Income and Wealth Effects of Italian Households 1

Transcription

1 Income and Wealth Effects of Italian Households 1 CharlesGrantTuomasPeltonen WORK IN PROGRESS Key Words: Consumption Housing Financial Wealth MPC Wealth Effect JEL codes: D12 E21 October The opinions expressed herein are those of the authors and do not reflect those of the European Central Bank. Corresponding author: Tuomas Peltonen European Central Bank and European University Institute. tuomas.peltonen@ecb.int. Charles Grant is affilated to Finance and Consumption European University Institute and Reading University. We would like to thank Michael Artis and Athanasios Tagkalakis for their support. All remaining errors are our responsibility.

2 Abstract The study quantifies the stock market and housing market wealth effects on households non-durable consumption using Italian household panel (SHIW) data of Homeowners and stockholders increase their consumption when the change in income is predictable. All households react to current changes in income and stockholders react to changes stockmarket wealth consistently with theory. However we found statistically significant housing wealth and negative effects for the oldest households when the wealth shock was positive. This can be explained be the regulation of the Italian housing market and that older households often provide the downpayment that enables their children to own their own home.

3 1 Introduction In the recent years there has been a considerable interest in investigating household portfolios as survey data on households asset allocations has become available. Although housing and other real assets still constitute the largest share of households wealth there has been a rapid increase in the households participation into risky asset markets. During the last decade many OECD countries have witnessed a significant increase in the households stock market investments either directly or indirectly through mutual funds. By the end of the 1990s about 50 percent of households in the US and Sweden and over 30 percent in the UK were investing directly or indirectly in the stock market. In the Netherlands Italy France and Germany the proportion was between 15 and 25 percent but it has increased significantly even in these countries (Guiso et al. 2002). There has also been a concomitant and substantial decline in private sector saving rates in several OECD countries and this reduction in saving rates has caused considerable debate. Policy makers have noticed the large increase in asset values in housing and equity markets and rightly have questioned if the reduction in saving rates and the increase in asset values are connected. In common with other industrialised countries Italy enjoyed a stock market boom in the late 1990s and a bust at the beginning of the new century. House prices were also increasing sharply during the period of study hence a study of Italy can complement those of other countries. The objective of the study is to analyze both direct and indirect wealth effects of housing and equity assets on non-durable consumption using the Survey of Italian Household and Wealth (SHIW) published by the Bank of Italy. This data contains detailed information on the consumption income and asset position of Italian households from and is constructed as a panel. This is almost unique as previous studies such as Attanasio Banks and Tanner (2002) have had to construct synthetic cohorts since their data was collected for a different set of households each year or else to impute the households wealth holding from a set of observable characteristics. In either case as will be explained in the main part of the text there arises an important identification problem since it becomes difficult to distinguish between a wealth shock and a shock to the cohorts future income which are likely to be correlated. For example highly educated households have steeper earnings profiles and are more likely to hold assets in the form of housing and equities. The 1

4 richness of our data will mean that we can address this problem. We will identify the stockowners and houseowners and will separate capital gains and losses from changes in saving behaviour. The data will also allow us to investigate the changes in consumption behaviour as other household characteristics differ. A major drawback of the study is that the direct stock market participation rate in Italy is still fairly low hence more effort will be given evaluating the housing market wealth effects. This seems sensible since even in anglo-saxon countries housing typically forms a larger proportion of the households overall asset position than equities. The information that can be obtained from SHIW documented by Guiso and Jappelli (2002a) among others is almost unique in containing detailed data on income consumption and wealth. Therefore it is possible directly to estimate the wealth effects on households consumption without constructing cohort summary statistics or making compromises on consumption or wealth measures as is the case for many survey data sets that have been used earlier for these purposes. Most importantly the richness of the dataset also allows us identify the stockowners and houseowners as well as separate capital gains and losses from changes in saving behaviour. To our knowledge this is the first study investigating real and financial wealth effects in Italy using panel household data 1. Italian consumers are also interesting in their own right. While the behaviour of asset values has been similar to other countries their are some unique features of the Italian market which make comparisons between the behaviour of Italian households and households elsewhere particularly interesting. Fewer Italian households participate in the stockmarket either directly or indirectly. Moreover the housing market is rather different as mortgage terms are lower and mortgage credit more difficult to obtain. While homeownership rates are not especially low by international standards households typically buy their first house at a later age and young Italians are more likely to live with their parents into their thirties. As we will discuss it seems likely that there is considerable unmet housing demand. Moreover unless the property is sold it is much more difficult to realise any of the capital gains or losses. That is the housing market seems relatively illiquid 1 A contemporaneous study by Paiella (2004) investigates wealth effects in Italy using the SHIW dataset. However her approach differs from ours as she estimates a consumption function using pooled cross-sectional data. 2

