174 *

0.178 *

0.223 **

0.205 ***

0.246 ***

(0.088)

(0.090)

(0.100)

(0.043)

(0.089)

(MA3Gov.ba/GDP)−1

0.538 ***

0.626 ***

0.593 ***

0.612 ***

(0.102)

(0.109)

(0.135)

(0.105)

(Gov debt/GDP)−1

−0.032

−0.165 ***

(0.028)

(0.032)

(Gov debt/GDP)−4

0.108 ***

(0.018)

Standard error × 100

1.195

1.195

1.356

0.910

1.179

0.873

0.873

0.840

0.603

0.876

Durbin Watson

1.93

2.03

2.01

2.24

1.91

AR1/MA1 (p-value)

0.828

0.877

0.899

0.475

0.789

AR2/MA2 (p-value)

0.258

0.149

0.138

0.125

0.118

Heteroscedasticity

0.001

0.003

0.035

0.286

0.002

(p-value)

Chow (p)

0.849

0.674

0.509

0.706

0.860

RESET(p)

0.821

0.284

0.638

0.494

0.748

Note: Standard errors are given in parentheses. For the equations (1), (2), (3) and (5) whose heteroscedasiticity tests have failed, reflecting the greater volatility of pre-1975 growth, the robust standard errors are given in parentheses. ***, ** and * indicate the statistical significance of independent variables, at the 1 percent, 5 percent and 10 percent levels, respectively.

The table was created with the help of Autometrics software, Doornik (2007).

(p.198) Real oil prices are not significant, though they are if US log GDP is omitted. The influence of the US as leading economy and key trading partner for Japan is confirmed. The change in nominal interest rates over the previous three years has a negative effect on next year’s household income growth, suggesting that, overall, monetary policy has some effect on growth. However, the real interest rate is insignificant, and positively signed. Columns 2 and 3 show the income forecasting equation fitted over alternative samples, showing remarkable parameter stability.

The actual and fitted values and residuals from the parsimonious model (the second column) are shown in figure A6.1, and the recursive stability tests in figure A6.2 in the annex. The latter are quite satisfactory.

## 3.3 Results for Euler Equations

The unpredictable-news feature of the residual in a consumption Euler equation is a key implication of the theory. However, Campbell and Mankiw (1989, 1991) showed that this prediction was rejected in virtually every country they studied because predictable income growth, or “excess sensitivity,” proved highly significant in explaining consumption growth. They are careful to note that since time aggregation and transitory noise in consumption induce first-order autocorrelation in the residuals, instruments dated t − 2 should be used to test the news hypothesis. We therefore use a simplified version of the income forecasting equation used in table 6.1 to generate the income growth forecast.

In this version, log real per capita income and the change in the nominal interest rate are lagged one year more. Then the interest rate term and log US GDP become insignificant and are omitted. The fitted value from this equation is lagged one year, thus embodying information lagged two years. Similar results are obtained whether the equation variants using government surplus to GDP or government debt to GDP are used.

Strictly speaking, the Euler equation does not apply to expenditure on durable goods. Hence table 6.2 and table 6.3 report results for the excess sensitivity test for Euler equations for non-durable goods and services as well as for total consumption.

In both tables the instrumented log change in real per capita income is highly significant—see columns 2 to 4 showing variants including (p.199)

Table 6.2 Estimates of Consumption Euler Equations for Non Durables Plus Ser vices

Dependent variable: Change in log non-durable consumption

1963–2006

1963–2006

1963–2006

1963–2006

1981–2006

1981–2006

(1)

(2)

(3)

(4)

(5)

(6)

Intercept

0.001

0.001

0.006 **

0.001

0.001 **

0.009 ***

(0.008)

(0.004)

(0.003)

(0.004)

(0.000)

(0.003)

Change in log y

0.641 ***

0.762 ***

0.653 ***

0.347 *

0.596 ***

(0.078)

(0.061)

(0.079)

(0.178)

(0.171)

Real interest rate_1

0.657 ***

0.226 **

0.180 **

0.207 ***

(0.143)

(0.088)

(0.088)

(0.064)

Income growth

0.167

volatility 2

(0.138)

Standard error × 100

2.956

1.246

1.221

1.234

0.738

0.795

0.227

0.790

0.798

0.782

0.523

0.441

Durbin Watson

1.04

1.74

1.59

1.61

2.52

1.83

Table 6.3 Estimates of Consumption Euler Equations for Total Consumer Expenditure

Dependent variable: Change in log consumption

1960–2006

1961–2006

1961–2006

1961–2006

(1)

(2)

(3)

(4)

Intercept

0.008

0.006

0.009

0.005

(0.008)

(0.004)

(0.003)

(0.004)

Change in log y

0.689 ***

0.770 ***

0.696 ***

(0.082)

(0.062)

(0.083)

0.614 ***

0.152

0.101

(0.154)

(0.092)

(0.093)

Income growth volatility_2

0.184

(0.145)

Standard error × 100

2.752

1.314

1.252

1.302

0.163

0.781

0.801

0.771

Durbin Watson

1.01

1.70

1.69

1.58

(p.200) and excluding the real interest rate and an income volatility measure. Column 1 shows the Euler equation including only the real interest rate. Though the real interest rate is very significant, columns 2 and 4 suggest that the coefficient is grossly upward biased because of its correlation with instrumented income growth omitted from column 1. Estimated Euler equations of the kind shown in column 1 therefore offer no basis for comparing estimates of the elasticity of inter-temporal substitution across countries.

