Get 20M+ Full-Text Papers For Less Than $1.50/day. Subscribe now for You or Your Team.

Learn More →

Measuring the impact of insurance on urban earthquake recovery using nightlights

Measuring the impact of insurance on urban earthquake recovery using nightlights Abstract We measure the longer-term effect of a major earthquake on the local economy, using night-time light intensity, and focus on the role of insurance payments for damaged residential property in the recovery process. The destructive Canterbury Earthquake Sequence (2010–2011) in New Zealand is our case study. Uniquely, for this event, >95% of residential housing units were covered by insurance and almost all incurred some damage. However, insurance payments were staggered over 5 years, enabling us to identify their local impact on recovery. We find that night-time luminosity can capture the process of recovery; and that insurance payments contributed significantly to the process of local economic recovery after the earthquake. Cash settlement of claims was no more effective than insurance-managed repairs in generating local recovery. Notably, delayed payments were less affective in assisting recovery; this suggests an important role for the regulator in making sure insurance payments are made promptly after disaster events. 1. Introduction New Zealand is prone to earthquakes. Recent destructive earthquakes in 2010, 2011, 2013 and 2016 have demonstrated the seriousness of this risk, and have shown that the local recovery from such events is often difficult and prolonged. In recent years, numerous papers have looked into the immediate restoration of activity after disasters, often from a microeconomic, single household, perspective or by focusing on a specific case-study (Rose et al., 1997; Sawada and Shimizutani, 2008; Chang, 2010; duPont Iv et al., 2015; Cole et al., 2017). However, the absence of detailed and sufficiently frequent microeconomic data has hindered attempts to shed more light on the dynamics of recovery over longer periods of time except for at the macroeconomic level. Moreover, the insurance sector is often thought to play a significant role in recovery post-disaster, but analysis of its precise functioning during the recovery process is rarely if ever pursued. Insurance is frequently mentioned as (almost) a panacea for disaster risk, and it is singled out as an important part of international disaster risk reduction efforts. Yet, except for Von Peter et al. (2012) and Poontirakul et al. (2017), there is no research that attempts to look into the role of insurance claim payments in post-catastrophe recovery. Our aim here is to provide a first attempt at measurement of the longer-term economic effect of insurance payments after a major earthquake event, using satellite night-time light as a proxy measure for economic activity. Put differently, we investigate how insurance claim payments for damaged residential property, and specifically their timing, affect the recovery process of the local economy. We focus on the destructive Canterbury Earthquake Sequence (CES) in 2010–2011 as our case study. We chose this event due to the availability of the insurance claim payment data, and specific characteristics of the earthquake and the insurance market in New Zealand that allow us to identify the impact of insurance payments. These are detailed in the next section. We first find that night-time luminosity can capture the earthquake damage and the process of recovery. We then find that the insurance payments contributed significantly to the process of economic recovery after the earthquake; and further emphasize the importance of the timing of payments, with delayed payments being a lot less effective. This specific earthquake sequence is an attractive case study for several reasons: First, the event is unique as >95% of residential housing units were covered by insurance, and practically all submitted a claim. Thus, unlike almost all other disasters where insurance penetration rates are much lower, there is no problem of selection bias (i.e., households that purchase insurance are different from those that do not). Second, these were really big events, from an insurance perspective. Two of five earthquakes in this sequence are listed as some of the costliest insured events, globally, ever. Several geographic aspects of Christchurch make it especially feasible to conduct the analysis by using night-time luminosity—especially noteworthy are the fact that the city is composed of mostly low rise, spread out residential neighborhoods (so that the nightlight sensors are not saturated) and there are many nights of low or no cloud cover (making the measurements more consistent). The remainder of this paper is structured as follows. In the next section we provide information about the earthquake events, insurance in New Zealand and the recovery process. We next discuss the use of nightlight luminosity as a proxy for economic activity and the history of its use in the analysis of disaster impact and recovery. After covering these literatures, we describe the data and methodology used in this paper. In the penultimate section, we present our empirical results; and we end with some further comments about future research. 2. The Canterbury Earthquake Sequence 2010–2011 On 4 September 2010, a M7.1 earthquake occurred, epi-centered close to Darfield village, a rural area not far from the city of Christchurch (the biggest city in the South Island of New Zealand, with a population of about 400,000). The earthquake damaged nearby towns and the eastern suburbs of the city which were vulnerable to liquefaction. Many old unreinforced masonry and heritage buildings were affected. This event was followed by a shallower M6.3 aftershock to the southeast of the city on 22 February 2011. This event resulted in intense ground motions which were directed toward the city center (GeoNet, 2011). Many buildings in the Central Business District (CBD) and elsewhere in the city were severely damaged. There were 185 fatalities in the February 2011 earthquake. Practically all residential buildings in the city experienced at least some damage, with many thousands eventually requiring complete rebuilds. Some areas around the Avon River suffered heavily from subsidence. The flood and liquefaction risk of this area was eventually found to be unacceptably high, and the government decided to re-zone it for non-residential use (a total of around 8000 homes were located in these Red Zones). Following all this, there were numerous aftershocks, which mostly led to additional damage to previously damaged buildings, and to delays in reconstruction. New Zealand has a very high insurance penetration ratio, with >95% of residences being insured for earthquakes (Nguyen and Noy, 2019). The New Zealand Earthquake Commission (EQC) is a public entity providing the first layer of residential insurance cover for earthquakes. The EQC was liable for residential claims that cover dwelling damage up to USD 67,000, content damage up to USD 13,400, and some land damage.1 The residential over-cap (over the EQC cap) and out-of-scope claims for damages (for example to driveways) were handled by the private insurers. Approximately 25,600 residential building over-cap claims were transferred to private insurers to be resolved. EQC (2017) reports that the public insurance scheme has settled over 167,000 and 73,000 valid dwelling and land claims, respectively. These claim settlements cost the EQC approximately USD 7.2 billion (EQC, 2016). The number of submitted claims was twice as large as the EQC planned for as a ‘worst foreseeable event.’ Private insurance companies also had limited experience handling such a large number of claims prior to this event, and almost no experience coordinating their work with the EQC. Further complications were the large number of aftershocks spread all over the city, many previously unacknowledged ambiguities in insurance contracts, complex cover for land damage that is not available in other jurisdictions and a legal system that was overwhelmed post-earthquake. These complications led to an insurance settlement process that has taken over 7 years to complete, and as of this writing (end of 2019) there are still some claims yet to be settled. These spatially random delays in claim settlements allow us to identify the effect of insurance payments on recovery. Three other aspects of the insurance system in Christchurch are helpful in establishing the identification of the causal channel from insurance payments to recovery: (1) Almost everyone had residential insurance in Christchurch. No one knows the exact number of uninsured properties, but from the Residential Red Zone properties for which we do have data, and which we view as a random sample of 8000 properties, the insurance penetration rate was about 98%. (2) Almost all properties incurred some damage, even if minor, and the excess (deductible) in the EQC contract was very low (USD 134). Therefore, almost everyone made a claim to the public insurer in Christchurch. (3) Earthquake risk was considered very low in Christchurch; in the pre-2010 New Zealand Government’s seismic risk zone maps (maps that determine the benchmark for building standards), Christchurch City was labeled as low risk (out of three levels, it had the lowest risk level). We therefore do not expect that seismic preparedness, beyond what is always mandated by building standards, was undertaken by the majority of households, so that damages were not endogenous to the characteristics of the households/owners. Several other research projects have looked at this event. Similar to residential insurance, commercial insurance claim settlement also faced delays due to the scale of claim handling, the complexity of claims, the ongoing seismicity and the shortage of experienced loss assessors. Additional reasons for delay in the assessment process include poor information management by both sides, slow decision-making and the use of brokers as intermediaries (Brown et al., 2013, 2016a; Seville et al., 2014). Using surveys, Stevenson et al. (2011) found that affected organizations financed their recovery primarily with their cash-flow instead of from the proceeds of claim payments. With these same surveys, Poontirakul et al. (2017) find no short-run difference in likelihood of business survival between insured and uninsured firms. However, later on, firms which had prompt and full claim payment experienced better recovery—in terms of performance and profitability—than those that had insufficient cover or delayed claim settlements; and the latter firms performed marginally worse than uninsured firms. 