Economic impact of large earthquakes: lessons from residential property values

Abstract

1. Introduction

The incidence of severe natural disasters has increased in frequency in recent years. These environmental shocks have both short-term and long-term repercussions on economic growth and development (Caruso and Miller 2015; Paudel 2023b). Disasters result in economic disruptions, causing property damage, loss of livelihoods and disparities in labor market outcomes (Boustan et al. 2020; Paudel 2022b; Rayamajhee and Paudel 2024). On the one hand, they cause poverty-environment traps that worsen the economic well-being of vulnerable communities (Carter et al. 2007; Van den Berg 2010; Gray and Mueller 2012; Paudel 2024a). On the other hand, they provide opportunities for reinvestment and capital upgrades, resulting in labor supply reallocation and technological development (Gignoux and Menéndez 2016; Deryugina, Kawano, and Levitt 2018).

Among different natural disasters, earthquakes have received significant attention among researchers. While earthquake risk can reduce housing values (Bernknopf, Brookshire, and Thayer 1990; Beron et al. 1997; Hidano et al. 2015), prior literature has focused largely on property markets in urban areas of large economies (Naoi, Seko, and Sumita 2009; Deng, Gan, and Hernandez 2015; Hanaoka, Shigeoka, and Watanabe 2018; Singh 2019). An influx of studies has delved into risks of induced seismicity caused directly by wastewater injection (Weingarten et al. 2015; Metz, Roach, and Williams 2017; Ferreira, Liu, and Brewer 2018), which is different from the natural incidence of high-intensity earthquakes. Although developing countries in Asia endure about 40 per cent of global economic losses from natural disasters (Mizutori and Guha-Sapir 2020; Kamble, Paudel, and Mishra 2024), whether large earthquakes influence residential markets in the developing world remains inconclusive. Understanding how economic agents respond to seismic shocks in low-income settings with inadequate risk-mitigating mechanisms is crucial to quantifying the true economic costs of natural disasters around the world.

In this article, I make use of the quasi-random variation in ground tremors to evaluate the economic impact of the April 2015 earthquake in Nepal on self-assessed residential property values. The 2015 earthquake in Nepal led to about 9,000 deaths and affected approximately 8 million people, causing economic losses of about 10 billion US dollars, equivalent to about a half of Nepal’s gross domestic product (Paudel and Ryu 2018). To quantify the severity of seismic intensity, I employ spatial variation in peak ground acceleration (PGA), which gives the maximum ground acceleration during earthquake shaking in a given location. Using the median district-level PGA threshold, I construct a binary indicator for a household’s exposure to high-intensity (“treated”) and low-intensity (“control”) seismic activity. Through the difference-in-differences (DID) research design, I evaluate changes in self-assessed residential values between households exposed to high and low levels of seismic shocks before and after the earthquake. Specifically, I use nationally representative, cross-sectional household data available from four different waves of household surveys during a four year-long period from 2014 to 2017. My empirical model incorporates a suite of fixed effects to account for both time-varying differences in property values common across households in villages, while controlling for month-by-year varying unobservable shocks to residential markets. I also employ an event study research design to estimate dynamic treatment effects on self-assessed home values associated with the earthquake.

Results indicate that seismic shocks have a strong negative short-term impact on residential markets. DID estimates show that residential property values declined by about 40.52 percentage points in areas with high intensity of seismic activities. Findings from an event study specification illustrate that this decline in residential property values lasted for about 24 months after the incidence of the earthquake. Subsequent analysis rules out the possibility that the economic impact of the earthquake on residential property values could be confounded by other channels. These channels include mortality selection, historical seismic shocks, population change, and the incidence of other natural disasters. Additional robustness checks, which involve the use of placebo tests and alternate indicators of seismic intensity, strengthen the validity of the estimated parameters. Although there are no systematic differences in baseline characteristics between treated and control households, it is reassuring that my treatment effect estimate is similar in magnitude to the one derived from a “synthetic difference in differences” estimator presented in Arkhangelsky et al. (2021).

I further explore heterogeneity in the impact of the earthquake across different property characteristics to examine the underlying mechanism of physical damage that could have caused the effect on residential property values. I hypothesize that well-built homes comprising outer walls with cemented bricks, concrete roofs, and pillar-based foundations are likely to have minimal damage from the earthquake, resulting in no significant effects on property values. Relatedly, weaker homes that possess non-concrete roofs, non-cemented outer walls, and foundations without pillars are likely to suffer from earthquake damage, causing significant changes in property values. These heterogenous treatment effect estimates conclude that earthquake-induced changes in property values are negative and statistically significant among homes with weaker foundations for outer walls and roofs. The effects of the earthquake on property values among homes with cemented bricks in the outer walls and concrete in the roof materials, however, are positive and statistically insignificant. These findings suggest that the introduction of policies aimed at replacing earthquake-affected homes built on weaker foundations with new ones under stringent building code regulations could substantially minimize future economic damage from earthquakes.