5 not only compared to the stockmarket but also compared to housing markets in Britain and the US. By estimating marginal propensities to consume in Italy and comparing estimated coefficients with those in the US and the UK the likely effect of the different regulatory environment can be assessed. As we will see the Italian results are different. The paper is organised so that section 2 discusses the permanent income hypothesis and how it motivates the regressions that we will run. In section 3 the results estimated in the other literature are discussed. Section 4 explains the particular features of the Italian market that makes it interesting and the data is discussed in section 5. Our results are reported in section 6. 2 Theoretical considerations Many studies have estimated the marginal propensity to consume (henceforth MPC) out of changes in income and changes in wealth. The two most common assets that have explicitly been studied are housing wealth and equities. Our aim is to compare the MPC of Italian households with those elsewhere (and particularly with the US and the UK) hence we will recapitulate the theory with the assumptions made elsewhere before describing our results. Most studies have either implicitly or explicitly employed an Euler equation format. Assuming that utility is additive separable and is discounted at rate β then the consumer problem for household i can be written: " T # X max E t u (c it it ) β s t s=t (1) such that JX JX A j is = A j is 1 1r j s yis 1 c is 1 (2) j=1 j=1 where E t denotes expectations at time t c it denotes the households (non-durable) consumption and y it denotes their labour income r j it denotes the rate of return of asset j (and Aj it denotes how much of that asset is held). Finally it denotes a set of taste-shifters and other factors that affect 3

6 the marginal utility of consumption. If asset j is held then the first order condition for this problem is: u(c it it ) c it ³ = E t β 1rt1 k u(cit1 it1 ) c it1 (3) Most studies impose that the utility function is isoelastic with a coefficient of relative risk aversion γ and separable taste shifters and hence estimate an euler equation of the form: ln c it = a 0 1 γ rj it εj it1 (4) where ε j it1 is the forecast error and if we assume the higher order terms are time invariant we can write: a 0 =lnβ γ ³ h ³ i 2 var ( ln c cit)varln 1r j t γcov ln 1r j t (c it) (5) Rational agents and arbitrage opportunities imply that any changes in the value of assets at time t is not predictable at t 1. Changes in income on the other hand may well be predictable. In the excess sensitivity test as formulated by Hall and Mishkin (191) and Attanasio and Weber (1993) the income growth in the current period is predicted by past income growth and is added as an extra term in the regression in equation 4 so it becomes: ln c it = a 0 1 γ rj it εj it1 αe ( y it I it 2 ) it (6) Theory argues that households should not react to these predictable changes in income hence a test of the significance of α is a test of the theory. Because we want to compare the behaviour of Italian households with those elsewhere we will report some results based on this equation and compare our estimated coefficient with those estimated elsewhere. Of equal interest is the response of households to unpredictable changes in income or in the value of their assets. We will thus also 4

7 report results for the regression of changes in consumption against changes in income and wealth. That is for those households that hold the relevant wealth item the regressions take the form: where W h it ln c it = a 0 1 γ rj it εj it1 α 1 y it α 2 W h it α 3 W s it it (7) is housing wealth and W s it is wealth held in equities. Because the MPC may well be different at different points in the lifecycle some regressions will compare the estimated coefficients for different age groups. Not all studies have estimated the MPC of households in this form. In the literature not all papers have estimated equation 7 using household level data. Instead they have estimated a version of this equation on cohorts constructed from observable characteristics. For instance Mankiw and Zeldes (1991) estimate a version of this regression using the Panel Study of Income Dynamics (PSID) and constructed cohorts based on asset holdings in 194. They find stockholders are much more sensitive to changes in the SP500 index than non-stockholders. Attanasio et al (2002) argue this does not account for changes in the composition of stockholding. Their paper used UK data but only had a time-series of cross sections. They used a set of observable characteristics to construct a cohort of likely stockholders and likely non-stockholders and regressed the average stock return against the average change in consumption of the two groups. However those households who are predicted to own stocks have other characteristics that are likely to affect their consumption behaviour. For instance they predict that educated households are more likely to own stocks. But suppose that there was some education biased productivity shock. In which case the future wages of educated workers is likely to increase (and hence their current consumption) at the same time as the profits of firms (and hence the value of equities). This indirect channel makes simple interpretation of the regression results more difficult. Similar arguments are true for housing wealth. Thus we believe that it is important to estimate the MPC of households using household level data where there is likely to be considerable heterogeneity in the cross-section of changes in income and wealth that can usefully be exploited. Nevertheless these indirect effect can also be interesting hence we evaluate the following model 5

