最近的房價暴漲,以及其隨后的崩潰,為更好的理解房價與消費的關系提供了一個簡單的推動。(原文Drevalence寫錯了,應該是prevalence )在英國生活的業主意味著大多數家庭都有明顯的房產,這些資產的價值的變化可能會影響家庭的消費行為。我們考慮出一個在周期框架內可能影響房價和消費的三種機制:(1)財富效應;(2)間接影響;(3)常見的因果關系。
回顧文獻,關注最近的Campbell和Cocco (2007)的辯論,發現財富效應的主要原因是房價和消費之間的關系,阿塔納西奧,布羅,漢密爾頓和萊斯特(2009),他們是通過收入期望找到支持常見的因果關系的四個人,而后者的文章已經在學術界流傳開來,并且這項研究前不久已經被出版了。
我們使用家庭數據來擬建一個基于(假定)人口和經濟因素的消費行為模型,通過添加房價模型,比較不同年齡和時期的影響家庭的狀態,我試圖將其與競爭理論區分開來。
Contents
1 Abstract.................................................................................................2
2 Introduction..........................................................................................5
3 Literature Review.................................................................................8
3.1Theory...........................................................................................8
3.1.1 Wealth Effect...........................................................................9
3.1.1.1 House Price Changes...............................................10
3.1.1.2 Consumption of Housing Services..........................14
3.1.2 Liquidity Constraints..............................................................12#p#分頁標題#e#
3.1.3 Common Causality.................................................................14
3.1.3.1 Income Expectations....................................................14
3.1.3.2 Aggregate Financial Conditions...................................15
3.1.3.3 Demography.................................................................16
3.2 Related Literature.....................................................................18
3.2.1 Theoretical Literature........................................................18
3.2.2 Empirical Literature..........................................................19
4 Analysis, discussion and interpretation............................................21
4.1 Data.........................................................................................21
4.1.1 Household....................................................................21
4.1.2 Housing.......................................................................23
4.1.3 Economic Conditions..................................................24
5 Methodology.......................................................................................25
6 Results.................................................................................................29
6.1 Age effects........................................................................... 29
6.2 Tenure Effects.......................................................................32
6.3 Financial Conditions.............................................................34
6.4 Robustness to Cell Size Changes..........................................35
7 Conclusion..........................................................................................37
8 Appendix.............................................................................................40
9 Reference............................................................................................42
2 Introduction
The recent house price boom and its subsequent collapse provide ample motivation for a better understanding of the relationship between house prices and consumption. The Drevalence of owner-occupation in the UK means that most households have significant housing assets,and changes in the value of these assets might be expected to affect Households' consumption behaviour.We consider the theoretical impact of house prices in a lifecycle framework,and discuss three mechanisms which might relate house prices and consumption:(i)wealth effects;(ii)collateral effects;and (iii)common causality.#p#分頁標題#e#
The literature is reviewed,focussing on the recent debate between Campbell and Cocco (2007),who find that wealth effects are the primary cause of the relationship between house prices and consumption,and Attanasio,Blow,Hamilton,and Leicester(2009),who find support for common causality through income expectations.Whilst the latter article has covered much of the ground we had hoped to consider, being published shortly before the completion of this work,it only covers the period to 2001/02.I extend this to 2007,thus capturing the effect of a further 65% rise in real house prices[1].
We use household data to estimate a model of consumption behaviour,conditional on demographic and economic factors.By adding house prices to my model and comparing the effects on households of different ages and tenure statuses, I attempt to distinguish between the competing theories.
My main contribution is to extend the debate through a more formal treatment of tenure status,with a particular focus on the consumption of different categories of renter.Due to lacking of good quality data on the level of rents faced,I exploit differences between the consumption of social and private assisted tenants,and private unassisted tenants.The former studies have relatively inflexible net rents(rents less Housing Benefit) which are infrequently adjusted,so they provide a counter-factual for the behaviour of private assisted renters if rental costs had remained unchanged,given certain assumptions as to How selection between tenure statuses is determined.This allows me to further distinguish Which mechanism(s) relate house prices and consumption.I also extend the current analysis using more recent data,including the post-2000 house price boom,allowing me to determine whether the recent boom can be explained in the same way as past booms.
3 Literature Review
3.1 Theory
Theoretically there are a number of channels through which house prices might affect consumption. The two most likely to directly affect individual behaviour and have a significan effect at an aggregate level are wealth effects and liquidity constrains. A third possibility which might explain co-movements in house prices and consumption is that there is some other underlying factor which drives changes in both house prices and consumption expenditure (common causality).
