Nominal Loss Aversion
An empirical study of the effect of nominal loss aversion on the price setting of households in the Dutch housing market
Abstract
This study analyses the effect of nominal loss aversion on the price setting of households in the Dutch housing market. A well-known pattern within real estate is that prices and trading volume seem to correlate with each other. A conventional explanation for this pattern is nominal loss aversion. This theory predicts that price changes cause changes in trading volume because investors dislike losses more than they like equal size gains. In order to study this phenomenon, this thesis examines the outcome of two surveys under Dutch households, held in 2009 and 2012. This thesis found evidence for the existence of nominal loss aversion in both 2009 and 2012. Price-volume correlation can also be explained by down payment requirements in the mortgage market. This reduces mobility and liquidity and thereby influences the prices and the number properties that are sold. This thesis found evidence for this explanation in both 2009 and 2012 as well. However, loss aversion plays a larger role in the price setting of Dutch households than equity constraints.
Date: 16-6-2014 Name: Mathilde Dirven Student number: 5883164 Supervisor: dhr. dr. M.A.J Theebe Faculty Economics and Business University of Amsterdam
Table of Contents 1. Introduction 2 2. Literature 4 2.1 Price-volume Correlation 4 2.2 Loss Aversion 6 2.2.1 Housing Market 7
2.2.2 Commercial Real Estate Market 9
2.2.3 Stock Market and M&A 10
2.3 Equity Constraints 11
2.4 Alternative Explanations 13
2.5 Summary Literature 14
3. Methodology and Hypotheses 15
3.1 Methodology 15
3.2 Hypotheses 20
4. Data and Descriptive Statistics 21
4.1 Data 21
4.2 Descriptive Statistics 23
5. Results 25
5.1 Basic Regression model 25
5.2 Extended Regression Model 26
5.3 Results in Perspective of Literature 28
5.4 Results in an Economic Perspective 30
6. Validity Check 31
7. Conclusion and Recommendations 33
References 36
1. Introduction
From the mid-80’s until early 2008, house prices raised almost continuously in the Netherlands. In 2007, the Netherlands was hit by the global financial crisis. The global financial crisis affected the Dutch housing market. Because of uncertainty and bad financing conditions consumers delayed buying and selling their homes. This resulted in stagnation of the Dutch housing market: both price levels and transaction volume decreased. This is consistent with a well-known pattern within real estate. Prices and trading volume seem to correlate with each other. For when prices are rising, trading volume tends to be higher than when prices are falling. This movement is inconsistent with a movement you would expect in case of an efficient market. In an efficient market, prices depend on the present discounted values of the future income streams. A conventional explanation for this inconsistency is nominal loss aversion. The theory of nominal loss aversion predicts that price changes cause changes in trading volume because investors dislike losses more than they like equal size gains. Kahneman en Tversky (1979) explained loss aversion by use of the prospect theory. Three essential features characterize this theory. First, gains and losses are examined relative to a reference point. Second, the theory holds that the value function will be steeper for losses, than it will be for equivalently sized gains. Third, the marginal value of gains or losses will diminish with the size of the gain or loss (Kahnemen and Tversky, 1979, p. 279-280). This results in a utility function shaped like a kinked “S” and is defined over changes in value relative to a reference point. A problem with the application of the prospect theory to empirical studies is that the reference point is rarely observed. An influential exception is a study by Genesove and Mayer (2001), who developed an empirical framework that captured the effect of loss aversion. Their model studied the relationship between list prices and prospective losses. Genesove and Mayer (2001, p. 1255-1256) found that property owners, faced with prospective loss, set a higher asking price and in fact do sell at a higher price than other sellers. As a result, they will suffer less sale frequency or, in effect, a longer time on the market. These results indicate that homeowners are averse to realizing nominal losses and can explain the price-volume correlation. Also the studies from Engelhardt (2003) and Anenberg (2010) found evidence for the presence of loss aversion in the US housing market. Although there seems to be an academic agreement on the existence of loss aversion in the US housing market, little research has been dedicated to other housing markets. Thereby previous studies tested the influence of loss aversion by examining transaction data, while this study developed an empirical framework that examined survey data. To contribute to existing literature this
study examines loss aversion in the Dutch housing market by using a new developed empirical model.
For these reasons, the aim of this study is to investigate the effect of nominal loss aversion on the price setting of households in the Dutch housing market. As stated, this research will contribute to existing literature for various reasons. First, this research studies nominal loss aversion in a different market setting than previous studies have done. Previous studies on nominal loss aversion were undertaken in the US housing market, while this study examines the Dutch housing. The Dutch housing market and the US housing market have different characteristics. However while executing this research, the Dutch Central Bank (DNB, 2014) presented a paper that studied loss aversion in the Dutch housing market. The DNB (2014) found that a large part of Dutch households have “rosecoloured glasses”: they found that the median homeowner reports a price that is 13% above the actual value and that homeowners are more optimistic about the price development of their own house than about general price developments. Based on these numbers, they suppose loss aversion plays a role in the Dutch housing market. However a shortcoming of this research is that this phenomenon is not studied by means of an empirical model. Thus by studying nominal loss aversion by an empirical model, comparable to the model of Genesove and Mayer (2001), previous literature would be enhanced by studying loss aversion in a different market setting. Second, a very interesting time period is studied. This study will examine the outcome of two surveys, held in 2009 and 2012. The Dutch housing market experienced a major drop in transaction volume and prices in 2008. By studying the outcome of the survey in 2009, the effect of loss aversion is examined right after a major price drop. By studying the outcome of the survey in 2012, the effect of loss aversion is examined in a period where households are probably more
anticipated to the changed market conditions. A third feature of this research is the development of an empirical model that examines survey data instead of transaction data. Transaction data from the real estate market is really difficult to obtain, since it is non-transparent. Especially, information about mortgage values is hard to acquire. Therefore this study uses data from a survey, held under Dutch households. However existing models for examining loss aversion such as the model of Genesove and Mayer (2001) were not applicable for this reason. Therefore, an adapted version of the model of Genesove and Mayer (2001) that can be applied to survey data is developed.
An Ordinary Least Squares (OLS) regression is used to study the effect of loss aversion on the expected sales price compared to the market value. Genesove and Mayer (2001) examine the effect of loss aversion on asking price. However the respondents of the
survey are not necessarily planning to move. Therefore, the expected sales price is used as a proxy for the asking price. An expectation is considered a loss if the expected sales price is smaller than the original purchase price. The control variable loan-to-value (LTV) is added to the regression model to control for the another possible explanation of price-volume
correlation: equity constraints. When house prices drop below the mortgage value,
homeowners become equity constrained and buying a new home becomes more difficult. Rather than moving to a smaller house, they may rationally choose to stay where they are. A household is considered as equity constraint when the LTV is larger than 1.
The results of this resarch indicate that both loss aversion and equity constraints play a role in explaining price-volume correlation in the Dutch housing market. For 2009 and 2012 the results for the loss variable are very similar. If a households faces a loss and if this loss increases with €10.000, the ratio of the expected sales price and the market value increases with 1,36% point for 2009 and 1,67% point for 2012. For the LTV, the effects in 2009 and 2012 differ more. For 2009, the results show that if a household is equity constrained and has an LTV of 1,1, the ratio of the expected sales price and the market value is 2,23% point higher than when a household has an LTV of 1. For 2012, the results show that if a household is equity constrained and has an LTV of 1,1, the ratio of the expected sales price and the market value is 1,16% point higher than when a household has an LTV of 1.
