Introduction

Housing is a basic human need that determines households’ welfare. In their quest to satisfy this need, households make financial efforts to buy a residential property, and housing tends to capture the largest share of households’ wealth. After deciding to become a homeowner, any fluctuation in the residential property price shapes households’ wealth perception. Wealth is the most important determinant of households’ portfolio composition (Campbell 2006; Carroll 2002) even if, as showed by the prospect theory, the evaluation of one’s wealth is often subjective, assessed by comparing the assets’ initial cost to their current estimated price, in which the initial price exerts an anchoring effect on wealth perception (Kahneman 2011). The housing wealth effect was demonstrated in studies pointing to homeowners feeling richer when the real estate market signals housing appreciation, responding by increasing consumption spending (Case et al. 2005), and investment in riskier assets (Cocco 2004). Housing is a medium risk non-liquid asset with significant economic and physical transaction costs, whose appreciation while contributing to increase homeowners’ future expected return, also raises their exposure to risk. This return-risk duality is likely to affect choices of other assets to include in homeowners’ portfolios.

When assessing the impact that housing has on portfolio composition, however, the literature rarely assessed the relative housing wealth perception, nor its endowment effect, but instead the focus was on the role played by the absolute value of housing. Given that the housing cost typically represents a significant financial burden, households may not feel compensated for the risk taken with housing acquisition until the housing return is beyond a threshold that is informed by its initial cost, the anchor. In that case, “rich-feeling” households will be those for whom the real estate market signals a valued residential property with respect to its initial valuation, represented by the acquisition cost. Ignoring this endowment effect may prevent a complete understanding of the housing wealth effect on households’ portfolio investment decisions.

Within this context, this paper studies the empirical relationship between portfolio composition and wealth perception from housing price appreciation of Portuguese households. As the housing rental market in Portugal is narrow, the majority of the population are homeowners, and the residential property is faced as a form of saving and bequests. The paper follows the household finance literature, which stated that being wealthy and owning housing wealth, is positively related to owning risky assets (Campbell 2006). This strand of the literature predicted that there should be a positive relationship between owning highly priced residences and the share of risky assets held in the portfolio. While this claim can be directly tested, it will nonetheless lack the ability to explain how households may suffer from an endowment effect from the main asset they own, a feature claimed by the prospect theory. To fill that gap, this paper assumes that households perceive a wealth gain when the housing price increases with respect to its acquisition cost, controlling for the number of years since tenure, and considering that wealth allocation decisions are interdependent.

The empirical strategy in the present work tests how housing wealth perception from housing increases homeowners’ exposure to risk by shaping their investment decisions. The empirical evidence relies on microdata drawn from the Portuguese Household Finance and Consumption Survey (HFCS) for 2010 (Finance and Network 2013a) and assumes as dependent variable three shares of real and financial portfolio, estimating a fractional multinomial logit model to find if there is a positive statistical significant relationship between the rate of housing valuation and the preference for risky assets. Further, a similar model is also estimated taking as dependent variable three shares of a purely financial portfolio. The preference for the 2010 survey is justified based on the deep crisis period that took place in Portugal during the IMF adjustment memorandum starting in 2012. During this period, Portuguese households’ income decreased sharply and the relative value of all types of assets fell, with the decline being especially marked in the housing market. Relying on data from that period could compromise a sound analysis of assets’ valuation.

The estimations found a statistically significant effect between the rate of housing valuation and the three different shares of the portfolio, indicating that the residential property appreciation profiles the portfolio composition. The main effect from a one unit increase in the rate of housing valuation was to increase households’ portfolio diversity, given households were predicted to hold larger shares of low and high risk assets in their real and financial portfolios. The findings suggest that housing is a basic need that once satisfied releases wealth to other investments, where liquidity has a central role.

This paper contributes to the literature by studying portfolio composition of Portuguese households relying on the HFCS, and assuming wealth perception from homeownership as a driver of portfolio investment choices. Moreover, the analysis is innovative insofar as it measures perceived wealth by the average annual growth rate of house price since tenure, and by discussing portfolio categories as shares of total wealth. On top of this, the housing wealth effect is assessed as a relative concept, seeking to enrich the household finance literature on households’ portfolio composition with the predictions from the prospect theory on households’ decisions.

There are several reasons why it is important to understand the effect of housing wealth on portfolio composition. First, given an almost nonexistent rental market in Portugal, the residential property absorbs a large slice of households’ wealth. Second, if homeowners overestimate their relative gain, they may allocate wealth to risky assets, increasing their exposure to less safe investments and compromising their financial stability. Third, this paper’s policy implications are that to avoid households’ wealth concentration in a single asset and their overexposure to the inherent financial risk, the government should develop policies to encourage the expansion of the rental market, allowing households to take their time accumulating sound savings before buying a house.

