key: cord-0937431-ib26nqjz
authors: Coulson, N. Edward; McCoy, Shawn J.; McDonough, Ian K.
title: Economic Diversification and The Resiliency Hypothesis: Evidence from the Impact of Natural Disasters on Regional Housing Values
date: 2020-08-30
journal: Reg Sci Urban Econ
DOI: 10.1016/j.regsciurbeco.2020.103581
sha: d8242cdb172661b354d026d17196095720e4b7c0
doc_id: 937431
cord_uid: ib26nqjz

We estimate the effect regional economic diversification has on the resiliency of the U.S. housing market treating the spatial and temporal variation in natural disasters as exogenous shocks to regional economies. Our study demonstrates that diversity dampens both the magnitude and the duration of the effects of a disaster on local real estate values. Implications of our findings for the potential benefits of diversification in regional economies are discussed.

Volatility in local housing markets is the result of many factors. Prominent among these are of course speculative demand (Goodman and Thibodeau, 2008) , but the condition of the local economy perhaps matters even more. For instance, the extraordinary decline in Las Vegas home prices may have had its origin in the bursting of the housing bubble, but the momentum of that decline was no doubt exacerbated by the national recession that followed, which particularly took its toll on tourism, upon which the Las Vegas economy particularly thrives. The lack of diversity in Las Vegas' economic base created singular difficulty for its economic recovery. This trend continues. In their analysis of cities most likely to be negatively impacted by a coronavirus-induced recession 1 , authors at the Brookings Institution (Munro et al., 2020) single out Las Vegas as particularly vulnerable (once again) for its relative lack of diversity in its industrial base.

It has been long argued that economic diversity plays a key role in promoting both regional economic growth as well as regional economic stability (Parr, 1965; Kort, 1981; Siegel et al., 1994; Siegel et al., 1995; Attaran, 1986; Wagner and Deller, 1988) .

Commonly, stability is defined in terms of the "absence of variation in economic activity over time" (Malizia and Ke, 1993 p. 222) . These concepts are related to the concept of resiliency. Formally defined, resilience refers to "the ability or capacity of a system to absorb or cushion against damage or loss" (Rose and Liao, 2005 p. 78; Holling, 1973) .

For local economies it has been defined as, "the ability of a regional economy to maintain or return to a pre-existing state (typically assumed to be an equilibrium state) in the presence of some type of plausibly exogenous (i.e. externally generated) shock" (Hill et al., 2012 ; see also Martin and Sunley, 2015) .

The purpose of this paper is to therefore document the link between diversity of the local economy and the resiliency of housing prices. Our focus is on housing prices due to their importance in measuring the health of the local economy (Hwang and Quigley, 2006) ; their manifestation of the area's quality of life (Albouy, 2008) ; and perhaps most important, their reflection of the future path of both of those factors.

The fundamental empirical challenge to identifying the link between diversity and 1 As we write this in May 2020, the recession is just now unfolding. resilience is that regional housing market downturns are rarely ever exogenous to local labor market conditions. Economic diversification is no exception. To circumvent this difficulty, we investigate the link between diversity and resilience by estimating the effect economic diversification has on explaining real estate price dynamics to natural disasters.

Natural disasters provide a useful context for studying the effect of diversification on resiliency for several reasons. First, the exogenous nature of disasters gives us a unique setting to study housing market responses to shocks net of concerns stemming from potential endogeneity between real estate market performance and regional business cycles. Second, with the frequency and severity of natural disasters on the rise, studying the economic impacts of disasters is an economically meaningful pursuit in its own right.

Through this lens, our paper contributes to broader research efforts in the environmental economics literature on climate change and human adaption by adding a complementary discussion centered on steps local policymakers might take to improve the economic resilience of their locality to climate-induced shocks.

We focus our empirical work on housing market responses to hurricanes and typhoons using a panel dataset of purchase-only house price indices at the MSA level that are maintained by the Federal Housing Finance Agency (FHFA). We link these data to FEMA's National Emergency Management System (NEMIS) which indicates the month, day, year and impacted MSA for the universe of federally declared disasters. Finally, we compute the usual measure of economic diversification for each MSA -a fractionalization index of labor market income across NAICS supersectors -using industry level data from the Quarterly Census of Employment and Wages (QCEW).

We estimate the impact of a disaster using a difference-in-differences (DID) approach. This method allows us to identify the average price effect due to a shock by estimating changes in home prices before and after a disaster hits impacted MSAs relative to home price dynamics across non-impacted MSAs. To test the hypothesis that regional economic diversification is a catalyst for resiliency, we estimate the effect that regional economic diversity has on attenuating the impacts of natural disasters on local home prices. Estimating the causal effect of diversification on growth presents a set of challenges that originate from the presence of endogeneity stemming from latent J o u r n a l P r e -p r o o f confounders that are both correlated with the outcome of interest and diversity.

However, and drawing on a recent study by Nizalova and Murtazashvili (2016) , we can recover the causal effect of diversity on resilience if we are willing to assume that both regional economic diversity and any potential latent confounders are jointly independent of the occurrence of natural disasters conditional on time-invariant geographic fixed effects.

To preface our main findings, our empirical results show that the impact of a disaster depends both on the level of diversity and the time elapsed since a shock. Highly concentrated regions experience price declines as large as -4.7% in the year immediately following a disaster. These initial impacts persist for as long as two years. Economic diversity has the effect of dampening the immediate price response due to a shock as well as the persistence of these initial price declines. We estimate that a one standard deviation increase in diversification (relative to the mean level of diversification in the U.S. economy) offsets the immediate (one to two year) price effects of a disaster as much as 1.96% to 2.3%.