The remainder of this paper is structured as followed. Chapter two discusses previous literature related to the topic of this research. In chapter three, the methodology of this research is described and hypotheses are stated. The next chapter describes the data that is used to examine loss aversion in the Dutch housing market. Chapter five shows the results of this study and chapter six shows some validity checks. The last chapter contains the
conclusion and a recommendation for further research. 2. Literature
2.1 Price-volume correlation
A well-known pattern within real estate is that prices and trading volume seem to correlate with each other. When prices rise, trading volume tends to be higher than when prices fall. This pattern does not correspond with a pattern you would expect in the case of an efficient market. In an efficient market, prices solely depend on the present discounted values of the future income streams. In general the correlation between prices and trading volume is
explained by two theories. First, the price-volume correlation can be explained by behavioural theories, such as loss aversion and irrational belief in mean aversion. Nominal loss aversion
predicts that price changes cause changes in trading volume because investors dislike losses more than they like equal size gains. Irrational belief in mean reversion occurs when an investor chooses to hold their losers and sell their winners because they believe today’s losers will soon outperform today’s winners. However there is not always a clear distinction
between this explanation and loss aversion. A second explanation for the positive price-volume correlation is down-payment requirements in the mortgage market. This reduces mobility and liquidity and thereby influences the prices and the number properties are sold. An alternative third explanation for price-volume correlations is that the real estate market fails to clear instantaneously, because it is a search market. Another alternative explanation for the price-volume correlation is the option value of homeowners: homeowners wait to sell when the benefits exceed net carrying costs.
Many studies found empirical evidence for the price-volume correlation hypothesis in housing markets. Miller and Sklarz (1986), for instance, found that sales predict price changes in the Hawaii condominium market, while Stein (1995) found evidence in the opposite
direction. Stein (1995) showed a positive relation between current house sales volume and last year’s rate of price change on city-level data in the US. Berkovec and Goodman (1996) reported a significant positive correlation when they regressed the change in median sales price of houses on the simultaneous change in turnover. They used national data from the US. Andrew and Meen (2003) studied the owner-occupied market in the UK. They found a
positive correlation between prices and the amount of transactions, especially in the short run. Also Ortalo-Magné and Rady (2004), who studied transactions of residential properties in England and Wales, showed a positive relationship between price and volume. Hort (2000), on the other hand, did not find a significant relation between price changes and turnover changes in the Sweden housing market. Follain and Velz (1995) found in contrast to most studies a negative correlation between house prices and volume sales in metropolitan areas in the US.
In the commercial real estate market the positive price-volume correlation is observed by Crane and Hartzell (2010) and Bokhari and Geltner (2011). Crane and Hartzell studied the movement of Real Estate Investment Trusts (REITs) and Bokhari and Geltner studied the commercial property market in the US. In contrast Leung and Feng (2004) did not find an significant relation between prices and volume in the Hong Kong commercial real estate market.
Besides evidence from real estate markets, price-volume correlation is a widely observed phenomenon in stock markets. Both Gallant, Rossi and Tauchen (1992) and
Campbell, Grossman and Wang (1993) found significant results that indicate a price-volume correlation on the New York Stock Exchange (NYSE). Gallant et al. (1992) found significant positive results for both the unconditional distribution of price changes and volume and the conditional distribution given past price changes and volume. Campbell et al. (1993) found that stock returns are lower on high-volume days than on low-volume days. The determinants that cause the price-volume correlation in the stock market are different from the real estate market because other market characteristics have to be considered.
2.2 Loss Aversion
Loss aversion is considered as a conventional explanation for price-volume correlation. The theory predicts that price changes cause changes in trading volume because investors dislike losses more than they like equal size gains. Loss aversion was first demonstrated by
Kahneman and Tversky (1979). Kahneman en Tversky (1979) explained loss aversion through the prospect theory. Three essential features characterize this theory. First, gains and losses are examined relative to a reference point. Second, the value function is steeper for losses than for equivalently sized gains. Third, the marginal value of gains or losses
diminishes with the size of the gain or loss (Kahnemen and Tversky, 1979, p. 279-280). This results in a utility function shaped like a kinked “S” and is defined over changes in value relative to a reference point.
Figure 1: Prospect theory value function
This theory was developed on the basis of survey questions and experiments. Shefrin and Statsman (1985) put this behaviour pattern in a wider theoretical framework by introducing a general disposition effect. According to this effect, investors tend to sell winners too early and
hold losers too long. This can be explained by five factors. Shefrin and Statman (1985) introduced mental accounting, regret aversion, self-control and tax considerations as
explanations for the disposition effect. Mental accounting is the act that individuals split their future and current assets into separate accounts and give different values to different asset groups. This will influence their purchasing decisions. Regret avoidance occurs when the strength of regret is stronger than the strength of pride. This results in a disposition to realise gains and defer losses. Self-control is explained by the feeling of pride for having chosen correctly in the past. Because of tax considerations a concentration of loss realisation occurs in December, which is consistent with the behavioural framework of Shefrin and Statman
(1985).
The following three paragraphs (2.2.1, 2.2.2 and 2.2.3) discusses some empirical evidence found for nominal loss aversion. Empirical evidence for the housing market, commercial real estate market, stock market and mergers and acquisitions (M&As) are described in these sections. It is interesting to compare the different methods and results of this topic. The outcome of this comparison helps to develop an appropriate methodology to study nominal loss aversion in the Dutch housing market. Lastly, the results of this paper will be compared with the findings of previous literature.
2.2.1 Housing Market
A problem with applying the prospect theory to empirical studies is that the reference point is rarely observed. An influential exception is a study by Genesove and Mayer (2001); they developed an empirical framework that captured the effect of loss aversion. Their model studied the relationship between list prices and prospective losses. To analyse the
effectiveness of prospective losses, the variable loss is regressed on the original asking price. Genesove and Mayer (2001) defined loss as the greater of the difference between the previous price and the estimated value in the quarter of entry, and zero. The research of Genesove and Mayer (2001) was focused on the Boston housing market between 1990 and 1997. Genesove and Mayer (2001, p. 1255-1256) found that property owners, faced with prospective loss, set a higher asking price and in fact do sell at a higher price than other sellers, suffering as a result less sale frequency or, in effect, a longer time on the market. They also found that investors were less loss averse than owner-occupant counterparts. A previous paper of
Genesove and Mayer (1997) studied whether liquidity constraints determine list prices, selling prices and the time on the market. Genesove and Mayer (1997) found evidence for this but it
appeared less important than loss aversion (Genesove and Mayer, 2001).