3. Insurance and disaster recovery elsewhere The literature on the economics of disasters has grown significantly in recent years, especially in its investigation of the varied immediate impacts of disasters. Yet, relatively less is known about the post-disaster recovery process and the factors that shape it—see Noy and duPont (2018) for a survey of the existing literature. Very few papers have closely looked at the role of insurance. This research has largely focused on explaining insurance penetration, while the insurance companies’ own research has concentrated on estimating expected disaster loss rather than on measuring their role in the recovery process (Kusuma et al., 2019). Melecky and Raddatz (2015) find that high- and middle-income countries, which have high insurance penetration, are affected less and experience better economic recovery following a disaster; and similar findings are reported in Von Peter et al. (2012)—but both only use aggregate country level data on insurance penetration. Platt et al. (2016) described the use of a wide range of data sources to identify the speed and the quality of recovery after major earthquakes. These include satellite imagery, crowd-sourced data, ground and household surveys, official statistics and insurance data. They conclude that remote sensing appears to provide accurate and reliable information. 4. Night-time luminosity in economic research In the past decade, night-time light has been increasingly used in the social science literature as an indicator for economic activity and human development. Because most consumption and household activities require illumination in the evening, using changes in light intensity as a proxy for per capita GDP growth appears to be feasible. When household income increases, its light usage also increases (i.e. lighting is a normal good); and studies showing the relationship between night-time luminosity and socioeconomic information are numerous (Sutton and Costanza, 2002; Doll et al., 2006; Sutton et al., 2007; Elvidge et al., 2009; Ghosh et al., 2009,, 2010; Chen and Nordhaus, 2011; Kulkarni et al., 2011; Michalopoulos and Papaioannou, 2013; Hodler and Raschky, 2014a; Pinkovskiy and Sala-i-Martin, 2016). In all these papers, night-time luminosity data are obtained from the DMSP/OLS or VIIRS DNB satellites.2 Luminosity data have been used to measure income at the sub-national level at various grid-cell sizes (Besley and Reynal-Querol, 2014; Montalvo and Reynal-Querol, 2016; Storeygard, 2016; Bruederle and Hodler, 2017; Henderson et al., 2018), projected onto cities and municipal boundaries (Brown et al., 2016b) and for administrative regions (Hodler and Raschky, 2014a, 2014b; Bickenbach et al., 2016). The correlation between the night-time light and economic activity tends to be weaker at very small unit levels (e.g. one pixel), so some aggregation appears to be necessary. Some have used night-time light in order to investigate the economic losses and recovery post disaster event. For instance, Klomp (2016) explores how large-scale disasters affect economic activity, using night-time light intensity and historic data on 1000 natural hazard events between 1992 and 2008. He finds that geophysical and meteorological events reduce night-time illumination in developed countries while hydrological and climatic disasters lead to a short-term decline in the light intensity in developing countries, and that earthquakes have more prolonged negative effects.3 Similar findings, using geo-coded indicators of disaster intensity, are reported by Felbermayr et al. (2018). Gillespie et al. (2014) used household survey data (2004–2007) in Sumatra during the recovery from its 2004 earthquake/tsunami to reveal the link between night-time luminosity and spending per capita at the community level, and Skoufias et al. (2017) did something similar for more disasters in Indonesia at the district level. They both suggest that satellite night-time imagery is a useful tool for assessing the post disaster impacts.4 5. Data Greater Christchurch includes Christchurch city and its satellite towns. According to the 2006 Census, the region’s resident population count was nearly 425,000 with 82% living in Christchurch City. We aggregate and analyze all data at the area unit (AU) level.5 Based on the 2016 Geographic Boundary of Statistics New Zealand, there are 183 AUs in Greater Christchurch, containing 158 AUs defined as residential areas. 5.1. Night-time light data We use night-time light data derived from images taken by DMSP/OLS and VIIRS DNB satellites. We convert the images to integer format at the pixel level. Because each AU has different size and can cover several pixels, we calculate the nightlight intensity weighted mean within each AU polygon. The relative scales of nightlight pixels and AU are illustrated in Online Appendix Figure 1. The average spatial area for an AU in Greater Christchurch is approximately 55 km2. Thus, even within the city, where AUs cover less ground area, each AU may contain more than 10 pixels. Figures 1 and 2 present the night-time light images of Greater Christchurch from the 2013 cloud-free composite DMSP/OLS and VIIRS DNB satellites, respectively. The brightly lit area in the figures corresponds to Christchurch City. It is noticeable that the DMSP data have saturation centered on the city area while the VIIRS shows much more detail. The latter has a better spatial resolution, about 750 m to the 2.7 km resolution of the former (NOAA, 2013). Due to the difference in time coverage of the two datasets, both are used in this paper, but for the main results on insurance and recovery, we use the higher-resolution VIIRS data.6 Specifically, the DMSP data of satellite F16 and F18, from 2009 to 2012, are used to capture the reduction in nightlights as the indicator of short-run disaster impact. The DMSP data are publicly available at annual frequency only. We then use quarterly VIIRS data for the period from 2012 to 2016 for each AU for the recovery trajectory.7 Figure 1 Open in new tabDownload slide Raw image of average DMSP/OLS night-time light in 2013. Figure 1 Open in new tabDownload slide Raw image of average DMSP/OLS night-time light in 2013. Figure 2 Open in new tabDownload slide Raw image of average VIIRS DNB night-time light in 2013. Figure 2 Open in new tabDownload slide Raw image of average VIIRS DNB night-time light in 2013. Figure 3 Open in new tabDownload slide Average VIIRS DNB night-time light in 2013 aggregated to the area unit level. Figure 3 Open in new tabDownload slide Average VIIRS DNB night-time light in 2013 aggregated to the area unit level. 5.2. Insurance claim data To measure the payments provided by the insurance sector during the recovery, we use the geo-coded payment data from the EQC. The dataset includes individual claims for earthquake events during 2010–2011. For each insured event, the data provide the actual amount that EQC has spent on each property and the estimated total damage cost as it was apportioned for each earthquake.8 We use records of approximately 220,000 claims for nearly 100,000 properties in Greater Christchurch. More than 85% of these claims came from Christchurch city. Three-fourths of the claims are for building structure and the rest are for land and content exposure.9 Table 1 provides summary statistics of quarterly claim payment data at the AU level. In Greater Christchurch, the average total of quarterly claim payments for each exposure per AU are USD 462,696, USD 17,347 and USD 60,240 for structure, content and land, respectively. The standard deviation of land claim payments is high, relative to its mean, as there are claims with high land remediation cost due to land movement and cliff collapse.10 Table 1 Summary statistics of claim payment data, by asset damage type Variables Building Content Land Total (N = 143,545) (N = 68,324) (N = 73,123) (N = 220,898) Mean Standard deviation Mean Standard deviation Mean Standard deviation Mean Standard deviation Total claim payment (USD) 462,695 696,423 17,347 43,840 60,240 1,424,564 540,284 1,642,358 Total exposed value of the assets (USD) 6,680,840 7,645,051 274,319 520,982 694,532 2,844,193 7,651,877 9,406,143 Proportion of cash paid/total settlement 0.73 0.26 1.00 0.00 0.55 0.41 0.71 0.35 Time to settlement (days) 845 538 489 439 688 514 984 542 Variables Building Content Land Total (N = 143,545) (N = 68,324) (N = 73,123) (N = 220,898) Mean Standard deviation Mean Standard deviation Mean Standard deviation Mean Standard deviation Total claim payment (USD) 462,695 696,423 17,347 43,840 60,240 1,424,564 540,284 1,642,358 Total exposed value of the assets (USD) 6,680,840 7,645,051 274,319 520,982 694,532 2,844,193 7,651,877 9,406,143 Proportion of cash paid/total settlement 0.73 0.26 1.00 0.00 0.55 0.41 0.71 0.35 Time to settlement (days) 845 538 489 439 688 514 984 542 Notes: Claim payments are aggregated at AU level and quarterly frequency. Summary statistics of variable ‘time to settlement’ are first calculated at the individual claim level. Open in new tab Table 1 Summary statistics of claim payment data, by asset damage type Variables Building Content Land Total (N = 143,545) (N = 68,324) (N = 73,123) (N = 220,898) Mean Standard deviation Mean Standard deviation Mean Standard deviation Mean Standard deviation Total claim payment (USD) 462,695 696,423 17,347 43,840 60,240 1,424,564 540,284 1,642,358 Total exposed value of the assets (USD) 6,680,840 7,645,051 274,319 520,982 694,532 2,844,193 7,651,877 9,406,143 Proportion of cash paid/total settlement 0.73 0.26 1.00 0.00 0.55 0.41 0.71 0.35 Time to settlement (days) 845 538 489 439 688 514 984 542 Variables Building Content Land Total (N = 143,545) (N = 68,324) (N = 73,123) (N = 220,898) Mean Standard deviation Mean Standard deviation Mean Standard deviation Mean Standard deviation Total claim payment (USD) 462,695 696,423 17,347 43,840 60,240 1,424,564 540,284 1,642,358 Total exposed value of the assets (USD) 6,680,840 7,645,051 274,319 520,982 694,532 2,844,193 7,651,877 9,406,143 Proportion of cash paid/total settlement 0.73 0.26 1.00 0.00 0.55 0.41 0.71 0.35 Time to settlement (days) 845 538 489 439 688 514 984 542 Notes: Claim payments are aggregated at AU level and quarterly frequency. Summary statistics of variable ‘time to settlement’ are first calculated at the individual claim level. Open in new tab We also exploit other information in the claim data; in particular, we focus on two variables: time it took to settle the claim and proportion of cash in settlement. The first is the average number of days to claim settlement, since the day the claim was launched, for each quarter in each AU. Online Appendix Figure 7 presents the temporal distribution of payments (per AU) for Q2 of each of the years following the earthquakes.11 There is no apparent spatial pattern in the temporal distribution of the claim payments, suggesting there is most likely no selection problem in the time-to-payment variable. It took 1–4 years for most claims to be resolved (average is nearly 3 years). The second variable is the proportion of cash payment amount relative to the total claim settlement in each AU. Table 1 illustrates that 71% of the amount the insurer paid to claimants was in cash, while the number of cash-paid claims was 60% (EQC, 2017). Figures 4 and 5 describe the association between the insurance claim data and the measured nightlights, both for the immediate impact phase in Figure 4, and for the longer-term recovery period (in Figure 5). Figure 4 Open in new tabDownload slide Change in nightlight from 2009 to 2011 and damage ratio. Note: The y-axis variable is described in Equation (1) and the x-axis variable is defined in Equation (2). Figure 4 Open in new tabDownload slide Change in nightlight from 2009 to 2011 and damage ratio. Note: The y-axis variable is described in Equation (1) and the x-axis variable is defined in Equation (2). Figure 5 Open in new tabDownload slide Quarterly change in NTL and same quarter insurance payments. Notes: The y-axis variable is described in Equation (4) and the x-axis variable is defined in Equation (5). Figure 5 Open in new tabDownload slide Quarterly change in NTL and same quarter insurance payments. Notes: The y-axis variable is described in Equation (4) and the x-axis variable is defined in Equation (5). 