The remainder of the paper is organized as follows. The next section briefly reviews the empirical literature on households’ portfolio composition. The subsequent section introduces the HFCS, describes the model’s dependent and explanatory variables and presents descriptive statistics on these variables. The following section displays and discusses estimations for total and financial portfolios, and robust checks based on different total portfolio composition. The final section concludes.

Literature Review

Modern portfolio theory (Markowitz 1952; Merton 1972; Tobin 1958) has investigated how, given a certain amount of individual wealth, private investors decide on which financial assets to include in a single portfolio among a group of alternatives that show diversification in return and risk. In particular, rational agents invest in efficient markets to maximize expected return from a combination of assets, ultimately aiming to increase total wealth. Since assets with higher expected returns display higher risk, agents should allocate a larger share of their personal wealth to riskier assets to increase the reward from investing. Risk-averse agents should avoid isolated decisions on wealth allocation, but instead compare the price evolution of alternative assets, and evaluate their return and risk trade-off to decide which assets to combine in the portfolio.

The household finance literature extended the debates from portfolio theory to the household level, and found households holding poorly diversified portfolios, mostly composed of cash deposits and a residential property accounting for the largest share of their total wealth (Campbell 2006). The empirical composition puzzles were identified with the predominance of under-diversified and home-biased portfolios, and with households owning a low share of equity on average and avoiding risky investments (Guiso et al. 2002; Haliassos and Michaelides 2002). If wealth were faced as a central determinant of the type of assets held in the portfolios, and richer households made up the group with the greatest appetite for riskier investments (e.g., Campbell 2006; Gollier 2002; Wachter and Yogo 2010), limited participation in equity markets (even by richer households), is one of the main stylized facts from the household finance.Footnote 1 Households that participate in financial markets were shown to exhibit concentrated portfolios, holding few stocks besides those from their employer, and showing a local market bias (Campbell 2006).

King and Leape (1998) studied households’ asset demands and pointed out the presence of incomplete portfolios. They estimated greater than unity wealth-elasticities for risky financial assets among US households’ portfolios, and rejected the presence of constant relative risk aversion in the portfolio composition data and supported the existence of portfolio composition puzzles. Carroll (2002) described the mean US household portfolio comprising cash deposits and a home with a mortgage, in opposition to the typical portfolio of the rich, which exhibited a much higher proportion of different types of risky assets. Risk aversion and capital market imperfections were identified as the most likely causes of US households’ composition puzzles. Arrondel et al. (2014) compared European households’ portfolios retrieving data from the HFCS. Their estimations pointed to heterogeneity throughout countries across wealth and income distributions, with recognized regularities such as real assets being the largest category of assets held by European households, and the main residence playing a dominant role in total net wealth allocation. European households tended to prefer safer types of financial assets, especially deposits and savings accounts. Nevertheless, a small minority of households owned risky financial assets. These authors also reported that asset ownership rates were increasing in wealth for all assets categories, but risky assets were more likely to be held by wealthier households, single households, and households in which the head of the household had higher educational attainment.

A strand of the literature on portfolio composition explicitly scrutinized the impact of including real assets within the household portfolio, placing real estate at the forefront. Brueckner (1997) emphasized how by being both a durable consumption good and a capital good with return and risk, housing comprises a consumption-investment duality that determines rational overinvestment in housing. Cocco (2004) treated the house price risk and the nonliquid nature of the housing investment as main determinants of the irrelevant presence of stocks in households’ portfolios, particularly among households with low net worth. Housing represented an important financial effort to homeowners, acting to reduce net wealth, especially of the youngest generations. Those who owned higher shares of housing in the portfolios tended to hold smaller shares of risky assets in a sort of crowding-out effect from housing to equity, an outcome also reported by Yao and Zhang (2005). Hu (2005) concluded that the decision to purchase a house biased financial portfolios toward safer assets, partially due to frictions associated with housing’s physical nature and transaction costs. Homeowners with a higher share of wealth concentrated in housing held fewer stocks to reduce their greater exposure to risk from housing and mortgage debt. Opposing to Hu (2005), Cardak and Wilkins (2009) studied Australian households’ homeownership and found it to increase their holdings of risky assets by providing easy access to cheap credit for other investments, acting as a sort of collateral.

On top of choice theory, prospect theory (Kahneman and Tversky 1979) questioned expected utility theory’s ability to explain decision making under risk. Individuals establish a reference point from which they measure gains and losses, but they do not react symmetrically to these gains and losses, being loss-averse, and tend to overweight losses with respect to gains. These individuals also prefer results obtained with certainty to results that are less probable. Bringing these features into the analysis of portfolio composition could explain the bias toward safer assets (or at least the preference for assets whose expected return is easier to track) and assets with less uncertain perceived returns such as housing.