We then position our empirical findings into broader discussions in the literature centered on the potential "dual effects" of diversifying a regional economy. Researchers and policymakers often debate the value of diversification in terms of the direct effect of diversification on growth. In part, the tension in the literature exists given the competing views about the role diversity plays in influencing economic growth. For example, some view diversification as a movement away from potential efficiency gains resulting from specialization and, perhaps, mitigating economic growth (see e.g. Izraeli and Murphy, 2003) . Others argue that diversification moves the economy towards an environment where knowledge spillovers can occur between industries, thus catalyzing economic growth (see e.g. Glaeser et al., 1992) . Further complicating this matter, there is a fundamental empirical challenge to estimating the direct effect of diversification on market outcomes. To identify a causal link, one would have to acknowledge the possibility that unobserved determinants of the market outcome of interest may also be correlated with diversity. Motivated by this observation, we build off the earlier work of Bartik (1991 ), Card (2001 ), and Ottoviano and Perri (2006 and propose an instrumental variables estimation strategy capable of controlling for this level of endogeneity. We then J o u r n a l P r e -p r o o f show that the benefits of diversification expressed in terms of resiliency do not appear to be offset by any potential costs stemming from a corresponding departure from industrial specialization.

As we explain in more depth below, we draw upon several literatures to motivate the theoretical mechanisms we have in mind. The literature has previously established a connection between city-level labor outcomes and the diversity of the labor force, largely focusing on standard insights from portfolio theory that a diverse array of industries will suffer less from shocks that have sector-specific effects. 2 In addition, the literature has discussed the connections between housing and labor markets in a city. As such, resiliency is a natural outcome to examine, in that the recovery from a (negative) shock would be quicker when there is an industrial base that is diverse.

It has also been recognized that migration (even only temporary) in response to a shock could influence local labor markets and house price dynamics (Boustan et al., 2020) . As such, and if after a disaster migration tends to occur from concentrated to more diverse areas, migratory dynamics could represent an additional mechanism explaining why diversity drives resiliency. Using data on county-to-county migration flows we formally test this hypothesis. Like Boustan et al. (2020) , we first show that natural disasters lead to statistically significant increases in out-migration and net-migration.

Additionally, using a difference-in-differences estimation framework we also find that conditional on experiencing a disaster, more diverse economies experience a smaller degree of outmigration and net-migration. Viewed through the lens where relative demand for housing falls more in areas with concentrated employment, the migration channel is perhaps one important explanation of our results.

We proceed by providing a background on related works in Section 2. We summarize our study area and data in Section 3. We present our empirical methodology in Section 4 and our findings in Section 5. In Section 6 we discuss potential threats to the identifying assumptions of our model. In Section 7 we utilize an instrumental variables approach to estimate the direct effect of diversification on housing values and summarize and conclude in Section 8.

2 See, e.g., Barth et al. (2015) , Malizia and Ke (1993) , Izraeli and Murphy (2003), Siegel et al. (1998) , Hammond and Thompson (2004), and Wagner and Deller (1998) . A full description of this literature is provided in Section (2). The analysis brings together several literatures that deal with the various connections between employment diversity, labor market performance, real estate prices, and the economic impact of natural disasters. We take up the connections between various pairs of these concepts in order to provide context for this analysis and the hypotheses that we wish to explore. These include the relationship between (1) diversity and volatility, the effect diversity may have on (2) economic performance and (3) resilience to economic shocks, (4) the links between home prices, diversity, and economic fundamentals, and (5) the impact of natural disasters on real estate markets.

The argument that industrial diversification may lead to reduced volatility and resilience in metropolitan economies is long-standing. Barth et al. (2015) note that, as would be suggested by standard portfolio theory, a diversified portfolio of industry employment yields lower overall volatility in metro employment. The emphasis on the use of portfolio theory as a lens through which to view employment volatility led to the insight that what mattered was not simply diversity as such, but the covariances of sectoral employments. Diversity is simply a means to tamp down the effect of these covariances on aggregate employment variability. The portfolio approach was pursued in much of the subsequent literature, including Malizia and Ke (1993) and Izraeli and Murphy (2003) with respect to unemployment changes and Siegel et al. (1994) and Wagner and Deller (1998) in the context of an input-output model. Hammond and Thompson (2004) also note that greater industrial specialization yields greater employment volatility but also emphasized the role of local demographic characteristics.

Interestingly, Carvalho (2014) notes the idea that disaggregating the economy into smaller sectors will serve to dampen the overall effects of a disturbance to any one sector was perpetuated by the earlier work of Lucas (1977, page 20) who writes:

"In a complex modern economy, there will be a large number of such shifts in any given period, each small in importance relative to total output. There will be much 'averaging out' of such effects across markets." J o u r n a l P r e -p r o o f Inspired by the lessons of the 2011 earthquake in Japan, Carvalho (2014) provides a new and more sophisticated perspective on the role of diversification in the national economy by advancing a multisector general equilibrium model. The key insight of Carvalho (2014) is that whether or not diversity catalyzes resiliency may ultimately depend on the complexity of the input-output linkages between sectors in an economy. In effect, Carvalho (2014) argues that cyclical fluctuations may arise from small shocks working their way through or across input linkages. In a diversified but horizontal economy, disaggregation leads to decreases in aggregate volatility. In contrast, once Carvalho (2014) relaxes the assumption that intermediate producers work in isolation from each other, shocks to one sector may propagate through other sectors. In this paper we abstract away from modeling input-output linkages across sectors within our small regional economies, but instead focus our efforts on modeling the degree to which labor market activity is fractionalized across sectors.