A few studies have built on the insights of Genesove and Mayer (2001). One of them is Engelhardt (2003) who used the model specification of Genesove and Mayer (2001) to study loss aversion in metropolitan areas in US in the period between 1985 and 1996. Three different transitions are studied: intermetropolitan1 mobility, intrametropolitan2 own-to-own mobility and intrametropolitan own-to-rent mobility. Engelhardt (2003) reported five principal findings. First, he found that intrametropolitan own-to-own mobility responds differently to nominal housing losses than to gains, which confirmed the findings of Genesove and Mayer (2001). Second, nominal loss aversion is less noticeable in own-to-rent and
intermetropolitan mobility. For these types of mobility, losses have little statistical or
economic effect on mobility. However also for these types of mobility, households hold on to nominal housing market gains. Third, in the intrametropolitan own-to-own mobility the study showed some evidence for binding equity constraints. Fourth, in intermetropolitan mobility and own-to-rent mobility, the evidence for low equity constrains was little. Also, equity constraints had little statistical or economic effect on mobility for these types of mobility. Fifth, and complementary to the findings of Gensove and Mayer (2001), he found nominal loss aversion has a more dominant effect than equity constraints in restricting household
mobility.
Subsequently, Anenberg (2010) also used the model of Genesove and Mayer (2001) to study the effect of loss aversion and equity constraints on selling prices using data from the San Fransisco Bay Area housing market over the period between 1988 and 2005. The model Anenberg (2010) used is less parametric than the original model of Gensove and Mayer (2001). Where Genesove and Mayer treated unobservable quality as fixed, Anenberg (2010) used a more flexible version of the repeated sales estimator. In addition a locally linear regression is applied to allow for the time effect to vary by house. Thereby Anenberg (2010) used a more rich and diverse sample of housing transactions. In line with the findings of Genesove and Mayer (2001), Anenberg found evidence that both loss aversion and equity constraints affect housing prices. The main difference is that Anenberg (2010) found larger effects. Furthermore he found that the relationship between seller characteristics and prices is casual. In addition, Anenberg (2010) reported also a new finding. The effects of loss aversion and equity constraints are smaller for homes that are surrounded by other homes. This is possible due to a more competitive market.
1
Intermetropolitan mobility is mobility within a metropolitan area
Seslen (2004) studied the loss aversion on MSA-level in the US in the period between 1985 and1997. A structural estimation model is used to examine rational behaviour of
individuals in their housing consumption. The key variable, expected return, is not directly observable in the choice equation. Therefore the expected return is estimated separately from the decision to trade. The expected return is estimated as a function of other economy-level observables, which is not related to the choice of the individual. The model shows that individuals are more likely to trade in the up-swing and when prices are rising they are more likely to trade-up. When the prices are declining the opposite is true: individuals are less likely to trade when the price are down and when the prices are decreasing they are not willing to trade-down. From these results Seslen (2004) concludes that irrationality in the housing market is likely to be concentrated on the downside.
Lastly, the DNB published a paper about the sentiment of the Dutch housing market in April 2014. The DNB (2014) examined the Dutch housing market from 2003 until 2012. Like this thesis, their research studied the outcome of a survey. The study found evidence for the presence of loss aversion and an endowment effect. The endowment effect is the effect that the value of an object increases with the duration of ownership. A random-effects generalized least squares (GLS) regression is executed to explain the variation in home value bias. The home value bias is the ratio of the perceived house value and the actual value. To examine endowment effects, their regression model includes dummies for the number of years respondents are living in their homes. The findings of the study by the DNB (2014) are supportive of the endowment effect. To control for equity constraints the model contains dummy variables for different LTV ratios. In addition they found that a household with an LTV larger than 1 has a home value bias that is 14% higher than households without a mortgage. Whether loss aversion is effecting the home value bias is not tested by the GLS regression model. The study compares the median of the perceived value with the actual value. The study of the DNB (2014) found that the median homeowner reports a perceived value that is 13% higher than the actual value. In addition they found that in 75% of homeowners have a rosy picture of the current value of their home.
2.2.2 Commercial Real Estate Market
Bokhari and Geltner (2011) applied the model of Genesove and Mayer (2001) on commercial real estate in the US. Bokhari and Geltner (2011) improved the model of Genesove and Mayer (2001) by including the prospect theory reference point in the value function. The advantage
of this improvement was that they were able to estimate a single unbiased coefficient measuring loss aversion. Genesove and Mayer (2001) provided a range, they only produced an upper and lower bound. The preference towards commercial real estate in contrast to housing comes from the view that property owners have a sentimental attachment to their homes and, as a result, could be overly influenced by emotions in their listing and sales behaviour. Bokhari and Geltner (2011, p.635) found that loss aversion plays a major role in pricing of commercial property, and it varies both across the type of market participants and across the cycle. They found that investors were most loss averse during the peak of the cycle (2007). The effect of loss aversion in the commercial property market is of similar magnitude as the effect Genesove and Mayer (2001) found on the housing market. Bokhari and Geltner (2011) enhanced the research of Genesove and Mayer (2001) by studying whether more experienced investors are less loss averse than their counterparts. They found evidence that the more sophisticated investors are at least as loss averse as their counterparts.
Crane and Hartzell (2010) applied to model of Genesove and Mayer (2001) on real estate investment trusts (REITs). REITs are companies that own, and in most cases operate, income-producing real estate. REITs are characterised by not paying corporate taxes and they are required to distribute almost all of their income. Crane and Hartzell (2010) studied the disposition effect by examining a sample of individual properties held by large, publicly traded REITs in the US over the period between 1996 and 2006. The findings indicate that REITs managers are more likely to sell properties that have performed better, compared to those that have performed worse. They controlled for market volume, property type and size and the performance and size of the REIT itself. The results are both economically and statistically significant. Next they found REITs are significantly less likely to sell properties that have a loss relative to a reference point that evolves over time, based on plausible
benchmark for returns: the rate of inflation and price changes on similar properties (Crane and Hartzell, 2010). Third, they found that the firms that appear to have investment decisions that are most subject to the disposition effect, tend to realize lower prices when they sell winners. These findings provide further support for the disposition effect. Crane and Hartzell (2010) found little support for alternative explanations (optimal tax timing, mean reverting property-level returns and asymmetric information) for selling winners and holding losers.
2.2.3 Stock Market and M&A
Loss aversion is also studied in other fields than real estate markets. Odean (1998) studied trading records for 10.000 accounts at a large US discount brokerage house in the period
between 1988 and 1993. Odean (1998) tested the disposition effect by looking at the
frequency with which they sell winners and losers relative to their opportunities to sell each. Odean (1998) found significant results that indicate a disposition effect. Individual investors realised their profitable stocks at a much higher rate than their unprofitable stocks (Odean, 1998). This behaviour was not found in December, probably caused by tax-reasons: December is the deadline for realising losses for tax benefits. Therefore investors sell their
loser before the deadline.
Baker, Pan and Wurgler (2012) studied the effect of reference point prices in M&As in the period between 1984 and 2007. Baker et al. (2012) examined this by using peak prices as reference point. The most importing result from their paper is that psychological pricing has real effects. Baker et al. (2012, p. 30) reported that bids exceeding the 52-week high,
discontinuously increase the probability of deal success and thus the distribution of capital across firms’ alternative investment policies.