5.3. Other variables We also use data from Statistics New Zealand on households, at the AU level, from the most recent census conducted before the earthquakes (in 2006)—this data are summarized in Table 2. Online Appendix Table 3 illustrates the correlation between the 2006 nightlight and control variables from the 2006 Census, matched at the AU level. There are positive correlations between light luminosity as a measure of economic activity and most explanatory variables. Online Appendix Table 3 also shows that light intensity captures the density variables better. For instance, the correlation between nightlight and population is 0.59, while the correlation between nightlight and population density is 0.71. There is also positive relationship between light brightness and income density; and nocturnal light is negatively correlated with the distance between the AUs and the city center. Table 2 Summary statistics of area units (AU) in Greater Christchurch Variables Christchurch city Waimakariri and Selwyn Mean Standard deviation Mean Standard deviation Area in squared kilometer 11.3 44.0 148.2 612.4 Night-time population 1755 1183 1643 1255 Night-time population density 2802 1296 514 698 Household income 68,420 18,220 74,171 18,345 Household income density 40,893 28,853 25,097 29,778 Variables Christchurch city Waimakariri and Selwyn Mean Standard deviation Mean Standard deviation Area in squared kilometer 11.3 44.0 148.2 612.4 Night-time population 1755 1183 1643 1255 Night-time population density 2802 1296 514 698 Household income 68,420 18,220 74,171 18,345 Household income density 40,893 28,853 25,097 29,778 Notes: Household income and night-time population are measured using 2006 Census data provided by Statistics New Zealand. The density variables are per km2. Open in new tab Table 2 Summary statistics of area units (AU) in Greater Christchurch Variables Christchurch city Waimakariri and Selwyn Mean Standard deviation Mean Standard deviation Area in squared kilometer 11.3 44.0 148.2 612.4 Night-time population 1755 1183 1643 1255 Night-time population density 2802 1296 514 698 Household income 68,420 18,220 74,171 18,345 Household income density 40,893 28,853 25,097 29,778 Variables Christchurch city Waimakariri and Selwyn Mean Standard deviation Mean Standard deviation Area in squared kilometer 11.3 44.0 148.2 612.4 Night-time population 1755 1183 1643 1255 Night-time population density 2802 1296 514 698 Household income 68,420 18,220 74,171 18,345 Household income density 40,893 28,853 25,097 29,778 Notes: Household income and night-time population are measured using 2006 Census data provided by Statistics New Zealand. The density variables are per km2. Open in new tab We also use Shake-maps for the September 2010 and February 2011 earthquakes, provided by the USGS.12 These maps provide the spatial distribution of the physical intensity for major earthquakes. We aggregate these macro-seismic intensities to the AU level. In all of our estimations, we exclude the CBD, because the area was cordoned off for 2 years, and its redevelopment was subject to a very different, complex and contentious regulatory regime.13 6. Methodology We now turn to the regression analysis where we explore the change in night-time light in the Greater Christchurch region, during and after the CES. The preliminary set of estimations, described in Section 6.1, is intended to establish the short-term impact of earthquake damage on local economic activity in Great Christchurch. The second, the main focus of this work, aims to estimate the effect of insurance payments on the recovery of residential areas in the region (Section 6.2). 6.1. Earthquake damage and the loss in night-time light We begin by using the immediate reduction in luminosity post-earthquakes as an indicator of the loss in economic activity in Greater Christchurch. The variable is calculated as follows: Economic_Lossieq=ΔNTLi2009-2011=ln⁡(NTLi,2009)-ln⁡(NTLi,2011), (1) where NTLi is our proxy for economic development based on the DMSP nightlight value (taken in logarithms) aggregated for each AU (denoted by i). We next aggregate the insurance claim payments over the whole period, to the AU level, to indicate the financial loss experienced by each AU due to earthquake damage. A number of papers in the literature have stressed that earthquake damage is correlated with income per capita (Kahn, 2005; Toya and Skidmore, 2007; Felbermayr and Gröschl, 2014). Hence, even in the spatially confined study at hand, cross-AU heterogeneity in damage may be driven by cross-AU differences in income per capita. To reduce the endogeneity of the financial loss indicator, we create a damage ratio variable from these aggregate figures (in Equation 2). Damagei,k=Claim_paymenti,k∑kAsset_valuei,k. (2) Damagei represents the total earthquake financial loss on all exposures (k = building, content, land or total) as a ratio of the total exposure value for all dwellings for which there were claims.14 The property value data are from the New Zealand Quotable Value (QV). In the first set of results, we use Economic_Lossieq  as a dependent variable indicating the change in economic activity due to the earthquakes. We hypothesize that AUs that have high ΔNTL experienced large economic losses because of the large amount of damage to property (assets), as follows: Economic_Lossieq=α+βkDamagei,k+γXi+εi, (3) where Damagei,k is earthquake damage (building, content, land or total) as a ratio of exposed value for each AU as specified in Equation (2). We use robust standard errors in order to control for the heteroscedasticity in the error terms. In addition, for robustness we include several control variables (⁠ Xi ⁠) that might also affect the measured economic loss in our regressions such as average household income, night-time population, average number of bedrooms in each household and surface area (taken in logarithms). For robustness, because of endogeneity concerns, we also implement a two-stage least squares instrumental variable (2SLS-IV) method. We use the ShakeMap’s earthquake physical intensity measure (Zi) as the instrumental variable for  Damagei,k ⁠. The correlations between damage ratio measured by property damage and the exogenous seismic intensity of February 2011 earthquake are over 50%, except for land damage. Thus, we expect to have strong first stage where the instrument is highly correlated with the endogenous variable.15 6.2. Insurance settlement and Christchurch recovery In the second set of regressions, we estimate the effect of insurance payments on local recovery, following the earthquakes. In this regression, we use the night-time VIIRS light dataset from April 2012 to August 2016 (VIIRS data are not available before that). We convert the nightlight data from monthly to quarter frequency. In order to identify the economic recovery in Greater Christchurch, we take the proportional change in the night-time radiance value for each quarter. In the specification, the variable is used as the dependent variable. Economic_Recovery i,tPost=ΔNTLi,tQ2.2012-Q3.2016=ln⁡(NTLi,t)-ln⁡(NTLi,t-1). (4) The main explanatory (RHS) variable is the insurance payment. It is the total insurance claim pay-out which an AU received at quarter  t ⁠, as described in Equation (5). Insi,t,k=ln⁡(Claim_paymenti,t,k). (5) The regression model is written as follows: Economic_Recovery i,tPost=αi+τt+βkInsi,t,k+γXi,t+εi,t. (6) whereby Insi,t,k is our measure of insurance payments described in Equation (5). Here, we hypothesize that the insurance payments are associated with the quarterly change in nightlight in the years following the earthquakes (Q2 2012–Q3 2016). We included AU and quarter-fixed effects to control for unobserved variations across individual AUs and over time. We also include AU cluster-robust standard errors to control for heteroscedasticity. Other insurance-related variables such as ‘settlement time’ and ‘proportion cash amount’ are also included. In some AUs, there are quarters without claim payments. The value of the insurance-related variables in this case is set as one.16 We also investigate the interaction term between insurance payment and settlement time and cash payments. By all accounts, EQC and private insurers resolved insurance claim on a first come, first serve basis without prioritizing any certain demographic groups or geographic location. Hence, we assume that the claim settlements were processed randomly across locations. In addition, we carry out spatial panel data analysis. Spatial econometric modeling helps us control for spatial effects. They may reduce the unobserved estimation bias which arises from both spatial and time dependence. More importantly, spatial regression methods permit us to identify spillover effects coming from neighboring AUs over time (Anselin et al., 2008; Lee and Yu, 2010; Elhorst, 2014).17 Following this literature, we implement four different spatial specifications including spatial autoregressive model (SAR), spatial error model (SEM), spatial Durbin model (SDM) and spatial autocorrelation model (SAC)—the estimated equations are described Online in Appendix B. We employ spatial panel maximum-likelihood estimation for the set of regression models with AU and quarter-fixed effects as described below. W is a non-negative spatial weighted matrix that describes the spatial structure of dependence between AUs in the sample. These models include three different types of interaction effects among units: (i) endogenous spatial interaction effects for the dependent variable  (WYi,t) ⁠; (ii) exogenous spatial interaction effects among the explanatory variables WXi,t and (iii) spatial interaction effects among the error terms  (Wϑi,t) ⁠. We employ a contiguity weighted matrix. The elements ωij of matrix W equal to 1/the number of neighbors of AU i if AU i and j share a border, otherwise  ωij=0 ⁠. In another version, we also use a matrix where each element ωij is the inverse distance between AU i and j ⁠. 