The estimated 40.52 percentage point decline in home values, equivalent to a $7,124 decrease in average value, is both economically and statistically significant. Large economy-based studies focused on the USA and Japan document a 13–35 per cent reduction in housing values from the earthquake (Beron et al. 1997; Naoi, Seko, and Sumita. 2009). Contrary to settings studied in prior literature, Nepal is a remittance-based economy with inadequate access to risk-mitigating mechanisms such as earthquake insurance and seismic hazard maps. In fact, countries with lower levels of income, weaker financial systems, and smaller degrees of openness incur substantial losses from natural disasters (Toya and Skidmore 2007). In this context, the estimated impact on home values can be interpreted as a “business-as-usual” scenario in which economic agents do not undertake any adaptive actions to mitigate the negative effects of seismic shocks, which has direct public policy implications. A conservative empirical exercise (more on Section 4.3.2) concludes that this earthquake-led decline in home values corresponds to an external cost of $71.2 million, which is arguably a much smaller estimate. I interpret the estimated loss of $71.2 million as a lower bound of the true economic cost of seismic shocks because it is beyond the scope of this study to account for earthquake-led changes in health, education, and labor market outcomes.

Figure 1

Event study analysis on changes in residential property values between treated and control districts.

Note: This figure plots the point estimate and 95% confidence interval of the coefficients on TreatedXPost from the event study specification in Equation (4). The dependent variable in the model is a self-assessed residential property value (in logs) for an individual household. The empirical specification includes a full set of dummies extending from 15 months before the earthquake to 24 months after and their interaction with the treatment indicator, district-specific quadratic monthly time trends, year-by-village fixed effects, demographic and housing characteristics such as household size, location type (urban or rural), number of rooms, number of years since the construction of the house, water supply type and access to electricity. Standard errors are clustered at the district level. Source: Author’s calculations.

Open in new tabDownload slide

This article is broadly related to a rich literature on the linkage between earthquake risk and housing values (Bernknopf, Brookshire, and Thayer 1990; Beron et al. 1997; Hidano et al. 2015). The majority of studies explore the incidence of large earthquakes in urban areas of large economies, including the USA, Japan, and China (Naoi, Seko, and Sumita 2009; Deng, Gan, and Hernandez 2015; Hanaoka, Shigeoka, and Watanabe 2018; Singh 2019). Notably, Beron et al. (1997) show that real estate price fell between 26 per cent and 35 per cent after the 1989 Loma Prieta earthquake in the six-county San Francisco Bay area. A different study reports a 13 per cent reduction in housing values from the earthquake in Japan, which corresponds to a discount of 3.8 million yen (Naoi, Seko, and Sumita 2009). There also exists growing evidence on risks of induced seismicity from wastewater injection and subsequent changes in home values in Oklahoma (Weingarten et al. 2015; Metz, Roach, and Williams 2017; Ferreira, Liu, and Brewer 2018). Contrary to prior emphasis on human-induced earthquakes, my focus on naturally occurring high-intensity earthquakes in a setting devoid of adequate risk-mitigating mechanisms that are prevalent in well-developed economies highlights the unique feature of this study. My approach is similar in spirit to Paudel (2022b) that uses satellite data on real-time active fire locations in a hedonic analysis framework to evaluate the economic costs of forest fires on poor households in rural locations. Environmental shocks such as forest fires occur more frequently throughout the year, while large earthquakes are rather unanticipated. To the author’s knowledge, this is the first study that exploits quasi-experimental research design to quantify changes in self-assessed residential property values in response to variation in seismic intensity in a developing country setting, with a special emphasis on the mechanism of physical damage.

The article further contributes to a growing literature on capitalization of risk perception into housing prices in response to natural disasters (Bin and Polasky 2004; Bin and Landry 2013). For example, Bin and Landry (2013) use multiple storm events within a DID framework and show a significant risk premium ranging between 6 per cent and 20 per cent for homes sold in the flood zones of North Carolina. A body of work has explored economic and behavioral changes in response to environmental shocks across diverse settings (Hanaoka, Shigeoka, and Watanabe 2018; Cole et al. 2019; Shakya, Basnet, and Paudel 2022; Paudel 2022a; Aguirre et al. 2023; Kim and Lee 2023). Recent studies highlight prospects of leakage in aid distribution during earthquakes across different levels of governance (Eichenauer et al. 2020; Paudel 2023a), implying that aid provision likely does not partially mitigate the economic costs of natural disasters in a developing country setting. Findings of this study are also directly related to how earthquakes influence welfare outcomes and resilience in labor markets across diverse settings (Gignoux and Menéndez 2016; Kirchberger 2017).