2.3 Equity Constraints
The basic theory of equity constraints was first demonstrated by Stein (1995). The effect of down payment constraints on selling behaviour is best understood through an example. Consider a household that purchases a home worth €100.000 with a 10% down payment. If house prices raise by 10%, the home will be worth €110.000 and the household will have €20.000 in equity. For the same down payment requirement the household could trade up significantly: it could now use that equity to purchase a €200.000 home with 10% down payment. However, would the price decline with 10%, the home would be worth €90.000 and the household would have no equity. The household could not make the down payment on the same home without other wealth. With no other wealth, the household could not move and remain homeowner3. But if house prices would fall by 5%, the household would only have enough to make a down payment of €5.000. Rather than moving to a smaller house, they may rationally choose to stay in their current house. Another option for the household would be “fishing”, meaning that they are listing there current house for a price above market value. They then hope to get lucky and raise enough money to make a reasonable down payment and be able to trade up for a larger home. These arguments suggest that both the time on the market and the volume of trade will be related to the level of prices. In conclusion, Stein’s
3
Example provided by Stein, J.C (1995), Prices and trading volume in the housing market: A model with down-payment effects, Quarterly Journal of Economics, 110, 379–406
model (1995) presents that a positive shock to market fundamentals improve the equity position of property owners, thereby allowing them trade up for a larger house. This will increase mobility and transaction volume . An increase in mobility and transaction volume will cause an even further increase in prices.
Stein (1995) did not develop an explicit equilibrium model. The theory of Stein (1995) was extended upon by Ortalo-Magné and Rady (1999, 2006). Ortalo-Magné and Rady (1999, 2006) integrated down payment constraints within an overlapping generations-model, where households move from starter homes to trade-up homes depending on income shocks and funding possibilities. In equilibrium this model generates a positive correlation between sales and prices: households are trading up if they can afford the down payment.
Several studies examined whether equity constraints can explain price-volume correlation in the housing market. Genesove and Mayer (1997) studied the Boston
Condominium in the period between 1990 and 1992. Genesove and Mayer (1997) developed a model that measured the effect of LTV ratio on the time on the market, asking price and sales price. The results of this study showed that homeowners with a higher LTV ratio set a higher asking price, have a longer time on the market and if they sell, receive a higher price than a homeowners with a lower LTV ratio. A homeowner with an LTV of 1 has a list price 4% higher than a homeowner with an LTV of 0,8. If a homeowner sells his house the sales price is also 4% higher for a homeowner with an LTV of 1 than a homeowner with an LTV of 0,8. Furthermore, the time on the market is 15% longer for homeowners with an LTV of 1 compared to homeowner with an LTV of 0,8. This can be explained by the strategy of ‘holding out for a high price’. Setting a higher asking price will consequently result in a
longer time on the market.
Another paper that examined equity constraints is written by Chan (2001). Chan (2001) studied the behaviour of homeowners in the New York metropolitan from 1989 until 1994. The effect of equity constraints was measured by estimating a hazard model of housing spell durations in which the key explanatory variable was the households contemporaneous LTV ratio. The study considered a homeowner equity constrained when the homeowner has less than 20% equity. Chan (2001) found that constrained households experience a 24% reduction in mobility relative to unconstrained owners in the four years after the decline in prices.
The effect of equity constraints on the price-volume correlation is also studied in the Netherlands. De Wit, Englund and Francke (2010) examined the Dutch housing market between 1985 and 2007. They developed a VEC-model that allowed them to study the
mechanism giving rise to the price-volume correlation. The findings showed that an interest shock had a gradually increasing effect on price and an immediate temporary effect on sales, which indicated that the credit constraint theory does not explain the correlation between prices and quantity. This conclusion is in contrast with the findings of Genesove and Mayer (1997) and Chan (2001). The results are more in line with the concept of a gradual adjustment of expectation to new information about fundamentals. This will be explained in more detail in the next section (2.4). De Wit et al. (2010) also stated that they did not study a period with falling prices, a situation the equity constraint theory might be the most relevant.
In addition, Stein’s (1995) model can also be applied to firms instead of households. Housing equity plays a role similar to collateral for firms. An economic shock can reduce the value of an assets used for productive purposes and collateral can decline the net worth of the firm. This can reduce the asset demand for constrained firms and then result in lower asset prices. This topic is more broadly studied by Kiyotaki and Moore (1997) and Schleifer and
Vishny (1992).
This research examines the effect of equity constraints, in addition to the effect of nominal loss aversion. However most of the previous called studies examine the US housing market, where down-payments are required. This is not the case in the Netherlands, but since the global financial crisis in 2007 credit markets have dried up. Therefore it is very important to examine equity constraints in the Netherlands as well. Thereby de Wit et al. (2010) propose to study equity constraints in a period with falling prices. By analysing data from 2009 and 2012, this proposition is complied with.
2.4 Alternative Explanations
An alternative explanation for price-volume correlation is that the real estate is a search-market and fails to clear instantaneously. According to Berkovec and Goodman (1996) sellers and buyers set reservation prices that correspond with their own economic situation and their expectations of the reservation prices of the buyers. A transaction occurs when the sellers’ asking price is lower than the buyers’ reservation price. Assumed is that buyers and sellers observe transaction prices but have no knowledge about the underlying market conditions. Buyers and sellers adjust their perception of the market price level on observed prices until a new stationary equilibrium is reached. From here a demand increase will result in an
immediate increase in transactions. This will follow in a gradual increase in reservation and transaction prices. The turnover rate will stay above normal until the new (and higher) equilibrium price is set. De Wit et al. (2010) studied this phenomenon in the Dutch housing
market for the period between 1985 and 2007. They found a pattern that is consistent with a search-market where buyers and sellers gradually learn about changes in market conditions
(de Wit et al., 2010, p.1).
Another alternative explanation for the price-volume correlation is the option-value of homeowners. The cost of buying the option can be seen as the down payment of mortgage at time of purchase. To carry on the option monthly principal and interest payments are
necessary. At the time the house is sold and the mortgage balance is paid, the option is exercised. When the homeowner defaults the option is expired. Because of this homeowners losses are limited and the gains are unlimited. However this is only the case with non-recourse loans. Cauley and Pavlov (2002) estimated the option-value of an owner’s interest in a
property in a specific real estate market (Los Angeles single family dwellings). They found that homeowners wait to sell their homes when the benefits exceed the net carrying costs. Therefore the possibility of an otherwise mutually advantageous transaction would be eliminated by the option-value of potential seller’s interest. In the Netherlands, non-recourse loans are not the case. Therefore, this explanation is not further considered in this paper. 2.5 Summary Literature
Several empirical studies show that within real estate, price and volume correlate with each other. Trading volume is often higher in times when prices are rising and lower in periods of falling prices. This pattern is inconsistent with a pattern you would expect in the case of an efficient market. This pattern is in general explained by two theories: nominal loss aversion en equity constraints. The theory of nominal loss aversion predicts that price changes cause changes in trading volume because investors dislike losses more than they like equal size. A problem with applying the prospect theory is that the reference point is rarely observed. However an exception is a study by Genesove and Mayer (2001). They developed an empirical framework that captured the effect of loss aversion. By applying this model on a housing market they found empirical evidence that supported the prospect theory. Other papers that applied this model or an adapted version of this model also found empirical evidence in favour of the prospect theory. Thereby empirical evidence for the prospect theory is found in commercial real estate markets, REIT markets, stock markets and in M&As. Also evidence for equity constraints is found in previous literature. The theory of equity constraints indicates that a positive shock to market fundamentals improves the equity position of
evidence for the effect of equity constraints for homeowners in the US. Evidence for the existence of equity constraints in the Netherlands is not found.