The overwhelming evidence for excess sensitivity is consistent with violations of some of the key assumptions behind the Euler approach, including rational expectations and the absence of credit or liquidity constraints. This is another powerful reason for preferring the augmented solved out consumption function approach set out in section 2.6.

Habits have sometimes between proposed as a potential reason for excess sensitivity. A Euler equation with habits includes lagged consumption growth as a regressor. However, when instrumented current income growth is included in runs of table 6.2 and table 6.3, lagged consumption growth is completely insignificant, with a coefficient close to zero.

## 3.4 Results for Aggregate Consumption

Our aim is to estimate for Japan variants of equation (6.17) discussed in section 2.6. The results are in table 6.4.

We use annual data from 1961 to 2006. In slight modification, we also include the lagged log real land price. It quickly becomes apparent that the ratio of physical assets to income and the real land price have negative coefficients. We therefore report equations in which each is included separately. Further, we cannot reject the hypothesis that the marginal propensities to spend out of deposits and illiquid financial assets are the same, and are equal to minus the coefficient on household debt. This may be because “deposits” includes a substantial amount of longer-term time deposits which are therefore not so liquid. At any rate, we can work with net financial wealth, which is always very significant and with a long-run marginal propensity to consume of around 0.05 to 0.07. This is consistent with estimates reported by Ogawa et al (1996) and Hori and Shimizutani (2003) discussed in section 2.7. The fitted values obtained from table 6.1 column 2 are taken as proxies for expected income growth, and are strongly significant. The measure of income volatility is significant, as shown in table 6.4 column 1.

(p.201)

Table 6.4 Estimates of the Solved-Out Consumption Function

Dependent variable: Change in log c

1961–2006

(1)

(2)

(3)

(4)

Intercept

−0.055 ***

−0.057 ***

−0.058 ***

−0.063 ***

(0.017)

(0.016)

(0.015)

(0.015)

log y − log c−1

0.356 ***

0.345 ***

0.347 ***

0.359 ***

(0.067)

(0.064)

(0.063)

(0.064)

Change in log y

0.289 ***

0.321 ***

0.323 ***

0.332 ***

(0.070)

(0.068)

(0.067)

(0.068)

Forecast income

0.367 ***

0.347 ***

0.348 ***

0.350 ***

growth rate

(0.083)

(0.079)

(0.078)

(0.079)

Income growth

−0.225 **

−0.024

volatility

(0.094)

(0.128)

Income growth

−5.666 **

−6.007 ***

−5.648 ***

volatility × forecast

(2.574)

(1.782)

(1.796)

income growth rate

Change in

−0.008

−0.007

−0.007

unemployment rate

(0.005)

(0.005)

(0.004)

Real interest rate

0.346 ***

0.346 ***

0.350 ***

0.367 ***

(0.062)

(0.059)

(0.054)

(0.054)

Net financial

0.022 ***

0.022 ***

0.023 ***

0.024 ***

wealth_1/income

(0.006)

(0.005)

(0.005)

(0.005)

log real land price_1

−0.014 ***

−0.015 ***

−0.015 ***

−0.016 ***

(0.004)

(0.004)

(0.004)

(0.004)

Standard error × 100

0.681

0.648

0.640

0.650

0.941

0.947

0.948

0.946

Durbin Watson

2.14

2.20

2.20

2.22

AR1/MA1 (p-value)

0.621

0.386

0.417

0.396

AR2/MA2 (p-value)

0.742

0.711

0.717

0.726

Heteroscedasticity

0.737

0.849

0.829

0.955

(p-value)

Chow (p)

0.255

0.191

0.298

0.635

RESET(p)

0.066

0.445

0.576

0.827

In table 6.4 column 2 the cross term of income volatility and expected income growth rate is added, as the theory discussed in section 2.6 suggests that greater income uncertainty should lead to a bigger discount on expected growth. When both are included, the cross term is found to be significant, while income volatility is insignificant and so excluded in columns 3 and 4.

(p.202) The change in the unemployment rate is not significant at the 5% level, probably because of its more limited variability in Japan, in contrast to its far more significant role in the UK and the US (see Muellbauer 2007). However, the sign is negative and the magnitude of the coefficient is not far below UK and US estimates.

The change in the nominal interest rate is always insignificant, unlike in the UK, but the level of the real rate has a strongly significant positive effect. This is not a disguised inflation effect, as the inflation rate is insignificant when included, while the real rate remains significant. Aron et al (2008) and Muellbauer (2007) find negative real interest rate effects in similar specifications estimated for the UK and US.

The log change in income has a positive and significant effect. This is also in contrast to UK and US findings, where this effect is not significant. The argument comes from applying the Campbell-Mankiw aggregation of credit constrained and unconstrained households to a solved out consumption function (see Muellbauer and Lattimore 1995). On this interpretation, the proportion of total income in income-constrained households n is given by (1 − β)π=β1 where the coefficient on the change in log income is β1 and β is the speed of adjustment.

