Prices drop on average 20% for units that become haunted (houses so declared because of a murder, suicide, or other unnatural death), 10% for units on the same floor, 7% for units in the same block, & 1% for units in the same estate

Bhattacharya, Utpal and Huang, Daisy J. and Nielsen, Kasper Meisner, Spillovers in Prices: The Curious Case of Haunted Houses (August 24, 2020). Review of Finance, 2020, SSRN: http://dx.doi.org/10.2139/ssrn.3679828

Abstract: Exploiting the unique institutional setting of Hong Kong’s real estate market, we uncover a curious ripple effect of haunted houses on the prices of nearby houses. Prices drop on average 20% for units that become haunted, 10% for units on the same floor, 7% for units in the same block, and 1% for units in the same estate. Our study makes two contributions. First, we provide an estimate of a large negative spillover on prices caused by a quality shock. Second, we find that the demand shock rather than the fire sale supply shock explains most of the spillover.

Keywords: fire sales, negative spillovers, haunted houses

JEL Classification: D62, H23, R21, R31

VI. Discussion and Alternative Specifications

There are a few issues to address. First, how valid is our definition of a haunted house? We define a haunted house in our paper as a house where an unnatural death occurred. According to this definition, a unit will not be considered haunted if the unnatural death occurred outside the unit. We now test this proposition by doing a placebo test. The treatment sample of 1,032 unnatural deaths is from our baseline specification in Column 1 of Table 4. The placebo sample (coded in similar fashion as the treatment sample) are 235 unnatural deaths that occur outside the residence of the deceased. We obtain information on these deaths from the Coroner’s Court as they do not feature on the real estate websites tracking haunted houses. The placebo sample consist of 183 deaths in traffic accidents, 23 accidental deaths during medical procedures, and 29 accidental deaths due to drowning while swimming in the ocean. We estimate Equation (1) using a joint specification for the treatment sample and for the placebo sample. Column (1) of Table 9 gives the coefficients of the treatment sample. The signs and magnitudes of the θ coefficients are similar to the signs and magnitudes of the θ coefficients in column (1) of Table 4. Thus, adding the indicators for the placebo effects to Equation (1) do not affect the estimated spillover effects. Column (2) of Table 9 gives the coefficients of the placebo sample. The θ coefficients are either insignificant or positive and significant. We take the placebo test one step further by excluding murder cases from the treatment group. As a result, the sample size declines with about 44,000 transactions in estates affected by murder. This insures that we compare the effect on prices of deaths due to accidents and suicides that occur at home (treatment group) to deaths due to accidents and suicides that occur outside the residence of the deceased. We find slightly smaller treatment effects and no placebo effects in Columns 3 and 4, respectively. Collectively, Table 9 tells us that when the owner of a unit dies unnaturally outside the unit, there is no discount on the unit’s price and no spillover effect on nearby houses. This is consistent with our conclusion that a house becomes haunted and is discounted only when there is an unnatural death in the house. Second, can the effects of idiosyncratic shocks be persistent? Yes. Unnatural deaths in a home cause a house to be declared haunted, and unnatural deaths in a home are, by definition, home-specific and unpredictable. This leads to a sudden drop of perceived quality. We find that house price recovery is slow, which implies that our idiosyncratic shock to perceived quality has persistent effects. This is different from the typical fire sales shock where effects on prices tend to be transient. Third, one might challenge whether it is rational for owners of neighboring units to sell at a discount. They could postpone sales until prices recover. We cannot address this concern directly because we do not have data on the amount of time that a unit has been on the market. So we address this concern with the following argument. As seen in Table 5, house price recovery is slow. Given this, it would seem that there is no point postponing sales until prices recover. So liquidity should not change much before and after the event for neighbors. Figures 6 and 7 show that it does not.

Fourth, how can discounts of 20% be sustained in equilibrium? The answer is straightforward. The belief in Feng Shui is quite strong among Chinese, and the population in Hong Kong is about 94% Chinese. Caucasians make up at most 5% of the population in any district, implying that most buyers have a large disutility for haunted houses. The few prospective buyers that do not themselves get disutility for haunted houses will, however, care about the resale value of their houses, and the resale value is expected to be low because many future buyers would dislike haunted houses. Fifth, a potential concern with the specification in equation (1) is that haunted houses are located in different areas than houses that are unaffected. If the demand for houses and/or the supply of houses in these locations are different, time-varying location effects might drive the results. For instance, a suicide might reveal economic hardship in the neighbourhood, or a murder might reveal crime in the neighbourhood. If the house price growth is slower in these districts, this effect might confound the estimated spillover effects in prices. We address the concern that our results are driven by slower price growth in affected districts by introducing high-dimensional fixed-effects to control for time-variant location effects. Table 10 reports the results, when we change the estimation (1) to 𝑦௜௧ ൌ 𝛼௜ ൅ 𝛽௝௧ ൅ 𝛾′𝑋௜௧ ൅ 𝜃′𝐻௜௧வ௞ ൅ 𝜖௜௧, (7) Here 𝛼௜ captures unit fixed effects, 𝛽௝௧ captures location-time fixed effects, 𝑋௜௧ is a vector of unit characteristics that change over time, and 𝐻௜௧வ௞ is a vector of indicators due to an unnatural death occurring before year t. Table 10 reports results. Column (1) shows the results from the baseline specification in Column (1) of Table 4. We note that although the coefficients in Columns (2) to (5) are slightly smaller than in Column (1), the results are qualitatively similar when we include locationtime fixed effects, both at the territory and district level. We conclude that our results are unaffected when we include high-dimensional fixed-effects that effectively controls for the development in house prices at different locations over time. These high-dimensional fixed effects also addresses potential concerns about pre-trends in house prices due to geographic location, because the spillover effects in Table 10 are estimated using variation in house prices within a location at a given time. Sixth, and finally, how relevant are our research findings for other parts of the world? We address this question in four ways. First, we find anecdotal evidence of a 25 percent discount on haunted houses in Australia, United Kingdom, and United States in a sample of 101 newspaper articles from Australia, United Kingdom and United States. For example, New York Times (Nov 24, 2016) interviewed Randall Bell, an economist who has consulted on the appraisals of notorious properties, like the homes of O. J. Simpson and Jon Benet Ramsey. According to Bell, the stigma can result in 25 percent lower prices. In comparison, we find that affected units in Hong Kong decline by 19 percent following an unnatural death. Second, we note that the U.S. legal system, as in Hong Kong, makes it illegal for a seller to hide the fact that the property being sold has a reputation of being haunted.19 Third, it is possible for prospective buyers in the U.S. to check whether anyone has died at a given address using web-based services like the website, www.diedinhouse.com. Fourth, Hong Kong is not an outlier in terms of suicide rates. The annual suicide rate in Hong Kong during our sample period is 14 per 100,000 people and is fairly stable over time. The corresponding numbers in Australia, China, Japan, United Kingdom and the United States are 13, 11, 23, 8 and 13, respectively. The annual homicide rate in Hong Kong during our sample period is a very low 0.6 per 100,000 people and is fairly stable over time. The corresponding numbers in Australia, China, Japan, United Kingdom and the United States are 1.3, 1.3, 0.4, 1.3 and 5.3, respectively. We therefore conclude that house price discounts due to unnatural deaths are relevant outside of Hong Kong. In Hong Kong, this may be due to Feng Shui, but in other parts of the world, the reason would be more universal: few like to buy a house where a recent unnatural death occurred.