There exists a separate debate on the role that diversity plays in promoting productivity and growth. Empirical analyses by Frenken et al. (2007) notes that a broad variety of industries increases the possibility of Jacobs (inter-industry) productivity spillovers (Glaeser et al. 1992) . Note, however, that our focus on resilience to shocks has little to do with productivity spillovers and much more to do with the ability of a broader based economy to handle stress. Reliance on a small set of industries for economic health can be detrimental when shocks are specific to those sectors.

More closely related to our research here is work that deals with the ability of local economies to absorb shocks, and the extent to which diversity aids in the return to the previous steady state. While not concerned with natural disasters, Hill et al. (2012) investigate drivers of resilience with quantitative case studies of metropolitan areas and show that diversity may attenuate an economic downturn. Similarly, Feyrer et al. (2007) ask whether employment diversity helped cities recover from the impacts of the "Rust (2013) show that more diverse counties witnessed relatively larger increases in employment following the flood than less diverse counties; a finding which is consistent with the hypothesis that diversity is a driver of economic resiliency.

It seems almost a truism that local home prices should depend on local economic conditions, but this truism has substantial empirical support (Hwang and Quigley, 2006; Gallin, 2006; Zabel, 2012) . Local income and unemployment rates are standard determinants of local home prices.

Beyond these standard determinants, and recognizing the previous work that relates diversity and economic fundamentals discussed above, Coulson et al. (2013) provide the first empirical evidence demonstrating that increases in economic diversification effectively leads to decreases in home price volatility. In a related work, Barth et al. (2015) show that home prices in Metropolitan Statistical Areas (MSAs) starting with a relatively lower degree of diversification tend to rise as these MSAs become less diversified.

Our study contributes to various strands of the environmental economics literature centered on real estate market responses to natural disasters (Harrison et al., 2001; Bin and Polasky, 2004; Hallstrom and Smith, 2005 Importantly, none of these studies is concerned with the recovery period as such, and do not examine the dynamics of real estate prices in the wake of the disaster, much less the economic factors that might speed such recovery. But it is the speed of such recoveries that are inherent in the discussion of "resilience". Note that those dynamics are not predetermined by theory. A natural disaster can destroy much of the existing housing stock, shifting the supply curve to the left, but at the same time cause both labor and firms to exit as well, which would cause a simultaneous shift of the demand for housing to the left, leaving the effect on housing price ambiguous. In the event we see that the initial path of home prices after a disaster is in the negative direction, the demand effect outweighs the supply effect. It is then of some interest to ask what factors can return home prices more quickly to the previous steady state, and our above discussion clearly suggests that economic diversity is one of those factors. The literature on diversity and local economies, along with the smaller body of work on diversity and home prices discussed above suggest that a diverse portfolio of industries will mitigate the particular impact that a disaster has on any one sector, leading to quicker recovery overall. The above example of Las Vegas is evocative of this, but more salient to our use of natural disasters would be a coastal economy reliant on tourism suffering more and for a longer period of time if a disaster has an asymmetric impact on coastally located tourist services (Kim and Marcouillier, 2015) .

Overall, our summary of the literatures above suggests that the value of a diverse portfolio of industries can ameliorate the effects of negative economic shocks. We measure that amelioration through the impact of a disaster on housing prices, and whether the return to the steady state is quickened by greater diversity. We find that it does, and in that sense, conclude that such areas are more resilient.

The primary dataset utilized in this paper is the quarterly, purchase-only house price In order to construct a measure of economic diversity, we obtain industry level wage data from the Bureau of Labor Statistics' Quarterly Census of Employment and Wages (QCEW). 5 Of particular interest to us are total wages, which track total compensation paid during the calendar quarter to employees within each NAICS supersector of each U.S. county. 6 NAICS supersectors, which are synonymous with two digit NAICS codes, represent the twenty, top-level industry groupings in the United States. 7 We measure economic diversity by first aggregating total wages within each NAICS supersector, , across counties residing within the same MSA, , at each year-quarter time-step, . We employ the usual measure of economic diversification that is based on the fractionalization index of labor market income across NAICS supersectors,

where ℎ denotes the share of labor market income for industry within MSA at time . The index is further standardized to be mean zero with a standard deviation equal to one. Thus higher values in the index represent more diverse regions and lower values in the index represent less diverse regions. To make clear how the index is interpreted, a hypothetical value of plus two in the diversity index would represent an area that is two standard deviations above the mean level of diversity, and a hypothetical value of minus two would represent an area that is two standard deviations below the mean level of diversity.

We treat the universe of federally declared hurricanes and typhoons as exogenous shocks to local real estate markets. Data describing these events are maintained by indicates the impacted county and records the month, day, and year each disaster began.

We use the County to MSA crosswalk provided by the United States Census Bureau to map impacted counties into impacted MSAs. Using these data, we can effectively identify the entire history of natural disasters impacting each MSA in our study area.