Just over half of Japanese consumption comes from households who are, or behave as if they were, income constrained if one uses the speed of adjustment (0.359) and β1 (0.332) from table 6.4 column 4. This is not far from previous estimates (see Hayashi 1997). However, given the somewhat unsatisfactory micro foundations for the Campbell-Mankiw story, it is probably a mistake to interpret this too literally in terms of credit constraints (see Carroll 2001; Aron et al 2008). The fit of the equation over the full sample is shown in figure A6.3 (see annex).

## 3.5 Co-integration Results

To investigate the co-integration properties of the data on the provisional hypothesis that current income growth (DLRY) is weakly exogenous, we set up a five-equation system with the log ratio of consumption to non-property income (LRCY), forecast income growth (EDLRY), real interest rate (tax adjusted) (TRCR), the relative price of land (LRPLAND), and the net financial asset to income ratio (NFAY), and include DLRY and the cross term of income volatility and forecast income growth (SEDLRY) as unrestricted I(0) variables. With the constant in the cointegrating space, we find there are two cointegrating vectors, one of which can be interpreted as a consumption function: (p.203)

$Display mathematics$
(6.18)

Unadjusted tests for the number of cointegrating vectors are marginal for the hypothesis of two cointegrating vectors versus three. However, the inclusion of I(0) variables in the model biases up the standard test statistics (Rahbeck and Mosconi 1999). Hence two is almost certainly correct. The significance of α for the second cointegrating vector ensures that the rank cannot be less than two.

The coefficients obtained in (6.18) were tested and found to be statistically not different from the long-run coefficients obtained in table 6.4 column 4, though the point estimate for the real land price effect is marginally more negative, while that for the ratio of net financial assets to income is marginally smaller. Since theory suggests an upper bound of 1 on the expected income growth coefficient, a value of 1 was imposed on the coefficient, an easily acceptable restriction.

In obtaining (6.18), α in the second vector, was assumed to be zero, an acceptable restriction. In addition, following Harbo et al (1998), to check the weak exogeneity of current income growth (DLRY), the following regression (see (6.19)) shows DLRY is unrelated to the cointegrating vector, and so passes the test (b0 is not statistically significant):

Table 6.5 Co-integration Results

Eigenvalue

0.782

0.545

0.311

0.210

0.150

Hypothesis

r = 0

r〈=1

r〈=2

r〈=3

r〈=4

λmax

62.5**

32.28*

15.26

9.67

6.66

λtrace

126.4**

63.86**

31.58

16.32

6.66

Beta (co-integrating vectors):

LRCY

TRCR

LRPLAND

FNFAY

DLRY1FIA

Constant

1.0000

−1.0432

0.0477

−0.0715

−1.0000

0.1824

0.4115

1.0000

−0.0164

−0.0453

0.0192

0.0693

alpha

LRCY

−0.2934

0.0000

TRCR

0.2303

−0.1674

Standard errors of alpha

LRCY

0.04603

0.0000

TRCR

0.0881

0.0421

LR test, rank = 2: Chi2(1) = 1.324 [0.259]

(p.204)

Table 6.6 Robustness Check for the Solved-Out Consumption Function

Dependent variable: Change in log c

1961–1992

1975–2006

1961–2006

1975–2006

1961–2006 (IV)

(1)

(2)

(3)

(4)

(5)

Intercept

−0.084 ***

−0.087 ***

−0.069 ***

−0.089 ***

−0.077 ***

(0.018)

(0.020)

(0.016)

(0.022)

(0.021)

log y − log c−1

0.435 ***

0.483 ***

0.372 ***

0.476 ***

0.409 ***

(0.073)

(0.094)

(0.067)

(0.098)

(0.088)

Change in log y

0.267 ***

0.269 **

0.320 ***

0.276 **

0.378 ***

(0.073)

(0.126)

(0.070)

(0.131)

(0.106)

Forecast income growth

0.446 ***

0.129

0.354 ***

0.131

rate

(0.089)

(0.121)

(0.080)

(0.123)

Change in log y+1

0.309 ***

(0.117)

Income growth volatility

−4.970 ***

0.630

−5.610 ***

0.991

−5.270 **

× forecast income growth

(1.728)

(6.277)

(1.809)

(6.521)

(2.213)

rate

Real interest rate

0.406 ***

0.574 ***

0.359 ***

0.565 ***

0.398 ***

(0.067)

(0.108)

(0.055)

(0.114)

(0.077)

Net financial wealth−1/

0.032 ***

0.033 ***

0.026 ***

0.034 ***

0.029 ***

income

(0.007)

(0.007)

(0.006)

(0.008)

(0.007)

log real land price−1

−0.017 ***

−0.006

−0.018 ***

−0.013

−0.019 ***

(0.006)

(0.009)

(0.005)

(0.025)

(0.005)

log real land price−1

0.012

0.011

× Step dummy with 1

(0.018)

(0.038)

since 1991

Standard error × 100

0.611

0.618

0.655

0.630

0.733

0.952

0.804

0.946

0.797

0.934

Durbin Watson

2.23

2.13

2.19

2.15

2.21

AR1/MA1 (p-value)

0.374

0.709

0.439

0.664

AR2/MA2 (p-value)

0.888

0.874

0.828

0.852

Heteroscedasticity

0.218

0.237

0.891

0.229

(p-value)

Chow (p)

0.020

0.361

0.797

0.536

RESET(p)

0.772

0.064

0.872

0.073

Note: Standard errors are given in parentheses. ***, ** and * indicate the statistical significance of independent variables, at the 1 percent, 5 percent and 10 percent levels, respectively. See Appendix 2 for the equation (5).