7. Results The first set of regressions results, examining whether earthquake damages explain the reduction in economic activity in Greater Christchurch, is shown in Table 3. The damage ratio variable (for buildings, contents and land damage) is the main explanatory variable in the regressions. In Columns 1–3 of the table, we estimate the effect of residential building damage on the economic activity in the immediate aftermath of the earthquake events. Other columns focus on the damage for content, land and total damage (sum of the three asset classes) in regressions 4–6, 7–9 and 10–12, respectively. In these specifications, the coefficients of damage variables for buildings are always positive and significant (they are mostly positive but insignificant for the content and land damage variables—see Columns 4–9). For instance, in Column 1 of Table 3, the economic loss will be 0.56% higher, when the residential buildings damage over property value increases by 1%. When controlling for other variables (taken in logarithms), the building damage indicators are still statistically significant and of the same order of magnitude. Maybe not surprisingly, the residential building damage appears to explain some of the economic loss immediately after the disaster; and it is the only variable that consistently has explanatory power. Table 3 Short run economic impact of the earthquakes using the damage ratio variable Variables Dependent variable: Change in night-time light between 2010 and 2011 Building Content Land Total (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) Damage ratio 0.559*** 0.416** 0.957** 0.757** 0.379 0.379 0.016 −0.006 0.747 0.474*** 0.343** 0.912** (0.186) (0.171) (0.401) (0.367) (0.389) (0.389) (0.051) (0.068) (0.493) (0.181) (0.162) (0.415) Household income 0.008 0.003 0.007 0.007 0.012 −0.025 0.007 −0.000 (0.037) (0.045) (0.036) (0.036) (0.038) (0.049) (0.037) (0.041) Night-time Population 0.018 0.017 0.019 0.019 0.019 0.021* 0.019 0.017 (0.012) (0.012) (0.012) (0.012) (0.012) (0.012) (0.012) (0.012) Number of bedrooms −0.060 −0.019 −0.071 −0.071 −0.093 0.023 −0.061 −0.011 (0.094) (0.119) (0.092) (0.092) (0.092) (0.144) (0.092) (0.114) Area square km 0.006 0.005 0.006 0.006 0.006 −0.012 0.005 0.003 (0.008) (0.008) (0.008) (0.008) (0.009) (0.018) (0.008) (0.009) Constant −0.086*** −0.237 −0.251 −0.076*** −0.210 −0.210 −0.052*** −0.228 0.011 −0.079*** −0.224 −0.221 (0.016) (0.312) (0.367) (0.016) (0.301) (0.301) (0.008) (0.309) (0.362) (0.0151) (0.305) (0.335) Observation 158 158 158 158 158 158 158 158 158 158 158 158 R-squared 0.045 0.097 0.058 0.022 0.079 0.079 0.000 0.074 0.031 0.037 0.093 0.043 IV 40.349 35.301 3.171 22.328 Variables Dependent variable: Change in night-time light between 2010 and 2011 Building Content Land Total (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) Damage ratio 0.559*** 0.416** 0.957** 0.757** 0.379 0.379 0.016 −0.006 0.747 0.474*** 0.343** 0.912** (0.186) (0.171) (0.401) (0.367) (0.389) (0.389) (0.051) (0.068) (0.493) (0.181) (0.162) (0.415) Household income 0.008 0.003 0.007 0.007 0.012 −0.025 0.007 −0.000 (0.037) (0.045) (0.036) (0.036) (0.038) (0.049) (0.037) (0.041) Night-time Population 0.018 0.017 0.019 0.019 0.019 0.021* 0.019 0.017 (0.012) (0.012) (0.012) (0.012) (0.012) (0.012) (0.012) (0.012) Number of bedrooms −0.060 −0.019 −0.071 −0.071 −0.093 0.023 −0.061 −0.011 (0.094) (0.119) (0.092) (0.092) (0.092) (0.144) (0.092) (0.114) Area square km 0.006 0.005 0.006 0.006 0.006 −0.012 0.005 0.003 (0.008) (0.008) (0.008) (0.008) (0.009) (0.018) (0.008) (0.009) Constant −0.086*** −0.237 −0.251 −0.076*** −0.210 −0.210 −0.052*** −0.228 0.011 −0.079*** −0.224 −0.221 (0.016) (0.312) (0.367) (0.016) (0.301) (0.301) (0.008) (0.309) (0.362) (0.0151) (0.305) (0.335) Observation 158 158 158 158 158 158 158 158 158 158 158 158 R-squared 0.045 0.097 0.058 0.022 0.079 0.079 0.000 0.074 0.031 0.037 0.093 0.043 IV 40.349 35.301 3.171 22.328 Notes: ***, **, * Significant at 1%, 5% and 10% levels, respectively. AU clustered, robust standard errors are shown in parentheses. All regressions are estimated with OLS. Columns (1)–(3) use a damage ratio measure for buildings, Columns (4)–(6) for contents, Columns (7)–(9) for land damage and Columns (10)–(12) aggregate all damages per AU over time. In all columns, the dependent (LHS) variable is the same. The instrumental variable (IV), in the bottom row, is the robust Kleibergen–Paap rk Wald F-statistic for test of weak instruments. IV regressions have overidentification p-values approximately equal to zero, except for land regression. Open in new tab Table 3 Short run economic impact of the earthquakes using the damage ratio variable Variables Dependent variable: Change in night-time light between 2010 and 2011 Building Content Land Total (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) Damage ratio 0.559*** 0.416** 0.957** 0.757** 0.379 0.379 0.016 −0.006 0.747 0.474*** 0.343** 0.912** (0.186) (0.171) (0.401) (0.367) (0.389) (0.389) (0.051) (0.068) (0.493) (0.181) (0.162) (0.415) Household income 0.008 0.003 0.007 0.007 0.012 −0.025 0.007 −0.000 (0.037) (0.045) (0.036) (0.036) (0.038) (0.049) (0.037) (0.041) Night-time Population 0.018 0.017 0.019 0.019 0.019 0.021* 0.019 0.017 (0.012) (0.012) (0.012) (0.012) (0.012) (0.012) (0.012) (0.012) Number of bedrooms −0.060 −0.019 −0.071 −0.071 −0.093 0.023 −0.061 −0.011 (0.094) (0.119) (0.092) (0.092) (0.092) (0.144) (0.092) (0.114) Area square km 0.006 0.005 0.006 0.006 0.006 −0.012 0.005 0.003 (0.008) (0.008) (0.008) (0.008) (0.009) (0.018) (0.008) (0.009) Constant −0.086*** −0.237 −0.251 −0.076*** −0.210 −0.210 −0.052*** −0.228 0.011 −0.079*** −0.224 −0.221 (0.016) (0.312) (0.367) (0.016) (0.301) (0.301) (0.008) (0.309) (0.362) (0.0151) (0.305) (0.335) Observation 158 158 158 158 158 158 158 158 158 158 158 158 R-squared 0.045 0.097 0.058 0.022 0.079 0.079 0.000 0.074 0.031 0.037 0.093 0.043 IV 40.349 35.301 3.171 22.328 Variables Dependent variable: Change in night-time light between 2010 and 2011 Building Content Land Total (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) Damage ratio 0.559*** 0.416** 0.957** 0.757** 0.379 0.379 0.016 −0.006 0.747 0.474*** 0.343** 0.912** (0.186) (0.171) (0.401) (0.367) (0.389) (0.389) (0.051) (0.068) (0.493) (0.181) (0.162) (0.415) Household income 0.008 0.003 0.007 0.007 0.012 −0.025 0.007 −0.000 (0.037) (0.045) (0.036) (0.036) (0.038) (0.049) (0.037) (0.041) Night-time Population 0.018 0.017 0.019 0.019 0.019 0.021* 0.019 0.017 (0.012) (0.012) (0.012) (0.012) (0.012) (0.012) (0.012) (0.012) Number of bedrooms −0.060 −0.019 −0.071 −0.071 −0.093 0.023 −0.061 −0.011 (0.094) (0.119) (0.092) (0.092) (0.092) (0.144) (0.092) (0.114) Area square km 0.006 0.005 0.006 0.006 0.006 −0.012 0.005 0.003 (0.008) (0.008) (0.008) (0.008) (0.009) (0.018) (0.008) (0.009) Constant −0.086*** −0.237 −0.251 −0.076*** −0.210 −0.210 −0.052*** −0.228 0.011 −0.079*** −0.224 −0.221 (0.016) (0.312) (0.367) (0.016) (0.301) (0.301) (0.008) (0.309) (0.362) (0.0151) (0.305) (0.335) Observation 158 158 158 158 158 158 158 158 158 158 158 158 R-squared 0.045 0.097 0.058 0.022 0.079 0.079 0.000 0.074 0.031 0.037 0.093 0.043 IV 40.349 35.301 3.171 22.328 Notes: ***, **, * Significant at 1%, 5% and 10% levels, respectively. AU clustered, robust standard errors are shown in parentheses. All regressions are estimated with OLS. Columns (1)–(3) use a damage ratio measure for buildings, Columns (4)–(6) for contents, Columns (7)–(9) for land damage and Columns (10)–(12) aggregate all damages per AU over time. In all columns, the dependent (LHS) variable is the same. The instrumental variable (IV), in the bottom row, is the robust Kleibergen–Paap rk Wald F-statistic for test of weak instruments. IV regressions have overidentification p-values approximately equal to zero, except for land regression. Open in new tab The control variables ‘household income’ and ‘number of bedrooms’, which indicate the size of a dwelling, have small and insignificant coefficients across regressions of all asset classes. Moreover, when using the earthquakes’ physical intensities as instrumental variables (Columns 3, 6, 9 and 12) we obtain similar results to the OLS regressions (with higher magnitudes for the coefficients). The magnitude of the coefficient is largest for building and total assets regressions. For instance, a 1% increase in the total asset damage over total asset value is associated with 0.91% reduction in residential economic activity. For the second set of regressions, our primary focus in this paper, we examine the effect of insurance payment on local economic recovery post-earthquakes. Table 4 provides the results for the estimations of Equation (6) including AU and quarter-fixed effects. The insurance payment variables are estimated for each exposure separately (Columns 1–6) and total exposure is provided in the last three columns. Table 4 Economic recovery following the earthquakes (Claim payment)—AU and quarter-fixed effect Variables Dependent variable: quarterly change in night-time light Building Land Total (1) (2) (3) (4) (5) (6) (7) (8) (9) Insurance payment 0.004 0.005 0.071* 0.000 −0.000 0.009 0.004 0.004 0.094*** (0.004) (0.004) (0.038) (0.001) (0.001) (0.009) (0.004) (0.004) (0.031) Settlement time 0.017 −0.112* 0.011*** −0.001 0.031 −0.121** (0.033) (0.058) (0.004) (0.012) (0.030) (0.054) Prop. Cash settlement −0.060*** −0.018 0.011 −0.012 −0.056*** −0.110 (0.022) (0.157) (0.010) (0.028) (0.021) (0.125) Ins. payment* settlement time −0.001*** −0.001 −0.001*** (0.000) (0.001) (0.000) Ins. payment* prop. cash settlement 0.007 0.000 0.005 (0.014) (0.003) (0.010) Constant −0.146*** −0.203 0.664 −0.101*** −0.171*** −0.106 −0.145*** −0.283 0.857* (0.052) (0.224) (0.495) (0.013) (0.024) (0.072) (0.053) (0.195) (0.436) Observation 2686 2686 2686 2686 2686 2686 2686 2686 2686 No. area units 158 158 158 158 158 158 158 158 158 R-squared 0.704 0.705 0.706 0.703 