Data and Model Variables

The HFCS

The subsequent empirical analysis was based on a sample of 4404 Portuguese households who participated in the first wave of the Eurosystem HFCS with 2010 as the reference year (Finance and Network 2013a). The HFCS was run by the European Central Bank intending to be representative of each country’s population. The first wave of the Portuguese survey was fielded by personal interviews throughout 2010 although part of the data reported to 2009, and the second wave was released in 2016 having as reference period 2013. The period between the first and the second wave was marked by a sharp reduction in net wealth, driven by a reduction in the value of asset holdings in households’ portfolios, and especially in the value of real estate. Portugal was one of the countries mentioned as experiencing substantial declines in housing prices (Finance and Network 2016). To avoid biased results, and given that 2013 is also a period of marked decrease in Portuguese households’ income, the empirical evidence in this work relied only on the first wave of the HFCS.

The survey for 2010 provides detailed information on household composition, socio-demographic features of the household representative, and a variety of economic and financial variables, including sources of household income and household’s wealth allocation. It also provides comprehensive information on their balance sheets, along with facts on consumption expenditures and details on credit constraints. Financial instruments identified in the survey are: deposits; total mutual funds; bonds; the value of the non-self-employment private business; publicly traded stock shares; managed accounts; money owed to households; voluntary pension and whole life insurance; and other assets. Real instruments correspond to the value of: household’s main residence, other real estate property, household’s vehicles, valuables, and self-employment businesses.

The unit of observation of the empirical model was the household. However, much of the socio-economic data were collected at the individual level and used in estimations reporting to the household representative. Following Costa and Farinha (2012), the reference person of the household was considered the adult male whenever possible, to increase the harmonization of results, and reducing the arbitrariness that occurs when leaving this decision to the household itself. A less positive feature of following this strategy was to lose gender as an explanatory variable in the model.

To capture the effect of housing valuation on portfolio composition, the narrower population of homeowners was distinguished from total population covering a sub-sample of 2986 households. To estimate the model with shares of financial portfolio as the dependent variable, 129 households that reported not possessing financial assets were further eliminated, and the observations were reduced to a total of 2857.

Dependent and Explanatory Variables

Dependent Variable: Portfolio Categories

The dependent variable in this model was a set of shares of assets bearing similar risk and held by households in their portfolios, namely three categories whose shares sum up to one were delineated in the portfolios: low risk assets, medium risk assets, and high risk assets. The household’s assets were pooled by risk and aggregated into a portfolio category. Whenever possible this study followed suggested classifications from the main literature on the subject to avoid arbitrariness in the assets’ classifications.

Financial and real assets included in the survey were previously enumerated. From the list, deposits, that is, deposits from both sight and savings accounts, were the only asset classified as low risk, and formed the category low risk assets (Carroll 2002). Medium risk assets included the value of the household’s main residence and of another real estate such as secondary homes, the value of self-employment businesses, mutual funds, bonds, money owed by other households, and voluntary pension and whole life insurance. All remaining assets were included in the high risk assets category. The shares of total portfolio were then converted into proportions of total wealth. Furthermore, a set of purely financial categories was built using former assets classifications and their proportions in total financial wealth were calculated to obtain the variables low risk financial assets, medium risk financial assets, and high risk financial assets. Table 1 summarizes the categorization of assets by risk.

Table 1 Classification of assets by risk
Full size table

Explanatory Variables: Housing

The HFCS distinguishes whether a household bought, built, or inherited the residential property, besides other less frequent possibilities. The dummy variable, housing acquisition, controls for households that had to allocate wealth to the acquisition of housing, taking the value 1 when the head of the household declared having bought or built his residence, and 0 otherwise. In a few cases, the dummy was also monitoring the households’ ability to calculate its rate of housing valuation.

The rate of housing valuation was captured using the household insight on the residential property valuation. The respondents indicated the price at which they bought the residential property and the price at which they would hypothetically sell it if they did so when the survey was taken. The ratio between current price and initial price was used as a measure of housing valuation and combined with n, years since homeownership, to calculate the rate of housing valuation as the nth root of the residential property price appreciation.

Change in housing equity was calculated as the change in the difference between the residential property current price and outstanding debt from purchasing, renovating, or refurbishing the residential property, and the residential price initial price and initial outstanding debt.

The liabilities related to the housing acquisition, in particular its concomitant outstanding debt, was captured by the dummy variable mortgage debt, equaling 1 when the household responded positively to having mortgage debt, and 0 otherwise.

Housing tenure years captured the number of years since homeownership.

Explanatory Variables: Socio-economic Features

Control variables included in the model comprised the difference between household total real and financial wealth and household debt (net worth); the age of the household head (age); dummies for marital status of the household representative (married, divorced); the number of dependents in the household (dependents); dummies for the work status of the representative (unemployed, retired, and other not working); dummies for educational attainment of household head (secondary education, higher education); and total income measured on a monthly basis (monthly income).