(p.205)

$Display mathematics$
(6.19)

where ECM = LRCY − 1.0432 * TRCR − 1.00 * EDLRY − 0.04768 * LRPLAND+ 0.07151 * NFAY; D73 is a dummy variable with one in 1973 and zero elsewhere; and S73 is a step dummy with zero up to 1972 and one onwards.

Similarly, the cross term of income volatility and forecast income growth (SEDLRY) passes the test using the same variables shown on the right side of (6.19).

Table 6.6 column 5 shows the results estimated by instrumental variables, further to test whether the results are biased because of possible endogeneity problems, particularly with respect to income and the real interest rate. The coefficients are similar to those from OLS and from the cointegration analysis, suggesting no endogeneity problem, consistent with the weak exogeneity tests. The instrumenting strategy takes great care to use efficient and plausible instruments (annex 6.2).

## 3.6 Other Robustness Checks

Other robustness checks on the table 6.4 results are also reported in table 6.6. Parameter stability is seen in estimates over different samples and in the recursive betas shown in figure A6.4 (see annex). These provide clear support for the long-term relevance of the model, though in short samples the real land price effect loses significance, given its lack of short-term variability.

This raises the question of whether there may have been a structural break in the coefficient on log real land price. We test for this by interacting the lagged log real land price with two step dummies. The first is zero until 1980 and one from 1981; the other is zero until 1990 and one from 1991. To avoid a jump in the interaction effect in, for example, 1991, the 1991 step dummy is multiplied by the lagged log land price index minus its 1990 value.

The results hardly alter when the step dummy beginning in 1991 is replaced by one beginning in 1981. The coefficient on the step dummy interaction effect is not significant, with a t-ratio of 0.7. The point estimate is consistent with a small amelioration in the negative impact of land prices on consumption after 1991 (and indeed after 1981). But we can easily accept the hypothesis of constancy of the negative real land price effect.

In Japan’s national accounts, the series based on 1993 SNA are available back to 1979. That series and the 1968 SNA were spliced with the (p.206) ratio in 1980 (see annex 6.1). In order to test the possibility that the results were biased due to this method of data adjustment, the equations shown in table 6.1 column 2 and table 6.4 columns 1 and 2 were estimated to 1998 using the 1968 SNA series. The results were not statistically different from those shown in the corresponding equations estimated to 2006. (We are grateful to Charles Horioka for suggesting this robustness check.)

Lower income growth and the uncertainty indicators explain some of the dramatic decline in the consumption to income ratio in the 1970s. The long-run contributions of the four I(1) explanatory variables—the net financial wealth to income ratio, the log real land price, the real interest rate, and the forecast growth rate of income—are shown in figure 6.7 and figure 6.8.

It is clear from these figures that the rise of the consumption to income ratio is very much driven by the rise in net financial assets owned by households, only somewhat offset by the rise in real land prices. Interestingly, net financial assets relative to income shows rather little cyclical variation, as the pension fund component is not very sensitive to the stock market, though its decline in the early 1990s also contributed to the drop in the consumption ratio then.

Figure 6.7 Long-run Contribution to Log Consumption/Income of Net Financial Assets/Income and the Log Real Price of Land.

(p.207)

Figure 6.8 Long-run Contribution to Log Consumption/Income of Real Interest Rate and Forecast Income Growth.

It is a striking fact that the four demographic shares plotted in figure 6.2 are jointly and individually insignificant when included in the consumption equation. Moreover, the signs often make little sense. What does make more sense is the inclusion of the proportion of the adult population aged 25 to 44. These are the main savers for a housing down payment. A rise in their proportion tends to lower consumption relative to income, though the effect still only has t = −1.1. Micro evidence indicates that older Japanese households tend to carry on saving until their 80s (Hayashi 1997), but perhaps the 25 to 44 group are the biggest savers.

This does not mean demographic developments are irrelevant for aggregate consumption in Japan. Accumulation of financial wealth has surely been, in part, driven by the aging of the population and lengthening of life-expectancy. Consumption or saving, conditional on such portfolio accumulations, is always less likely to be so sensitive to demographic structure.

We also carried out a calibration exercise based on equations (6.11) and (6.12), defining three age groups and making plausible assumptions about differences by age in the marginal propensities to consume (p.208) out of income and assets. Using annual household survey data, we constructed time varying aggregate marginal propensities:

$Display mathematics$

Calibrations with small age differences in the micro propensities fit better than those with larger age differences. But the results differ very little from those reported for time-invariant marginal propensities in table 6.4.

## 3.7 Overall Interpretation

There is a wide-spread view in the profession that the Euler equation is the “structural relationship,” while solved out models are “reduced forms.” One response is that solved out models defined by the life-cycle model are combinations of the Euler equations and the life-cycle budget constraint, and are therefore just as structural as an Euler equation. The second response is that the Euler equation is overwhelmingly rejected by the data, while the augmented solved out model fits the data far better and has stable parameters for a 47-year period.