We provide a graphical illustration of the 100 MSAs included in our sample in Figure   7 The set of NAICS supersectors includes: 1. To visualize differences in the degree of diversification across MSAs, Figure 1 illustrates the average level diversification within each MSA over the study period, again, standardized to have a mean of zero and standard deviation equal to one. Henceforth, MSAs in dark green represent the least diversified regions (e.g. those with long-run average diversity levels 3.95 to 2.14 standard deviations below the mean MSA).

Likewise, MSAs in bright red represent the most diversified regions; those with long-run average diversity levels .98 to 1.45 standard deviations above the mean MSA. Lastly, we provide a list of every MSA in our sample in appendix Table A1 Despite these overall trends in diversity, we learn that there exists an immense amount of heterogeneity in the trajectory of our small regional economies. To further illustrate the variation in the data, we compute the growth rate, , , in economic 

To study the impact of regional growth shocks on residential housing prices, we estimate a difference-in-differences model exploiting the random nature of regional disasters.

More specifically, we employ the fixed effects estimator, ln &'( = ) *+ , • . , + 0 , • . , × 2(3 4 5 ,6 5 + 7 • 2(3 + 8 + 9 + : , # 2

where ln &'( is the log transformed housing price index for MSA in time , 2(3 is the fractionalization index for regional economic activity for MSA in time , and . , is a treatment-indicator < years prior to or after a disaster-related event. To fix ideas, .

is equal to one if at time MSA is within 1 to 2 years of a disaster. Likewise, .

is equal to one if at time t MSA is within -1 to -2 years of a disaster.

Additionally, 8 captures MSA-specific, time-invariant unobserved heterogeneity.

Finally, 9 includes an exhaustive set of year-quarter fixed effects and MSA specific linear time trends. Note, the causal interpretation of + , and 0 , stems from the assumption that whether or not a region is hit by a natural disaster is random conditional on MSA and

year-quarter fixed effects.

We specify equation (2) with a series of three pre-disaster and three post-disaster event indicators and, similar to Gallagher (2014), bin each . , for any time period < < −3 and for any time period < > 3 by creating single event indicator variables for the end periods of the event study, . 5 and . 5 . The inclusion of . 5 and . 5 simply serve the practical purpose of allowing us to study how home prices evolve in the years shortly after a disaster (e.g. < ∈ [1,3] as well as in the years shortly before a disaster; we normalize all coefficient estimates relative to the year immediately before a J o u r n a l P r e -p r o o f disaster (< = −1).

The classical difference-in-differences estimator is typically operationalized by excluding the interaction terms . , × 2(3 from the model. In this case the researcher relies on estimates of + , to identify the average impact of the event of interest. However, the inclusion of . , × 2(3 allows us to estimate how the average effect of a disaster is influenced by economic diversity. Letting 2(3 D denote a particular value of 2(3 , the relevant parameter of interest to us is,

For the sake of clarity, with 2(3 D set at the mean G HIJ , parameter estimates of E <, G HIJ = + , + 0 , • G HIJ represent the average impact of a disaster in the < K year following a shock. Thus, the interpretation of E <, G HIJ parallels what is typically reported in related works focused more exclusively on estimating the average impact associated with an event.

In our empirical work we first present parameter estimates of E <, 2(3 D evaluated across the distribution of 2(3 . We use the superscript "j" to refer to the "j th " percentile of 2(3 . This approach also allows us to evaluate the economic significance of diversification on resiliency by reporting the magnitude of the estimated impact of a shock at different values of diversity, L and L′ (e.g. E <, 2(3 D vs. E <, 2(3 DN ). We formally test the hypothesis that economic diversification is a catalyst for resiliency by examining whether or not the impact of a disaster on home prices is affected by changes in diversity. More precisely, if economic diversity catalyzes price resiliency to disasters then,

Finally, one of the underlying identifying assumptions of our model is that the average change in home values across impacted MSAs would have been proportional to the average change in prices in non-impacted MSAs in the absence of treatment. While we cannot directly test whether or not this assumption holds, we provide supporting evidence of parallel trends by investigating estimates of E <, 2(3 D in the periods leading up to a disaster (e.g. < < −1). For instance, focusing on the 5 th percentile of diversity, model estimates for E −3, 2(3 R and E −2, 2(3 R are small in magnitude and statistically insignificant. Turning attention to columns 2 through 7, parameter estimates for E"−3, 2(3 D # and E"−2, 2(3 D # are also statistically insignificant and close to zero in magnitude. Independent of the level of diversification, we find no statistical evidence suggesting that home price trends among impacted regions differ from home price trends in non-impacted regions in the years leading up to a shock; an empirical finding that lends credence to the underlying identifying assumption of the model.

Next, we turn our attention to coefficient estimates of the post-disaster treatment indicators. As indicated in column 1, at the 5 th percentile of diversity, we estimate that disasters induced a statistically significant reduction in housing prices of 4.8% and 5.2%

in the first two years following a shock, respectively. After two years, we do not detect a statistically significant impact of a disaster on housing prices, which suggests that the immediate market impacts of a disaster are economically relevant but nonetheless transitory.

Next, we focus our attention to parameter estimates of E"+1, 2(3 D #, which capture the immediate, first year impact of a shock. Table 1 shows that as we move from the 5 th to the 10 th percentile of diversity, the first-year impact of a disaster decreases in magnitude (in absolute value) from -4.8% to -2.8%. Estimates further decline as we move J o u r n a l P r e -p r o o f to the 25 th percentile but remain statistically significant. At the average level of diversity in the data (column 4), model estimates indicate a 0.8% reduction in home prices.