3.1.1 Wealth Effect
The wealth effect argument considers households that optimise consumption decisions over their lifecycle to maximise utility.An unexpected permanent increase in the value of an asset owned by the household will increase lifetime wealth,which the household wants to run down over the remainder of its life. Since housing costs will also rise,the effect on non-housing expenditure will be dampened. If I assume that households are Finitely-lived, i.e. I abstract from dynastic considerations and bequest motives,this should lead to an unambiguous increase in consumption by households whose holdings of household equity exceed the level of housing services required over the remainder of their life. The impact on other households depends on the form of the rise in house prices, and the manner in which they expect to consume housing services over their lifetime.#p#分頁標題#e#
3.1.1.1 House Price Changes
Considering first the issue of house prices,we must distinguish between the effect of a Once-and-for-all increase in prices,and a change in the growth rate of house prices. Houses can be bought as a consumption good, for owner-occupation, or as an investment good,to be rented out.The supply of houses is highly inelastic,even in the medium run,so the total housing stock is approximately constant.
In my first stylised scenario I consider all exogenous,overnight doubling in the price of houses.I assume a constant exogenous rate of household formation,equal to the rate of household destruction.Ceteris paribus,a doubling in the purchase price of houses means young households will find it cheaper to rent rather than buy a house,expanding the demand for rental property.To increase the number of properties available for renting, landlords must buy houses at the new market price.If I assume the housing market is competitive,rent on the marginal rental property will equate the cost of rental housing services and the cost of owner-occupied housing services.This occurs when the cost of renting has also doubled.Renters are worse off, as they now face higher lifetime rents. Owner occupiers may be better off,if the increase in wealth exceeds the rise in their housing costs.
I notice that I have assumed house price changes to be exogenous. To be confident that house price changes cause changes in consumption growth,I would prefer to know the underlying cause of a change in house prices,So that I could be certain it was unrelated to consumption.
3.1.1.2 Consumption of Housing Services
The existence of transaction costs and heterogeneity between houses create significant housing market frictions and lagged responses.If there is a positive (exogenous) shock to house prices this period,equilibrium will not be immediately reached,and house prices will continue to adjust in subsequent periods[2][3].Hence households that were previously indifferent between renting and owning now have all incentive to buy,hoping that any
house they purchased would rapidly rise in value,and providing a hedge against the risk of rising rents.Such an action is rational,and forms a Nash equilibrium[4]:if renters have mixed strategies over future tenure choice,and they increase their probability of homeownership,the aggregate increase in demand for houses will increase house prices, justifying the decision to purchase. Some homeowners will also look to move,using their increased net housing equity to fund a deposit on a larger house.Again such an action is a Nash equilibrium,as other agents behaving in the same way will increase future prices. Expectations of higher future prices imply an upwards-sloping short-run demand curve,and a “rational bubble’’ is created.Speculators may also increase their exposure to the housing market,buying properties in the anticipation of asset price gains,further fuelling the house price boom.#p#分頁標題#e#
3.1.2 Liquidity Constraints
Households are constrained in the amount they can borrow.An increase in house prices increases the collateral of homeowners,so they should be able to borrow more,allowing them to bring forward future expected income.Homeowners can therefore finance a higher level of current consumption,particularly of durable[5].Consumer durables are large purchases as a proportion of monthly income,so households may need to borrow or dip into precautionary savings to fund their purchase.Since they are long-lasting, intertemporal substitution of purchase is feasible,so an increased ability to borrow allows homeowners to purchase these goods sooner.Expenditure on non-durables may rise if there is an increase in expectations of future income,but the size of this effect will be much smaller.
It is important to distinguish this from hypotheses like wealth effects,which involve an increase in spending out of net assets.If house prices double,the wealth effect is the spending down of the increased asset value over the household’s lifespan.The lifting of liquidity constraints,on the other hand,allows the household to spend out of future expected income,with the house simply providing collateral.
Typically we expect that young households are most likely to face liquidity constraints, as they have not had the time to build up sufficient savings with which they can smooth consumption. Carroll (2001) shows that “the implications of precautionary savings and liquidity constraints for consumption growth are virtually indistinguishable”[6], so low wealth consumers are likely to voluntarily behave as if liquidity constrained,to smooth the effects of potential income shocks. Young homeowners,who are likely to have low wealth,will therefore benefit from a rise in house prices,as it increases their collateral. increasing their ability to borrow. For older homeowners we expect the effect to be muted,or to not exist at all,as they would no longer have higher future incomes,so do not wish to borrow to smooth consumption. They often also have net housing equity against which they could already have borrowed. Renters should not change their expenditure as they do not experience any collateral effect.