3. Methodology and Hypotheses 3.1 Methodology
A problem with applying the prospect theory to empirical studies is that the reference point is rarely observed. An influential exception is a study by Genesove and Mayer (2001); they developed an empirical framework that captured the effect of loss aversion. Genesove and Mayer (2001) developed two feasible models. The first model is specified as followed:
(1)
This model specifies that the log asking price ( ), is a linear function of the observable attributes and the quarter of listing and the difference between the purchase price and the predicted price from a hedonic equation ( . The is the vector of observable attributes and is a time-effect that shifts expected price proportionally. The second feasible model is specified as followed:
(2)
This model specifies that the log asking price ( ), is a linear function of the observable attributes and the quarter of listing , the noisy proxy for unobserved quality
and the difference between the purchase price and the predicted price from a hedonic equation ( . is the previous selling price, is the vector of the
observable attributes and is the quarter of original purchase. In a later stadium Genesove and Mayer (2001) also incorporate an LTV variable to control for another possible
explanation for price-volume correlation: equity constraints. The studies by Engelhardt (2003), Anenberg (2010) and Bokhari and Geltner (2011) applied an adapted version of the model of Genesove and Mayer (2001) on their data.
Based on the models from Genesove and Mayer (2001), an adapted version for this study is developed. A large difference between their study and this study is the fact that Genesove and Mayer (2001) used transaction data and this study is based on a survey. This means the respondents are not necessarily planning to move. Therefore this paper examines the effect of loss aversion on expected sales price instead of asking price. The expected sales price is used as a proxy for the asking price. In order to study nominal loss aversion in the Dutch housing
market, two models are constructed. First a model with dummy variables of loss and the LTV is constructed to determine if there is any effect of loss aversion and equity constraints on the dependent variable.
The first model is specified in the following way:
(3)
This model specifies that the log of the ratio of expected sales price and WOZ value is a function of loss and LTV. The dependent variable is the ratio of the expected sales price and WOZ value. The expected sales price is the price the respondent from the survey expects to receive if the current house was sold at the time the survey was completed. The WOZ value is used as a proxy for market value. The WOZ value is the market value of a property
determined by municipality on specific date. For the dataset WOON2012 this is 1 January 2011, for the dataset WOON2009 this is 1 Januri 2009. This WOZ value is used as an approximation of the market value at the time the survey was completed. Therefore it is assumed there is no large fluctuation in price between the time the survey was completed and the specific date (maximum of 1 year). By using a ratio of the expected sales price and the WOZ value, the dependent variable captures the expectation if the property is sold below or above market value. The specification is comparable with the study of the DNB (2014). Their dependent variable is the home value bias, which is the ratio of the perceived house value and the actual value. They determined the actual value by multiplying the original purchase price and a price development factor. The specification of the independent variable does not correspond with the model of Genesove and Mayer (2001). Their independent variable is the asking price at time of listing. However they controlled for housing characteristics, which is not included in the model of this study. Housing characteristics are already incorporated in the WOZ value.
To capture the effect of loss aversion, the first regression model uses a dummy variable. An expectation is considered as loss if the expected sales price is smaller than the original purchase price. A dummy variable is constructed, which has the value of one if the difference between the original purchase price and the WOZ value is smaller than zero and zero if the difference between the original purchase price is larger than zero.
The control variable LTV is added to the regression to control for the another possible explanation of price-volume correlation: equity constraints. As stated in the previous section, it is also important to control for the effect of equity constraints in the Netherlands, even
while down payments are not required in the Netherlands. Since the global financial crisis in 2007, credit market have dried up, which affected the accessibility of new mortgages. The LTV ratio measures the households’s home equity stake. The LTV is the mortgage value at time of purchase divided by the WOZ value4.
(4)
In the Netherlands, saving mortgages, investments mortgages and interest-only mortgages are very popular compared to repayment mortgages (DNB and AFM, 2009). Therefore the mortgage value at time of purchase is used as a proxy for the mortgage value at time the survey was completed. Figure 2 shows the distribution of different mortgage types from 1998 until 2010. As stated before the WOZ value is used as an approximation of the market value at the time the survey was completed.
Figure 2: % of all mortgage types in the Netherlands
Most literature (Genesove and Mayer, 2001 and 1997, Anenberg, 2010 and Engelhardt, 2003) considers households with an LTV greater than 0,8 equity constrained in their empirical analysis. The reason for this is that 80% is the maximum share that can be borrowed upon purchase in the US (Genesove and Mayer, 1997). All these papers studied the US housing market. However this papers studies the Dutch housing market. In the Netherlands, LTV ratios are higher than in the US on average. The relatively high LTV ratios in the Netherlands
4
The WOZ value is present in both sides of the equation. The WOZ value is part of the independent variable and the LTV. This could lead to reverse causality. Reverse causality can be tested by using an instrumental variable to isolate the part of independent variable that is uncorrelated with the error term. However a valid instrumental variable is not available for this research and therefore the test cannot be applied.
0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100% 1998 2000 2002 2004 2006 2008 2010
% of all mortgage types Netherlands
Interest only Other Life insurance Savings Investments Traditional Bron: ABN Amro 2012can be explained by tax deductibility combined with lenient mortgage regulation (ABN Amro, 2012). Households are not constrained with a maximum LTV of 0,8. However since 2010 there is a strict LTV-limit of 104% plus applicable transfer tax. In the future the LTV-limit is scheduled to decrease further to 100% in the future (ABN Amro, 2012). For these reasons this study considers households equity constrained when the LTV ratio is larger than 1. The first regresion model uses a dummy variable which has the value one if the LTV is larger than 1 and zero if the LTV is smaller than 1.
The second regression model is developed to determine the scale of the effect. In this case the dependent variable is regressed on the interaction variable loss, the LTV and a spline function of the LTV.
The second model is specified in the following way:
ƞ (5)
Again, the dependent variable is the ratio of the expected sales price and the WOZ value. By using this ratio, the dependent variable captures the expectation if the property is sold below
or above market value.
For the second regression model, the interaction variable loss is constructed. The first part is the difference between the WOZ value and the original purchase price. The second part is the dummy variable, that is also used in the first regression. This specification is based on a paper by Baker, Pan and Wurgler (2012) who studied the effect of reference point prices on M&As. This specification captures the effect of prospective losses more clearly than the specification used by Genesove and Mayer (2001).
(6)
The following variables, the LTV and the spline function of the LTV are incorporated to control for equity constraints. This specification is based on the paper of Genesove and Mayer (1997). Both LTV and the spline function of the LTV are added to make the function continuous. As described earlier the LTV is the ratio of the mortgage value at time of
purchase and the WOZ value. Then a spline function of the LTV is added to the regression model to capture the threshold effect. The spline function is created in order so that the log of the ratio of the expected sales and WOZ value is piecewise lineair and continuous in LTV. This allows the sensitivity of the log of the ratio of expected sales price and WOZ value to differ on either side of a threshold of 1 (Genesove and Mayer, 1997).