We believe our consumption function is one very useful element in a system of equations, which include equations for income, portfolio allocation, asset prices, the government budget, and the conventional ingredients of a macro-econometric model. On the basis of such a multi-equation model, one can examine the impact of shocks and policy changes of various types.

Our approach is preferable to the recently fashionable New Keynesian version of a DSGE model with the consumption Euler equation at its core, given the latter’s rejection by the data, and its omission of the credit channel, and of any economic role for asset prices. And, Pierre-Olivier Gourinchas’s concerns about the endogeneity of income are handled in our model.

As far as estimation bias is concerned, we have shown that there is no evidence of a bias by appropriate instrumentation. While annual income is clearly endogenous for current consumption, it is far less obvious that the current growth rate of income is endogenous for the log ratio of consumption to income, or indeed in which direction any endogeneity bias might point. The cointegration evidence is consistent with weak exogeneity of current income growth for the log ratio of consumption to income. (p.209)

# 4 Empirical Results for Debt

We argue in the previous section that inter-temporal consumer theory can explain a positive effect of the short-term real interest rate on consumption, and found strong evidence for such an effect in Japan. One alternative hypothesis that might explain a positive interest-rate effect is the rise in real interest rates that can accompany credit market liberalization, as in the UK between the end of the 1970s and the mid-1980s. Credit market liberalization raises consumption, so omitting such an effect could bias the coefficient on the real interest rate upwards. Indeed, this probably explains why almost a generation of UK researchers had difficulty finding the negative interest-rate effect they were expecting in UK consumption functions.

The question therefore arises of whether Japan may have gone through a similar liberalization episode. We have seen circumstantial evidence against this view. Japan shows a continued rise in the ratio to income of liquid assets minus debt (figure 6.4), unlike in the US and UK where these declined from the 1980s. We also found stable parameter estimates for the consumption function. Had there been significant liberalization of credit markets, we would expect at least some of the parameter shifts discussed below equation (6.17), for which UK evidence is strong.

Now we develop a model for the household debt to income ratio to see if a stable model can be found without the sorts of shifts one requires to make sense of the growth of UK household debt (see Fernandez-Corugedo and Muellbauer 2006). In fact, a relatively simple and stable model explains the rise of the ratio of debt to income in Japan in terms of two asset-to-income ratios, the nominal interest rate, the change in the unemployment rate, and the rate of acceleration of the log consumer expenditure deflator. (One interpretation of the last factor is an indicator of risk of rises in nominal interest rates.)

The model includes a pre-1974 trend to reflect the earlier development of the financial system, which seems to have reached a fairly mature level by the mid-1970s. The ratios to income of physical assets and illiquid financial assets are all highly significant—as in the UK, South Africa, and US.

The model, shown in table 6.7, was developed from a general equation also including real interest rates, income uncertainty indicators, forecast income growth, the income level, and more complex dynamics. The fit, and the recursive stability tests, are very satisfactory (see (p.210)

Table 6.7 Estimates of the Equation for Ratio of House hold Debt to Income

Dependent variable: Debt/y

1961–2006

1961–1992

1975–2006

(1)

(2)

(3)

Intercept

−0.046 **

−0.072 **

0.073 **

(0.019)

(0.028)

(0.015)

Ratio of debt−1 to y

0.506 ***

0.407 ***

0.459 ***

(0.061)

(0.084)

(0.080)

Trend pre-1974

0.0063 ***

0.0075 ***

(0.0008)

(0.0012)

Change in unemployment rate

−0.013 **

−0.008

−0.007

(0.006)

(0.015)

(0.007)

Rate of acceleration of log

−0.202 ***

−0.198 ***

0.033

consumer expenditure

(0.061)

(0.066)

(0.142)

deflator

Nominal call rate−1

−0.364 ***

−0.352 ***

−0.264 **

(0.086)

(0.125)

(0.127)

Physical wealthy/income

0.030 ***

0.029 ***

0.030 ***

(0.002)

(0.006)

(0.003)

Net financial wealths/income

0.059 ***

0.088 ***

0.070 ***

(0.014)

(0.022)

(0.017)

Standard error*100

0.853

0.854

0.962

0.998

0.998

0.996

Durbin Watson

2.240

2.140

2.100

AR1/MA1 (p-value)

0.220

0.582

0.439

AR2/MA2 (p-value)

0.341

0.684

0.699

Heteroscedasticity (p-value)

0.098

0.055

0.242

Chow (p)

0.763

0.211

0.125

RESET(p)

0.739

0.958

0.598

annex,figures A6.5 and A6.6). While the model could probably be improved with more directly apposite interest rate measures and other sophistications, it provides strong evidence against serious structural breaks in debt growth in Japan. The only slight evidence, using step dummies, we can find for a structural break is for a small liberalization for debt in 1987, reversed in 1991–2, so leaving no longer term trace. A corresponding shift indicator in the consumption function is completely insignificant.