3.1.3 Common Causality
3.1.3.1 Income Expectations
In a lifecycle framework, expenditure in a given period is a fraction of permanent income. The size of this fraction depends on the marginal utility of expenditure that period, which may be based on a number of factors such as relative size of household compared with the future, relative prices, and the state of health of the household. An increase in expectations about future income raises the permanent income available to the household to spend over its life. King (1990) emphasises the role of improved productivity, and hence rising income expectations in the 1980s, as the cause of the mid-1980s consumption and house price boom.#p#分頁標題#e#
This higher expenditure will be reflected in a higher level of current consumption, and a greater demand for housing. Since housing supply is relatively inelastic in the short and medium run, an increase in demand will lead to an increase in house prices, so house price changes and consumption growth will be correlated.
Since the increase in permanent income will be largest for households with longest remaining working lives, we would expect a change in income expectations to have a larger effect on young households, of all tenures, than on older ones.
3.1.3.2 Aggregate Financial Conditions
When financial conditions in an economy are "loose", so that credit is relatively cheap, this will stimulate borrowing. Agents can bring forward the replacement of consumer durables, and pay a lower cost to finance this expenditure. Since most households will not have the wealth to purchase a house without a mortgage, the decision to purchase a house will also depend on the interest rate at which the household can borrow. Hence aggregate financial conditions may act to affect both house prices and consumption.
This mechanism is different from liquidity constraints. In the latter, an increase in house prices makes it easier to borrow to fund consumption. A general reduction in interest rates, oh the other hand, will increase both current expenditure (through secured and unsecured borrowing) and the ability to fund larger house purchases, driving up the level of house prices.
A loosening of aggregate financial conditions also widens the range of people who have access to credit. Muellbauer and Murphy (MM, 1990) noted that financial deregulation in the 1980s meant that "even the retired...can borrow up to 25% of the value of their home and never pay interest, the cumulated interest and debt simply being paid from the value of their estate at death."[7] More recently, financial institutions were willing to provide mortgages that were a much greater proportion of the house's value, with Northern Rock going so far as to lend 125% of the house's value[8].
If overall financial conditions cause the aggregate effects we observe, then we would expect the young to benefit the most, for the same reason as with liquidity constraints. However, households of all tenure statuses will be able to benefit, and hence be better off. Another similarity to liquidity constraints is that households will mostly borrow to fund purchases of durables. Looking at aggregate data, we would also expect to find an increase in the number of households who are homeowners, as a greater proportion of renters become able to borrow enough to buy a house.
3.1.3.3 Demography
Changes in population size or structure could both have effects at an aggregate level, even if households do not individually change their behaviour. An increase in the number of households would increase aggregate consumption: the growth rate of aggregate consumption would be equal to the sum of average household consumption growth and the rate of net household formation. If the supply of houses is relatively inelastic, house prices will be positively correlated with the difference between the net rate of household formation and the growth rate of the housing stock.#p#分頁標題#e#
A change in population structure could also affect the mechanism which relates consumption and house prices. Campbell and Cocco (CC, 2007) note that "as the population ages and becomes more concentrated in the old homeowners group, aggregate consumption may become more responsive to house prices"[9] through wealth effects. Since such a change would not be observable in household level data, the housing market must, at least in part, be considered at a macroeconomic level.
Since I focus on household level data, the impact of household formation will not be relevant here, except insofar as it changes the size of households sampled. This problem is solved by controlling for household size directly. However, I note that other studies[10][11] find household formation to be endogenous and heavily influenced by income and housing costs. Bramley et al. (2006) find "a central estimate [of income elasticity] of 0.15"[12] and of -0.08 for housing cost elasticity. I therefore expect much of the underlying cause of changes in household formation to be represented elsewhere in my model.
3.2 Related Literature
A full treatment of the literature on house prices, and the nature of their relationship with consumption, is beyond the scope of this thesis. Instead I focus on a few key papers and consider the relevance of their conclusions to my analysis. My focus is on the recent debate, as this provides results against which I will compare my analysis.
3.2.1 Theoretical Literature
A major contribution to the theoretical debate was made by Muellbauer and Murphy (MM, 1990), and their discussants, King (1990) and Pagano (1990). MM suggest consumption growth in the late 1980s was caused by rising house prices and "financial liberalisation [mat] allowed households to cash it in as consumer expenditure financed by borrowing."[13]They find "an important extrapolative component"[14] in the determination of house prices, and suggest this would increase the willingness of owner-occupiers to spend out of housing wealth.