This results in the following function:
1 ∗ 1 (7)
In a less technical way , this can be described as the amount of LTV above the equity
constrained level of 1 in case the household is equity constrained. The second regression model, is executed in his basic form as stated in equation 4 and in an extended form. The extended form of this regression model contains a few extra control variables. First, dummy variables for the month the survey is completed are added to control for the difference in the date the survey was completed and the WOZ value was determined. Second, dummy variables for the year the house was purchased in are added to control for the overestimation of the loan value. To determine the LTV, the mortgage value at time of
purchase is used. But the longer ago the house is purchased, the higher the probability the loan value is overestimated. Subsequently, the log of taxable income is included besides the dummy variables for the month the survey is completed and the dummy variables for the year the house is purchased. This control variable is added based on the statement of Berkovec and Goodman (1996). According to Berkovec and Goodman sellers and buyers set reservation prices that correspond with their own economic situation. Lastly, a dummy variable for the moving tendency is added to the second regression model. The moving tendency dummy has a value of 1 if the household took action to move in the past half year. This control variable is added to the model to control for the fact that data from a survey is studied instead of
transaction data. If a household would like to move, they probably have more knowledge about the value of their house. It is expected that household who have more knowledge about the value of their house, expect a sales price closer to the market value.
This study also considered to take up housing characteristics as control variables. However, this is already incorporated in the WOZ value and adding housing characteristics will lead to multicollinearity. Previous studies such as Genesove and Mayer (2001) did incorporate housing characteristics. They studied the effect of loss aversion on asking price in itself and therefore controlling for housing quality was necessary. This paper examines the expected sales price compared to the market value and is therefore not impressionable by the arguments of Genesove and Mayer (2001).
3.2 Hypotheses
This section describes hypotheses that are constructed to study whether the phenomenon loss aversion plays a significant role on the price setting of households in the Dutch housing market.
The first hypothesis is:
H01: The coefficient that represents a loss will have a positive sign
The research of Genesove and Mayer (2001), Engelhardt (2003), Anenberg (2010) and Seslen (2004), who all studied loss aversion in the US housing market, found evidence that supports this hypothesis. Also Bokhari and Geltner (2011), who studied the commercial real estate market in the US, supports this hypothesis. In addition, Crane and Hartzell (2010) found evidence for this hypothesis in the REITs market.
Moreover, this research also examines whether the alternative explanation equity constraints plays a role on the price setting of households in the Dutch housing market. Therefore the second hypothesis is:
H02: The coefficient that represents the LTV will have a positive sign
This hypothesis is consistent with the findings of Genesove and Mayer (1997, 2001), Chan (2001) Anenberg (2010) and the study of the DNB (2014). However there are also studies who did not find supporting evidence for this hypothesis. Engelhardt (2003), who found that loss aversion has a significant effect on intra-metropolitan mobility, did not find evidence of a connection between equity and homeowner mobility. De Wit et al. (2010) found that the credit constraint theory does not explain the correlation between price and quantity. This thesis also aims to investigate if loss aversion plays a larger role on the price setting of households in the Dutch housing market than equity constraints. Therefore a third hypothesis is stated:
H03: The coefficient that represents LTV, will be significantly smaller (in absolute value) than
This hypothesis is only tested in the first regression model because in this regression model the variables have the same functional form (dummy variables) and can easily be compared. The hypothesis is consistent with the findings of Genesove and Mayer (2001) and Anenberg (2010). They found that liquidity constraints, though still significant, appear less important than loss aversion.
These hypotheses are tested for the two datasets (WOON2009 and WOON2012). An important part of this research is the comparison of the outcome of the different datasets. Until early 2008, house prices rose almost continuously in the Netherlands. However when the Netherlands was hit by the global financial crisis and the housing market got affected, house prices and trading volume went down (Graph 2 & 3). In 2009, households just started to notice the effects of the global financial crisis. It is expected that by 2012 the households were more anticipating of the changed market conditions. Therefore it is expected that a larger effect of loss aversion will be found in 2009 than in 2012. Another result of the global
financial crisis were the bad financing conditions. Banks were less willing to provide mortgages after the global financing crisis. In 2009, the financial market was still in shock from the crash in 2007 and credit markets had dried up. By 2012, financial markets recovered a bit. Therefore it is expected that equity constraints also play a larger role in 2009 than in 2012. However since 2010 there is a strict LTV limit of 104% plus applicable transfer tax (ABN Amro, 2012). This might increase the effect of equity constraints in 2012.
4. Data and Descriptive Statistics 4.1 Data
For this research two datasets are used: WOON2009 and WOON2012. These datasets contain the results of a survey held under Dutch citizens. The government holds this survey every three year to research how people live and how people would like to live in the future. The government looks into the composition of households, housing, housing costs, housing needs and environment. In 2009 78.000 people were interviewed and in 2012 70.000 people were interviewed.
By using the datasets WOON2009 and WOON2012 a very interesting time period is studied5. To study the effect of loss aversion it is very important to study a period that captures a large fall in prices (Bokhari and Geltner, 2011). From the mid-80’s until early 2008, house prices raised almost continuously in the Netherlands. In 2007, the Netherlands was hit by the global financial crisis. The global financial crisis affected the Dutch housing market. Because of the uncertainty and bad financing conditions consumers delayed buying or selling a home. This resulted in a stagnation of the Dutch housing market. According to data from Kadaster and NVM, both transaction volume and price levels decreased after 2008 (Figure 3 and 4).
Figure 3: Transaction volume in the Netherlands
Figure 4: Average price level in the Netherlands
5 Bokhari and Geltner (2011) found that investors were most loss averse during the peak of the cycle. Therefor it would be
interesting the study the outcome of WOON2006 as well. However the results of this dataset are not comparable with the outcome of WOON2009 and WOON2012. Up until the crisis, WOZ values highly underestimated the true market value. Many municipalities rectified this diversification from market value during the financial crisis. The very low WOZ value resulted in a unusual low ratio of expected sales prices and WOZ value for WOON2006. Therefore the results of this dataset are left out.
0 50.000 100.000 150.000 200.000 2008 2009 2010 2011 2012 2013 Number of properties
Transaction Volume in the
Netherlands
Source: Kadaster 180000 190000 200000 210000 220000 230000 240000 250000 2008 2009 2010 2011 2012 2013 Eu roAverage Price level in the
Netherlands
Source: NVM4.2 Descriptive Statistics
In this section the descriptive statistics for each dataset are described. First, households that rent rather than buy are dropped from the dataset. Only households that are owner occupied are left. Second all observations that have missing values are deleted from the datasets. Last, outliers are removed for the variables expected sales price, WOZ value, loss and the LTV. Outliers are removed because large outliers can make ordinary least squares (OLS) regression results misleading. An observation is considered an outlier when it is two times larger than the standard deviation (Stock and Watson, 2012). This resulted in 28.323 observations in 2009 and 27.698 observations in 2012. The sample is more than halved by taking these steps. Most observations ( 40.000) are dropped because the households are renters instead of
homeowners. Furthermore, a large number of observations are dropped because the mortgage value was unknown.