8 ln c it = a 0 1 γ rj it εj it1 α 1 y it α 2 W h it α 3 W s it α 4 index it () where index is the change in the stock market or housing market index. 3 Literature survey 3.1 Housing wealth effects Given that home is the largest form of household wealth holding even small responses to fluctuations in its value can generate large macroeconomic effects. However the empirical evidence of the MPC of housing wealth is mixed. The time series literature which estimates aggregate versions of equation 7 concentrating on house price effects is reviewed by Poterba (2000). The US studies have generally found the MPC of housing wealth to be (Peek 193 and Skinner ). In a recent multi-country study Case et al. (2001) found the MPC of housing wealth to be for 14 developed countries in which seems to be rather high. In another study using data for 16 OECD countries Ludwig and Slok (2002) reported that the sensitivity of consumption to changes in equity wealth in Italy estimated to be 0.03 is about two times as large as the sensitivity to changes in housing wealth. An example of studies using US household data is the paper by Engelhardt (1995) who found the MPCs of housing wealth to be of size In addition he found the response to housing wealth changes to be asymmetric with greater sensitivity of consumption to falls in wealth rather than increases. For the UK Carruth and Henley (1990) Miles (1997) and Disney et al. (2003) reported the MPC of housing wealth to be For Japan Hori and Shimizutani (2003) investigated the wealth effects of housing with Japanese Panel Survey of Consumption data and found that the MPC of housing gains to be around but not statistically significant. Finally in a contemporaneous study to ours Paiella (2004) investigated the real wealth effects using the SHIW. Her approach differed from ours as she estimated a consumption function using cross-sectional data and therefore her results will not give information whether wealth effects occur in the short run. However she found that the MPC of real wealth to be around 2.4 percent. 6

9 3.2 Equity wealth effects A classic paper that studies household consumption and stock market returns is Mankiw and Zeldes (1991). Mankiw and Zeldes (1991) and later Poterba and Samwick (1995) split their sample into stockholders and non-stockholders and assessed the differences in their consumption behaviour. The first two papers used the US PSID which has a long panel of households but unfortunately does not measure consumption expenditure particularly well. Nevertheless both studies found that stockholders consumption is more volatile and more highly correlated with the excess return of stock market than the consumption of non-stockholders. Attanasio et. al (2002) used the UK Family Expenditure Survey (FES) which does not contain a panel component. Hence they developed a method for separating the likely stockholders from non-stockholders enabling them to control for changing composition of stockholders as well as the selection to the group. They also found the consumption of stockholders was more highly correlated with stockmarket returns than the consumption of non-stockholders. They also found they could not reject the predictions of the consumption capital asset pricing model for the group of households predicted to own both bonds and stocks. The shortcoming of their analysis was the lack of real panel data for consumption and asset holdings. The stock market wealth effect has been studied with micro datasets by Dynan and Maki (2001) and Maki and Palumbo (2001) for the USA. The approache of Dynan and Maki (2001) is closest to our study while Maki and Palumbo (2001) combined the Flow of Funds Accounts data ( ) and Survey of Consumer Finances for their cohort -analysis. They found the marginal propensity to consume out of net worth to be 3-5 percent. Dynan and Maki (2001) used in their study quarterly household level data from Consumer Expenditure Survey (CEX) between 193 and The drawback of using the CEX data set is that not only is the panel short (households are interviewed for four quarters) but the balance sheet information is reported only once net purchases of assets are measured with a major error and the assets are aggregated. In addition the wealth and other variables are topcoded for as many as 16 percent of households in some years. Nevertheless they estimate the MPC to be between 5-14 percent. Hori and Shimizutani (2003) investigated the equity and housing wealth effects in Japan. However their data lacked information of the value of stocks 7

