RDP 2006-08: A Survey of Housing Equity Withdrawal and Injection in Australia 5. Characteristics of Households Withdrawing and Injecting Equity

Having identified the various methods through which households withdrew and injected equity during 2004, it is of interest to consider whether there are common characteristics across households that withdrew or injected equity.

5.1 Key Bivariate Relationships

The survey data confirm that age and income are key variables in distinguishing households that altered their housing equity from the rest of the population. The results are consistent with previous work that show age and income to be important determinants of the incidence of homeownership with debt (see Ellis et al 2003). They also confirm that households that own property, particularly those with housing debt, are most readily able to withdraw or inject equity.

Figure 2 shows the age profile of households in the survey – where age is determined by that of the household head, defined as the main income earner. Those aged between 40 and 49 accounted for the highest proportion of households that changed housing equity, and the highest proportion of property owners with housing debt. In comparison, the age profiles for all households and all property owners are much flatter. Also, withdrawers and injectors tended to have higher household incomes than the general population, as did property owners – particularly indebted property owners.

Age also differed notably between households that withdrew equity and those that injected, with withdrawer households typically older. The breakdown of average net housing equity flows from the survey data by age shows that, over 2004, households with a household head aged between 20 and 49 years were typically equity injectors (Figure 3). In contrast, older households were typically net withdrawers, with the size of the average net withdrawal increasing with age. This is consistent with the typical life-cycle pattern whereby younger households inject equity when they purchase their first home and trade up to more expensive housing in mid-life, before withdrawing equity when they sell property in their later years. Such a profile is also implied by the use of housing as an investment vehicle, given households will typically accumulate equity in their peak earning years. Indeed, of households that engaged in a property transaction and withdrew equity, just over half were 50 years of age or older, and they accounted for 61 per cent of the value of equity withdrawn by property transactors. In comparison, the same age bracket accounted for less than 40 per cent of total net injections.

5.2 Empirical Modelling

In this section, we present some formal empirical results that help to further evaluate the relative influence of different household characteristics on their propensity to inject or withdraw housing equity and on the value of such flows. We aim to address three questions, which together build towards an understanding of the drivers of aggregate housing equity withdrawal. First, what characteristics influence a household's decision to alter their housing equity? Second, for households that did alter equity, what influenced whether they injected or withdrew? Third, what factors affect the average value of such adjustments? Throughout this section, we separate households that transacted in property from those that did not (transactors and non-transactors). This treatment, supported by the data, reflects that the decision to alter equity through a property transaction is typically undertaken as part of a change in dwelling ownership, which involves a much larger set of considerations than the decision to alter equity without a property transaction.

Modelling transactor withdrawals and injections

Assessing the characteristics that influence whether transacting households adjust or maintain their housing equity turns out to be a trivial exercise, as no household in the survey that made a property transaction maintained a constant level of housing equity. Given this, we move directly to the second question of what characteristics influence whether such households inject or withdraw.^[7]

A logit model is an appropriate tool for modelling the discrete choices of property transactors. The random variable, y, is defined so that it is 0 if the household injected equity and undertook one or more transactions, and 1 if the household withdrew equity and transacted. The probability that a household withdrew equity, given it transacted property, is given by:

where x is a vector of household characteristics and β a vector of coefficients.^[8]

Results of this logit model are shown in Table 6. The model is able to identify which households injected and which households withdrew equity, with an overall accuracy rate of 77 per cent.

The role of the life-cycle is clearly evident, consistent with the bivariate analysis in Section 5.1. Households whose main income earner was in their 30s, 40s or 50s predominantly injected equity following a property transaction, while households whose head was in their 60s or 70s predominantly withdrew equity.^[9]

The results also suggest that portfolio rebalancing plays a part in determining the likelihood of withdrawal. For example, households with greater housing equity were more likely to withdraw equity following a transaction than those with less housing equity. Households with relatively easy access to housing equity as a source of funds were also found to be more likely to withdraw than inject, as evidenced by a positive coefficient on households with a larger number of properties (such that they were more readily able to liquidate part of their holdings). However, some surprising results are also evident; retirees that transacted in property were found to be less likely to withdraw than were other households, as were property-transacting couples.

