RDP 2014-04: Home Price Beliefs in Australia 7. Financial Decisions and Home Valuation Differences

7.1 Homeowners

We now examine whether home valuation differences are associated with the economic decisions made by homeowners. In other words, in a regression framework, we now treat the valuation differences across postcodes as an independent variable. Specifically, we assess whether valuation differences are correlated with household spending, leverage and portfolio decisions. For instance, if optimistic homeowners typically overestimate the value of their homes we might expect that they spend more than pessimistic homeowners because they believe their lifetime wealth to be relatively high. Moreover, we might expect these optimistic homeowners to hold relatively more housing debt, on average, and to allocate a higher fraction of their financial portfolios to risky assets.^[10]

To demonstrate the effect of valuation differences on household decisions, we first estimate postcode-level regressions of the following form:

where Y_pt is a measure of financial decisions for the average homeowner in postcode p in period t. The measures of household decisions include a spending measure – the (log) level of household consumption expenditure (SPENDING_pt) – measures of household leverage – the (log) level of total debt (DEBT_pt) and the (log) level of total housing debt (HDEBT_pt) – as well as portfolio allocation measures – the average share of wealth held in financial assets (FINSHARE_pt) and the average share of wealth held in equities (EQSHARE_pt). The key explanatory variable is the average self-assessed home value in each postcode and year.

We also estimate similar regressions in which we decompose the average self-assessed home value into the average home valuation difference () and the average home sale price :

Equation (8) is a more general specification than Equation (7) because we allow the estimated coefficients on the average valuation difference (β₁) and the average home sale price (β₂) to potentially differ from each other.

The key explanatory variable is the average home valuation difference made by homeowners in each postcode and year . The specification includes a set of control variables for the average household in each postcode and year (X_pt), which is similar to that used in Equation (6). The estimates of Equations (7) and (8) are shown in Table 3 for each dependent variable. For brevity, we do not report the coefficient estimates on the control variables. A more detailed analysis of the regression results, including a discussion of the estimated effect of the control variables, is available in Appendix D.

For postcode-level spending, the positive coefficient on average self-assessed home values (column 1, block 1) indicates that household spending and self-assessed home values are positively correlated. A 1 per cent increase in the level of housing prices is associated with a 0.31 per cent increase in household spending. This result implies a marginal propensity to consume of around five cents for a one dollar increase in the self-assessed value (using an average spending-to-housing wealth ratio of 16 per cent), which is broadly in line with the magnitude of ‘housing wealth effects’ documented in previous studies for Australia (see, for example, Dvornak and Kohler (2007) and Windsor et al (2013)).

Decomposing the average self-assessed home value into the average home valuation difference () and the average home sale price (column 2), we find that the positive wealth effect reflects two factors: first, that higher market values are associated with more spending and, second, that greater overvaluation is also associated with more spending. A 1 percentage point increase in the valuation difference is associated with a 0.34 per cent increase in household spending, on average. To gauge the economic significance of this result, the regression estimates imply that a 1 percentage point increase in the valuation difference would generate about the same effect on spending as a 2 per cent increase in household income, so the effect seems reasonably large.

We also find that postcodes in which the average homeowner appears to overestimate the value of their homes are those that typically have more debt (column 2, blocks 2 and 3). This is true for total debt, (DEBT_pt , block 2) and, more specifically, housing debt (HDEBT_pt , block 3).^[11] For example, the estimates indicate that a 1 percentage point increase in the valuation difference is associated with a 0.8 per cent rise in debt, on average.

We find that postcodes in which households generally overestimate the value of their homes are those in which households allocate a relatively high share of their wealth to financial assets, such as equities (column 2, blocks 4 and 5). The estimates indicate that a 1 percentage point increase in home valuation differences is associated with a higher average share of wealth held in financial assets of about 16 basis points. In terms of economic significance, this is roughly the same effect as 2 per cent more household income. We also find that higher average housing prices do not significantly affect the share of wealth held in financial assets (block 4), but do affect (increase) the share of wealth held in equities (block 5).

In sum, we find that, at the postcode level, valuation differences are positively correlated with the level of spending, debt and the share of wealth held in financial assets. In other words, postcodes that appear to overvalue their homes typically spend more, have higher leverage and choose riskier portfolios than postcodes that do not. Since this effect exists after controlling for the average level of prices, it suggests that these valuation differences may be capturing underlying optimism or pessimism.

7.2 Renters

To check whether home valuation differences are likely to be capturing optimism or pessimism among homeowners, we estimate the same postcode-level regressions on the decision variables of renting households. If valuation differences are capturing true differences between homeowner beliefs and sale prices then the valuation differences should only affect the decisions of homeowners and not renters. If, instead, we find that renters' decisions are also affected by our measure of valuation differences then this could be evidence that the measures are capturing unobserved heterogeneity in the characteristics of properties sold or valued in each postcode.

To do this, Equations (7) and (8) are re-estimated for each relevant decision variable, constructed using only data obtained from renting households. Specifically, we consider the average level of renters' spending, non-housing debt (credit cards and personal loans) and the average share of wealth held in equities (Table 4).

At the postcode level, the spending of renters appears to be positively associated with market-inferred prices (column 2). This could reflect the fact that housing prices and spending are commonly associated with a third factor, such as income expectations (this is consistent with the findings in Attanasio et al (2009), for example). However, the postcode-level spending of renters is negatively correlated with homeowner valuation differences (column 2). Moreover, there is no significant relationship between renters' debt levels and either market-inferred housing prices or valuation differences. Finally, we find that the share of renters' total assets held in equities is responsive to market-inferred housing prices, which could be due to equity holdings and housing prices being commonly associated with a third factor. But, unlike homeowners' portfolio allocations, valuation differences do not explain the share of renters' wealth held in equities.

Taken together, these results are not consistent with the alternative explanation that our estimates of the homeowner valuations differences are contaminated by unobservable factors that affect the beliefs and decision-making of all households. Instead, our results support the notion that our estimated home valuation differences are capturing only the sentiments of homeowners, insofar as these valuation differences affect the decision-making of homeowners and not renters.

Footnotes

This is consistent with the findings of Brunnermeier, Gollier and Parker (2007). [10]

The results also hold for other measures of leverage, including the ratio of total debt to total assets, the ratio of housing debt to housing assets and the ratio of housing debt to household income. These unreported results are available upon request. [11]