10 which they imputed. After all they did not find statistically significant wealth effects in Japan. Finally Paiella (2004) found using pooled cross-sections of SHIW that the MPC to consume out of financial wealth is 9.2 percent in Italy. As far as the macro studies are concerned Ludvigson and Steindel (1999) found in their study with the U.S. data that the MPC out of stock market wealth is but suggested that there is some instability in the coefficients over time. Recently Ludwig and Slok (2002) and Bertaut (2002) investigated the stock market wealth effect using international macro panels. Bertaut (2002) found evidence of a significant wealth effect in the USA in the United Kingdom in Canada and in Japan. She also found emerging evidence of more important wealth effects in some smaller European countries where equity issuances are more common. Ludwig and Slok (2002) studied the stock and housing wealth effects using panel cointegration methods with a quarterly panel of 16 countries divided into bank-based and market-based economies. They found that the long-run elasticity of consumption in stock market wealth is twice as large in market-based economies as in bank-based economies and that the elasticities have increased significantly through time. 4 Household Portfolios in Italy According to Banca D Italia (2004) the median household net wealth the sum of real assets (property companies and valuables) and financial assets (deposits government securities equity etc.) net of financial liabilities (mortgages and other debts) was euros. Real assets constituted the largest share of household net wealth with a median value of euros while the median value of financial assets was 7066 euros. Higher values of financial assets were observed for households where the heads were university graduates (2240 euros) managers (25696 euros) or self-employed (155 euros). The value of financial assets varied significantly with the geographical location: 50 percent of households in the south and in the islands owned less than 2732 euros in financial assets against euros in the north and 9743 euros in the centre of Italy. On average the amount of durable goods owned by households was 1750 euros of which 73 euros were in vehicles in Finally the households median net income was 276 euros in 2002.

11 4.1 Housing wealth In the last three decades the home-ownership ratio in Italy has increased notably (from 46 percent in 1961) and is currently above the European Union average. According to Banca D Italia (2004) 6.5 percent of households are owner-occupiers; 20.9 percent are tenants; 10.6 percent are occupying under other arrangements. Although the share of home-owners in Italy is similar to other major economies the Italian housing market is very different to the UK and the USA. Firstly the household age-tenure profile in Italy is sharply different from the profile in Anglo-Saxon countries and in Scandinavia. Chiuri and Jappelli (2003) report that in Italy individuals buy their first house at older ages than elsewhere. By the age of 30 the home-ownership ratio in the Anglo-Saxon countries is around percent but is only percent. They also report the differences in the mortgage market. For example between 196 and 1996 the outstanding mortgage loans were on average 5.49 percent of GDP whereas in UK and USA the ratios were 51.7 percent and percent respectively. Of the EU countries only in Austria were the outstanding loans of GDP lower than in Italy. Moreover the average downpayment ratio was 40 percent in Italy but averages only percent in the EU 11 percent in the US and only 5 percent in the UK. In 197 the minimum downpayment ratio was regulated at 25 percent (a reduction on previous levels) although this restriction was abolished in The period over which the debt was repaid was also shorter in Italy typically being over years and transaction costs were higher. Lastly the spread between saving and lending rates were higher than elsewhere. Part of the explanation for this under-development of the Italian mortgage market may be due to the difficulties in foreclosing. On average it took lenders 4 months to foreclose a mortgage in Italy but only 9 months in the US and less than 5 months in the UK. Judicial inefficiency is likely to make lenders more reluctant to offer mortgages. Despite these facts relatively few Italian households rent rather than own their home. In common with many other countries homeownership confers tax advantages. Moreover the rental market is highly regulated which makes it difficult to alter the rent or to evict sitting tenants. This contributes to a reluctance to supply rental property. The fact that the mortgage market is underdeveloped but homeownership rates are high suggest that households finance their house purchases through their own means rather than through 9

12 financial intermediaries in the formal lending sector. There must be offsetting factors such as intergenerational transfers that alleviate the impact of mortgage market imperfections on households housing decision. Engelhardt and Mayer (199) and Guiso and Jappelli (2002b) both found that bequests gifts and other inter vivos transfers shorten the saving period before home ownership and increase the value of the house purchased. Guiso and Jappelli (2002b) report that about one-third of Italian home-owners reply that they have received the house itself as a gift or as a bequest or have received financial support for purchasing an apartment. For the US Engelhardt and Mayer (199) find that one in five first-time home buyers receives a financial transfer from a friend or relative to help fund the first downpayment and these transfers typically account for more than half of the down payment. However they conclude that gifts are a poor substitute for efficient credit markets. 4.2 Equity wealth In Italy the household portfolios (mainly from SHIW data set) are well documented by Guiso and Jappelli (2002a) among others. They report a large shift towards riskier portfolios and an increase in stock market and mutual funds participation in Italy during the 1990s. They also find that the increasing role of stock market investments is due not only to the increase in the participation rate but also (and of equal importance) to a sharp increase in the share of wealth invested. According to them the portfolio shift towards direct and indirect stock holding originated in the reduction of Italian Treasury bill returns in the changes in the social security system that lowered the expected future income of workers in the privatization process of public companies and in the Italian stock market growth. Also the increase in competition among the financial firms offering investment services (reduction in entry costs and financial information costs) and the availability of new financial products increased the Italian s interest in stocks. Banca D Italia (2004) reports that the direct stock market participation rate in Italy was about 9.6 percent in With the indirect participation through mutual funds and pension funds the total participation rate increases to just over 20 percent in Guiso and Jappelli (2002) found that the age profile of stock market participation is hump shaped with a peak around 10