Household wealth comprises real and financial wealth measured as derived variables calculated according to the HFCS definitions. In the HFCS, real wealth corresponds to Total real assets 1, and financial wealth is denominated total financial assets 1. Net worth was obtained by subtracting the category total outstanding balance of household’s liabilities to the sum of their real and financial wealth. These liabilities cover mortgage debt, from household main residence (HMR) and from mortgages on other proprieties, and non-mortgage debt.

Married included married individuals and those who declared having a consensual union on a legal basis. The number of dependents followed the definition from the survey identifying as a dependent any individual under 25 years old who is not the head of the household and does not cohabit with him/her in connubial terms. The work status distinguished through dummies those who were not working because they were either unemployed, retired, or other not working meaning they were students or homemakers. Secondary education implied having successfully completed upper secondary, and higher education stood for households’ representatives having completed at least a first stage tertiary degree.

Lastly, monthly income was calculated dividing by 12 the derived variable total household gross income that covers employee income, self-employment income, income from public and private pensions, income from unemployment benefits and from regular social transfers.

Explanatory Variables: Financial Environment

The household financial background may limit the access to credit markets, inhibiting their participation in assets markets (Cocco et al. 2005). The dummy variable credit constraints that captures households’ limited access to finance, assumes the value 1 when the respondent responded having applied for a loan in the last 3 years and having been either turned down or received credit in an amount less than what they had applied for. This dummy also equaled 1 if for the same period the household reported not having applied for a loan due to perceived credit constraints. These classifications followed the definitions from Finance and Network (2013b).

To capture households’ appetite for risk the paper used self-declared risk-aversion to build the dummy variables high risk taker, and average risk taker. Financial respondent and spouse were questioned about their attitudes when saving or making investments. A high risk taker was someone who indicated the willingness to either take substantial financial risks expecting to earn substantial returns or take above-average financial risks expecting to earn above-average returns. An average risk taker was a financial respondent who responded that she/he would take average financial risks expecting to earn average returns.

Additionally, the survey questioned households about their future expectations regarding income, net worth, work environment, and prospects of becoming unemployed or of being forced to work fewer hours. According to life-cycle and permanent income theories (e.g. Friedman 1957; Modigliani and Brumberg 1954) household consumption and savings patterns change with income and wealth, households tending to increase savings when, as during recessions, income is expected to fall, given that pessimism driven by the fear of unemployment triggers cautious behaviors (Carroll 1997). Symmetrically, savings would decrease before positive prospects. Two independent dummies controlled for these effects. Optimistic expectations equaled 1 when the household indicated expecting future income to increase more than prices, when the household predicted receiving a substantial gift or inheritance, or anticipated a positive change in its net worth. Pessimistic expectations captured the set of households having declared negative prospects by expecting either future income to increase less than prices, net worth to decrease, or having to accept changes in employment leading to lower wages. In those cases, the variable took the value 1.

Past adverse changes captured household’s reporting a downturn in financial conditions. This dummy assumed the value 1 when the household stated that at least one of its members had unfavorable job changes, a substantial reduction in their net worth in the 3 years that preceded the interview, an unusually low income during the year reported in the interview, or an increase in regular expenses in the same time span. The variable equaled 0 when the household had been affected by any of these events.

Evidence on Portuguese Portfolios

Table 2 shows summary statistics for homeowners and for total and financial portfolios. Similar calculations were performed for total population but the corresponding values are not presented. The evidence revealed the average homeowner to have a riskier behavior than the average household (total population), a result mostly explained by the ownership of real assets, and especially by housing. Risky financial investments were found to be poorly appealing to Portuguese households, as both medium risk and high risk assets were nearly absent from their financial portfolios.

Table 2 Portfolio shares summary statistics (homeowners)
Full size table

Homeowners’ participation rates by asset and the average share of total wealth by asset are in Table 3. The percentage who reported owning any form of deposits was very high (96%) disclosing Portuguese households’ preference for liquidity. As expected, the residential property accounted on average for 73% of total wealth, while real assets such as vehicles and other real estate property were among preferred assets. The most predominant financial investment, voluntary pension and whole life insurance, was held by 16.6% of these households, but accounted for only a modest share of about 0.98% of their total wealth. The generally low participation rates in further assets suggest a not very diversified portfolio, and highlight the insignificant role played by equity. On top of this, the significant gap that prevails between participation rates and wealth shares points to precautious behaviors from homeowners when investing in categories of assets having more uncertain return.

Table 3 Homeowners participation rates and shares of total wealth
Full size table

Table 4 shows descriptive statistics of the explanatory variables used in the econometric model. In comparison with total population, homeowners earned a 1.1 fold monthly income, and owned a 1.3 fold net worth. The rate of housing valuation clearly contributed to wealth dissimilarities, differing by a factor of about 1.4 between samples. Net worth was more unevenly distributed than income among homeowners, and that disparity was even more pronounced than the distribution of the rate of housing valuation. Socio-economic factors did not indicate important disparities between the two samples, but confronted with total population, homeowners had slightly fewer respondents with credit constraints and marginally more respondents self-classified as high risk takers.