The remainder of the article is structured as follows. Section 2 presents a brief background on the empirical setting and hedonic valuation, along with the nature of data used in the study. Section 3 presents the empirical strategy, and Section 4 describes the main findings and discusses policy implications. Section 5 concludes with a summary of the article.

2. Background

2.1 Empirical setting

Nepal, a land-locked country surrounded by India on three sides and China to the north, is a federal republic consisting of seven provinces and 77 administrative districts that are further subdivided into 777 urban and rural municipalities. While each province includes 8–14 districts, each district has an average population of over 300,000 individuals and an average size of approx. 740 square miles (Paudel 2022b). Nepal is divided into three ecological zones: mountains (35 per cent), hills (42 per cent) and terai or plains (23 per cent), and forests constitute almost 40 per cent of the country’s land (Paudel 2018). Nepal comprises 103 caste and ethnic groups that are predominantly Hindus and Buddhists (Paudel 2021b). According to the Asian Development Bank, Nepal’s social protection system is ranked much lower than the average level (Paudel 2024b).

Nepal is vulnerable to natural disasters, including earthquakes and forest fires. Recent satellite data report that Nepal faces the recurring threat of forest fires that damage around 200,000 hectares of land every year (Paudel 2021a). Nepal also ranks 11th among the world’s most earthquake-prone countries and predominantly relies on remittances that comprise of over a quarter of Nepal’s Gross Domestic Product (Shakya, Basnet, and Paudel 2022). Natural disasters in Nepal have been linked with poverty and demographic changes that, in turn, have led to both ethnic fractionalization and shifts in political regime (Paudel and de Araujo 2017; Paudel 2021b; Shakya, Basnet, and Paudel 2022). In theory, Nepal has adopted the Sendai Framework for Disaster Risk Reduction into the National Disaster Risk Reduction Strategic Action Plan 2015–2030. The provincial government coordinates between federal and local governments, while the local government engages with relevant stakeholders during the disaster management cycle as per the Local Government Operation Act 2017. Yet market provision for earthquake insurance and dissemination of seismic hazard maps do not exist in Nepal. To the author’s best knowledge, there is unavailability of any relevant data on the establishment and operation of disaster management funds in Nepal, including details on how these resources are mobilized.

The 2015 earthquake in Nepal affected the lives of 20 per cent of the country’s population, causing an economic loss of 10 billion US dollars that amount to almost 50 per cent of Nepal’s gross domestic product (Paudel and Ryu 2018; Eichenauer et al. 2020). The shock possessed a magnitude of 7.8 (on the Richter Scale) on April 25, 2015, that killed about 9,000 people and damaged over 700,000 homes across the country (Shakya, Basnet, and Paudel 2022). More than 100 aftershocks followed the major earthquake in April. Supplementary Appendix Figure A1(a) includes a red dot to illustrate the epicenter of the earthquake in the eastern Gorkha District located northwest of the capital city of Kathmandu.

Although Nepal requested financial assistance from external donors during the first 5 months after the earthquake, research concludes that the implementation of aid distribution has not been effective. For example, Eichenauer et al. (2020) report that the majority of aid commitments in the aftermath of the earthquake favored municipalities dominated by higher castes and deprived those in need residing further away from the capital city of Kathmandu. Literature has further pointed out supporting evidence of leakage in aid distribution across different levels in the aftermath of the earthquake (Eichenauer et al. 2020; Paudel 2023a).

2.2 Hedonic framework

Although prices of housing units for hedonic analyses should ideally come from transactions in the competitive market where the demand price of a housing unit equals the offer price at equilibrium (Rosen 1974; Taylor 2003), there are different challenges associated with a study of housing markets in the Nepalese rural setting. First and foremost, Nepal, Nepal, and Berrens (2017) state that “reported market data, even if available, do not provide the actual price of housing units in Nepal as both sellers and buyers have incentives to understate the actual prices to avoid the stamp duty that both sides are required to pay at the given rates.” Second, confidential administrative data from property registration offices do not truly reflect the actual value of residential properties because of existing rampant under-reporting of the sales price for tax purposes (Nepal et al. 2020). Finally, while anecdotal evidence indicates that developers have slowly entered the housing market in recent years, there does not exist any developer survey yet that can be used in empirical research (Paudel 2022b).

The use of self-reported residential property values in the aftermath of a large natural disaster in a hedonic valuation framework is well documented in the literature (Naoi, Seko, and Sumita 2009; Paudel 2022b). According to Kiel and Zabel (1999), the use of owner’s valuations can provide reliable estimates of housing prices and neighborhood characteristics, especially in settings where it is not feasible to collect property value data from market transactions (Nepal et al. 2020). The magnitude of the estimated bias from using self-reported housing values is relatively small, averaged between 3 per cent and 8 per cent (Goodman and Ittner 1992; Kiel and Zabel 1999; Agarwal 2007; Gonzalez-Navarro and Quintana-Domeque 2009, 2016; Nepal et al. 2020).