13 the age of 50 and that the participation is generally correlated with education. Stock market participation is limited below median financial wealth levels and even in the fourth quartile of the wealth distribution it is only slightly above 50 percent. On the other hand the correlations between the amount invested in stocks and age education or financial wealth are generally weak. The strong correlation between wealth and stock market participation points to the importance of fixed participation costs (minimum investment requirements transaction costs and information costs) as a crucial element in understanding the portfolio choice of Italian investors. Due to the increased investments in the stock market total Italian stock market participation is approximately equal to the participation rate in other major European countries such as France 23 percent and Netherlands 24 percent. However it is still far from the total investment rates of Sweden 54 percent UK 34 percent and USA 4 percent as reported by Guiso Haliassos and Jappelli (2002). Mutual funds and other managed investments in Italy represented another 15 percent (peak in percent). In the 190s the direct stockholding accounted for only about 15 percent of households financial assets and indirect holding through mutual funds was virtually absent (the mutual funds were introduced in 194) Guiso and Jappelli (2002a). 5 The data 5.1 Description of the data The study utilizes the Survey of Households Income and Wealth (SHIW) published by the Bank of Italy. It is a representative sample of Italian households where the households are sampled in alternate years and since 197 there has been a panel component to the survey. In this study we focus on the panel section of SHIW having data from seven years: and Each year between 4000 and 000 new households were sampled. While many households were only sampled once a subsection of households of over 1000 households were resampled in the next survey allowing us to construct a household panel. The survey includes questions about individual characteristics and their occupational status; 2 The sample design was somewhat different in 197 and hence we omit this year. 11

14 sources of income (payroll and self-employment income pensions transfers and property income); expenditure (durables and non-durables) the properties lived in or owned by the household; financial assets and liabilities. It also includes a set of sampling weights adjusted to bring some socio-demographic marginal distributions into line with the corresponding distributions found in Istat s population statistics and labour force survey. This data set has a number of advantages compared to US and UK data. Unlike the PSID it contains consumption information on a broader range of items than just food expenditure. It also contains comprehensive income and wealth information unlike the CEX and hence it negates the need to construct cohort averages and match cohorts across different data sets. Finally unlike the FES it contains a panel component. Hence this dataset is almost unique in that it allows a household level study of the MPC from changes in income and wealth. 3 The data set used in our estimations has been cleaned in the following way. First households with fewer than two observations during the period were deleted. Secondly in order to focus on working age households and not to be concerned with composition effects and mortality risk only families with the head of household of age were included in the sample. Thirdly families with zero or negative real total consumption and real non-durable consumption were deleted as well as households from the lowest real disposable income percentile. Finally standard consistency checks were made and missing observations were deleted. 5.2 Identification of the wealth effects In contrast to earlier studies on wealth effects with the exceptions of Dynan and Maki (2001) and Hori and Shimizutani (2003) the study attempts to identify the pure wealth effects from the changes in the household saving behavior. Following the above authors the change in household non-durable consumption is regressed on changes in income and on the capital gains from owneroccupied housing and equities. Ideally the capital gains for the asset could be calculated from the reported change in the value of the asset net of any sales and purchases. However the value of the 3 Other valuable datasets that include real and financial information are e.g. DNB Household Survey (DHS) and the Japanese Panel Survey of Consumption. 12

15 Indices 199=100 sales and purchases is not reported in non-survey years (although there is sufficient information to identify when sales and purchases took place). This resulted in a small number of households being excluded. Nevertheless we are left with 1150 observations for the owner-occupied housing capital gains. 4 Figure 1 below depicts a stock market index together with a residential property price index for Italy The average (nominal) annual capital gain on Italian housing market was.0 percent in To our knowledge this is the first study investigating the pure wealth effect of owner-occupied housing on non-durable consumption since e.g. the approach by Hori and Shimizutani (2003) using households which household ownership status has not changed is not valid because households could have changed their residences even though their ownership status has not changed House price index Share price index Figure 1: Italian house price (whole country existing dwellings) and share price indices Sources: Banca d Italia based on data of Il Consulente Immobiliare and IMF IFS September Furthermore the availability of data on the net purchases of equities (publicly listed) limited 4 We focus on the owner-occupied housing where according to Cannari-D Alessio (1990) measurement error is likely to be the smallest. 13