Coefficient estimates reported in columns 2 and 3, which reveal the market impacts of a hurricane at the 10 th and 25 th percentiles of diversity, are not suggestive of a statistically meaningful reduction in prices in the second year following a shock. In contrast, we do estimate a statistically significant price effect two years after a shock when evaluated at the 5 th percentile of diversity. We visualize these findings in Figure 5 . Specifically, panel (a) of Figure 5 plots coefficient estimates on the y-axis against years since a shock on the x-axis for non-diversified MSAs. Panel (b) plots coefficient estimates for the diversified MSAs.

[ Figure 5 : About Here]

Note that in the baseline log-linear specification described in equation (2) Table A2 are qualitatively similar to the estimates reported in Table 1 .

We also consider the following variant of estimating equation (2) Note, we superscript all model parameters to indicate we are estimating a different model. Given estimates of equation (5), the price effect of a disaster < years after a disaster impacts a region expressed as a function of diversity is, E′ <, 2(3 = +′ , + 0′ , • ln 2(3 . # 6

Parameter estimates of E′ <, 2(3 which we report in Table A3 are also qualitatively J o u r n a l P r e -p r o o f similar to parameter estimates of E <, 2(3 in Table 1. 9 To summarize our main findings, recall that the parameter E <, 2(3 represents the price impact of a disaster in the < K year after a shock conditioning the level of diversity 2(3 . Estimates of E <, 2(3 allow us to evaluate differences in the degree to which housing prices change in response to a disaster at any point in time and at any level of regional diversity. Model results reported in Table 1 show that highly concentrated regions (e.g. those lying below the 25 th percentile of diversity) experience negative and statistically significant price declines in the first two years following a shock. However, as diversity increases, these immediate price responses attenuate towards zero.

Collectively, these findings indicate that diversification attenuates both the magnitude and the duration of the impacts of a disaster on regional housing values.

The empirical findings presented in the preceding section lend credence to the resiliency hypothesis. To the extent that economic diversification attenuates the immediate impact and the persistence of a shock, our estimates suggest that diversification has an economically meaningful impact on an MSAs level of resiliency. Here, we formally test if diversification has a statistically significant effect on resiliency.

We formalize a test of the resiliency hypothesis by first recalling that the estimated price effect of a disaster < years after a disaster hits expressed as a function of diversity is given by, E <, 2(3 = + , + 0 , • 2(3 . # 7

This expression allows us to derive the direct effect that a unit increase in diversity has on attenuating home price responses due to a disaster, OE <, 2(3 O2(3 = 0 , . # 8

Note that 2(3 is bounded above by one. As such, it is useful to consider estimates of Table 2 reports estimates of 0 [ , derived from coefficient estimates of equation (2) as well as p-values associated with hypothesis tests specified in (9). As shown in Table 2, for post-treatment time periods we reject the null hypothesis that 0 [ , ≤ 0 in favor of the alternative hypothesis that 0 [ , > 0. For the sake of completeness, we also report estimates of 0 [ , for each pre-treatment period (e.g. periods -3, -2). In the absence of a shock, our hypothesis that these estimates should not be statistically different from zero is supported by the statistical evidence in Table 2 . 10 [ These findings provide statistical evidence allowing us to reject in the null hypothesis in favor of the resiliency hypothesis. Yet, whether or not these tests are valid ultimately depends on the underlying identifying assumptions of our empirical model. We proceed by discussing the potential threats to the identifying assumptions of our modeling exercise.

In this section, we elaborate in more detail the underlying identifying assumptions of our main empirical model along with a series of robustness checks. We also investigate whether or not migration is a potential mechanism driving our results. 10 For robustness, we also tested the sensitivity of our model estimates to the exclusion of larger known disasters. To do this, we leveraged the Housing Assistance Dataset published by FEMA (https://www.fema.gov/medialibrary/assets/documents/34758). For each disaster, the data document the total number of FEMA applicants (e.g. homeowners) who received an inspection following a major presidential disaster declaration. Again, these inspections are required to qualify for assistance options. The data also record the total number of applications who received an inspection but had no damage recorded by the inspector. We use both variables to compute the number of homes that were (a) inspected and (b) determined to be damaged by each disaster. We then flagged outliers by identifying disasters that fell above the 95th percentile with respect to the number of homes inspected and determined to be damaged by each disaster and obtained estimates of 0 [ , after excluding this set of flagged disasters; our model results are qualitatively unaffected by this data restriction.

A key identifying assumption of our empirical model is that conditional on MSA and time fixed effects, the shocks we introduce to regional economies appear random. This assumption cannot be explicitly tested. Instead, we rely on the conditionally random nature of a natural disaster as one piece of supporting evidence

Another identifying assumption of our empirical model is that non-impacted MSAs serve as valid control groups for impacted MSAs. That is, to interpret our estimates as causal requires one to assume that price trends in impacted MSAs would have been proportional to price trends in non-impacted MSAs in the absence of treatment. While

this assumption cannot be explicitly tested, our empirical findings in section (5) provide evidence supporting it. More specifically, model estimates of equation (2) reported in Table 1 demonstrate that in the period of time leading up to a shock, there does not exist economically or statistically meaningful differences in pre-treatment price trends between impacted and non-impacted regions.