2.3 Data

The core analysis of the study is based on detailed household-level data from four sets of nationwide household surveys, known as the Annual Household Surveys (AHS), conducted in different months of 2014, 2015, 2016, and 2017 in Nepal. This dataset, a nationally representative random sample of households from National Population Census 2011, is provided by the Central Bureau of Statistics in Nepal. The surveys involved a complete list of wards with number of households provided by National Population Census 2011 for their sampling procedure. The study also employs seismic intensity map of the 2015 earthquake provided by Paudel (2023a), which was originally made publicly available at the Humanitarian Data Exchange (https://data.humdata.org/) and has been used extensively in previous literature (Shakya, Basnet, and Paudel 2022).

2.3.1 Residential property values and characteristics

AHS provides detailed characteristics on demographics, housing characteristics and property values. Supplementary Appendix Table A1 provides summary statistics of demographic characteristics of households, residential property values and housing attributes. The household survey explicitly asks a homeowner, “If you would like to buy a dwelling just like the one you own today, how much money would you have to pay?” In the empirical sample, 76.56 per cent identify themselves as males and the average household size comprises of 6.1256 members. On an average, log-transformed residential properties are valued at approximately 13.32 Rs., consist of 5.06 rooms and are 16.45 years old. Indicators of material used for outer wall reveal that 35.03 per cent have cement bonded bricks or stones, 38.77 per cent have mud bonded bricks or stones, 5.28 per cent have wood, 17.41 per cent have bamboo or leaves, 1.27 per cent have unbaked bricks, and 2.23 per cent have other materials.

In relation to foundations of dwelling, houses are pillar bonded (23.16 per cent), cement bonded (13.36 per cent), and mud bonded (40.47per cent) and consist of wooden pillar (20.48 per cent) and other foundation (2.53 per cent). The majority of the residential properties use galvanized iron (33.25 per cent), concrete or cement (26.30 per cent), tiles or slate (21.17 per cent), and straw or hatch (11.89 per cent) for roof. Among households, 80.08 per cent have access to electricity and 42.77 per cent have access to piped water. The empirical sample includes 18,832 households interviewed in 2014, 13,448 households interviewed in 2015, 4,655 households interviewed in 2016, and 2,928 households interviewed in 2017, respectively. Supplementary Appendix Figures A2 and A3 present geographical variation in household size, self-reported residential property values, share of male-headed households, and average grades of schooling completed across districts of Nepal.

2.3.2 Seismic intensity

To quantify seismic intensity of the 2015 earthquake, I use district-level variation in PGA. PGA is the maximum ground acceleration that occurs during earthquake shaking at a location. The unit of PGA is g, where g is the gravitational acceleration, or 9.8 m²/s. The larger the value of PGA, the higher the degree of economic damages from earthquakes. PGA values in the empirical sample average approximately 0.0581 with a standard deviation of 0.0563 and range between a minimum of 0 and a maximum of 0.3600. PGA reflects the immediate ground motion experienced during the earthquake, which indicates a higher likelihood of damage to buildings and infrastructure. While PGA provides a standardized measure for comparing the intensity of earthquakes across time and space, it does not directly account for the duration of the shaking that may cause further structural damage. Consistent with prior literature (Fekrazad 2019; Shakya, Basnet, and Paudel 2022; Paudel 2023a), I employ variation in PGA to proxy for the seismic intensity of the 2015 earthquake.

The choice to use PGA as the preferred indicator of seismic intensity is crucial from a causal inference perspective. This is important because the exogeneity of earthquake-related deaths and physical damage across geographical locations is disputable. Evidence indicates that reported earthquake-related damage can be potentially correlated with a number of unobservable characteristics of a region (Paudel and Ryu 2018; Shakya, Basnet, and Paudel 2022). To alleviate potential bias with damage-related indicators, I use variation in PGA as a proxy for the intensity of seismic shocks. Supplementary Appendix Figure A1(b) provides a geographical map of Nepal with different values of PGA associated with the 2015 earthquake of Nepal. Darker shades of gray represent higher values of PGA and thus reflect large potential economic damages. I also use distance to the epicenter of the earthquake as an alternate proxy for a household’s exposure to seismic shocks from the 2015 earthquake in the empirical sample (more on Section 4.2).