16 us to focus only on households where the number of equities has remained the same between two surveys. This means that a small number of observations of equity wealth were left out the analysis because pure wealth effectscouldnotbeidentified. Also some observations with indirect stockholding through mutual funds were excluded because the wealth effects could not be identified. The average (nominal) annual capital gain for the Italian stock market in was 9.0 percent. One should note that the capital gain on stocks might be measured with error because although we limited ourselves to households whose number of stocks has not changed there still might be portfolio reallocations which might bias the wealth effect estimate. Hori and Shimizutani (2003) avoided this problem because their dataset included the data on net purchases of securities. However their dataset did not distinguish between the security types and their stock capital gain was based on assumption that certain percentage of assets were stocks. Dynan and Maki (2001) instead used capital gains calculated from the average return of the stock market because of measurement error issues. Our paper will instead use the values reported by households themselves: we would expect households to react rather to the changes in their portfolio (direct wealth effect) than to some representative portfolio (indirect wealth effect). 6 Empiricial results 6.1 Excess sensitivity test There were 3074 households for which there was a sufficiently long panel to conduct the excess sensitivity test. In this regression non-durable consumption was regressed against predictable changes in disposable income the interest rate and some household characteristics. Income and the interest rate were predicted using dummies for area of living interacted with occupation and with the employment sector income level and change all lagged twice as instruments as well as the real interest rate lagged once and twice. The regression was estimated on the whole sample and for subsamples of households (defined by their status in year t 1). These groups were self-employed households and all employed households stockholding employed households homeowning but not stockholding employed households and non-stockholding non-homeowning employed households. The log of 14

17 non-durable consumption was regressed against the log of non-financial disposable income the real interest rate and the following control variables: age age squared age cubed schooling dummies dummy for female head dummy for married log size of the household and its square and birth cohort dummies. The regression results reported in table 1 show that for the whole sample predictable changes in labour income are not significant at least changes in labour income that are predicted by the econometrician. However the interest rate is significant and the estimated effect is quite large. Of the other variables only married and log-family size are significant at the 10 percent level. From this regression the permanent income hypothesis that is motivated by equation 6 can not be rejected. However when only employed households are included in the regression in the second column of results then predictable changes in income has a coefficient of 0.30 which is significant at the one percent level. Column 3 shows that for self-employed households predictable changes in income are not significant. This result is not completely surprising. Measuring income is problematic for the self-employed and hence may not be reported particularly accurately. This measurement error would downward bias the estimated coefficient. The last three columns investigate the effect of predictable changes in income for stockholders homeowners and renters. The results for stockholders are significant at the 5 percent level and for homeowners significant at the 1 percent level. Renters on the other hand are not significant at significant at conventional significant levels (although the estimated coefficient is significant at the 10 percent level). Moreover the pattern of the estimated coefficient shows that homeowners increase their consumption relatively more when they experience a predictable change in income compared to stockholders but that both groups are much more sensitive to these changes than renters. The lower level of statistical significance as well as the lower level of income coefficient for households that do not own their primary residences and listed equities is somewhat puzzling as these households might be expected to be most sensitive to predictable income changes. This pattern would be suggested if young households typically rented their homes because they were constrained and unable to afford the downpayment requirements to buy their own property. However such households would not spend their extra income on non-durable consumption but rather 15

18 would allocate any extra monies to the downpayment. Households who do not own a home will increase their saving in order to buy one later on. Homeowners however if they are constrained and unable to increase their borrowing when they already have a housing loan may well only increase their non-durable spending when they actually obtain the money. The real puzzle concerns the stockholding group. Holding equities would seem to preclude being credit constrained and yet the estimated coefficient for this group is two-thirds as high as for the homeowning households. 6.2 Direct wealth effects The direct wealth effects of housing and equity wealth were estimated based on equation 7. Regressions were run separately for all households and for households whose head was employed and self-employed. Within the employed sample separate regressions were run for homeowners for stockholders (who own listed equities) and for households who neither owned their home nor ownedanystock(whichwewillcallrenters). In addition the housing wealth effects were also estimated for different age groups as well as for positive and negative wealth gains. The observations were weighted to match the underlying population and White errors were calculated. Table 2 reports the estimated marginal propensity to consume estimated in logs. The first column shows the results when the whole sample is used. It finds that the MPC of 25 percent out of changes in labour income is 0.0 percent from changes in housing wealth and 0.65 percent from changes in stockmarket wealth. The regression also finds that the interest rate having a female head and being married are also significant. The estimated MPC for income and for stockmarket wealth is significant but for housing wealth it is not significant. This suggests that households are reacting to changes in stockmarket wealth but not to changes in housing wealth. The second and third column compare self-employed and employed households. Self-employed households have a lower MPC from labour income. For neither employed nor self-employed households is the MPC of housing wealth significant. Indeed for employed households the estimated coefficient is negative which can not be plausible. Changes in stockmarket wealth are significant for employed households but not (at the 5 percent level) for the self-employed. Nevertheless the estimated coefficient is similar in both cases. Comparing the interest rate effects the estimated co- 16