0.704 0.704 0.704 0.705 0.706 Total effect of Ins. payment 0.004 0.005 0.070* 0.000 −0.000 0.005 0.004 0.004 0.128*** (0.004) (0.004) (0.039) (0.001) (0.001) (0.009) (0.004) (0.004) (0.032) Variables Dependent variable: quarterly change in night-time light Building Land Total (1) (2) (3) (4) (5) (6) (7) (8) (9) Insurance payment 0.004 0.005 0.071* 0.000 −0.000 0.009 0.004 0.004 0.094*** (0.004) (0.004) (0.038) (0.001) (0.001) (0.009) (0.004) (0.004) (0.031) Settlement time 0.017 −0.112* 0.011*** −0.001 0.031 −0.121** (0.033) (0.058) (0.004) (0.012) (0.030) (0.054) Prop. Cash settlement −0.060*** −0.018 0.011 −0.012 −0.056*** −0.110 (0.022) (0.157) (0.010) (0.028) (0.021) (0.125) Ins. payment* settlement time −0.001*** −0.001 −0.001*** (0.000) (0.001) (0.000) Ins. payment* prop. cash settlement 0.007 0.000 0.005 (0.014) (0.003) (0.010) Constant −0.146*** −0.203 0.664 −0.101*** −0.171*** −0.106 −0.145*** −0.283 0.857* (0.052) (0.224) (0.495) (0.013) (0.024) (0.072) (0.053) (0.195) (0.436) Observation 2686 2686 2686 2686 2686 2686 2686 2686 2686 No. area units 158 158 158 158 158 158 158 158 158 R-squared 0.704 0.705 0.706 0.703 0.704 0.704 0.704 0.705 0.706 Total effect of Ins. payment 0.004 0.005 0.070* 0.000 −0.000 0.005 0.004 0.004 0.128*** (0.004) (0.004) (0.039) (0.001) (0.001) (0.009) (0.004) (0.004) (0.032) Notes: ***, **, * Significant at 1%, 5% and 10% levels, respectively. AU cluster and robust standard errors are shown in parentheses. All regressions are estimated with AU and quarter-fixed effect. Columns (1)–(3) measure the ‘insurance payment’ for building damages (per AU and quarter), Columns (4)–(6) measure the payments for land damage and Columns (7)–(9) sum up the building, contents and land payments together. In all columns, the dependent variable is the same. Open in new tab Table 4 Economic recovery following the earthquakes (Claim payment)—AU and quarter-fixed effect Variables Dependent variable: quarterly change in night-time light Building Land Total (1) (2) (3) (4) (5) (6) (7) (8) (9) Insurance payment 0.004 0.005 0.071* 0.000 −0.000 0.009 0.004 0.004 0.094*** (0.004) (0.004) (0.038) (0.001) (0.001) (0.009) (0.004) (0.004) (0.031) Settlement time 0.017 −0.112* 0.011*** −0.001 0.031 −0.121** (0.033) (0.058) (0.004) (0.012) (0.030) (0.054) Prop. Cash settlement −0.060*** −0.018 0.011 −0.012 −0.056*** −0.110 (0.022) (0.157) (0.010) (0.028) (0.021) (0.125) Ins. payment* settlement time −0.001*** −0.001 −0.001*** (0.000) (0.001) (0.000) Ins. payment* prop. cash settlement 0.007 0.000 0.005 (0.014) (0.003) (0.010) Constant −0.146*** −0.203 0.664 −0.101*** −0.171*** −0.106 −0.145*** −0.283 0.857* (0.052) (0.224) (0.495) (0.013) (0.024) (0.072) (0.053) (0.195) (0.436) Observation 2686 2686 2686 2686 2686 2686 2686 2686 2686 No. area units 158 158 158 158 158 158 158 158 158 R-squared 0.704 0.705 0.706 0.703 0.704 0.704 0.704 0.705 0.706 Total effect of Ins. payment 0.004 0.005 0.070* 0.000 −0.000 0.005 0.004 0.004 0.128*** (0.004) (0.004) (0.039) (0.001) (0.001) (0.009) (0.004) (0.004) (0.032) Variables Dependent variable: quarterly change in night-time light Building Land Total (1) (2) (3) (4) (5) (6) (7) (8) (9) Insurance payment 0.004 0.005 0.071* 0.000 −0.000 0.009 0.004 0.004 0.094*** (0.004) (0.004) (0.038) (0.001) (0.001) (0.009) (0.004) (0.004) (0.031) Settlement time 0.017 −0.112* 0.011*** −0.001 0.031 −0.121** (0.033) (0.058) (0.004) (0.012) (0.030) (0.054) Prop. Cash settlement −0.060*** −0.018 0.011 −0.012 −0.056*** −0.110 (0.022) (0.157) (0.010) (0.028) (0.021) (0.125) Ins. payment* settlement time −0.001*** −0.001 −0.001*** (0.000) (0.001) (0.000) Ins. payment* prop. cash settlement 0.007 0.000 0.005 (0.014) (0.003) (0.010) Constant −0.146*** −0.203 0.664 −0.101*** −0.171*** −0.106 −0.145*** −0.283 0.857* (0.052) (0.224) (0.495) (0.013) (0.024) (0.072) (0.053) (0.195) (0.436) Observation 2686 2686 2686 2686 2686 2686 2686 2686 2686 No. area units 158 158 158 158 158 158 158 158 158 R-squared 0.704 0.705 0.706 0.703 0.704 0.704 0.704 0.705 0.706 Total effect of Ins. payment 0.004 0.005 0.070* 0.000 −0.000 0.005 0.004 0.004 0.128*** (0.004) (0.004) (0.039) (0.001) (0.001) (0.009) (0.004) (0.004) (0.032) Notes: ***, **, * Significant at 1%, 5% and 10% levels, respectively. AU cluster and robust standard errors are shown in parentheses. All regressions are estimated with AU and quarter-fixed effect. Columns (1)–(3) measure the ‘insurance payment’ for building damages (per AU and quarter), Columns (4)–(6) measure the payments for land damage and Columns (7)–(9) sum up the building, contents and land payments together. In all columns, the dependent variable is the same. Open in new tab The estimated coefficients for the main variable are small and insignificant when other insurance variables are not included (Columns 1, 4 and 7), or when we also control for the time of claim settlements, and the proportion of claims paid in cash (Columns 2, 5 and 8). Nevertheless, the insurance payment variables are positive and significant when we control for the interaction between the amount of insurance payment and the other variables characterizing those payments (time to settlement, and proportion in cash—see Columns 3, 6 and 9). The estimated coefficients are positive and are statistically significant especially for the largest exposure (building damage). Not surprisingly, payments for damage to contents (not shown) and to land (Columns 4–6), which are relatively small, are associated with very small coefficient estimates, and do not have any statistically discernible impact on recovery. In the bottom row of each column (each regression) we aggregate the impact of insurance payments, by also accounting for the interaction effects with the time and cash variables. When the insurance payment for building damage increase by 1% in an AU, for that quarter, the economic recovery of residential areas increases by about 0.70%, on average. This finding is important. It is the first time, as far as we know, that detailed post-catastrophe insurance payments are empirically linked with better local economic recovery outcomes.18 The effect of the settlement time variable on the outcome variable is negative, as we hypothesized. Both the variable in levels, and its interaction term with payment size are negative and significant (Columns 3 and 9) for claims for building damage. In other word, the positive impact of the claim amount is reduced when the settlement process is delayed—i.e. delayed payments are less helpful in generating increased economic activity. This might be because with delayed payments the owner of a ‘delayed’ property may have already moved elsewhere or had already fixed her house without insurance funding, but to a lower standard. The coefficient of the proportion of cash settlement variable19 is negative and statistically significant for building structure and total assets (Columns 2 and 8). It was suggested, in New Zealand and elsewhere, that cash payments of insurance claims enable recipients to move away and not rebuild. While we find some supportive evidence of that in these results, once we include the interaction terms (between cash proportions and the aggregate amount of payments) we no longer find this negative effect. If anything, our regression results show evidence that payments in cash did not necessarily hinder (nor assist) in the process of recovery. In the specifications in Table 4, we also control for the variations across time using the quarter dummies. The coefficients of the quarter dummies are large and volatile for the first 2 years after the CES, their coefficient estimates become smaller in absolute term from 2014 onward. Economic recovery in residential areas occurred mainly in 2012 and 2013 and the recovery rate thus declines as time passes. To further test the robustness of our results, we re-ran similar specifications using spatial panel models—this allows us to control for the spatial dependencies in the regression set. Online Appendix Tables 4–6 report the estimation results examining the effect of insurance payment on local recovery for the different spatial econometric models described in the Online Appendix. The finding using the spatial models are quite similar to the results of the non-spatial regressions presented in Table 4. Building and land specifications have significant coefficients, while content regressions do not. The payment * time interaction term is, as was the case in previous specifications, negative and statistically significant. We carry out model selection tests (Anselin et al., 1996; Olivia et al., 2009; Belotti et al., 2016; Noy et al., 2016), these support the SDM model specification. In addition, we also implement the Hausman test for the spatial panel model to test whether random effect models are preferred. The estimated spatial autoregressive ρ and autocorrelation  (λ) coefficients are significant. The economic recovery of an AU is positively influenced by the recovery of other surrounding AUs. Because the estimation coefficients of the specifications cannot be compared with each other, we derive the direct and spillover effects.20 In general, a 1% increase in insurance payment directly leads to 0.4–0.5% increase in residential recovery. However, this direct positive effect would be reduced when the claim settlement was delayed. If the spatial regression models include the endogenous interaction term  WYi,t ⁠, the direct effects contain the feedback effects in their estimates. The feedback effects occur when the impact goes through neighboring AUs and back to the initial AU (LeSage and Pace, 2009). In our result, when taking the difference between direct effect and point estimate, the feedback effect only accounts for about 10–12% of the direct effect. Generally, higher insurance payment received in an AU does not only lead to better economic recovery locally, but it also increases the economic growth in neighboring AUs. The spillover effect of the delay in claim payment is also observable in these spatial models. 