King (1990) argues against this explanation, claiming that an increase in house prices reflects an increase in the cost of housing services, so although current homeowners are better off, non-owners now face higher housing costs. Rising house prices therefore only redistribute consumption. A subtlety missed in this argument is that a redistribution of wealth will redistribute lifetime consumption, so in the short run we may not see complete offsetting, depending on the relative marginal propensities to consume of the gainers and losers.
Instead, King (1990) proposes that the consumption boom should be explained by "a change in beliefs about the future growth rate of incomes"[15]. Pagano (1990) makes a similar argument: "the surge of house prices [is] a symptom rather than the cause of the consumption boom"[16] with increased expected future incomes as the underlying cause of both.#p#分頁標題#e#
Skinner (1989) elaborates on this, noting that "when the house lasts longer than does the owner, if there is a permanent increase in demand...the current owner can capture some of the future rent on the asset when it is sold to future generations"[17]. Nevertheless, he finds that "shifts in house value had no effect on consumption in the later 1970s"[18] in the US. Skinner proposes a number of mitigating factors to explain the lack of responsiveness: (I) the existence of a bequest motive (with reasonable parameter values increased house prices may generate higher savings); (II) an inability to access the wealth, something that may now have changed after thirty years of financial innovation; (III) a belief that the capital gain is only temporary; or (IV) the house price increase being expected and consequently already factored into spending plans.
3.2.2 Empirical Literature
Attanasio and Weber (AW, 1994) use Family Expenditure Survey (FES) data from 1974-88 to test the MM and King hypotheses. They use a lifecycle framework to estimate a regression that considers expenditure as a function of age and lifetime wealth, controlling for other demographic factors. Their results "cast considerable doubts on the MM hypothesis... show[ing] that capital gains in real estate cannot explain most of the surge in consumption of young households."[19] Instead they conclude that "much of the increase in the aggregate propensity to consume reflects a change in the perceptions by private agents of the dynamics of their permanent income"[20] i.e. they find support for the common causality hypothesis, acting through changes in income expectations.
In a paper published after the topic of this dissertation had been accepted, Attanasio, Blow, Hamilton, and Leicester (ABHL, 2009), update the AW paper using a longer (FES) data series, from 1978 to 2001/02. Their results support "the findings in AW...that the co-movements in consumption and house prices are generated not by a causal link running from the former to the latter, but by common factors."[21]
This contradicts work by Campbell and Cocco (CC, 2007), who take a different approach. They also use FES data - for 1988-2000 - to perform cohort analysis, but they find older house owners are most significantly affected by changes in house price, whilst the impact on young renters is insignificantly different from zero. This is consistent with the wealth effect, and does not support the AW/ABHL conclusion. However, they also find national house prices changes to be correlated with consumption changes and significant for all tenures, suggesting that perhaps causation comes through aggregate financial conditions.
A comparison of the methodologies used is provided in ABHL (2009)[22], but the key differences are (I) AW/ABHL specify their regression in levels whilst CC use first differences; (II) AW/ABHL do not control for current income; and (III) AW/ABHL do not split cohorts by tenure, but only by birth year of household head. I discuss these issues further when I justify our methodology.#p#分頁標題#e#
4 Analysis, discussion and interpretation
Although I observe a correlation between house prices and consumption at an aggregate level, in order to distinguish between the channels considered and draw conclusions as to which is most significant, I must analyse household level data. In the absence of appropriate panel data to test the hypotheses, I instead construct a pseudo-panel.
4.1 Data
4.1.1 Household
UK household data that include detailed consumption measures and housing variables are available in the Expenditure and Food Survey[23] (EFS, formerly the Family Expenditure Survey: FES). This is an annual cross-sectional survey, which provides monthly data on around 7000 households. I consider 24 years of data, from 1984 to 2007, as some series I use are not available prior to 1984. I also adjust the data from 1993/94, since when the data have been reported by fiscal rather than calendar year, using the month of interview to reclassify observations by calendar year.
Price data are deflated by the implicit consumption deflator[24], obtained from the ratio between real and nominal aggregate consumption in GDP. I use the aggregate consumption deflator series because it provides consistent monthly data, whilst not being directly affected by changes in interest rate, as with RPI. Consistent with these data, we take 2003 to be our base year.
Since my purpose is to better understand the relationship between aggregate consumption and house prices, EFS/FES consumption data can only be useful if it follows the same trend as aggregate consumption. For example, if demographic changes were driving the observed relationship, I would not necessarily see any link between household and aggregate data. Tanner (1998), and Blow et al. (2004), study the relationship between FES and National Accounts data. Both find a high correlation, although the former is more volatile, as one might expect with survey data. I note with some caution that such a study has not been undertaken with the more recent EFS data, but since the survey methodology is largely unchanged, I would not expect any serious divergence in the relationship. I am therefore comfortable using the survey evidence to draw conclusions about the aggregate economy.