Table 1 shows the summary of statistics from WOON2009 and WOON2012. The average expected sales price is slightly lower in 2009 than the average expected sales price in 2012. This does not correspond with our expectation based on the market developments of sales price (Figure 4). Also the average WOZ value in 2009 and 2012 does not correspond with the market development of sales price (Figure 4). But it is comparable with the average values of the expected sales price. This deviation could be explained by the fact that the two surveys do not cover the same respondents. It could be the case the survey of 2012 contains more larger
Table 1 Summary of statistics WOON2009
Variable Observations Mean Std.Dev
Expected Sales Price 27600 249180 90617
WOZ Value 27600 238755 83381 Loss Dummy 27600 0.1205 0.3255 Ltv Dummy 27600 0.2105 0.4077 Loss 3325 20745 25347 Ltv > 1 5811 1.1389 0.1210 WOON2012
Variable Observations Mean Std.Dev
Expected Sales Price 27698 255940 97390
WOZ Value 27698 247226 85084
Loss Dummy 27698 0.2142 0.4103
Ltv Dummy 27698 0.2877 0.4527
Loss 5934 19500 21690
homes than smaller homes, which results in higher values. However, for this study it is more interesting to compare the ratio of expected sales price and the WOZ value. Table 2 shows the ratios of the expected sales price divided by the WOZ value in 2009 and in 2012.
These ratios are in line with our expectations. As stated before, in 2009 households just started to notice the effects of the global financial crisis. In 2012, the households were probably more anticipated to the changed market conditions. This corresponds with the numbers shown in table 2. The ratio of expected sales price and WOZ value was lower in 2012 than in 2009. This means that households set a reservation price closer to the market value (WOZ value) in 2012. Another explanation could be that effect of loss aversion is larger in the peak of the cycle. This is in line with the findings of Bokhari and Geltner (2011), who found that investors were most loss averse during the peak of the cycle. From figure 3 and 4 can be concluded the market conditions were slightly better in 2009 than in 2012. However, 2009 cannot really be considered as a peak.
Table 1 also shows some information about the variables loss. A loss occurs when the original purchase price exceeds the market value (WOZ value). In 2009, 12% of the
households faced a loss when they sold at the moment of the survey and in 2012 21% of the household faced a loss when they sold at the moment of the survey. When a household faced a loss, the average loss for 2009 was €20.745 and for 2012 €19.500. It is particularly striking that in 2012 more households faced a loss, while the height of the loss was smaller than in 2009. It is difficult to compare these numbers because loss depends partly on the original purchase price and the height of the original purchase depends for a large amount on when the house is purchased. When the LTV is larger than one, the household is considered equity constraint. In 2009 the percentage of households that are equity constrained is 21%, while in 2012 this percentage is 29%. If a household is equity constrained the average LTV is 1,14 for 2009 and 1,18 for 2012. This does not suit the expectation. A lower LTV ratio for 2012 was expected because since 2010 there has been a strict LTV limit of 104% plus applicable transfer tax (ABN Amro, 2012). However a large amount of houses is purchased before this rule was introduced. Only 8% of the house was purchased after 2009 and suffered from this measurement (Appendix (2)).
Table 2 Ratio Expected Sales Price / WOZ Value
WOON2009 WOON2012
5 Results
5.1 Basic Regression model
In table 3 the output of the two regression models for WOON2009 and WOON2012 is shown. First the dependent variable (Log of (Expected sales price/WOZ value)) is regressed on the variables loss dummy and LTV dummy in order to determine if there is any effect of loss aversion and equity constraints on the dependent variable. The second regression model is executed to determine the scale of the effect. In this case the dependent variable is regressed on the interaction variable loss, the LTV and the spline function of the LTV.
From the first regression we can conclude that both loss aversion and equity constraints affect the dependent variable in 2009 and 2012. For both years the coefficients of loss dummy and LTV dummy are significant. The regression output shows that if a household expects a loss, the ratio of the expected sales price and the WOZ value is 5,8% point higher in 2009 and 4,3% point higher in 2012 than when the household does not expect a loss. In addition, the first regression model shows that when households are equity constraint (LTV>1), the ratio of the expected sales price and the WOZ value is 5,1% point in higher 2009 and 4,1% point
Table 3: The effect of loss aversion
Estimates of OLS regression to estimate the effect of loss aversion on the expected sales price. The dependent variable is the Log (Expected Sales Price/WOZ value)
WOON2009 WOON2009 WOON2012 WOON2012
Variable (1) (2) (1) (2)
Loss dummy .0563138*** .0421011***
(.0040679) (.0036379)
Loss*Loss dummy 1.36e-06*** 1.67e-06***
(1.48e-07) (1.26e-07) Ltv dummy .0493741*** .0403887*** (.0030422) (.0033128) Ltv .061564*** .0479778*** (.0047329) (.0052755) (Ltv-1)*(Ltv>1) .1628293*** .0637697*** (.0203888) (.0188527) Constant .0198147 -.0118617 -.0006358*** -.0255062 (.0014174) (.0034185) (.0016759) (.0038125) Adjusted R^2 0.0238 0.0332 0.0165 0.0243 Prob > F 0.0000 0.0000 0.0000 0.0000 Number of obs. 27600 27600 27698 27698
Coefficient estimates and their standard errors are reported. Standard errors include the White correction for heteroskedasticity.
higher in 2012 than when they are not equity constraint. The second regression model shows the scale of the effects of loss aversion and equity constraints. Again, the coefficients of the variables are significant. For 2009 and 2012 the results for the loss variable are very similar. If a household faces a loss and the loss increases with €10.000, the ratio of the expected sales price and the WOZ value increases with 1,36% point in 2009 and 1,67% point in 2012. For the LTV, the effect in 2009 and 2012 differs more. For 2009 the results show that if a household is equity constrained and has an LTV of 1,1, the ratio of the expected sales price and the WOZ value is 2,23% point
( . . ∗ . ) higher than when a household has an LTV of 1. For 2012 the results show that if a household is equity constrained and has an LTV of 1,1, the ratio of the expected sales price and the WOZ value is 1,16% point higher than when a household has an LTV of 1.
5.2 Extended Regression Model
Table 4 shows us the results of an extended regression model. To improve the model, a few control variables are added to second regression model of the previous section. The first regression model includes dummy variables for the month the survey is completed in and dummy variables for the year the house is purchased in. The dummy variables for the month the survey is completed in are added to control for the difference in the date the survey was completed and the WOZ value was determined. The dummy variables for the year the house was purchased in are added to control for the overestimation of the loan value. In the second regression model, besides the dummy variables for the month the survey is completed in and the dummy variables for the year the house was purchased in, the log of taxable income is included. Taxable income is included to control for the economic situation of the household. In the third regression model, dummy variable for the moving tendency is added to the second regression model. This control variable is added to the model to control for the fact that data from a survey is studied instead of transaction data.
In addition a fourth regression is executed that included variables, that controlled for the quality of the house. However, as expected this regression suffered a lot from
multicollinearity and many regression coefficients became insignificant. Therefore this regression is not included in this paper.