We highlight that economic diversification (2(3 ) appears in our empirical specification both by itself and interacted with a suite of disaster indicators, 2(3 × . , .

As such, one might raise the concern that if economic diversity and housing values are both related to some latent confounder, and since coefficient estimates on . , and 2(3 × . , are both used to test the resiliency hypothesis, model estimates of the impact of diversity on resiliency (e.g. coefficient estimates of the interaction terms) are potentially problematic due to the inconsistency of the estimator. Like the model set forth in Nizalova and Murtazashvili (2016), the econometric problem here is akin to a heterogeneous treatment effect analysis where the source of heterogeneity is perhaps endogenous. Given their theoretical findings, and key to addressing potential endogeneity here, estimating eq. (1) via OLS will yield estimated coefficients on . , and 2(3 × . , that are consistent so long as both regional economic diversity and any latent confounders are jointly independent-having controlled for MSA-specific fixed effectsof the occurrence of natural disasters. Said another way, if disasters are conditionally J o u r n a l P r e -p r o o f independent of both : and 2(3 , again controlling for geography that is fixed over time, then the coefficients on the interaction terms 2(3 × . , can still be consistently estimated whether 2(3 is independent of : , or not. Thus, even if there are other potential factors correlated with both diversity and home prices, the estimated effect on the interaction terms will not be confounded so long as the aforementioned assumptions hold. Henceforth, since we only rely on estimates of the coefficients on 2(3 × . , and . , , the conditionally random nature of disasters allows us to consistently estimate the effect that diversification has on catalyzing resiliency. (2) without the inclusion of any additional controls. With the goal of including variables measuring both local labor market conditions and firm structures, column 3 replicates column 2 but includes a full set of (linear) control variables measuring labor market income by industry. Likewise, column 4 replicates column 3 but also includes a full set of (linear) control variables measuring employment by industry.

The set of parameters estimates in column 2 are qualitatively unaffected by the inclusion of these controls. However, we take an additional step in assessing the robustness of our findings by instrumenting for economic diversity and estimating the model via two stage least squares (2SLS).

Employing an instrumental variable approach requires an instrument(s) that is correlated with changes in diversification in a given MSA, is otherwise exogenous to local economic conditions in said MSA, and is arguably excludable from the structural equation. As such, we construct an instrument by adopting the shift-share methodology used by Ottaviano and Perri (2006) We use 2(3 IJ to instrument for the level of diversification in each MSA. Here, the identifying assumption is that changes in the national growth rate of sector s are exogenous to the local economic conditions of a specific region . However, diversity enters our main estimating equation by itself, 2(3 , and also through the suite of interaction terms, . , × 2(3 . Along these lines, in order to obtain 2SLS estimates of 0 [ , , we first estimate equation (2) instrumenting 2(3 and . , × 2(3 with the set of instruments 2(3 IJ and . , × 2(3 IJ . When looking at the results presented in Table 3 , three things become apparent. First, the estimated effects in the pre-periods remain statistically indistinguishable from zero.

Second, the estimated impacts are similar in sign though slightly smaller in magnitude. In particular, the estimated effect one period out is 1.68% (compared to 1.96%), the estimated effect two periods out is 1.95% (compared to 2.26%), and the estimated effect three periods out is 1.63% (compared to 2.30%). Third, statistical significance at conventional levels still holds for the estimated effects in the first two post periods and the estimated effect is only marginally insignificant the third post period.

The stability of the results when instrumenting for the potential endogenous nature of diversity provides further confidence in our baseline results. In particular, the stability of these results having instrumented for both diversity and the interaction of diversity with the set of natural disaster indicators lends empirical strength to the underlying identifying assumption of disasters appearing random conditional on geography. This conclusion, of course, is only convincing if the exclusion restrictions are valid. The idea that changes in the national growth rate in the share of labor market income for particular sectors is exogenous to the local economic conditions, while at the same time not being a localized determinant of regional home prices, seems reasonable. This argument, however, does preclude the possibility of unobservables that are correlated with both the national growth rate in the share of labor market income from particular sectors and localized home prices conditional on regional economic diversity. Lastly, and as noted in the Table 3 , the null of underidentification is rejected (p-value = 0.00) and the estimated first-stage Kleibergen-Paap F statistic is equal to 12.33.

[ 

Boustan et al. (2020) show that migration in response to disasters could influence local labor markets and house price dynamics. Along these lines, migration may be a potential mechanism explaining our results if post-disaster net migration is in the direction of more J o u r n a l P r e -p r o o f diverse areas thus leading to more resilient home prices in more diverse areas. This logic motivates us to ask several questions: 1) Do natural disasters lead to increases in net outmigration (e.g. out-migration net of in-migration)? 2) Does there exist a tendency for diverse regions to experience more in-migration than out-migration? 3) Following a disaster, does diversity dampen the degree of net out-migration in a region?

We address these questions by obtaining the County-to-County Migration Flows database that is based on the 2011 to 2015 American Community Surveys. 11 The data are constructed based on identifying residents in each U.S. county, and whether or not they lived in the same residence one year ago; for respondents indicating they lived in a different residence the data records the county they currently reside in as well as the county they previously resided in. This information is then used to construct estimates of flow and counter flow migration for all county pairs.