Table 4 Explanatory variables summary statistics (homeowners)
Full size table

Portfolio allocation changed across both the wealth distribution and the rate of housing valuation distribution. Figures 1, 2, 3 and 4 show total and financial portfolio shares by quartiles of net worth and the rate of housing valuation. Quartiles of the rate of housing valuation were calculated after isolating all households that were not homeowners. Non-proprietaries concentrated 53% of their total wealth in the form of deposits, a share decreasing in net worth across the distribution. Focusing solely on financial assets, the analysis revealed richer households raising the proportion of higher risk assets, especially medium risk to the detriment of deposits. Compared to total population, homeowners invested less in low risk assets and marginally increased the weight of deposits in the portfolios with the rate of housing valuation. The initial significant financial effort and mobilization of low risk assets that buying a residential property typically requires, works as a barrier for the group of non-homeowners, the poorest fraction of households in the total population. In this context, the share of deposits behaves as an inferior good for total population, being held proportionally more by the fraction of poorest households, but as a luxury good among proprietaries, increasing in wealth and in the rate of housing valuation, and even being preferred to high risk assets, a regular feature in the two types of portfolios.

Fig. 1
figure 1

Shares of total portfolio by quartiles of net worth (total population)

Full size image
Fig. 2
figure 2

Shares of financial portfolio composition by quartiles of net worth (total population)

Full size image
Fig. 3
figure 3

Shares of total portfolio by quartiles of housing valuation (total population)

Full size image
Fig. 4
figure 4

Shares of financial portfolio by quartiles of housing valuation (total population)

Full size image

Empirical Model Results

The main hypothesis of this paper is that the wealth effect from the residential property appreciation molds the portfolio composition, while the household becomes relatively more exposed to risk by owning a valued medium risk asset. Prospect theory reported households suffering from an endowment effect, overvaluing gains from what they own. When the real estate market signals an appreciation of the asset representing their largest possessions, they experience a wealth effect that should lead them to invest in riskier and higher-return assets as suggested by the household finance literature. To test the housing wealth effect on households’ investment decisions, two portfolios were estimated, one of real and financial assets, and one purely financial portfolio. The model’s dependent variable was identified with the proportion of each risk category of assets held in the household portfolio, the sum of the three categories adding up to 1, regressed against the rate of housing valuation, and the set of covariates. Each fraction is bounded and can take any value between 0 and 1, thus standard estimation methods such as OLS are not suited since they would produce non-linearity of the conditional expectation. To solve the problem, Papke and Wooldridge (1996) have proposed the use of a nonlinear link function, as for instance the logistic function, to impose constraints on the conditional mean and produce predicted values that lie between the boundary values. By applying a quasi-maximum likelihood estimator of the coefficients, these turn out to be consistent and asymptotically normal. This study applied FMLogit assuming that the reallocation of wealth across shares is interdependent, and afterwards relaxed interdependence estimating each share using a fractional generalized model to assess initial estimations’ robustness.

Portfolio Estimations

This subsection followed Buis (2008), applying a multivariate generalization of the fractional logit model by Papke and Wooldridge (1996). The fractional multinomial logit model measures the simultaneous changes in proportions of multiple variables as a result of a group of covariates. The technique uses one of the shares as a pivotal variable. When regressing the multinomial logit model for the joint probability of holding each specific portfolio share, the independence in these estimations was attested through robust standard errors. In these regressions the estimated coefficients and their standard errors are not particularly informative, and the choice was thus to report marginal effects that were held at the sample mean for continuous variables and at zero for dummy variables. Given that the covariates determine a reallocation of wealth across assets from the same portfolio, the marginal effects sum up to zero through equations. Changes in different variables result in different substitution patterns between portfolio shares. The results for total and financial portfolio estimations are in Table 5. A fair part of the estimated marginal effects were statistically significant, especially for the proportions of low and medium-risk categories and for total portfolio estimations.

Table 5 Regression results for total portfolio and financial portfolio
Full size table

The rate of housing valuation was interpreted as having a positive wealth effect on the portfolio composition if a unit increase in this rate was found to increase wealth allocated to riskier assets while decreasing the share of low risk assets, given the set of control variables. Since the housing wealth effect could derive from the increase in its equity, that is, the difference between the residential property price and the outstanding debt from its acquisition, the variable change in equity that compares current equity to initial equity was understood in the same way. This understanding is in line with the measure of housing net worth from Wakita et al. (2000) who claimed that the change in net worth is a better measure of wealth accumulation since it is able to capture its dynamic nature. The estimated marginal effects confirmed that both the valuation of housing assets and the increase in housing equity impact portfolio shares, but a closer look revealed these variables predicting movements in opposite directions on each of the portfolio shares.