19 efficient is significant for the self-employed but smaller and no longer significant for the employed. These results are consistent with income and wealth being less well measured for the self-employed households. The fourth fifth and sixth columns compare the employed stockholders the employed householders and the employed households who neither own stocks nor their own home. The estimated MPC from changes in labour income is around 40 percent for the homeowning and stock-owning household but is significantly lower for the last group. Similarly to before changes in housing wealth are not statistically significant and the estimated coefficients are economically small. However for the stockholding group the estimated MPC from changes in stockmarket wealth is statistically significant. The size of the effect is one percent which is moderately large. By contrast the interest rate effect is only significant for the homeowing group. However the estimated coefficent is larger for the stock-owners. The estimated coefficent is negative for the renters but much smaller in absolute size and not significant. One interpretation of the fact that renters seem to react less to changes in income that homeowners or stockholders is that they are less sensitive to changes in income. That is they receive the same income shocks but react less. For instance if credit constraints are more often binding for homeowners rather than renters then this group would react more to the same change in income. This seemed to be the story from table 2. However current income changes may in part be predictable and it is possible that the predictable component of the change in income is higher for renters. Even if the change in income was not predicted it will be made up of permanent and temporary components and households ought to change their consumption by more in response to a permanent shock than a temporary shock. However we do not know how renters homeowners and stockholders differ in their income shocks or the predictability of their changes in income. The effect of changes in wealth are less ambiguous. If we assume that the riskiness of each type of asset does not change over time then arbitrage conditions would suggest that changes in the face value of the asset are lump-sum unanticipated wealth shocks. This is similar to obtaining a temporary income shock and we would expect the household if markets were perfect to spend only the annuity value of the wealth shock. If we look at stockmarket wealth then the MPC is 17

20 0.65 percent in the whole sample and is 1.0 percent if we only look at the stockholders. These coefficients are in line with what is predicted by theory. However the estimated effect for housing wealth is only one tenth of this effect (and even in some cases is negative). Note that the small size and non-significance of these coefficients can not be due to the small sample size since fewer households own stocks but the estimated coefficient on stockmarket wealth was statistically significant. This suggests that housing wealth has a negligible effect on consumption and is neither statistically nor economically significant. One explanation for this is that stockmarket wealth is liquid but the restrictions in the housing market in Italy compared to the UK and the US means that it is much more difficult to realise the appreciation in the value of the householder s home. The last two columns of table 2 report regression results for those households who reported a positive wealth shock and those households who reported a negative shock. For both positive and negative shocks the MPC from income changes is positive and around 30 percent. Households seem to react slightly more when they have a positive wealth shock but the difference is not very large. The effect of changes in changes in wealth held in stocks is significant for the positive shock households and is around 1.3 percent. For the negative shock households the effect is around much smaller being 0.09 percent and is far from being significant. The interest rate effect is positive for both groups but is larger (and significant at the 5 percent level) for those households whose wealth went down. More curious is the behaviour of changes in housing wealth. Households whose wealth went down show an MPC out of changes in housing wealth of 0.1 percent which is small and statistically insignificant. But for households who had a positive wealth shock their MPC from changes in housing wealth was negative was -0.6 percent and is highly significant. That is housing wealth seems to enter with the wrong sign. Recall that many more households own their home compared to households who own stocks so for most of these households their increase in net wealth is because their home has become more valuable. This seems to contradict one of the motivations of this study: the expectation that reductions in the saving rate were due to increases in the value of household assets. We will return to this discussion. 1