8. Conclusion Very few papers have examined economic recovery in the longer-term (beyond the first 2 years), and none have looked at the role of insurance in facilitating recovery at the local level. This lacuna is mainly due to the limited availability of the required data at the appropriate frequency and over the longer term. Our contribution to the empirical literature is 2-fold: First, we show one can measure the immediate economic impact and the economic recovery of local areas after a sequence of earthquakes using the change in night-time luminosity. Second, we used data on insurance claim payments to examine the effectiveness of these payments in facilitating recovery. We found that the earthquake damage significantly reduced the nightlight radiance in the immediate aftermath of these events, and that the amount of lights bounced back and even increased in the years that followed. Using the insurance payment information, we found that building claim payments contributed significantly to local residential recovery in the years following the earthquakes. However, prolonged settlement delays (in cases when these delayed occurred) reduced the benefits of these insurance payments. We also found that settling claims in cash (versus doing the required reconstruction) did not change the dynamics of recovery in any material way. We also quantify the positive spillover effects of insurance payments to the recovery of other neighboring AUs. As far as we are aware, the average time it took to settle claims was remarkably longer in Christchurch as almost every residential property that was damaged (and almost all were) was also insured. Yet, delays are by no means unique. Complaints about the time it takes to settle claims appear after almost every large insured event. As other countries increase the penetration rates for insurance for natural hazards, this problem may further exacerbate in other jurisdictions as well. It is also important to note that while public earthquake insurance is less prevalent, and less often used, there are many publicly funded programs for flood insurance in many different countries (and not only in high-income countries). Flood insurance programs may suffer from the same weaknesses as the risk is correlated across larger spatial areas than earthquake risk is. The recent events associated with the 2017 Atlantic Hurricane season (especially Hurricane Harvey, which was the most heavily insured) have amply demonstrated that risk can be correlated across a large densely populated region. The role of insurance in the recovery of Houston should clearly be of concern to policymakers and the residents there, and unfortunately, in future events that are bound to occur in many places around the world. Acknowledgements We thank QuakeCoRE (publication #0370) and the Resilience National Science Challenge for providing funding that supported this work. Funding Funding was received from QuakeCoRE and from the RNSC. Footnotes 1 Liability for land damage is capped at the market value of the land. The local currency cap amounts were NZD 100,000 for dwelling damage and NZD 20,000 for contents. These nominal caps were set in the 1993 Act, and modified in 2019. We convert all currency figures to USD, based on the 2016 yearly average exchange rate. 2 See the Online Data Appendix for more detail about the luminosity data. 3 Several research papers have used night-time light to capture the immediate economic impact of climate disasters (Tanaka et al., 2000; Bertinelli and Strobl, 2013; Mohan and Strobl, 2017; Del Valle et al., 2018). 4 Hashetera et al. (1999) used the illumination intensity before and after the 1999 Marmara earthquake in Turkey to identify the impacted areas. Kohiyama et al. (2004) assessed the immediate impact of the 2001 Gujarat earthquake using night-time light intensity, and claimed that the estimated loss is consistent with their fieldwork information. Other examples include Escudero et al. (2017) and Elliott et al. (2015). 5 AUs are aggregation of mesh-blocks (the smallest geographical unit used by Statistics New Zealand). In urban areas, AUs are often a collection of several city blocks while in rural areas, AUs may be similar to localities or communities according to Statistics NZ. 6 The DMSP and VIIRS data are not comparable; even after radiometric inter-calibration undertaken by NOAA, comparison is impossible as the images were acquired at different times at night. 7 We chose to aggregate the data to a quarterly AU panel in order to smooth out spatial and temporal volatility (see figure 3). In principle, we could have conducted the analysis per pixel or per mesh-block (the smallest spatial unit for which data are collected), but the existing evidence suggests that nightlight will not be a good proxy for economic activity at such a high resolution. For the monthly model, none of the variables of interest seems to be statistically significant at that high frequency (these results are available in the Online Appendix Table 7). 8 This estimated damage cost is the total payment that EQC and private insurers would have transferred to the claimants (as insurance liability was based on replacement costs rather than the value of damage). 9 The insurance claim payments across exposures (building, land and content) are highly correlated (Online Appendix Table 2). Online Appendix Figure 4 provides the breakdown of EQC claims across districts and the separate earthquakes in 2010–2011. Even though the epicenter of the first event was located further away from Christchurch City, the number of valid claims for the first earthquake is nearly as high as for the latter one. 10 EQC does not only covered for the visible land damage, but the scheme has also been found liable for ground improvement works or long-term reduction of property values due to increased flood and liquefaction vulnerabilities generated by the earthquakes. EQC is the only insurance scheme globally that offers compensation for such risks. 11 A fuller representation of all quarters is available as a short stop-motion clip at: https://www.youtube.com/watch?v=9wdjOcP9XGk. 12 Seismologists have started to produce detailed shake maps for major earthquakes. The maps capture the exact spatial extend of earth surface movements and their decay in magnitude across space (that decay is not linear in distance and depends on surface conditions). 13 Including the few residential CBD observations does not change any of the results reported below. 14 As almost all houses were insured, the deductible was very low, and almost all houses incurred some damage (even if minor), this sum approximates quite closely the total value of all residential assets in each AU. 15 We assume that the effect of earthquakes’ physical intensity (Zi) on Economic_Lossieq only come from our endogenous explanatory variables— Damagei,k ⁠. When we run the tests of endogeneity, the null hypothesis (H0: damage ratio variable is exogenous) is rejected. 16 So that their log value will be equal to zero. 17 Spatial models have been used in economic geography, urban and regional science (Baltagi and Li, 2004; Kelejian and Piras, 2014; Firmino et al., 2016; Noy et al., 2016). 18 Von Peter et al. (2012), in a widely cited paper, found an association between overall insurance coverage and post disaster GDP growth at the national level. 19 The variable is excluded in the content specification because all the content payments were settled in cash. 20 To obtain the direct and spillover effects estimates, we use the variation of 500 simulated parameter combinations drawn from the multivariate normal distribution implied by the maximum-likelihood estimation. This procedure is widely used in spatial statistic inferences (LeSage and Pace, 2009; Vega and Elhorst, 2015). References Anselin L. , Bera A. K. , Florax R. , Yoon M. J. ( 1996 ) Simple diagnostic tests for spatial dependence . Regional Science and Urban Economics , 26 : 77 – 104 . Google Scholar Crossref Search ADS WorldCat Anselin L. , LeGallo J. , Jayet H. ( 2008 ) Spatial panel econometrics. In Matyas L. , Sevestre P. (eds) The Econometrics of Panel Data, Fundamentals and Recent Developments in Theory and Practice , 3rd, pp. 901 – 969 . Dordrecht : Kluwer . Google Preview WorldCat COPAC Baltagi B. H. , Li D. ( 2004 ) Prediction in the panel data model with spatial autocorrelation. In Anselin L. , Florax R. , Rey S. J. (eds) Advances in Spatial Econometrics: Methodology, Tools, and Applications , pp. 283 – 295 . Berlin : Springer . Google Preview WorldCat COPAC Belotti F. , Hughes G. , Mortari A. P. ( 2016 ) Spatial panel data models using Stata . CEIS Tor Vergata - Research Papers Series , 14 . WorldCat Bertinelli L. , Strobl E. ( 2013 ) Quantifying the local economic growth impact of Hurricane Strikes: an analysis from outer space for the Caribbean . Journal of Applied Meteorology and Climatology , 52 : 1688 – 1697 . Google Scholar Crossref Search ADS WorldCat Besley T. , Reynal-Querol M. ( 2014 ) The legacy of historical conflict: evidence from Africa . American Political Science Review , 108 : 319 – 336 . Google Scholar Crossref Search ADS WorldCat Bickenbach F. , Bode E. , Nunnenkamp P. , Söder M. ( 2016 ) Night lights and regional GDP . Review of World Economics , 152 : 425 – 447 . Google Scholar Crossref Search ADS WorldCat Brown C. , Seville E. , Vargo J. ( 2013 ) The Role of Insurance in Organisational Recovery Following the 2010 and 2011 Canterbury Earthquakes. Resilient Organisations Research Report 2013/04, University of Canterbury. Brown C. , Seville E. , Vargo J. ( 2016a ) Efficacy of insurance for organisational disaster recovery: case study of the 2010 and 2011 Canterbury Earthquakes . Disasters , 41 : 388 – 408 . Google Scholar Crossref Search ADS WorldCat Brown M. , Guin B. , Kirschenmann K. ( 2016b ) Microfinance banks and financial inclusion . Review of Finance , 20 : 907 – 946 . Google Scholar Crossref Search ADS WorldCat Bruederle A. , Hodler R. ( 2017 ) Nighttime lights as a proxy for human development at the local level . PLoS ONE , 13 : e0202231 . Google Scholar Crossref Search ADS WorldCat Chang S. E. ( 2010 ) Urban disaster recovery: a measurement framework and its application to the 1995 Kobe Earthquake . Disasters , 34 : 303 – 327 . Google Scholar Crossref Search ADS PubMed WorldCat Chen X. , Nordhaus W. ( 2011 ) Using luminosity data as proxy for economic statistics . Proceedings of the National Academy of Sciences , 108 : 8589 – 8594 . Google Scholar Crossref Search ADS WorldCat Cole M. A. , Elliott R. J. R. , Okubo T. , Strobl E. ( 2017 ) Natural disasters and spatial heterogeneity in damages: the birth, life and death of manufacturing plants . Journal of Economic Geography, 19: 373 – 408 . WorldCat Del Valle A. , Elliott R. J. R. , Strobl E. , Tong M. ( 2018 ) The short-term economic impact of tropical cyclones: satellite evidence from Guangdong Province . Economics of Disasters and Climate Change , 2 : 225 – 235 . Google Scholar Crossref Search ADS WorldCat Doll C. , Muller J. , Morley J. ( 2006 ) Mapping regional economic activity from night-time light satellite imagery . Ecological Economics , 57 : 75 – 92 . Google Scholar Crossref Search ADS WorldCat duPont Iv W. , Noy I. , Okuyama Y. , Sawada Y. ( 2015 ) The long-run socio-economic consequences of a large disaster: the 1995 Earthquake in Kobe . PLoS ONE , 10 : e0138714 . Google Scholar Crossref Search ADS PubMed WorldCat Elhorst J. P. ( 