3. Methods

Three methodological issues are worth highlighting. First, the identifying assumption of the DID model in Equation (3) requires that residential property values in both treated and control districts evolve in a parallel way in the absence of the earthquake. Relatedly, Supplementary Appendix Table A2 shows that none of the baseline observable characteristics are significantly different between treated and control households. Second, estimated parameters may be biased if individuals move to different districts in responses to the earthquake. Finally, the binary treatment indicator for seismic shocks, $Treatedd$⁠, in Equation (3) above is based on a median threshold of 0.04, implying that estimated parameters could be sensitive to the choice of the threshold. Section 4.2 addresses each issue in detail and conducts different robustness checks to strengthen the validity of the empirical strategy presented above.

4. Results

4.1 Residential property values

Table 1 presents DID estimates of the economic impact of the 2015 earthquake on self-assessed residential property values. The most preferred specification in Column (3) includes district-specific quadratic monthly time trends, demographic and housing covariates, month-by-year fixed effects and year-by-sampling unit fixed effects. The coefficient on Treated shows that differences in residential property values between treated and control districts before the incidence of the 2015 earthquake are statistically insignificant. The coefficient on After indicates that residential property values in control districts after the earthquake are not statistically different from those before the earthquake. This indicates that seismic shocks from the earthquake do not have any direct impact on the local housing market through changes in overall demand and supply conditions (Naoi, Seko, and Sumita 2009). The coefficient on the interaction term, –0.51, shows that there is a significant decrease in property values of treated districts compared to control counterparts in response to seismic shocks from the 2015 earthquake. The magnitude of the coefficient implies a sizable 40.52 percentage point decrease ¹ in residential property values of treated districts after the incidence of the earthquake. This finding is consistent with the notion that a rare but “sharp” event may result in an overestimation of risk among individuals (Tversky and Kahneman 1992; Deng, Gan, and Hernandez 2015), leading to a decrease in residential property values in response to large seismic shocks from the earthquake.

Table 2 breaks down DID estimates across different sub-samples to explore the robustness of results shown above.² In Column (1), I exclude observations from 2016 and 2017. In Column (2), I exclude observations from 2017. Across both columns, the coefficient on the interaction term is negative and statistically significant. Columns (1) and (2) indicate that the slope coefficients of –0.5185 and –0.5212 translate to a decline in residential property values of 40.45 and 40.61 percentage points, respectively. Next, I exclude three districts in the Kathmandu Valley (Kathmandu, Bhaktapur, and Lalitpur), which tend to have higher valuations of residential properties compared to the rest of the country. Column (3) shows that the coefficient on the interaction term, –0.5153, is statistically significant, which is almost similar in magnitude to the most preferred specification in Table 1. Finally, I exclude districts with the highest reported incidents of forest fires from the empirical sample to explore the robustness of the baseline DID estimate. This is important because Nepal experiences forest fires during the dry season from November to June every year. Furthermore, forest fires can directly influence residential property values (Paudel 2022b). To ensure that other environmental disasters do not confound the results presented here, I exclude districts with reported cases of forest fire events from the sample. Column (4) shows that the coefficient of the interaction term is still negative and statistically significant, suggesting that the effect of seismic shocks from the 2015 earthquake is associated with a 37.55 percentage point decline in residential property values. Overall, these results show that baseline DID estimates reported in Table 1 are robust across different sub-samples.

I also adopt an event study approach to estimate a flexible DID model for residential property values. This approach examines whether an increase in residential property values declines over time as individuals realize that the recurrence probability of a large magnitude earthquake is small (Kreps 1984; Deng, Gan, and Hernandez 2015). Figure 1 plots the coefficients and the confidence intervals estimated from the event study regression on residential property values. This confirms that the coefficients denoting the differences between treated and control districts before the incidence of the earthquake are all non-significant and close to zero, suggesting that is no obvious trend in the pre-earthquake period. Figure 1 also shows that the effect of seismic shocks on residential property values is negative and statistically significant four months after the incidence of the earthquake. The negative effect appears to be statistically significant and pronounced between 12 and 24 months after the 2015 earthquake. Overall, the coefficients from the event study analysis support the baseline DID regression estimates in Table 1.

Finally, I investigate whether changes in residential property values in response to seismic shocks differ between property characteristics in the empirical sample. For example, Hidano et al. (2015) report that older buildings are more vulnerable to seismic hazards and show that prices of newly constructed properties with stringent building code regulations do not change significantly in response to information on seismic hazard risk. To the extent that a decline in property values indicates larger damage from seismic shocks, I expect that households living in residential properties with weaker foundations in treated districts are likely to report a significant decrease in property values. To explore this further, I estimate the baseline DID specification in Equation (1) across housing characteristics belonging to six different sub-samples: homes that have outer walls with and without cemented bricks, homes that have roof materials with and without concrete, and homes that have foundations with and without pillar. Table 3 shows that earthquake-induced changes in property values are negative and statistically significant among homes with weaker foundations for outer walls and roofs. For example, Columns (2) and (4) illustrate that property values in treated districts among homes with no cemented bricks in the outer walls and no concrete in the roof materials decline by 44.25 and 42.36 percentage points, respectively. On the other hand, the effect of the earthquake on property values among homes with cemented bricks in the outer walls and concrete in the roof materials is positive and statistically insignificant. This analysis provides strong suggestive evidence that decline in property values in the aftermath of the 2015 earthquake likely reflects physical damage from seismic shocks.