4.1.2 Housing
Quarterly data on national and regional house prices are obtained from the Nationwide House Price Index[25]. Data are based on residential mortgages at the approvals stage, and are "mix-adjusted" to give the price for a fixed representative house. Since the definitions of regions vary between Nationwide and the EFS, district level data from Nationwide are used to construct comparable regional house prices for the EFS regions.
#p#分頁標題#e#
4.1.3 Economic Conditions
Bank of England[26] data on base rate decisions are used to calculate the average annual nominal interest rate. The real interest rate is obtained by subtracting the (aggregate consumption) inflation rate from the nominal interest rate. I also consider the "Household External Finance (HEF) index"[27], which measures the "spread between the household specific interest rate and the Bank of England's policy rate"[28]. These provide some control for "aggregate financial conditions": the ability of a household to access future expected income.
5 Methodology
Since the EFS/FES is an annual cross-sectional study, I cannot follow particular individuals over time. I follow Browning et al. (1985) and Deaton (1985) in creating a pseudo-panel data structure. I can then follow "synthetic" households, indexed by birth cohort, region of residence and tenure status. Birth cohorts are defined by the year of birth of household head, in eight ten-year intervals from 1900-09 to 1970-79. Outside this range there are too few observations to provide reliable results. The data pre-classify households into one of ten UK regions; I maintain this classification, except to drop observations on households from Northern Ireland where the number of observations is low and Issues of data construction reduce comparability. Issues of sample size also lead us to exclude tenants living rent-free and homes bought through rental-purchase schemes from our tenure categories.
After converting prices into real terms, I average the value of all variables, conditional on birth cohort, region and tenure, such that in each calendar year the data are divided into "cells". Cells which only contain a relatively small number of households may be biased[29] by idiosyncrasies in a single observation. However, excluding cells below a certain threshold is costly, as I ignore some useful variation. I perform my base analysis using a cell size threshold of ten households. I then perform a comparison against thresholds of twenty and thirty households, and comment on differences in the results, where these exist.
Each cell can be considered as a "representative household" for a given cohort, region and tenure combination, and I can follow the representative households across calendar years, almost as I would with panel data. However, I must exercise caution in my analysis, due to the potential for selection bias. This arises because although birth cohort is truly exogenous, as it is based on birth year (which is unchanging[30]), selection into region and tenure are potentially endogenous. Suppose renters with the highest lifetime wealth, and hence the highest consumption, move from renting to owner-occupation when there is a positive shock to expected lifetime incomes. Then the average consumption of renters will appear to fall as a consequence of this shock even if there is no change in the income of those remaining in this group. The empirical evidence suggests such endogeneity for region is highly limited[31].#p#分頁標題#e#
We follow ABHL in using total expenditure excluding housing as our measure of consumption. We prefer this measure because it removes the potential for endogeneity bias due to the treatment of housing costs, whilst retaining durables which appear to be highly correlated with house prices[32]. We also follow the AW/ABHL lifecycle framework: expenditure in a given period is a proportion of lifetime wealth, with the size of this proportion determined by a set of demographic factors. By transforming their original equation and averaging across cohorts[33], ABHL find a relationship whose parameters can be estimated.
In our terms, the cell index represents a unique cohort-region-tenure combination, rather than simply birth cohort as in ABHL. Unlike ABHL, we take logs of averaged variables, rather than taking logs before averaging. The ABHL method puts less weight on households with highly variable values, being essentially a geometric mean. To provide greater consistency with national accounts data, which weight households equally, we compute an arithmetic mean, and then take logs.
Expected lifetime wealth is captured by the constant and demographic variables. I assume the effect, of these demographic variables on expenditure is constant across cells. I do not include current income, as this contains unexpected changes in income. By not controlling for this directly, I can test the extent to which house prices provide additional information about the expenditure of different groups, beyond the information from demographic factors. By analysing the effect of house prices on different groups, I hope to distinguish between the competing hypotheses.
6 Results
To identify which mechanisms are supported by the data, I consider the implications of the level of regional house prices on the consumption of groups, classified first by age, then by tenure status. I also consider the effects of changes in cell size threshold. In all cases, I regress total expenditure excluding housing costs on demographic and economic variables[34]; birth cohort, region, and tenure dummies[35]; and relevant house price terms. All regressions use heteroscedasticity-robust estimators, of the form proposed by White (1980). All regressions are specified as random-effects models, since cohort-averaging should remove household fixed-effects.