A cointegration analysis for a four-variable system comprising the ratio of current debt to current income (DEBTY), the ratio of current physical wealth to current income (RAY), the ratio of current net (p.211) financial wealth to current income, and the nominal call rate, with the pre-1974 trend in the cointegration space, suggests there is only one cointegration vector, consistent with the long-run solution of the debt equation in table 6.7 column 1.

The cointegrating vector was

DEBTY = −0.750 * RC + 0.157 * FNFAY + 0.040 * RAY + 0.012 * pre-1974Trend,

where the change in unemployment rate was included as an unrestricted I(0) variable. The coefficient of Rc was restricted at −0.75, acceptable at the 5% level.

# 5 Conclusion

The introduction summarizes the chapter. Here we wish to stress three main points regarding the role of monetary policy.

First, theory can explain why higher real interest rates might have positive direct consumption effects in Japan and the evidence is that they do. Second, when household credit markets are underdeveloped, and other institutional features are present (for example, inheritance taxes favoring land and housing), theory suggests a negative housing “wealth” effect. Evidence from Japan is of such a negative effect, implying that higher real interest rates could indirectly increase consumption through this channel. However, for the economy as a whole, our evidence is that monetary policy works in the conventional direction. Our point is that the household part of the transmission process is far weaker than in the US or UK.

Third, there is no evidence of serious long-term credit market liberalization for households in Japan. This contradicts the frequent assumption by casual Western observers of strong parallels between Japan’s “bubble economy” and credit-fuelled consumption booms experienced by some Western economies in the 1980s and in recent years. Between the mid-1990s and the early 2000s, the mounting bad loan problem of Japan’s banking system made household credit liberalization even more difficult.

Our research conclusions are therefore broadly consistent with the focus on problems in the banking sector by Hoshi and Kashyap (2005).

The evidence from the income forecasting model in section 3.2 has an interesting bearing on the role of fiscal policy in Japan’s lost decade. This model shows a strong negative effect from either a moving average (p.212) of the government deficit to GDP ratio or from the level and change in the government debt to GDP ratio. The rises in either measure from 1995 to 2005 account, according to the model, for a loss of non-property disposable income of the order of 1 percentage point per annum, around 10% over 10 years. Log income has a coefficient of 1 in the long-run solution for consumption, conditional on asset to income ratios, and the like. Hence, there was a similar long-run effect on consumption. In addition, according to the estimated consumption equations, there was also a negative short-run effect on consumption from the deterioration in the government finances.

These estimates are consistent with a Ricardian element in household perceptions of the government balance sheet, and suggest that fiscal policy had limitations during the lost decade. It is wise, however, to be aware of the possibility of some upward bias in the estimates of the size of these effects. As Comin (this volume) notes, there was a significant productivity slowdown in the period, possibly connected with the survival of zombie companies or poor investment allocation. An unexpected slowdown might well cause government revenue to deteriorate unexpectedly relative to sticky expenditure commitments, contributing to fiscal deterioration. If the underlying productivity growth rate has been omitted from the model for income, there could then be an upward bias in the size of the fiscal coefficients in the income equation.

Evidence in this chapter suggests a distinctively weak monetary transmission mechanism in Japan because of the special features of Japanese credit markets and tax institutions, the large net liquid financial asset holdings of Japanese households, and perhaps their distinctive preferences. It seems likely that this was a factor contributing to Japan’s lost decade. In contrast, US and UK households currently hold much higher levels of debt than of liquid assets. Provided credit markets can function again, interest rate policy in the US and UK should have far bigger effects on aggregate demand than was—and is—the case in Japan.

The following annexes, and additional figures, are available on the web. Please visit http://mitpress.mit.edu/japansbubble

1. Annex 6.1

2. Annex 6.2

# Acknowledgments

The authors acknowledge funding support from ESRI. We are grateful for advice from Janine Aron, John Duca, Noriki Hirose, and Wataru Takahashi, (p.213) as well as to participants at seminars at the Dallas Federal Reserve and the IMF. We have greatly benefited from the comments of Charles Horioka, Martin Lettau, Pierre Olivier Gourinchas, and the editors. We take responsibility for all remaining errors.

References

Bibliography references:

Ando Albert, Michiko Yamashita and Atsuyoshi Murayama, 1986, “Raifu Saikuru Kasetsuni Motozuku Syohi Chochiku no Kodo Bunseki (Analysis on Consumption and Saving Behaviour Based on the Lif-ecycle Hypothesis),” Keizai Bunseki Vol. 101, Economic Research Institute, Economic Planning Agency.

Aron, Janine, John Muellbauer and Anthony Murphy. 2008. “Housing Wealth, Credit Conditions and UK Consumption.” Presented at the Econometric Society European Meetings, Milan. August 2008. http://www.eea-esem.com/EEA-ESEM/2008/Prog/viewpaper.asp?pid=1705

Attanasio, Orazio, Laura Blow, Robert Hamilton and Andrew Leicester. 2009. “Booms and Busts: Consumption, House prices and Expectations,” Economica 76: 20–50.

Attanasio, Orazio and Guglielmo Weber. 1994. “The UK consumption boom of the late 1980s: aggregate implications of microeconomic evidence,” Economic Journal, 104 (427): 1, 269–302.