2014 ) Spatial Econometrics—From Cross-Sectional Data to Spatial Panels . Heidelberg : Springer . Google Preview WorldCat COPAC Elliott R. J. R. , Strobl E. , Sun P. ( 2015 ) The local impact of typhoons on economic activity in China: a view from outer space . Journal of Urban Economics , 88 : 50 – 66 . Google Scholar Crossref Search ADS WorldCat Elvidge C. , Sutton P. , Ghosh T. , Tuttle B. , Baugh K. , Bhaduri B. , Bright E. ( 2009 ) A global poverty map derived from satellite data . Computers & Geosciences , 35 : 1652 – 1660 . Google Scholar Crossref Search ADS WorldCat EQC. ( 2016 ) General Reports: Insurance Liability Valuation. Melville Jessup Weaver. EQC. ( 2017 ) Canterbury Earthquakes—Progress and Updates. Wellington, New Zealand: The Earthquake Commission. Escudero I. , Anil O. , Montero M. , Jose J. , Strobl E. ( 2017 ) The impact of Hurricane strikes on short-term local economic activity: evidence from nightlight images in the Dominican Republic. Policy Research Working Paper Series 8275, The World Bank. Felbermayr G. , Gröschl J. ( 2014 ) Naturally negative: the growth effects of natural disasters . Journal of Development Economics , 111 , 92 – 106 . Google Scholar Crossref Search ADS WorldCat Felbermayr G. , Gröschl J. K. , Sanders M. , Schippers V. , Steinwachs T. ( 2018 ) Shedding light on the spatial diffusion of disasters. CESifo Working Paper Series 7146 . WorldCat Firmino D. C. d. S. , Elhorst J. P. , Neto R. d. M. S. ( 2016 ) Urban and rural population growth in a spatial panel of municipalities . Regional Studies , 51 : 894 – 908 . Google Scholar Crossref Search ADS WorldCat GeoNet. ( 2011 ) M6.3, Christchurch, February 22, 2011. Available online at: http://info.geonet.org.nz/display/quake/M+6.3%2C+Christchurch%2C+22+February+2011. Ghosh T. , Anderson S. , Powell R. , Sutton P. , Elvidge C. ( 2009 ) Estimation of Mexico’s informal economy and remittances using night-time imagery . Remote Sensing , 1 : 418 – 444 . Google Scholar Crossref Search ADS WorldCat Ghosh T. , Powell R. , Elvidge C. , Baugh K. , Sutton P. , Anderson S. ( 2010 ) Shedding light on the global distribution of economic activity . The Open Geography Journal , 3 : 148 – 161 . WorldCat Gillespie T. W. , Frankenberg E. , Chum K. F. , Thomas D. ( 2014 ) Night-time lights time series of Tsunami damage, recovery and economics metrics in Sumatra, Indonesia . Remote Sensing Letters , 5 : 286 – 294 . Google Scholar Crossref Search ADS PubMed WorldCat Hashetera S. , Kohiyama M. , Maki N. , Hayashi H. , Matsuoka M. ( 1999 ) Use of DMSP-OLS images for early identification of impacted areas due to the 1999 Marmara Earthquake Disaster. Proceedings of the 20th Asian Conference on Remote Sensing, 2: 1291 – 1296 . Henderson J. V. , Squires T. , Storeygard A. , Weil D. ( 2018 ) The global distribution of economic activity: nature, history, and the role of trade . Quarterly Journal of Economics , 133 : 357 – 406 . Google Scholar Crossref Search ADS WorldCat Hodler R. , Raschky P. ( 2014a ) Economic shocks and civil conflict at the regional level . Economics Letters , 124 : 530 – 533 . Google Scholar Crossref Search ADS WorldCat Hodler R. , Raschky P. ( 2014b ) Regional favoritism . Quarterly Journal of Economics , 129 : 995 – 1033 . Google Scholar Crossref Search ADS WorldCat Kahn E. M. ( 2005 ) The death toll from natural disasters: the role of income, geography, and institutions. The Review of Economics and Statistics , 87 : 271 – 284 . Google Scholar Crossref Search ADS WorldCat Kelejian H. H. , Piras G. ( 2014 ) Estimation of spatial models with endogenous weighting matrices, and an application to a demand model for cigarettes . Regional Science and Urban Economics , 46 : 140 – 149 . Google Scholar Crossref Search ADS WorldCat Klomp J. ( 2016 ) Economic development and natural disasters: a satellite data analysis . Global Environmental Change , 36 : 67 – 88 . Google Scholar Crossref Search ADS WorldCat Kohiyama M. , Hayashi H. , Maki N. , Higashida M. , Kroehl H. , Elvidge C. ( 2004 ) Early damaged area estimation system using DMSP-OLS night-time imagery . International Journal of Remote Sensing , 25 : 2015 – 2036 . Google Scholar Crossref Search ADS WorldCat Kulkarni R. , Haynes K. , Stough R. , James R. ( 2011 ) Revisiting night lights as proxy for economic growth: a multi-year light based growth indicator (LGBI) for China, India and the U.S. GMU School of Public Policy—Research Paper. Kusuma A. , Nguyen C. , Noy I. ( 2019 ) Insurance for catastrophes: why are natural hazards underinsured, and does it matter? In Okuyama Y. , Rose A. (eds) Advances in Spatial and Economic Modeling of Disaster Impacts . Springer Science & Business Media . Google Preview WorldCat COPAC Lee L. F. , Yu J. ( 2010 ) Some recent developments in spatial panel data models . Regional Science and Urban Economics , 40 : 255 – 271 . Google Scholar Crossref Search ADS WorldCat LeSage J. , Pace K. ( 2009 ) Introduction to Spatial Econometrics . Boca Raton (FL ): CRC Press . Google Preview WorldCat COPAC Melecky M. , Raddatz C. ( 2015 ) Fiscal responses after catastrophes and the enabling role of financial development . World Bank Economic Review , 29 : 129 – 149 . Google Scholar Crossref Search ADS WorldCat Michalopoulos S. , Papaioannou E. ( 2013 ) Pre-colonial ethnic institutions and contemporary African development . Econometrica , 81 : 113 – 152 . Google Scholar Crossref Search ADS PubMed WorldCat Mohan P. , Strobl E. ( 2017 ) The short-term economic impact of tropical cyclone Pam: an analysis using VIIRS Nightlight Satellite Imagery . International Journal of Remote Sensing , 38 : 5992 – 6006 . Google Scholar Crossref Search ADS WorldCat Montalvo J. , Reynal-Querol M. ( 2016 ) Ethnic diversity and growth: revisiting the evidence. Economics Working Papers 1585, Department of Economics and Business, Universitat Pompeu Fabra. Nguyen C. , Noy I. ( 2019 ) Insuring earthquakes: how would the Californian and Japanese insurance programs have fared after the 2011 New Zealand earthquake? Disasters , forthcoming. NOAA. ( 2013 ) Visible Infrared Imaging Radiometer Suite (VIIRS) Sensor Data Record (SDR) User’s Guide. NOAA Technical Report NESDIS 142. Noy I. , duPont W. ( 2018 ) The long-term consequences of disasters: what do we know, and what we still don’t . International Review of Environmental and Resource Economics , 12 : 325 – 354 . Google Scholar Crossref Search ADS WorldCat Noy I. , Taupo T. , Cuffe H. ( 2016 ) Household vulnerability on the frontline of climate change in Tuvalu . Environmental Economics and Policy Studies , 20 : 705 – 739 . WorldCat Olivia S. , Gibson J. , Smith A. D. , Rozelle S. , Deng X. ( 2009 ) An empirical evaluation of poverty mapping methodology: explicitly spatial versus implicitly spatial approach. Paper presented at the Australian Agricultural and Resource Economics Society, Cairns, Australia. Pinkovskiy M. , Sala-i-Martin X. ( 2016 ) Lights, camera … income! Illuminating the national accounts-household surveys debate . The Quarterly Journal of Economics , 131 : 579 – 631 . Google Scholar Crossref Search ADS WorldCat Platt S. , Brown D. , Hughes M. ( 2016 ) Measuring resilience and recovery . International Journal of Disaster Risk Reduction , 19 : 447 – 460 . Google Scholar Crossref Search ADS WorldCat Poontirakul P. , Brown C. , Noy I. , Seville E. , Vargo J. ( 2017 ) Insurance as a double-edged sword? Quantitative evidence from the 2011 Christchurch Earthquake . Geneva Papers on Risk and Insurance , 42 : 609 – 632 . Google Scholar Crossref Search ADS WorldCat Rose A. , Benavides J. , Chang S. E. , Szczesniak P. , Lim D. ( 1997 ) The regional economic impact of an earthquake: direct and indirect effects of electricity lifeline disruptions . Journal of Regional Science , 37 : 437 – 458 . Google Scholar Crossref Search ADS WorldCat Sawada Y. , Shimizutani S. ( 2008 ) How do people cope with natural disasters? Evidence from the Great Hanshin-Awaji (Kobe) earthquake in 1995 . Journal of Money, Credit and Banking , 40 : 463 – 488 . Google Scholar Crossref Search ADS WorldCat Seville E. , Stevenson J. R. , Brown C. , Giovinazzi S. , Vargo J. ( 2014 ) Disruption and Resilience: How Organizations Coped with the Canterbury Earthquakes. ERI Research Report. Skoufias E. , Strobl E. , Tveit T. B. ( 2017 ) Natural disaster damage indices based on remotely sensed data: an application to Indonesia. Policy Research Working Paper Series 8188, World Bank. Stevenson J. R. , Seville E. , Kachali H. , Vargo J. , Whitman Z. ( 2011 ) Post-disaster organisational recovery in a central business district context: the 2010 & 2011 Canterbury Earthquakes. Resilient Organisations Research. Storeygard A. ( 2016 ) Farther on down the road: transport costs, trade and urban growth in sub-Saharan Africa . Review of Economic Studies , 83 : 1263 – 1295 . Google Scholar Crossref Search ADS PubMed WorldCat Sutton P. , Costanza R. ( 2002 ) Global estimates of market and non-market values derived from night-time satellite imagery, land cover, and ecosystem service valuation . Ecological Economics , 41 : 509 – 527 . Google Scholar Crossref Search ADS WorldCat Sutton P. , Elvidge C. , Ghosh T. ( 2007 ) Estimation of gross domestic product at sub-national scales using night-time satellite imagery . International Journal of Ecological Economics & Statistics , 8 : 5 – 21 . WorldCat Tanaka M. , Sugimura T. , Tanaka S. ( 2000 ) Monitoring water surface ratio in the Chinese floods of summer 1998 By DMSP-SSM/I . International Journal of Remote Sensing , 21 : 1561 – 1569 . Google Scholar Crossref Search ADS WorldCat Toya H. , Skidmore M. ( 2007 ) Economic development and the impacts of natural disasters . Economics Letters , 94 : 20 – 25 . Google Scholar Crossref Search ADS WorldCat Vega S. H. , Elhorst J. P. ( 2015 ) The SLX model . Journal of Regional Science , 55 : 339 – 363 . Google Scholar Crossref Search ADS WorldCat Von Peter G. , Von Dahlen S. , Saxena S. C. ( 2012 ) Unmitigated disasters? New evidence on the macroeconomic cost of natural catastrophes. BIS Working Paper 394, Basel: Bank for International Settlements. © The Author(s) (2019). Published by Oxford University Press. All rights reserved. For permissions, please email: journals.permissions@oup.com This article is published and distributed under the terms of the Oxford University Press, Standard Journals Publication Model (https://academic.oup.com/journals/pages/open_access/funder_policies/chorus/standard_publication_model) http://www.deepdyve.com/assets/images/DeepDyve-Logo-lg.png Journal of Economic Geography Oxford University Press

Measuring the impact of insurance on urban earthquake recovery using nightlights

Journal of Economic Geography , Volume Advance Article – May 1, 2020

Loading next page...