4.2 Robustness checks

4.2.1 Parallel trends

The identifying assumption of the DID model in Equation (3) requires that residential property values in both treated and control districts evolve in a parallel way in the absence of the earthquake. I formally test the parallel paths assumption in Supplementary Appendix Table A3 by checking for differences in pre-earthquake trends of residential property values between treated and control districts. Across all the columns, the coefficient on TreatedXMonths measures the pre-earthquake monthly trend in property values among treated districts compared to control districts. In all three specifications, the coefficient on the interaction term is small and not statistically significant, showing that the trends in residential property values are not statistically different between treated and control districts before the earthquake. I therefore fail to reject the null hypothesis of parallel trends.

4.2.2 Compositional changes

It is possible that using data from different periods may be potentially biased by compositional changes in treated districts. To the extent that individuals move in response to the earthquake, the estimated parameters may be biased. To address this issue, I make use of district-level population for each year and run double differences specification to evaluate changes in population in treated districts compared to control counterparts after the incidence of the earthquake. Supplementary Appendix Table A4 shows that the interaction term is statistically insignificant, implying that compositional changes in population are unlikely to contaminate estimated parameters in (1). Supplementary Appendix Figure A4 provides additional evidence and shows that the number of sampled individual households is not statistically different between treated and control districts before and after the incidence of the earthquake. Relatedly, Shakya, Basnet, and Paudel (2022) suggest that individuals exposed to the 2015 earthquake are less likely to migrate for labor-related purposes and reside in their areas to help families during the reconstruction and rebuilding phase. These findings provide further strengthen the validity of baseline treatment effect estimates in Table 1.

4.2.3 Mortality selection

It is possible that relative differences in survival rates of a specific gender after a large earthquake may possibly bias the treatment effect estimates. This is relevant because Caruso and Miller (2015) demonstrate that natural disasters can be worse for females, who completed almost 0.8 grade less schooling after the 1970 Ancash Earthquake. In the context of this study, if male household heads are less likely to survive in response to large seismic shocks, it is possible that those who survived seismic shocks may report worse economic outcomes. The same argument holds true for individuals belonging to different household sizes. To the extent that demographic outcomes change in response to a large earthquake, it is possible that mortality selection may confound the true treatment effect estimates.

Supplementary Appendix Table A5 examines whether demographic changes in gender and household size alter in response to seismic shocks from the 2015 earthquake. I estimate baseline DID specification of Equation (1) with two separate outcome variables: binary indicator for male in the first three columns, and household size in the last three columns. Across all six different specifications, none of the slope coefficients of interest are statistically significant. In particular, the coefficient on the interaction term indicates that there is no statistically significant difference in gender and household size in treated districts compared to control counterparts after the 2015 earthquake. These estimates rule out the possibility that mortality selection may confound the effect of seismic shocks on residential property values.

4.2.4 Historical earthquakes

Previous literature highlights the human capital repercussions of the 1988 earthquake³ in Nepal (Paudel and Ryu 2018). The sample of this study includes individuals from districts that saw at least ten deaths during the 1988 earthquake: Bhojpur, Dhankuta, Ilam, Khotang, Morang, Panchthar, Saptari, Sindhuli, Sankhushawa, Sunsari, Terhathum, and Udayapur. The 1988 earthquake may have induced potential unobserved long-term economic outcomes through aid-financed efforts in building better infrastructure. This may, in turn, contaminate the validity of treatment effect estimates presented here.

Table 4 presents results from baseline DID specification in a restricted sample that excludes districts with at least ten deaths from the 1988 earthquake. Controlling for demographic and housing characteristics, district-specific monthly time trends, and a suite of fixed effects, I find that the estimates on the effect of the 2015 earthquake on residential property values are robust across different specifications. The most preferred specification in Column (3) shows that the coefficient of the interaction term is –0.4326, implying that residential property values in areas exposed to large seismic shocks decline by 35.1180 percentage points after the incidence of the 2015 earthquake. This provides evidence that results in Table 1 are free of unobserved long-term effects of the 1988 earthquake in Nepal.

4.2.5 Placebo tests

To strengthen the validity of my identifying assumption, I conduct two different types of placebo tests across space and time. The first test involves simulating false locations of the 2015 earthquake and running the baseline DID specification among individuals from simulated locations. I select districts from the control group where the 2015 earthquake never occurred and create a false treatment binary indicator of the earthquake incidence. The second test involves creating a false time period of the earthquake incidence and running the baseline DID specification among individuals from pre-treatment period. I restrict the sample to the first eleven months before the 2015 earthquake and create a false binary indicator of earthquake incidence period. To the extent that seismic shocks truly influence residential property values, I expect results from these two types of placebo tests to yield statistically insignificant estimates.