Blinder, Alan and Deaton, Angus, “The Time Series Consumption Function Revisited,” Brookings Papers on Economic Activity, 2: 1985, 465–511.

Bosworth, Barry, Gary Burtless and John Sabelhaus. 1991. “The Decline in Saving: Evidence from Household Surveys.” Brookings Papers on Economic Activity 1991: 1.

Campbell, J. and Mankiw, N., 1989. “Consumption, Income and Interest rates: Reinterpreting the Time Series Evidence.” NBER Macroeconomics Annual, pp. 185–216.

Campbell, J. and Mankiw, N., 1991. “The Response of Consumption to Income: A Cross Country Investigation.” European Economic Review 35: 715–721.

Carroll, C. 2001. “Death to the Log-Linearized Consumption Euler Equation! (And Very Poor Health to the Second-Order Approximation),” Advances in Macroeconomics, Berkeley Electronic Press, 1(1): 1003–1003.

Comin, Diego A. 2011. “An Exploration of the Japanese Slowdown during the 1990s.” In Japan’s Bubble, Deflation, and Long-term Stagnation, edited by Koichi Hamada, Anil K Kashyap] and David E. Weinstein. The MIT Press, chapter 11 this volume.

Deaton, Angus. 1999. “Saving and Growth.” In Klaus Schmidt-Hebbel and Luis Serven, eds., The Economics of Saving and Growth. Cambridge: Cambridge University Press.

Deaton, Angus and John Muellbauer. 1980. Economics and Consumer Behaviour. Cambridge University Press.

Doornik, Jurgen 2007. An Introduction to OxMetrics 5, London: Timberlake Consultants Press.

(p.214) Engelhardt, Gary V. 1996. “Consumption, Down Payments and Liquidity Constraints,” Journal of Money Credit and Banking 28(2): 255–271.

Fernandez-Corugedo, Emilio and John Muellbauer. 2006. “Consumer Credit Conditions in the U.K.” Bank of England working paper, 314.

Friedman, M. 1957. A Theory of the Consumption Function. Princeton: Princeton University Press.

Friedman, M. 1963. “Windfalls, the ’Horizon’ and Related Concepts in the Permanent Income Hypothesis.” In Carl F. Christ, Milton Friedman, Leo A. Goodman, Zvi Griliches, Arnold C. Harberger, Nissan Liviatan, Jacob Mincer, Yair Mundlak, Marc Nerlove, Don Patinkin, Lester G. Telser, Henri Theil (eds.) Measurement in Economics: Studies in Mathematical Economics and Econometrics in Memory of Yehuda Grunfeld, Stanford University Press.

Gokhale, Jagadeesh, Laurence Kotlikoff and John Sabelhaus. 1996. “Understanding the Postwar Decline in U.S. Savings: a Cohort Analysis,” The Brookings Papers on Economic Activity, 315–344.

Gourinchas, Pierre-Olivier and Jonathan A. Parker, 2001. “The Empirical Importance of Precautionary Saving,” American Economic Review 91(2): 406–412, May.

Hall, R., 1978. “Stochastic Implications of the Life Cycle-Permanent Income Hypothesis.” Journal of Political Economy 96: 971–987.

Hamada, Koichi. 2004. “Policy Making in Deflationary Japan,” The Japanese Economic Review 55(3): September.

Hansen, L. P. and K. J. Singleton (1983), “Stochastic Consumption, Risk Aversion, and the Temporal Behavior of Asset Returns,” Journal of Political Economy 91(2): 249–265.

Harbo, Ingrid, S̶ren Johansen, Ben Nielsen and Anders Rahbe. 1998. “Asymptotic Inference on Cointegrating Rank in Partial Systems,” Journal of Business and Economic Statistics 16(4): 388–399.

Hayashi, Fumio. 1985. “The Permanent Income Hypothesis and Consumption Durability: Analysis Based on Japanese Panel Data,” The Quarterly Journal of Economics 100: 1083–113.

Hayashi, Fumio. 1997. “A Review of Recent Literature on Japanese Saving” in Fumio Hayashi, Understanding Saving: Evidence from the United States and Japan, MIT Press, Cambridge.

Hayashi, Fumio, Takahashi Ito and Joel Slemrod. 1988. “Housing Finance Imperfections, Taxation, and Private Saving: A Comparative Simulation Analysis of the United States and Japan,” Journal of the Japanese and International Economies 2: 215–238.

Hori, Masahiro and Satoshi Shimizutani. 2003. “Asset Holding and Consumption: Evidence from Japanese Panel Data in the 1990s,” ESRI discussion Paper No. 55.

Horioka, Charles Y. 1988. “Saving for Housing Purchase in Japan,” Journal of the Japanese and International Economies 2(3): 351–384.

Horioka, Charles Y. 1993. “Saving in Japan” in World Savings: An International Survey, ed. Heertje, 238–78, Blackwell Publishers, Oxford.

Horioka, Charles Y. 1996. “Capital Gains in Japan: Their Magnitude and Impact on Consumption,” Economic Journal 106(May): 560–577.

(p.215) Horioka, Charles Y. 1997. “A Cointegration Analysis of the Impact of the Age Structure of the Population on the Household Saving Rate in Japan,” Review of Economics and Statistics 79(3): 511–16.