6.1 Age effects
Table 1 summarises the key results for age effects. I divide households into three categories: under-40s are defined as "young"; 40-60 year-olds as "middle aged"; and over-60s as "old". For young and middle aged households I find that an increased level of regional house prices significantly increases expenditure in all cases. For each of these I cannot conclude that the coefficients on the young and middle aged terms are significantly different, with p-values ranging 0.15-0.93. For old households I find negative coefficients, although these are not statistically significant in all cases.#p#分頁標題#e#
In Specification A.II I additionally control for national house prices. Although not economically significant (i.e. the coefficients are close to zero), these are individually and jointly[36] statistically significant. Consequently their inclusion should reduce bias in other estimated coefficients. The coefficients on regional house price interaction terms are substantially increased with the inclusion of national house price terms.
To ensure my results are not biased by the recent house price boom, in Specification A.III I exclude the most recent data. Since the ABHL and CC data series end in 2001 and 2000 respectively, excluding post-2001 data provides the additional benefit of comparability. My results are robust to the reduction in sample size, so the recent data do not fundamentally change our conclusions, although they do improve the efficiency of our estimation. To the extent that my conclusions vary from those in the literature, this deviation is therefore due to our analysis, rather than the additional data available to us.
In Specifications A.IV and A.V, I use as my regress and durable consumption excluding housing costs, and non-durable consumption respectively. Although I would not expect the same numerical coefficients, the sign and significance of these provides a robustness-check of our results. I find broadly similar results, so they are not dependent on my measure of expenditure.
A positive expenditure response to the level of house prices for young and middle aged households could be supportive of either the liquidity constraints or income expectations channel. In both cases I expect the largest increase in expenditure to be for the youngest groups, either because they are furthest from their average lifetime income and have the lowest collateral, so will be most inclined to borrow, or because they have the longest working lives over which to benefit from an increase in future income.
The fall in expenditure of older households provides strong evidence against the wealth effects channel, under which they would be expected to benefit the most. There are a number of potential explanations for this. Skinner (1989) has shown that if bequest motives are very large, expenditure could actually fall. Skinner (1993) suggests instead that risk-averse households who have a precautionary savings motive, combined with heterogeneity in house price shocks and health/income shocks, can lead to increases in saving. I do not consider these arguments in detail because I believe the results are better explained by income expectations. Older households' incomes are often fixed in nominal terms, as in the case of unindexed private pensions. If house prices are seen as correlated with the general price level, an increase in the level of house prices can lower real wealth for such households.
6.2 Tenure Effects#p#分頁標題#e#
In Table 2 I consider housing tenure effects. Again there are clear patterns: social tenants and private assisted renters are broadly unaffected by the level of house prices, whilst private unassisted renters and all owners are better off. National house price terms are individually and jointly insignificant, but I again continue to include them as a control. The results are qualitatively robust to changes in sample period and in the measure of expenditure (the specifications follow the same pattern as Table 1).
These results may support either the non-causal channel through expected income, or the liquidity constraints channel, of which the collateral channel is a special case. If future income is expected to be higher, all groups should increase expenditure. This is seen for all groups except social tenants and private assisted renters. However, since selection into these groups is dependent on current income, it is plausible that those who experienced significant positive shocks have largely moved out of this group, if increases in expected future income are correlated with increases in current income. This is consistent with the ten percentage point fall in the size of this group over the sample period[37]. Also, since many in these groups depend on benefits, which have been fixed in real terms since the 1980s, they are unlikely to benefit from increases in productivity which raise future incomes.
Alternatively, rising house prices may have increased homeowners' collateral, allowing them to increase borrowing. This could explain why outright owners have smaller changes in expenditure[38]: they already have significant untapped collateral, so any change in expenditure is likely to be wealth related. ABHL consider the positive expenditure coefficient for renters as "prima facie evidence against the wealth hypothesis...[since] we might have expected renters' consumption to fall...as rents might be expected to move in line with house prices, and because renters may hope to move onto the housing ladder"[39]. However, as I have already argued, an increase in the expected growth rate (something we are unable to measure) is likely to change the time path of rents, reducing their current level, but increasing their growth rate. Since unassisted renters tend to be young, they are likely to expect higher future incomes. The deferral of some housing costs can therefore be welfare-improving, allowing non-housing expenditure to be increased today, with higher costs only being paid once incomes have risen. Assisted renters and social tenants are unlikely to see a change in housing costs, and tend not to move into the unassisted market (with the exception of right-to-buy), so are unaffected by house prices.