Among Portuguese homeowners, the rate of housing valuation positively predicted the share of low and high-risk total assets (by 0.129 and 0.091, respectively), and negatively predicted the share of medium risk deposits (− 0.221), while the change in housing equity negatively predicted the share of low risk assets (− 0.003), positively predicted the share of medium risk assets (0.004), and displayed a non-statistically significant result for the high-risk category. Thus, households that experience higher relative housing price appreciation choose to diversify their portfolios, holding on average greater proportions of both deposits and high-risk real and financial assets. These households dare to hold a higher share of high risk assets in their portfolios when at the same time, they hold proportionally more safe deposits. Knowing high risk assets concentrated 2% of homeowners’ riches (Table 2), the magnitude of these impacts is not trivial, representing a 4.5-fold increase, thereby involving an important decrease in safety with housing valuation. However, the combined wealth reallocation within categories predicts a portfolio in which the share of deposits dominates the high-risk real and financial assets, indicating that if households with higher rates of residential property valuation become less risk-averse, they nevertheless act cautiously when allocating their wealth, maintaining low-risk deposits significantly represented.

Those households whose housing equity changed positively were predicted to decrease their relative holdings of liquidity, a finding indicating a change in households’ sensitivity. This may signify that households do not perceive the residential property as a risk-bearing asset, but rather appraise it as being akin to deposits, and interchange wealth across these categories, given their financial conditions and convenience.

For the narrower financial portfolio, the model reported signs analogous to those found for total portfolio for both the rate of housing valuation and the absolute change in housing equity, although with loss of statistical significance for high-risk financial assets. Across households, a one unit increase in the rate of housing valuation entailed a transfer of about 30 percentage points of financial wealth from medium-risk to low-risk financial assets. The model also documented a negative association between the change in housing equity and holding low risk assets (− 0.0017), but a positive one with the share of medium-risk assets (0.0016). Thus, if rising housing equity is related to losing liquidity, it also connects to possessing medium-risk financial assets, a feature that could explain the reasonable presence of pension plans in Portuguese homeowners’ portfolios (a 16.6% participation rate as shown in Table 3). These households do not rely solely on deposits as a form of financial investment, instead they display a preference for moderate-risk ones.

The outcomes for the estimations of the change in equity are consistent with Cocco (2004), who showed that besides decreasing financial net worth, housing reduces liquid assets, and with the findings of Arrondel and Savignac (2015) who concluded that homeowners lessen their equity investment to reduce their exposure to risk. In the same line of reasoning, Flavin and Yamashita (2002) sustained that being both a consumption and investment good, housing exerts a constraint on the portfolio decision that can compromise portfolio diversification. In the present estimations, this effect is evident when analyzing the impact on portfolio composition from housing equity. Housing overinvestment tends to be the case in Portugal, other portfolio decisions are molded by this constraint considering their expected return and risk.

Mortgage debt discloses the effect on these households’ portfolios of holding a liability related to housing. As expected, the predicted change in portfolios was a reduction of liquidity, the share of deposits decreasing 2.5 percentage points transferred to medium-risk assets.

Households with older housing tenure held more bank deposits, and fewer medium-risk real and financial assets. This may reflect that those who owned housing for a longer time are older individuals, prone to liquidity, as it may signal they were able to pay off their mortgage debt and afterwards accumulate liquidity.

Socio-economic variables in the model revealed standard behaviors shaping household investments. Individuals with higher net worth were predicted to invest more in real and financial high-risk assets, and to cut back the fraction of wealth allocated to medium-risk assets. The similarity between the signs of the coefficients for these findings and those for the rate of housing valuation is noteworthy, reinforcing the suspicion that homeowners may rely on the relative increase in their housing value as an indicator of increased wealth.

Households in which the head of the household was older exhibited a preference for safety. Recent explanations (e.g., Spaenjers and Spira 2015) exposed the increase in life expectancy determining future uncertainty, and prompting savings to increase by the end of the life cycle, especially in the form of liquid assets. Being other not working, predicted a 2.9 percentage points higher propensity to hold high risk real and financial assets. If holding fewer low risk assets can derive from not having a work occupation, a greater share of high risk assets suggests that this group may include wealthier households that can afford to live from allowances, and are more predisposed to make risky investments.