Horioka, Charles Y. 2004. “Are the Japanese Unique? An Analysis of Consumption and Saving Behavior in Japan.” Discussion Paper No. 606, The Institute of Social and Economic Research, Osaka University.

Horioka, Charles Y. 2006. “The Causes of Japan’s ’Lost Decade’: The Role of Household Consumption,” Japan and the World Economy 18(4).

Hoshi, Takeo and Anil Kashyap. 2005. “Japan’s Financial Crisis and Economic Staganation,” Journal of Economic Perspectives 18: 3–26.

Ito, Takatoshi and Frederic S. Mishkin. 2004.”Two Decades of Japanese Monetary Policy and the Deflation Problem.” NBER Working Paper 10878, 2004.

Jackman, Richard and John Sutton. 1982. “Imperfect Capital Markets and the Monetarist Black Box: Liquidity Constraints, Inflation and the Asymmetric Effects of Interest Rate Policy.” Economic Journal 92: 108–128.

Jappelli, Tullio and Marco Pagano. 1994. “Saving, Growth and Liquidity Constraints.” Quarterly Journal of Economics 109(1): 83–109.

Kimball, Miles S., 1990. “Precautionary Saving in the Small and in the Large,” Econometrica 58(1): 53–73.

King, Mervyn. 1990. Discussion of “Is the UK Balance of Payments Sustainable?” John Muellbauer, Anthony Murphy, Mervyn King, and Marco Pagano. Economic Policy 5(11): 347–395.

Masubuchi, Katsuhiko. 2006. “Japan’s Household Savings Ratio,” Economic and Social Research Institute, Cabinet Office.

Modigliani, Franco and Richard Brumberg. 1954. “Utility Analysis and the Consumption Function: An Interpretation of Cross-section Data.” In Kenneth K. Kurihara (ed.) Post-Keynesian Economics. New Brunswick, NJ: Rutgers University Press.

Moriizumi, Yoko. 2003. “Targeted Saving by Renters for Housing Purchase in Japan,” Journal of Urban Economics 53: 494–509.

Muellbauer, John. 1988. “Habits, Rationality and Myopia in the Life Cycle Consumption Function.” Annales d’Economie et de Statistique 9: 47–70.

Muellbauer, John. 1996. “Income Persistence and Macropolicy Feedbacks in the US,” Oxford Bulletin of Economics and Statistics 58: 703–733.

Muellbauer, John. 2007. “Housing, Credit and Consumer Expenditure.” In Housing, housing finance, and monetary policy, A Symposium Sponsored by the Federal Reserve Bank of Kansas City, Jackson Hole, Wyoming, August 30-September 1, 2007: 267–334.

Muellbauer, John and Ralph Lattimore. 1995. “The Consumption Function: A Theoretical and Empirical Overview.” In Hashem Pesaran and Michael Wickens, eds., Handbook of Applied Econometrics. Oxford: Blackwell.

Murata, Keiko 1999. The Consumption Function in Japan, DPhil Thesis, University of Oxford.

(p.216) Murata, Keiko. 2003, “Precautionary Savings and Income Uncertainty: Evidence from Japanese Micro Data.” Monetary and Economic Studies 21(3). Institute for Monetary and Economic Studies, Bank of Japan.

Nakagawa, Shinobu, 1999, “90 nendai iri go mo nihon no kakeichochiku ritsu ha naze takai noka? (Why Does the Household Saving Rate Remain High after 1990 in Japan?),” Bank of Japan.

Nakagawa, S. I. Oshima, 2000, “Jisshitsu Kinri no Teika ha Kojin Shohii wo Shigeki Suru noka?—Jisyo bunseki wo chushin ni (Does the Decline in Real Interest Rate Stimulate Households’ Consumption?)” Working Paper 00–2, Research Bureau, Bank of Japan.

Poterba, J. M. 2000. “Stock Market Wealth and Consumption.” Journal of Economic Perspectives 14(2): 99–118.

Poterba, J. M. and J. M. Manchester. 1989. “Second Mortgages and Household Saving.” Regional Science and Urban Economics 19(2): 325–46.

Ogawa, Kazuo, Shinichi Kitasaka, Hiroshi Yamaoka and Yasuharu Iwata, 1996, “An Empirical Re-evaluation of Wealth Effect in Japanese Household Behavior.” Japan and the World Economy 8: 423–442.

Otsuka, Misuzu. 2006. Essays on household portfolio and current account dynamics. ProQuest/UMI.

Rahbek, Anders and Rocco Mosconi, 1999. “Cointegration Rank Inference with Stationary Regressors in VAR Models,” Econometrics Journal, Royal Economic Society 2(1): 76–91.

Saito, Makoto and Shigenori Shiratsuka. 2003. “Precautionary Motives versus Waiting Options: Evidence from Aggregate Saving Rate in Japan.” Monetary and Economic Studies 21(3). Institute for Monetary and Economic Studies, Bank of Japan.

Sekita, Mika. 2008. “Rishi Shotoku, Haito shotoku, kabushiki tou no joto shotoku nojikkou zeiritsu no keisoku (Estimates of Effective Tax Rates on Interest Income, Dividends, Transfer of Shares, etc.),” mimeo.