6.3 Financial Conditions
Throughout my results, the coefficients on real interest rates and the HEF index were close to zero. To the extent that these proxy financial conditions in the economy, this suggests that aggregate financial conditions do not provide an explanation for the house price and consumption booms. However, since nominal interest rates are adjusted by policymakers, I cannot disentangle the simultaneity between real interest rates and the level of consumption. Since financial markets and institutions tend to respond to changes in policy rate, a similar criticism can be made of the HEF index. I am therefore unable to draw strong conclusions about the effect of aggregate financial conditions.#p#分頁標題#e#
6.4 Robustness to Cell Size Changes
Specification C.I finds that results for age-regional house price interactions are robust to increases in cell size threshold. Specification C.III instead uses national house prices. For minimum cell sizes above ten, the coefficients are both economically and statistically significant, and the results follow the same pattern as for regional house prices. However, when national and regional house prices are included, national house prices are not individually statistically significant, although they remain jointly significant. This suggests they are only useful as a proxy, in the absence of data on regional house prices.
The importance of regional house prices tends to support causal channels, as regional prices will follow the household's actual house price better than national prices. Region-specific productivity shocks, which affect income expectations, are also possible. However, I can rule out aggregate financial conditions, which are the same nationwide.
Specification C.II considers interactions between age and the growth rate, rather than level, of regional house prices. With one exception, all coefficients are insignificantly different from zero. It may be that as with current income, inclusion of actual, rather than expected, growth rate introduces biases. Further work that estimates the expected growth rate, may find this to be significant.
Results for tenure interactions (Specification C.IV) are robust to an increase in the cell size threshold to twenty, although we lose all private assisted observations. At a threshold of thirty, only the owned with mortgage interaction term is still significant. For private unassisted renters this is likely due to the small remaining sample size (31 observations). The point estimate does not change substantially, but the reduction in sample size reduces the efficiency of estimation, so the standard error is larger. For owned outright the increase in threshold biases the tenure sample towards older cohorts, with only 3.2% of owned outright cells being households born after 1950. Since old age interaction terms have a negative coefficient, this is likely to explain the some of the fall in the coefficient estimate.
Whilst I risk some measurement error in using small cell sizes, since I have no reason to believe this error is systematic, and the relevant coefficients are positive, the errors should be negatively correlated with the observed value. This will tend to cause "Type II error", biasing inference towards incorrectly concluding insignificance. I therefore accept the results of my lower threshold regressions, viz. private unassisted renters and owners, with and without mortgage, have expenditures that vary positively with house prices.
7 Conclusion
#p#分頁標題#e#
There has been considerable disagreement amongst academics and policymakers as to the cause of recent fluctuations in house prices, and their aggregate effects. The appropriate policy response depends on the sustainability of such changes: whilst aggregate spending out of wealth might be seen as a Ponzi game between current and future generations[40], the ability to spend out of higher future income is welfare-improving.
I constructed a pseudo-panel, averaging households by birth cohort, region of residence and tenure status. Using the ABHL methodology, I considered whether house prices were significant in explaining the level of expenditure of different groups of households. This methodology was limited by being based on a partial equilibrium model; I did not consider the income- generating process, the underlying cause of changes in house price or the determination of real interest rates. Whilst omission of these added clarity to my argument, making the comparative statics easier to consider, a general equilibrium model would be desirable to ensure that important feedback mechanisms are not excluded. A further significant exclusion was that of uncertainty, which may be significant in explaining tenure selection[41].
Despite these shortcomings, our results allowed us to differentiate between the theoretical mechanisms considered. We found that the expenditure of younger households is positively related to house prices, whilst for the oldest households there was no relationship or, in some cases, a negative one. We also found that both owners and private unassisted renters were better off in a house price boom, and that private assisted renters and social tenants were no worse off. These results are consistent with the conclusion in ABHL: the data provide support for the income expectations hypothesis. However, we have shown that these results are also consistent with the liquidity constraints channel. ABHL reject this possibility on theoretical grounds, but this is because they consider only the special case of collateral as the binding constraint for the availability of household liquidity. By broadening this effect to consider the time-path of housing costs, we have shown that this a priori rejection is unjustified, and further, that the data are not inconsistent with this effect. I found no support for the CC conclusion that wealth effects are most significant, nor for the aggregate financial conditions channel. I also concluded that the mechanisms underlying the post-2001 boom were not significantly different from the earlier period.