To model the value of injections and withdrawals undertaken by property transactors, we use sub-sample ordinary least squares (OLS), with separate equations for injectors and withdrawers.^[10] The decision to use sub-sample OLS rests on a desire to model actual decisions, rather than possible decisions. In other words, our approach is to estimate what factors influenced the value injected or withdrawn, given that a household had already decided to inject or withdraw (the conditional probabilities). This is preferable to estimating the unconditional probabilities if the decision to inject or withdraw was taken prior to the decision regarding the amount, as we assume. The results are shown in Table 7.

Age appears to play an important role in determining the average value of withdrawals, in addition to the role it plays in influencing the propensity to withdraw. The value of withdrawals tended to be higher for households whose head was in their mid to late 30s, lower for those nearing retirement, and higher again for older households trading down or selling outright. In contrast, there is little variation in the value of injections as households aged. Diversification considerations seem to influence the values withdrawn and injected; households with large asset holdings tended to withdraw more, and those with more housing equity tended to inject less. Also, high levels of borrowing (measured by the LVR) tended to reduce the amount withdrawn, perhaps reflecting constraints against further borrowing or even that they had withdrawn substantial equity previously.^[11]

Modelling non-transactor withdrawals and injections

The appropriate framework for modelling non-transactors' propensity to inject or withdraw equity is less clear than for transactors. It is theoretically desirable that the three choices facing non-transactors – to inject, withdraw or maintain their equity – be modelled in a single framework to take account of the simultaneity of these decisions. However, estimates from a multinomial model that includes these three decisions indicate that there is little distinction between households that injected and households that withdrew.^[12] Consequently, we first model the decision of households to adjust their housing equity or maintain it using the logit framework represented by Equation (1) above. We restrict the sample to households that owned property at some time during the year, in order to abstract from households whose tenure choice precluded them from injecting or withdrawing equity. We then model the choice to either inject or withdraw equity for those households that made one of these choices, again using a logit framework. There is little loss of efficiency but a gain in clarity from this approach.

The fit of the model for the first regression is very good, with almost 90 per cent of households correctly identified. This partly reflects the fact that most households with a loan are required to make principal repayments irrespective of their other activities. Nevertheless, over 70 per cent of the households in the sample are still correctly identified in a model that removes all loan variables.^[13] Table 8 (left-hand side) presents the results from this model.

The most notable influence on the decision to adjust equity, rather than maintain it, is the age of the household. Consistent with the results for transactors and those shown in Section 5.1, middle-aged non-transacting households are found to be particularly likely to have adjusted their housing equity. In contrast, older households typically did not make such adjustments. The implied probability for households to adjust their housing equity peaks when the household head is aged 40–44 years, and remains above 50 per cent until the household head is beyond retirement age.

Portfolio-rebalancing motives again appear to be important, as investors in housing and households with larger annualised capital gains were more likely to adjust their housing equity. Furthermore, households that were more easily able to access their funds – due to loan features such as an offset account – were also more likely to adjust their housing equity. One of the potential benefits of using these facilities (as opposed to selling other assets for example) to access funds is that the household retains ownership of the (property) asset, and hence the potential to benefit from any capital gains. Finally, and somewhat surprisingly, households with lower incomes were found to be more likely to adjust their equity than those with higher incomes.

In contrast to the high prediction rate for the first regression, the second model cannot correctly identify non-transactor households as either injectors or withdrawers, with all but five households estimated to have injected – suggesting caution in interpreting the results.^[14] There are very few characteristics that are found to distinguish the two groups (Table 8, right-hand side), with many characteristics that were important in determining whether such households adjusted equity not found to be important in determining whether they injected or withdrew equity. Those households with easy access to funds (due to a line of credit or redraw facility) were more likely to withdraw housing equity, as were households that had experienced larger annualised capital gains on their property. This adds to evidence suggesting that an extended period of strong house price growth is likely to support aggregate housing equity withdrawal. In contrast, households that were ahead of schedule on their loan repayments were more likely to inject equity, perhaps indicating a pre-established preference towards investing in their homes. Retirees are also (counter-intuitively) found to inject more often than withdraw, reflecting the high incidence of renovation spending by such households.