For 2009, table 4 shows significant coefficients for loss aversion and equity constraints for all three regression models. For 2012 the table shows insignificant results for the first and second regression model. The coefficient estimates for both 2009 and 2012 are quite comparable with the results of the basic regression model discussed earlier. For 2009 and 2012, the coefficients for loss and LTV become smaller when more control variables are added. If the output of the third regression model is compared with the second regression output of the basic regression model, it shows the effect of loss aversion and equity constraints are smaller. The basic regression model shows that if a households faces a loss and the loss increase with €10.000, the ratio of the expected sales price and the WOZ value increases with 1,36% point for 2009 and 1,67% point for 2012. While the third regression of table 4 shows that if a household faces a loss and the loss increases with €10.000, the ratio of the expected sales price and the WOZ value increases with 1,26% point for 2009 and 1,46% point for 2012. The results differ not much, but the basic regression model slightly overestimates the effect of loss aversion.
For equity constraints, we found the effect of equity constraints is underestimated in 2009 and overestimated in 2012. For the basic regression we found that if a household is equity constrained and has an LTV of 1,1, the ratio of the expected sales price and the WOZ value is 2,23% point higher than when a household has an LTV of 1 in 2009 and 1,16% point
Table 4: The effe ct of loss ave rsion
Estimates of OLS regression to estimate the effect of loss aversion on the expected sales price. The dependent variable is the Log (Expected Sales Price/WOZ value). Regression (1), (2) and (3) includes dummys for the month the survey is completed and dummys for the year the house is purchased. The complete regression output can be found in the appendix (5).
WOON2009 WOON2009 WOON2009 WOON2012 WOON2012 WOON2012
Variable (1) (2) (3) (1) (2) (3)
Loss*Loss dummy 1.50e-06*** 1.47e-06*** 1.20e-06*** 1.86e-06*** 1.86e-06*** 1.47e-06*** (1.53e-07) (1.53e-07) (2.83e-07) (1.30e-07) (1.32e-07) (3.27e-07) Ltv .0929941*** .0920666*** .0808181*** .0724198*** .0711693*** .0672016***
(.0065595) (.0065822) (.0401851) (.006939) (.00696) (.0146064) (Ltv-1)*(Ltv>1) .1237181*** .1296903*** .14378*** .0288587 .0311362 .0783971**
(.021966) (.0219674) (.0136121) (.0203596) (.0205913) (.0369317) Log (Taxable Income) -.0000266 -.0017926 .0061098** .0011402
(.0026565) (.004203) (.0024238) (.0043388)
Move tendency dummy .0095214 .0114682**
(.0050901) (.0054575) Constant -.0219315 -.0211551 -.2293237 -.0586612 -.124315 -.1075253 (.0601311) (.0659983) (.5643381) (.0126095) (.0289289) (.0523972) Adjusted R^2 0.0375 0.0372 0.0439 0.0267 0.0268 0.0291 Prob > F 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 Number of obs. 26093 25886 5302 26404 26171 5723
Coefficient estimates and their standard errors are reported. Standard errors include the White correction for heteroskedasticity. * significant for 10%; ** significant for 5%; *** significant 1%.
higher than when a household has an LTV of 1 in 2012. While for the third regression in the extended regression model if a household is equity constrained and has an LTV of 1,1, the ratio of the expected sales price and the WOZ value is 2,27% point higher than when a household has an LTV of 1 in 2009 and 1,15% point higher than when a household has an
LTV of 1 in 2012.
The extended regression model has significant coefficients and similar values as the basic regression model. However when validity checks are applied, the basic regression output is preferred over the extended regression model. The extended regression model suffers from multicollinearity problems and an omitted variable bias. This is explained in more detail in chapter six. To put the results in perspective of previous literature and in an economics perspective the results from the basic regression outputs are used. However when a conclusion is drawn, we have to keep in mind these results might slightly overestimate the effect of loss aversion.
5.3 Results in Perspective of Literature
The results support the first hypothesis. The sign that represents a loss is a positive sign. When households face a loss, they set a higher expected sales price compared to the WOZ value than if do not face a loss. This result is in line with the findings of Genesove and Mayer (2001), Engelhardt (2003), Anenberg (2010) and Seslen (2004). The magnitude of the effect of loss aversion is best comparable with the findings of Genesove and Mayer (2001) and Anenberg (2010), because their model is most similar to the one used in this research. Both Genesove and Mayer (2001) and Anenberg (2010) found that buyers set a 3,5% higher asking price when the they are facing a loss of 10%. This result can be best compared with our findings by using an example. If a household purchased a home for €100.000 and the market value today is €90.000, they face a loss of 10% or €10 000. Based on the results of Gensove and Mayer (2001) and Anenberg (2010), buyers will set a 3,5% higher asking price. Thus instead of setting an asking price of €90.000, they will set an asking price of €93.150. Based on our results, the ratio of expected sales price and WOZ value will increase with 1,36% point in 2009 and 1,67% point in 2012. Thus instead of a ratio of 1 (€90.000/€90.000), they will have a ratio of 1,0136 (2009) and 1,067 (2012). The household will set an expected sales price of €91.224 in 2009 and €91.503 in 2012. From this example we can conclude that this
research finds a smaller effect of loss aversion. The differences between the studies can explain the differences in the magnitude of the results. There are multiple differences between
this study and the study of Genesove and Mayer (2001) and Anenberg (2010). First of all, Genesove and Mayer (2001) and Anenberg (2010) both examine transaction data, while this research studies a survey. This study examines the effect of loss aversion on expected sales price, while Genesove en Mayer (2001) and Anenberg (2010) studied the effect of loss aversion on the asking price. However this difference does not necessarily explain why a smaller effect for loss aversion is found. Actually it would be more logical that a larger effect for loss aversion was found. Setting an expected sales price in a survey is less committal than actually setting an asking price. Another difference is that Gensove and Mayer (2001) and Anenberg (2010) studied another market and time period. Both Gensove and Mayer (2001) and Anenberg (2010) studied the US housing market while this paper examines the Dutch housing market. There are many differences between the US housing market and the Dutch housing market. In addition, Genesove and Mayer studied the period between 1988 and 2005 and Anenberg the period between 1985 and 1996. By studying a different market and period different market conditions occurred, which could have led to different effects. For example Bokhari and Geltner (2011) found that investors are most loss aversion in the peak of a cycle. This indicates that the effect of loss aversion could be different under different market
conditions. In the research period of this study the Dutch housing market is characterised by bad financing conditions and low consumer confidence, which might explain why the effect of loss aversion is smaller than found in previous studies. However the results do support literature that there is a positive effect of loss aversion on the expected sales price and
therefore support the first hypothesis.
The second hypothesis is also supported. In addition to loss aversion, equity
constraints have a positive effect on the ratio of the expected sales price and the WOZ value. This outcome is in line with the findings of Gensove and Mayer (2001, 1997), Chan (2001) and Anenberg (2010). The results of this research are most comparable to the findings of Anenberg (2001), because they also used spline function for the LTV. Anenberg (2010) found that an increase in LTV of 0,8 to 1 is associated with a 3,3% increase in price. This is lower than our results, which indicate that an increase in LTV of 1 to 1,2 is associated with an increase of 4,59% point ( . . ∗ . of the ratio of expected sales price and WOZ value in 2009. However, for 2012 we found an effect that is smaller than the findings of Anenberg (2010). An increase in LTV of 1 to 1,2 is associated with an increase of 2,26% point of the ratio of the expected sales price and the WOZ value. Again, there are large
differences between the study of Anenberg (2010) and this research. As mentioned before, the time period and market differ. Another important difference is that in the research of