Our unit of analysis is a county to county pairing, (i,j). For each pairing, we measure out-migration from county i to county j denoted by the variable Out-migration ij as well as in-migration to county i from county j denoted by the variable In-migration ij . For the sake of clarity, both of these variables are always greater than or equal to zero. Thus, Outmigration ij measures the number of people that moved from i to j and In-migration ij measures the number of people that moved to i from j. We then construct the variable Net-migration ij = Out-migration ij -In-migration ij . Henceforth, if Net-migration ij >0 (<0) then the number residents that moved out of county i to j is greater (smaller) than the number of residents that moved into county i from county j.

Next, we compute the standardized diversity index at the county level and compute the difference in diversity between counties i and j (DIV i -DIV j ) for every county to county pairing. Finally, we leverage a useful aspect of the timing of disasters in our data. If natural disasters cause increases in Net-migration ij , then we would expect the coefficient estimate of + to be positive; that is, + represents the effect of a disaster in county i on the average level of net migration from county i to all other counties.

Second, note that county i is more diverse than county j when "2(3 − 2(3 D # > 0. Hence, if there is a tendency for diverse regions to experience more in-migration than outmigration -in effect, leading to a reduction in net-migration -we would expect estimates of + to be negative. Lastly, coefficient estimates of + r capture the effect that a relative standard deviation increase in the diversity of region i has on mitigating the average level of net migration from i in response to a disaster. If diversity serves to dampen the net migratory response to a shock, then we would expect coefficient estimates of + r to be negative and statistically significant.

We present results of estimating equation (14) in Table 4 . Here, robust standard errors are clustered at the level of treatment, which in our case is the county level, but model results are robust to not clustering as well. For completeness, columns 1 and 2 show estimates of equation (14) using outmigration from i to j as the dependent variable.

Likewise, columns 3 and 4 show estimates of equation (14) using net migration from i to j as the dependent variable. First, focusing on the model estimate shown in column 1 which excludes the terms "2(3 − 2(3 D # and 2 × "2(3 − 2(3 D #, we learn that disasters lead to a statistically significant increase in outmigration. Turning attention to column 3, we learn that the associated counter flow of residents in response to a disaster does not offset the degree of outmigration; that is, on average, natural disasters lead to a statistically significant (net) increase in the number of residents leaving a region. Our main estimating equation is presented in column 4. Coefficient estimates of "2(3 − 2(3 D # are negative and statistically significant which suggests than on average, residents tend to move to diverse regions. Lastly, estimates of the coefficients of 2 × J o u r n a l P r e -p r o o f "2(3 − 2(3 D # are negative and statistically significant in all models, indicating that heightened levels of economic diversity decrease out-migration and net-migration.

[ To aid in interpretation, the coefficient estimate of 2 is -4.59 (p-value <0.01) and the coefficient estimate of 2 × "2(3 − 2(3 D # is -1.689 (p-value < 0.01).

Thus, a one standard deviation increase in diversity is estimated to reduce net-migration by (-1.689 / 4.59) x 100 = 36.79%. Clearly, the migration channel, wherein relative demand for residence in areas with concentrated employment falls after disasters, is important for the explanation of our results.

Our findings show that regional economic diversification tamps down the effects of a disaster on housing values. These findings indicate that there may exist meaningful benefits from enhancing local, urban variety as a means to mitigating housing price responses to externally generated shocks. However, resiliency is only one of the three main objectives policy makers often seek to achieve through diversification; price stability and price appreciation are other relevant considerations. While Coulson et al.

(2013) demonstrates that economic diversity effectively decreases housing price volatility, less work has been dedicated to understanding the direct effect of economic diversity on housing values. We proceed by addressing this shortcoming of the literature.

On the theoretical front, a priori, the relationship between diversity and housing values is ambiguous. Some researchers have noted that diversification necessarily implies a departure from specialization. From a pure quantitative perspective, this is true.

Moreover, and to the extent that efficiency advantages stemming from specialization exist, some have argued that diversification may be an impediment to economic growth thus leading to decreases in home values. Izraeli and Murphy (p.2, 2003) summarize this sentiment quite succinctly:

"The theory of comparative advantage shows very clearly the gain from specialization and trade. In the context of a nation, the geographic concentration of production benefits sub-national units, i.e., regions. This rationale explains why regions specialize in one or few industries in which they enjoy a comparative J o u r n a l P r e -p r o o f advantage over their trade partners." 12

Taking a different view, Glaeser et al. (1992) emphasize the importance of knowledge spillovers that occur between industries. Their idea, which is consistent with the earlier work of Jacobs (1969), suggests that the "variety and diversity of geographically proximate industries rather than geographical specialization promote innovation and growth." (p. 1128) On this account, diversity may ultimately lead to increases in housing values.

On the empirical front, there are inherent difficulties in estimating the direct effect of diversity on home prices. As we note earlier, to establish a causal link one would need to confront the possibility that diversification is an endogenous covariate. An ideal but impractical experimental setting is one in which economic diversity in an MSA changes randomly and without regard to local economic conditions. Given the lack of this ideal setting, alternative approaches must be considered. We advance one such approach here by estimating variants of the following estimating equation, ln &'( = 8 + ! 2(3 ; + + 8 + 9 + : , # 15

using two stage least squares (2SLS) instrumenting for 2(3 with 2(3 IJ . Here, 8 is a complete set of MSA fixed effects and 9 an exhaustive set of year-quarter fixed effects.