21 6.3 Differences across Age-groups Table3looksatdifferences between different age-groups in the in the population. The first three columns repeat the analysis in the first column of table 2 but looking at households between 30 and 40 those between 41 and 50 and those between 51 and 65. The estimated MPC from changes in income is 16 percent for the youngest households 22 percent for middle-aged households and 26 percent for the oldest households. Notwithstanding that their may be differences in the predictability and the permanence of the changes in income older households are likely to have a shorter planning horizon and hence we might expect these households to react more to one-off shocks. This may partly explain the higher MPC from changes in income for the oldest group. Looking at changes in wealth the estimated effect is negligible and insignificant for the middle group. The youngest group a negligible MPC out of changes in housing wealth but the estimated effect for stockmarket wealth is larger at 0.6 percent and marginally not significant at the 5 percent level. The oldest group have the largest estimated effect. Although the effect of changes in housing wealth is not significant for stockmarket wealth the effect is over 1 percent and is significant. The next six columns look at positive and negative changes in asset values. As before the MPC from changes in income is smaller for young and middle-aged households than for old households whether the change in asset values has been positive or negative. However there does not seem to be a significant difference in the MPC from income changes for those households whose assets increased in value compared to those households whose assets fell in value. The MPC from changes in stockmarket wealth is around one percent for the oldest agegroup whether the change in overall wealth was positive or negative although only for the positive wealth change was the coefficient statistically significant. The estimated coefficient was much smaller for middle-aged households and not significant. For youngest households they did not react significantly to a negative shock but the positive shock had a large and significant effect at 3.6 percent. Neither young nor middle-aged households whose overall wealth went down reacted much to changes in their housing wealth. Young households did not react to changes in their housing wealth when the overall change in their asset position was positive either. The effect for middleaged households was to reduce their consumption by 0. percent which is marginally not significant 19

22 at the percent level. The oldest group reduced their consumption by 0. percent (the same as the middle-aged households) when their overall wealth shock was positive and their MPC was 0.7 percent when their overall shock was negative. Given that for most households who owned assets housing is by far its most important component this suggests that the oldest households reduce their consumption when the value of their house goes up or goes down. The behaviour of households when their stockmarket wealth changes seems to be consistent with standard theories but explaining their behaviour when house prices change is more problematic. The fact that older households are reacting to changes in housing wealth but younger households are not is puzzling. Moreover reducing consumption when house prices increase does not seem to be consistent with the story that was told earlier about households not being able to realise their asset gains when house prices increase. Here we seem to have households significantly reducing their consumption when house prices increase. For an explanation recall that in the Italian housing market for much of this period their were strict downpayment rules for mortgages and that the term of mortgage loans is relatively low. Consequently people tend to buy their first home at later ages than elsewhere. Moreover problems in the rental market means that these people often remain with their parents for a longer time. Recall that Guiso and Jappelli (2002) report that one third of households receive a gift from friends and relatives to enable them to pay their deposit. These transfers are typically from the parent. Hence increasing house prices would increase the size of the transfer that the parent would need to make to the child and enable them to move out of the parental home. This would be consistent with the increase in house prices reducing the level of consumption of older households. In contrast younger households who have no way of realising any change in the value of their home do not change their consumption behaviour. We believe this rationalisation requires further study. 6.4 Indirect wealth effects Unlike in this paper previous estimates of the MPC of changes in wealth have not been able to directly measure each households change in wealth. However as is well known changes in wealth are likely to be concurrent with changes in the wider economy. For example the value of stocks 20

23 reflects the future profitability of firms. If this profitability is due to productivity gains then at least part of these productivity gains will be captured by higher wages (either now or in the future) - which will also increase households consumption. Using an index uses the approach of Mankiw and Zeldes. They argued that stockowners react more to aggregate stock market return than nonstockowners and were therefore partially able to explain the equity premium puzzle. However if different types of households have different portfolios then again inference about the MPC of wealth gains can result in misleading inferences. For example if educated households are more likely to hold wealth on the stockmarket then skill biased technological progress will both increase these households stockmarket wealth since firms become more profitable and also the wages of educated households. Similarly house prices change with local economic conditions which will also affect current and future wages. This makes regressions of aggregate consumption against aggregate house prices or aggregate changes in the stockmarket difficult to interpret as pure wealth effects. Even partitioning the data between likely stockholders and unlikely stockholders as in Attanasio et. al. (2002) will not solve this problem if the distribution of wealth shocks is correlated with shocks to permanent income. Carroll et. al. (1994) argued that it is important to look at the behaviour of individual households since at the individual level their is likely to be enough heterogeneity in shocks to asset values and shocks to income to be approximately uncorrelated. Nevertheless looking at the change in the value of the assets that the household holds and in an index of asset values is interesting hence we report some results based on equation. The results will be reported shortly! 7 Conclusions This study has analyzed the effect of changes in the value of housing and equity assets on nondurable consumption using Italian household panel data from With this unique dataset containing detailed information on income consumption and wealth it is possible directly to estimate the wealth effects on households consumption without the need to use matching techniques to construct the data set for the analysis or to make compromises on consumption or wealth measures 21