The propensity of the higher educated to invest comparatively more in risky assets was well documented in the literature that relates it to households’ financial literacy (e.g., Barasinska et al. 2012). The estimations conveyed those with secondary and with higher education transferring wealth from deposits to high-risk financial applications, confirming that schooling increases financial market participation. However, for the enlarged portfolio, the model indicated the higher educated allocating wealth to bank deposits, in detriment to medium-risk assets. This is in accordance with the findings that indicated that the more educated tend to be located at the top quartile of the income distribution, and are able to keep enough liquidity in their portfolios, and to regularly reinforce it, holding a higher share of deposits within portfolios (Alves et al. 2010). Corroborating this possibility, a one unit increase in income was predicted to divert resources from medium-risk real and financial assets to low and high-risk ones, but to depress the average amount of deposits within financial assets. This indicates that low-risk deposits are a normal good among total portfolios since their relative demand increases with income, but are an inferior good within strictly financial portfolios.

The set of covariates controlling for financial conditions revealed interesting features of Portuguese homeowners. Credit constraints, as anticipated, were found to decrease liquidity while they raised the share of medium-risk real and financial assets. The implication is that households are forced to use liquidity when applying for credit and having it totally or partially refused. The attitude toward risk was related to portfolio allocations, households in which at least one of its elements was assumed to be a high-risk taker displaying negative marginal effects on the share of medium real and financial assets and positive marginal effects on the share of high risk assets within entirely financial portfolios, a result corroborating Finke and Huston (2003) who had shown that households with higher risk tolerance have also higher likelihood of holding financial assets. The statistical significance of the coefficients in financial portfolio estimations strengthens the idea that the attitude toward risk may be one of the strongest determinants of the demand for risky investments (Barasinska et al. 2012). Expectations were shown to affect household’s perception on future income, distressing over their investment decisions, but with different results on the two types of portfolio. Optimistic expectations contributed to raising the share of deposits (0.030) held in total portfolios but to shrinking (− 0.035) this share among total financial assets, the medium-risk category being the counter-part of these changes. Pessimism, on the other hand, reduced liquidity across households’ total portfolio. It seems that the “confident-feelers” are most likely those that have been able to accumulate deposits among their asset building, and the pessimists are those who have been eroding their liquidity, while there is no indication that either sentiment will contribute to increase their exposure in the financial market.

To sum up, wealthier households, with higher monthly income in which the head of the household is not working and/or is self-assessed as a high-risk taker, display positive predictions with respect to the share of high-risk real and financial assets. Combined, these findings settle investment on the residential property as a basic need for Portuguese households, reinforce the central role that liquidity has on their portfolio preferences, and disclose behavioral features as being decisive for riskier investments.

To check these results, additional estimations were performed using fractional response generalized linear models for each portfolio but dropping interdependence between choices regarding the allocation of wealth across portfolio categories. These estimations followed Williams (2016) and are also in line with Papke and Wooldridge (1996). The fractional estimations were used to verify coherence in households’ preferences, supporting or discarding the direction and magnitude of wealth reallocation detected in the fractional multinomial logit estimations. These estimations produced results in line with the fractional multinomial logit model estimations and are available upon request.

Robustness Checks

Since the residential property may be understood as a risk-free asset by households, even interchangeable with deposits, and given that homeowners’ portfolios have the highest shares of wealth allocated to housing, further estimations were performed for a model in which housing is classified in the category of low risk assets. The presence of housing on the households’ portfolios was shown to mold households’ preferences towards saving decisions. Fisher (2013) and Fisher and Montalto (2011) studied the effects of loss aversion on respectively the saving behaviors of Spanish and US households and reported opposite results for the two countries, finding loss aversion behavior among US households but not among Spanish households, a fact that Fisher (2013) associated with the higher portion of wealth held in real estate by Spanish households, a buffer for their income uncertainty.

In the revised portfolio, the residential property was classified as a low risk asset, dropped from the medium risk group, and three categories were still considered. Results for these estimations are in Table 6. Apart from few exceptions, as for mortgage debt and other not working, all the coefficients that were statistically significant in the two estimations reversed signs for the revised categories. In the new estimations, the rate of housing valuation was shown to decrease the relative holdings of low risk assets and to increase medium-risk ones, while the change in equity produced the opposite result. Across the distribution, households displaying higher changes in equity were also seen to decrease their holdings of high risk assets. This risk-averse behavior suggests that households facing an increase in the fraction of the residential property they own, consider their housing as a warranty for a future downturn such as losing their jobs, being severely ill or injured, or just elderly. They will also tend to reject risky investments of any sort. It seems that a positive change in equity is less perceived as a change in wealth but rather as a change in tenure status, moving from feeling a renter, when the bank owns a significant amount of the residential property, to feeling owner, when housing liabilities decrease in proportion.