We report OLS estimates of equation (15) in column 1 of Table 5 . For completeness, in column 2 we present estimates of equation (15) allowing diversity to enter the model non-linearly. For the sake of interpretation, in the log-linear specification, column 1, we present estimates of equation (15) after standardizing the diversity index mean zero standard deviation one. This allows us to interpret coefficient estimates as the effect of diversity on home prices due to a one standard deviation increase in diversity. Columns 3 and 4 report 2SLS estimates of columns 1 and 3, respectively. Additionally, relevant first stage statistics are also reported.

[ Column 1 suggests a one standard deviation increase in diversification may lead to a 1.34% reduction in price. Column 2 indicates that a 1% increase in diversification may lead to a corresponding 0.6% decrease in housing values; however, both effects are statistically insignificant. Further, as shown in columns 3 and 4, the magnitudes of these estimated price decreases are meaningfully attenuated toward zero after we instrument for diversity suggesting there is no economically discernable relationship between diversification and housing values.

Diversification is often regarded as a positive policy objective for local real estate markets in terms of improving price resiliency; albeit, this conventional wisdom has persisted in the absence of any formal empirical evidence. Our findings demonstrate that economic diversification has the two-pronged effect of attenuating the immediate impact and the relative persistence of a shock in a small regional economy to local housing values. Our modeling exercise shows that diversity catalyzes the resiliency of the housing markets to climate shocks.

There exists a long-standing debate in the literature on the potential "dual-effects" of diversification on the regional economy in terms of the direct effect of diversification on regional market performance. Through the lens of the housing market, we show that the concerns issued in previous studies regarding the potential downsides of diversification stemming from the microeconomic foundations of comparative advantage do not appear to be warranted. After instrumenting for diversity, we find no economically meaningful or statistically relevant relationship between diversity and regional housing values.

Considering these results, the policy goal of improving resiliency through diversification can likely be achieved net of ancillary concerns of impeding economic progress.

Albouy, D. (2). Standard errors are reported in parentheses and are clustered at the MSA level. E <, 2(3 D represents the estimated impact of a disaster on home prices < years since a disaster struck relative the year before a disaster (< = −1 conditional on the jth percentile of economic diversity, DIV j . Estimating equation (2) includes year by quarter fixed effects, MSA fixed effects, and MSA-specific linear time trends.

(1) (14). Robust standard errors in parentheses and are clustered at the county level. Out-migration ij measures the number of people that moved from i to j. Net-migration ij = Out-migration ij -In-migration ij where In-migration ij measures the number of people that moved to i from j. DIV i -DIV j represents the difference in the standardized diversity index between county i and j. Disaster i is a binary variable equal to one if county i experienced a disaster in 2011 or 2012 and zero otherwise. Coefficient estimates for Disaster i thus represent the estimated effect of a disaster on each outcome, holding the relative level of diversity between regions constant. Likewise, coefficient estimates for 2 × "2(3 − 2(3 D # represent the estimated effect that a relative standard deviation increase in the diversity of region i has on mitigating or exacerbating the average response to a disaster.

(1) J o u r n a l P r e -p r o o f Notes: This table reports parameter estimates exp[θ(τ, DIV j )]-1 obtained from estimating equation (2). Standard errors are reported in parentheses and are clustered at the MSA level. E <, 2(3 D represents the estimated impact of a disaster on home prices < years since a disaster struck relative the year before a disaster (< = −1 conditional on the jth percentile of economic diversity, DIV j . Estimating equation (2) includes year by quarter fixed effects, MSA fixed effects, and MSA-specific linear time trends. Notes: This table reports parameter estimates of E′ <, 2(3 obtained from estimating equation (5). Standard errors are reported in parentheses and are clustered at the MSA level. E′ <, 2(3 represents the estimated impact of a disaster on home prices < years since a disaster struck relative the year before a disaster (< = −1 conditional on the jth percentile of economic diversity, DIV j . Estimating equation (5) includes year by quarter fixed effects, MSA fixed effects, and MSA-specific linear time trends. • HIJ where HIJ refers to the sample standard deviation of diversity in the data and E <, 2(3 the estimated price effect of a disaster < years after a disaster hits expressed as a function of diversity obtained from estimating equation (2). We report p-values associated with the test & ] : 0 [ , ≤ 0 vs. &`: 0 [ , > 0 in brackets. The estimates reported in column 2 are based on estimating equation (2) without the inclusion of the set of income by industry controls. In contrast, the estimates reported in columns 3 and 4 estimate equation (2) including the income by industry controls and employment by industry controls, respectively.

(1) • We use natural disasters as a source of exogenous shocks to examine the effect of economic diversity on the resiliency of US housing markets. • Diversity dampens the magnitude and duration of the effects of disasters.

• One mechanism is suggested by our subsequent finding that outmigration after disasters is lower in more diverse cities.

J o u r n a l P r e -p r o o f

Did the Rust Belt Become Shiny? A Study of Cities and Counties That Lost Steel and Auto Jobs in the 1980s

Related variety, unrelated variety and regional economic growth

Industrial diversity, growth, and volatility in the seven states of the Tenth District

Learning about an infrequent event: Evidence from flood insurance take-up in the United States

The long-run relationship between house prices and income: evidence from local housing markets

Directions for diversification with an application to Saskatchewan

Growth in cities

Where are the speculative bubbles in US housing markets

Market responses to hurricanes

Employment risk in the U.S. metropolitan and nonmetropolitan regions: The influence of industrial specialization on population characteristics

Environmental determinants of housing prices: the impact of flood zone status