The non-causal and liquidity constraints hypotheses have different welfare implications, particularly for the least well-off. If the latter mechanism is the most significant, then those on low incomes who live in social housing may be unable to benefit during a house price boom simply because their rents are adjusted less frequently, and are "sticky-down". This is consistent with the observation (from the tenure dummies) that social tenants and private assisted renters have a significantly higher level of consumption than other groups, controlling for other factors. Social housing positively affects the level of welfare, but during a house price boom this welfare increases more slowly than if social rents fell in line with private rents. Unfortunately there are currently no data which allow me to measure household responses to changes in their housing costs, or the responsiveness of rents to changes in house prices, either of which would be sufficient to allow me to distinguish between the two remaining channels. A lack of appropriate data is thus the key constraint to further work in this area.#p#分頁標題#e#
Appendix
Table 1. Impact of house prices on expenditure, by age
Table 2. Impact of house prices on expenditure, by tenure
Standard errors in parentheses
*= p<0.05
Figure 1. Residuals against cell size for Table 1, Specification II
Table 3. Sensitivity of results to cell size
Standard errors in parentheses
*= p<0.05
Rerferences
[1] Attanasio, O., Blow, L., Hamilton, R., and Leicester, A. (2009) "Booms and Busts: Consumption, House Prices and Expectations." Economica 76: pp20-50.
[2] Attanasio, O., and Weber, G. (1994) "The UK Consumption Boom of the Late 1980s: Aggregate Implications of Microeconomic Evidence." Economic Journal 104, ppl269-1302.
[3] Bank of England (2004d) Inflation Report.
[4] Benito, A. (2009) "Who Withdraws Housing Equity and Why?" Economica, 76, pp51-70.
[5] Besley, T., Meads, N., and Surico, P. (2008) "Household External Finance and Consumption." External MPC Unit Discussion Papers 25.
[6] Blow, L., Leicester, A., and Oldf ield, Z. (2004) "Consumption Trends in the UK, 1975-99." Institute for Fiscal Studies.
[7] Bramley, G., Karley, N. K., and Watkins, D. (2006) "Local need and affordability model for * Scotland - Update (2005 based)." Communities Scotland Report 72: pp72-127.
[8] Browning, M., Deaton, A., and Irish, M. (1985) "A profitable approach to labour supply and commodity demands over the life cycle." Econometrica 53: pp503-544.
[9] Cameron, A. C. and Trivedi, P. K. (2009) Microeconometrics using Stata, Stata Press.
[ 10] Campbell, J. and Cocco, J. (2007) "How do house prices affect consumption? Evidence from micro data." Journal of Monetary Economics 54: pp591-621.
[11] Carroll, C. (2001) "A Theory of the Consumption Function, With and Without Liquidity Constraints (Expanded Version)." NBER Working Paper 8387.
[12] Deaton, A. (1985) "Panel Data from Tune Series of Cross Sections." Journal of Econometrics 30: ppl09-126.
[13] Hendry, D. (1984) "Econometric Modelling of House Prices in the UK." in D.Hendry and K.Wallis (eds.) Econometrics and Quantitative Economics, Basil Blackwell: Oxford, pp211-252.
#p#分頁標題#e#
[14] King, M. (1990) "Is the UK balance of payments sustainable: Discussion." Economic Policy 11: pp383-387.
[15] Muellbauer, J. and Murphy, A. (1990) "Is the UK balance of payments sustainable?" Economic Policy 11: pp348-395.
[16] Nash, J. (1950) "Equilibrium points in n-person games." Proceedings of the National Academy of Sciences 36: pp48-49.[
17] Nash, J, (1951) "Non-cooperative games." The annals of Mathematics 54: pp286-295.
[18] Pagano, M. (1990) "Is the UK balance of payments sustainable: Discussion." Economic Policy 11: pp387-390.
[19] Peterson A. W. A., Pratten, C. F., and latch, J. (1997) "An Economic Model of the Demand and Need for Social Housing." Department of the Environment, Transport and Regions.
[20] Sinai, T. and Souleles, N. (2005) "Owner-occupied Housing as a Hedge Against Rent Risk." Quarterly Journal of Economics 120: pp763-789.
[21] Skinner, J. (1989) "Housing Wealth and Aggregate Saving." Regional Science and Urban Economics 19: pp305-324.
[22] Skinner, J. (1993) "Is housing wealth a sideshow?" NBER Working Paper 4552.
[23] Tanner, S. (1998) "How much do consumers spend? Comparing the FES and National Accounts." in J.Banks and P.Johnson (eds.) How reliable is the family expenditure survey?, Institute for Fiscal Studies: London.
[24] White, H. (1980) "A heteroskedasticity-consistent covariance matrix estimator and a direct test for heteroskedasticity" Econometrica 48: pp817-838