Table 6 Estimation results for revised total portfolio
Full size table

The rate of housing valuation was predicted to increase medium-risk real and financial assets, once more moving along with the marginal effect from a change in net worth. The portfolio diversification was confirmed as the main outcome from housing wealth effect, diminishing the portfolio category that includes housing, while enlarging all other shares. After a certain threshold of perceived housing valuation, households appear to choose to hold other types of assets, beyond the residential property, never letting go of a fair share of low-risk deposits. Housing is thus as a central concern for homeowners, a financial problem to solve before taking decisions on other investments. Perceived housing wealth, although not instigating an undoubtedly risky behavior effect, is nevertheless shaping these homeowners’ investments, leading them to increase the combination of assets with different risk, possibly preventing movements that they feel can compromise their financial stability.

As before, independent estimations of these new portfolio shares were performed by running fractional generalized model estimations, and the findings corroborated the original estimations. Results of those estimations are available upon request.

Conclusion

This paper provides evidence that households take into account wealth perception from the valuation of their residential property when deciding how much wealth to allocate to assets bearing different risk and return. The relationship between three shares of a single portfolio and the rate of housing valuation is empirically tested with data retrieved from the 2010 HFCS. Specifically, a fractional logit model is estimated to capture the joint probability distribution of allocating wealth throughout three different asset risk categories belonging to either a mixed or a financial portfolio. The results indicate that across the distribution, the share of housing within homeowners’ portfolios is negatively related to the rate of relative appreciation of their residential property price. This evidence is robust to changes in portfolio shares definition. The study is limited by the lack of availability of adequate panel data, given portfolio allocations have a dynamic nature.

The prospect theory predicts that when making decisions, agents overvalue gains from what they own, whereas the household finance literature has shown that wealth encourages the holding of risky assets, with the housing wealth effect being an important driver of households’ consumption and investment decisions. This paper’s findings suggest that the endowment effect from the residential property appreciation affects homeowners’ investment decisions, boosting wealth reallocation to other categories of real and financial assets. The estimated marginal effects display identical signs to those for net worth, pointing to the rate of housing valuation containing information on wealth perception. Together, these variables suggest richer-feeling households preferring diversity in their portfolios, not giving up low-risk deposits even when they can afford to invest in uncertain-return high risk assets. The control for equity effects, captured by their absolute change along tenure years, reveals households choosing to hold less liquidity in their real and financial portfolios. There is the possibility that households understand housing as a safe investment similar to bank deposits.

Generally, our results contribute to the literature by suggesting that the endowment effect from housing is a driver of households’ portfolio decisions. The paper’s evidence also reinforces the composition puzzles by revealing a strong home bias in Portuguese households’ portfolios. The study’s main contribution to the literature is to assess housing wealth by homeowners’ perception of the residential property relative valuation along tenure years and to suggest that the cost of housing implicitly defines a threshold, from which households’ feel free to invest in different assets, and diversify their portfolio. The housing wealth effect biases households’ portfolios toward riskier assets in tandem with the reinforcement of bank deposits, countervailing a risky behavior with a simultaneous risk-averse investment. One main implication is that the housing wealth effect reduces the share of housing within portfolios.

This work is innovative by using microdata and joint probability estimations to find a role for housing valuation on portfolios choices. Nevertheless, there are several limitations in this study. First, the preference for a cross-section analysis discards available information on housing prices evolution. As the second wave of the HFCS covers a period of overall decline of the value of households’ assets holdings and average income, it was decided to retrieve data from only the first wave, especially since this effect was relatively stronger on real estate prices. Second, the focus on the subset of homeowners, a consequence of choosing housing valuation as the model’s main explanatory variable, is biased but defensible since households have been accused of irresponsible financial decisions encouraged by a prospering real estate market. Third, every asset held by the household may have an endowment effect, which is neglected in this study. This is justified by the residential property being the single asset with a significant participation rate of Portuguese population. Additionally, since any classification involves arbitrariness, and it is not clear how households perceive risk, one may question the extent to which additional changes in the assets categories would significantly impact the estimation results.

Its limitations aside, this study is important given the recent concern with households’ financial stability and its connection to housing decisions. Homeowners typically concentrate the largest share of their wealth and net worth on the residential property, and monitor its price evolution as a measure of their savings and riches. This study shows that households do not necessarily behave carelessly before real estate appreciation, but that owning a valued residential property contributes to diversify their portfolios, leading them to apply their wealth to both low and high-risk assets.

This work’s results have important policy implications. First, booms and busts in the real estate market can have consequences that go beyond it, impacting households’ investment decisions, and ultimately affecting their riches and welfare. The greater the number of owners in the total population, the more the population investment decisions will be related to the evolution of housing prices. Second, the relationship between housing valuation and portfolio diversification denotes that households are overinvesting in housing, a behavior that is relaxed only after a certain housing valuation threshold. Policy measures should be taken with the aim of relaxing this constraint, such as better legislation or specific tax policies encouraging owners to rent their dwellings and deepen the housing rental market. Additionally, financial literacy campaigns, involving the government and financial institutions, could alert households to the risks inherent to owning a residential property and to the range of other possible financial and real investments, enhancing the framework in which households’ decisions are made.