The first three columns in Supplementary Appendix Table A6 present results from the first placebo test across space. Across all three columns, the coefficients on the interaction term are negative and statistically insignificant. Similarly, the last three columns in Supplementary Appendix Table A6 show findings from the second placebo test across time. In the most preferred specification, the coefficient on the interaction term is 0.3705 and statistically insignificant. Findings from Supplementary Appendix Table A6 strengthen the validity of the main identifying assumption used in the study.

4.2.6 Alternate treatment indicators and methodology

I note that the binary treatment indicator for seismic shocks, $Treatedd$⁠, in Equation (3) is based on a median PGA threshold of 0.04. To ensure that results from Equation (3) are not sensitive to the threshold used in generating the binary treatment indicator, I show that primary findings in Table 1 are robust to alternate indicators of measuring the seismic shock from the 2015 earthquake. First, I employ a continuous measure of PGA to define the treatment of earthquake in the main empirical specification. This involves replacing a binary indicator of seismic shock with PGA in the baseline DID specification provided in Equation (3). Column (1) of Table 5 shows that the coefficient of the double interaction term is strongly negative and statistically significant, suggesting that the decline in residential property values in treated districts increases more in absolute value in response to an increase in PGA. Specifically, a standard deviation increase in PGA from the 2015 earthquake corresponds to a sizable slope coefficient of –0.76, implying a decline of 53.45 percentage points among residential property values in treated districts. This estimate is consistent with baseline findings from Table 1, which suggests that as the intensity of seismic shocks increases in magnitude, the negative impact on residential property values increases more in absolute value.

Second, I combine the location of the epicenter of the 2015 earthquake with village-level geographical coordinates to construct a distance-based treatment of earthquake incidence. Due to unavailability of longitude and latitude information for each household in the sample, I construct distance between a village-level centroid and the epicenter of the 2015 earthquake. This allows me to estimate baseline DID Equation (3) with a continuous treatment measure of proximity to the epicenter. The coefficient on the interaction term in Column (2) of Table 5, –0.10, shows that a standard deviation increase in proximity to the epicenter of the earthquake induces a 9.71 percentage point decrease in residential property values. This confirms that the negative effect of the seismic shocks on residential property values is larger in absolute value among treated areas that are closer to the epicenter of the 2015 earthquake.

Finally, I employ a “synthetic difference in differences” estimator presented in Arkhangelsky et al. (2021) to explore the robustness of my baseline findings from Table 1. Because this estimation requires balanced panel data, I aggregate my empirical sample to the district level. While this process leads to a reduced sample size, the estimated parameter is similar in magnitude and direction. For example, Columns (1) and (2) of Supplementary Appendix Table A8 show that the average treatment effect estimates are roughly 0.53, which corresponds to a 41.32 percentage point decline in self-assessed residential property values in the aftermath of the earthquake. It is reassuring to see that this estimated impact is slightly larger than the baseline estimates of 40.52 percentage points in Table 1.

4.3 Discussion

4.3.1 Comparison to existing studies

This article is broadly related to a large number of studies exploring the economic impact of earthquake risks on housing prices in urban areas from the USA, Japan, and China. For example, Naoi, Seko, and Sumita (2009) show that post-quake changes in the effect of earthquake risk probability are significantly negative in Japan, inducing a 13 per cent reduction in housing values approximately equal to a discount of 3.8 million yen. They further argue that households are likely to underestimate earthquake risk especially when no recent seismic shocks are reported. Relatedly, Hidano et al. (2015) use individual residential apartment transactions in 23 wards of Tokyo between 2008 and 2012 to show that the unit prices of units in low-risk zones were about 13,970–17,380 JPY higher than those in high-risk zones. In a different study, Deng, Gan, and Hernandez (2015) explore the 2008 Wenchuan earthquake in China to show that relative prices of units located in the first and second floor compared to high floor units increased significantly after the earthquake before returning to pre-earthquake levels after 360 days.