Our difficulty in modelling decisions regarding injecting versus withdrawing equity may reflect our inability to proxy what are likely to be significant distinguishing characteristics. For example, we have no proxy for households' tolerance for risk – those that are less risk averse are more likely to be willing to make withdrawals. Similarly, we do not have information on whether households suffered temporary shocks to their income during the year, with adverse shocks likely to encourage withdrawals and positive shocks encouraging injections. A second (potentially related) possibility is that middle-aged households tend to both inject and withdraw in regular succession, depending on their spending needs at the time. This would be consistent with the finding that households that can access their housing equity relatively cheaply (through loan features such as an offset account) are more likely to adjust their equity.

To model the value injected and withdrawn by non-transactor households, we use the same methods as for transactors – that is, sub-sample OLS. This regression is better able to distinguish between injectors and withdrawers than the previous logit regression. For households injecting equity, the value of these injections tended to be largest for those in their middle years, while there was little effect of age on the value of withdrawals (Table 9). Injector households whose heads were employed full-time also tended to make larger injections, consistent with consumption-smoothing motives, although there is no evidence that larger withdrawals were made by those not working. Households with multiple incomes were found to inject less and withdraw more, perhaps reflecting the greater stability of their incomes. However, some surprising results are also evident; for example, households with high LVRs were found to adjust their equity by large amounts, regardless of whether injecting or withdrawing.

Looking at both the propensity and value of injections and withdrawals by non-transactors, it is clear that total non-transaction-based housing equity withdrawal was underpinned by households in their middle years. Such households were more likely to inject and withdraw, and, when they did inject, tended to inject larger amounts than other households. Given the similarity of both injectors and withdrawers, it is also not surprising that those with relatively cheap access to their funds were more likely to adjust (and particularly withdraw) housing equity. Portfolio-rebalancing motives appear to have had a smaller, but still important, influence on non-transaction-based housing equity withdrawal. However, it is difficult to distinguish households that injected from those that withdrew, although age, income stability and gearing ratios do appear to have had different effects on the average value of injections and withdrawals.

Footnotes

While it is probable that households purchasing property inject and households selling property withdraw, and hence that our model partly captures factors influencing the decision to buy or sell property, there are a number of households for which this is not true. [7]

For details on the construction of variables used in the regressions, see Appendix D. [8]

The age variable is categorical, with an open-ended ‘70 years and older’ bracket. The income variable is similarly constructed (with ‘$130,000 or more’ the open-ended response). The results are relatively insensitive to the use of larger intervals or dummy variables for these variables, and to the exclusion of households in these categories, suggesting that the results would be robust to the use of better-measured age and income variables. [9]

The value of injections or withdrawals is specified in log terms. [10]

Capital gains was excluded as an explanatory variable as this information is only available for properties still owned at the end of 2004. [11]

An alternative to the multinomial logit model is the ordered probit approach. However, this method suffers to an even greater extent from the similarity in character of injectors and withdrawers, given that it treats withdrawals as negative injections. [12]

An alternative approach would be to exclude non-indebted households from the regression. However, it is possible for such households to have withdrawn equity by taking out a loan during 2004, or to have injected through renovations, so we feel it is better to include this variable as a control, rather than restrict our sample. [13]

The model predicts most households to be injectors, rather than withdrawers, because the number of injectors by far exceeds the number of withdrawers. Excluding small withdrawals and injections (those under $20,000 in absolute value) modestly improves our ability to separate these two groups, with 33 per cent of withdrawers correctly identified. Under this alternative specification, income, LVR and housing assets all become significant. [14]