In the USA, Bernknopf, Brookshire, and Thayer (1990) investigate earthquake and volcano hazard notices issued for the Mammoth Lakes, California, and document a perceived loss in market value of homes. Beron et al. (1997) use real estate sales transactions from the six-county San Francisco Bay area and show that hedonic price fell between 26 per cent and 35 per cent after the 1989 Loma Prieta earthquake. According to Beron et al. (1997), expending more resources to assess and disseminate possible losses from earthquake induces gains in welfare. More recently, Fekrazad (2019) investigates the hypothesis that earthquake-risk salience increases in a housing market in response to the news of out-of-the-market earthquakes, and documents that home value index and median listing price decrease by 6 per cent and 3 per cent in California after the incidence of high-casualty earthquakes outside of California. In a different study, Singh (2019) exploits revisions of earthquake fault maps in California over time to show that average property values decline by 6.6 per cent after the delineation of the fault zone. Similarly, Ferreira, Liu, and Brewer (2018) exploit the timing of earthquakes and the distance of properties to injection wells to conclude that large human-induced earthquakes (of magnitude larger than 4) worsen perceived risk of groundwater contamination, estimated at 12.4 per cent of the price of the average home in Oklahoma. Bin and Landry (2013) use multiple storm events within a DID framework and show a significant risk premium ranging between 6 per cent and 20 per cent for home sold in North Carolina in the flood zone, with the documented effect diminishing over time.

4.3.2 Policy implications

In this article, I find that residential property values declined by 40.52 percentage points in districts that experienced large seismic shocks from the 2015 earthquake. This represents a $7,124 decrease in average residential property values. Based on the empirical sample, each district in Nepal, on an average, comprises of 54,112 residential properties. Even if 20 per cent of these properties are prone to large seismic shocks from the earthquake, approximately 10822 homes would be affected. The number of affected homes, combined with an average decline in property values, results in an external cost of $71.2 million. This back-of-the-envelope calculation represents a lower bound to the true external cost of a large earthquake. It is beyond the scope of this study to directly account for consequences of the 2015 earthquake on loss of public infrastructure and changes in health, education and labor market outcomes.

From a policy perspective, a comprehensive understanding of how risks of natural disasters are capitalized into residential property values is important. To the extent that households use information on earthquake risk and update their risk perceptions over time, areas with low seismic risk experience higher housing prices compared to areas with high seismic risk (Singh 2019). Some alternate policy solutions include incentivizing households to buy earthquake insurance⁴ that covers home repairs and property damage from the earthquake, promoting tax deductions to upgrade dwellings, and providing seismic hazard maps for estimates of potential ground shaking and structural damage. Findings from this article show that the impact of seismic shocks on residential property values is negative and primarily driven by homes with weaker foundations. This implies that the introduction of policies aimed at replacing homes comprising of inadequate foundations in areas of high seismic risk with new ones under stringent building code regulations would substantially minimize future economic damage from earthquakes.

5. Concluding remarks

This article examines the short-term economic impact of a major earthquake on residential property values in a developing country setting. Exploiting information on plausibly exogenous PGA, I employ a quasi-experimental research design using the DID framework to show that self-assessed residential property values declined by 40.52 percentage points in districts with high-intensity seismic activity during the 2015 earthquake. Results from an event study specification indicate that the negative effect on residential property values lasted for 12–24 months after the incidence of the earthquake. The heterogeneity in earthquake-induced changes in property values across different home characteristics indicates that a decline in residential property values reflects increased damage from large earthquakes.

Two limitations of the article are worth highlighting. First, the methodology relies on self-reported residential property values collected from household surveys. A comparison of my estimates with those derived from residential market transactions will contribute to an estimation of potential bias when quantifying the economic costs of natural disasters in a developing country setting. Second, the study does not shed light on potential channels that may help mitigate a portion of earthquake-induced damages among affected households. Future research may further explore the channels of aid and migration and their interaction with economic damages from seismic shocks.

Among different risk-mitigating mechanisms, aid-financed efforts may play a key role in partially mediating the negative effect of seismic shocks on residential property values. This is important because provision of aid has the potential to mitigate the negative effect of natural disasters on economic outcomes among poorer households in a developing country setting (Andrabi and Das 2017; Eichenauer et al. 2020). While an investigation on efficacy of aid is beyond the scope of this study, an influx of studies has highlighted prospects of leakage in aid distribution across different levels of governance in the context of Nepal (Paudel 2023a). Specifically, aid allocation could be associated with little regard for the specific socioeconomic and physical vulnerabilities, inducing difficulties in disaster recovery among victims (Eichenauer et al. 2020; Spoon et al. 2020). According to Angeles and Neanidis (2009), the probability of aid misuse increases if the elite with access to economic and political power shows little concern for other social groups. These results call for an effective implementation of risk-mitigating mechanisms to account for physical damage from future seismic shocks in a developing country setting.

Footnotes

Supplementary material

Supplementary material is available at the Oxford Economic Papers Journal online. These are the datasets, replication files, and the online appendix.

Funding

This research did not receive any specific grant from funding agencies in the public, commercial, or not-for-profit sectors.

Conflict of interest statement. The author does not have any affiliation with or involvement in any organization or entity with financial or non-financial interests in the subject matter or materials discussed in this manuscript.

Acknowledgements

The author thanks Francis Teal and two anonymous referees for helpful comments in the review process.

References