 
/lp/oxford-university-press/measuring-the-impact-of-insurance-on-urban-earthquake-recovery-using-NeZ8CLhNpS

References (66)

Publisher
Oxford University Press
Copyright
© The Author(s) (2019). Published by Oxford University Press. All rights reserved. For permissions, please email: journals.permissions@oup.com
ISSN
1468-2702
eISSN
1468-2710
DOI
10.1093/jeg/lbz033
Publisher site
See Article on Publisher Site

Abstract

Abstract We measure the longer-term effect of a major earthquake on the local economy, using night-time light intensity, and focus on the role of insurance payments for damaged residential property in the recovery process. The destructive Canterbury Earthquake Sequence (2010–2011) in New Zealand is our case study. Uniquely, for this event, >95% of residential housing units were covered by insurance and almost all incurred some damage. However, insurance payments were staggered over 5 years, enabling us to identify their local impact on recovery. We find that night-time luminosity can capture the process of recovery; and that insurance payments contributed significantly to the process of local economic recovery after the earthquake. Cash settlement of claims was no more effective than insurance-managed repairs in generating local recovery. Notably, delayed payments were less affective in assisting recovery; this suggests an important role for the regulator in making sure insurance payments are made promptly after disaster events. 1. Introduction New Zealand is prone to earthquakes. Recent destructive earthquakes in 2010, 2011, 2013 and 2016 have demonstrated the seriousness of this risk, and have shown that the local recovery from such events is often difficult and prolonged. In recent years, numerous papers have looked into the immediate restoration of activity after disasters, often from a microeconomic, single household, perspective or by focusing on a specific case-study (Rose et al., 1997; Sawada and Shimizutani, 2008; Chang, 2010; duPont Iv et al., 2015; Cole et al., 2017). However, the absence of detailed and sufficiently frequent microeconomic data has hindered attempts to shed more light on the dynamics of recovery over longer periods of time except for at the macroeconomic level. Moreover, the insurance sector is often thought to play a significant role in recovery post-disaster, but analysis of its precise functioning during the recovery process is rarely if ever pursued. Insurance is frequently mentioned as (almost) a panacea for disaster risk, and it is singled out as an important part of international disaster risk reduction efforts. Yet, except for Von Peter et al. (2012) and Poontirakul et al. (2017), there is no research that attempts to look into the role of insurance claim payments in post-catastrophe recovery. Our aim here is to provide a first attempt at measurement of the longer-term economic effect of insurance payments after a major earthquake event, using satellite night-time light as a proxy measure for economic activity. Put differently, we investigate how insurance claim payments for damaged residential property, and specifically their timing, affect the recovery process of the local economy. We focus on the destructive Canterbury Earthquake Sequence (CES) in 2010–2011 as our case study. We chose this event due to the availability of the insurance claim payment data, and specific characteristics of the earthquake and the insurance market in New Zealand that allow us to identify the impact of insurance payments. These are detailed in the next section. We first find that night-time luminosity can capture the earthquake damage and the process of recovery. We then find that the insurance payments contributed significantly to the process of economic recovery after the earthquake; and further emphasize the importance of the timing of payments, with delayed payments being a lot less effective. This specific earthquake sequence is an attractive case study for several reasons: First, the event is unique as >95% of residential housing units were covered by insurance, and practically all submitted a claim. Thus, unlike almost all other disasters where insurance penetration rates are much lower, there is no problem of selection bias (i.e., households that purchase insurance are different from those that do not). Second, these were really big events, from an insurance perspective. Two of five earthquakes in this sequence are listed as some of the costliest insured events, globally, ever. Several geographic aspects of Christchurch make it especially feasible to conduct the analysis by using night-time luminosity—especially noteworthy are the fact that the city is composed of mostly low rise, spread out residential neighborhoods (so that the nightlight sensors are not saturated) and there are many nights of low or no cloud cover (making the measurements more consistent). The remainder of this paper is structured as follows. In the next section we provide information about the earthquake events, insurance in New Zealand and the recovery process. We next discuss the use of nightlight luminosity as a proxy for economic activity and the history of its use in the analysis of disaster impact and recovery. After covering these literatures, we describe the data and methodology used in this paper. In the penultimate section, we present our empirical results; and we end with some further comments about future research. 2. The Canterbury Earthquake Sequence 2010–2011 On 4 September 2010, a M7.1 earthquake occurred, epi-centered close to Darfield village, a rural area not far from the city of Christchurch (the biggest city in the South Island of New Zealand, with a population of about 400,000). The earthquake damaged nearby towns and the eastern suburbs of the city which were vulnerable to liquefaction. Many old unreinforced masonry and heritage buildings were affected. This event was followed by a shallower M6.3 aftershock to the southeast of the city on 22 February 2011. This event resulted in intense ground motions which were directed toward the city center (GeoNet, 2011). Many buildings in the Central Business District (CBD) and elsewhere in the city were severely damaged. There were 185 fatalities in the February 2011 earthquake. Practically all residential buildings in the city experienced at least some damage, with many thousands eventually requiring complete rebuilds. Some areas around the Avon River suffered heavily from subsidence. The flood and liquefaction risk of this area was eventually found to be unacceptably high, and the government decided to re-zone it for non-residential use (a total of around 8000 homes were located in these Red Zones). Following all this, there were numerous aftershocks, which mostly led to additional damage to previously damaged buildings, and to delays in reconstruction. New Zealand has a very high insurance penetration ratio, with >95% of residences being insured for earthquakes (Nguyen and Noy, 2019). The New Zealand Earthquake Commission (EQC) is a public entity providing the first layer of residential insurance cover for earthquakes. The EQC was liable for residential claims that cover dwelling damage up to USD 67,000, content damage up to USD 13,400, and some land damage.1 The residential over-cap (over the EQC cap) and out-of-scope claims for damages (for example to driveways) were handled by the private insurers. Approximately 25,600 residential building over-cap claims were transferred to private insurers to be resolved. EQC (2017) reports that the public insurance scheme has settled over 167,000 and 73,000 valid dwelling and land claims, respectively. These claim settlements cost the EQC approximately USD 7.2 billion (EQC, 2016). The number of submitted claims was twice as large as the EQC planned for as a ‘worst foreseeable event.’ Private insurance companies also had limited experience handling such a large number of claims prior to this event, and almost no experience coordinating their work with the EQC. Further complications were the large number of aftershocks spread all over the city, many previously unacknowledged ambiguities in insurance contracts, complex cover for land damage that is not available in other jurisdictions and a legal system that was overwhelmed post-earthquake. These complications led to an insurance settlement process that has taken over 7 years to complete, and as of this writing (end of 2019) there are still some claims yet to be settled. These spatially random delays in claim settlements allow us to identify the effect of insurance payments on recovery. Three other aspects of the insurance system in Christchurch are helpful in establishing the identification of the causal channel from insurance payments to recovery: (1) Almost everyone had residential insurance in Christchurch. No one knows the exact number of uninsured properties, but from the Residential Red Zone properties for which we do have data, and which we view as a random sample of 8000 properties, the insurance penetration rate was about 98%. (2) Almost all properties incurred some damage, even if minor, and the excess (deductible) in the EQC contract was very low (USD 134). Therefore, almost everyone made a claim to the public insurer in Christchurch. (3) Earthquake risk was considered very low in Christchurch; in the pre-2010 New Zealand Government’s seismic risk zone maps (maps that determine the benchmark for building standards), Christchurch City was labeled as low risk (out of three levels, it had the lowest risk level). We therefore do not expect that seismic preparedness, beyond what is always mandated by building standards, was undertaken by the majority of households, so that damages were not endogenous to the characteristics of the households/owners. Several other research projects have looked at this event. Similar to residential insurance, commercial insurance claim settlement also faced delays due to the scale of claim handling, the complexity of claims, the ongoing seismicity and the shortage of experienced loss assessors. Additional reasons for delay in the assessment process include poor information management by both sides, slow decision-making and the use of brokers as intermediaries (Brown et al., 2013, 2016a; Seville et al., 2014). Using surveys, Stevenson et al. (2011) found that affected organizations financed their recovery primarily with their cash-flow instead of from the proceeds of claim payments. With these same surveys, Poontirakul et al. (2017) find no short-run difference in likelihood of business survival between insured and uninsured firms. However, later on, firms which had prompt and full claim payment experienced better recovery—in terms of performance and profitability—than those that had insufficient cover or delayed claim settlements; and the latter firms performed marginally worse than uninsured firms. 3. Insurance and disaster recovery elsewhere The literature on the economics of disasters has grown significantly in recent years, especially in its investigation of the varied immediate impacts of disasters. Yet, relatively less is known about the post-disaster recovery process and the factors that shape it—see Noy and duPont (2018) for a survey of the existing literature. Very few papers have closely looked at the role of insurance. This research has largely focused on explaining insurance penetration, while the insurance companies’ own research has concentrated on estimating expected disaster loss rather than on measuring their role in the recovery process (Kusuma et al., 2019). Melecky and Raddatz (2015) find that high- and middle-income countries, which have high insurance penetration, are affected less and experience better economic recovery following a disaster; and similar findings are reported in Von Peter et al. (2012)—but both only use aggregate country level data on insurance penetration. Platt et al. (2016) described the use of a wide range of data sources to identify the speed and the quality of recovery after major earthquakes. These include satellite imagery, crowd-sourced data, ground and household surveys, official statistics and insurance data. They conclude that remote sensing appears to provide accurate and reliable information. 4. Night-time luminosity in economic research In the past decade, night-time light has been increasingly used in the social science literature as an indicator for economic activity and human development. Because most consumption and household activities require illumination in the evening, using changes in light intensity as a proxy for per capita GDP growth appears to be feasible. When household income increases, its light usage also increases (i.e. lighting is a normal good); and studies showing the relationship between night-time luminosity and socioeconomic information are numerous (Sutton and Costanza, 2002; Doll et al., 2006; Sutton et al., 2007; Elvidge et al., 2009; Ghosh et al., 2009,, 2010; Chen and Nordhaus, 2011; Kulkarni et al., 2011; Michalopoulos and Papaioannou, 2013; Hodler and Raschky, 2014a; Pinkovskiy and Sala-i-Martin, 2016). In all these papers, night-time luminosity data are obtained from the DMSP/OLS or VIIRS DNB satellites.2 Luminosity data have been used to measure income at the sub-national level at various grid-cell sizes (Besley and Reynal-Querol, 2014; Montalvo and Reynal-Querol, 2016; Storeygard, 2016; Bruederle and Hodler, 2017; Henderson et al., 2018), projected onto cities and municipal boundaries (Brown et al., 2016b) and for administrative regions (Hodler and Raschky, 2014a, 2014b; Bickenbach et al., 2016). The correlation between the night-time light and economic activity tends to be weaker at very small unit levels (e.g. one pixel), so some aggregation appears to be necessary. Some have used night-time light in order to investigate the economic losses and recovery post disaster event. For instance, Klomp (2016) explores how large-scale disasters affect economic activity, using night-time light intensity and historic data on 1000 natural hazard events between 1992 and 2008. He finds that geophysical and meteorological events reduce night-time illumination in developed countries while hydrological and climatic disasters lead to a short-term decline in the light intensity in developing countries, and that earthquakes have more prolonged negative effects.3 Similar findings, using geo-coded indicators of disaster intensity, are reported by Felbermayr et al. (2018). Gillespie et al. (2014) used household survey data (2004–2007) in Sumatra during the recovery from its 2004 earthquake/tsunami to reveal the link between night-time luminosity and spending per capita at the community level, and Skoufias et al. (2017) did something similar for more disasters in Indonesia at the district level. They both suggest that satellite night-time imagery is a useful tool for assessing the post disaster impacts.4 5. Data Greater Christchurch includes Christchurch city and its satellite towns. According to the 2006 Census, the region’s resident population count was nearly 425,000 with 82% living in Christchurch City. We aggregate and analyze all data at the area unit (AU) level.5 Based on the 2016 Geographic Boundary of Statistics New Zealand, there are 183 AUs in Greater Christchurch, containing 158 AUs defined as residential areas. 5.1. Night-time light data We use night-time light data derived from images taken by DMSP/OLS and VIIRS DNB satellites. We convert the images to integer format at the pixel level. Because each AU has different size and can cover several pixels, we calculate the nightlight intensity weighted mean within each AU polygon. The relative scales of nightlight pixels and AU are illustrated in Online Appendix Figure 1. The average spatial area for an AU in Greater Christchurch is approximately 55 km2. Thus, even within the city, where AUs cover less ground area, each AU may contain more than 10 pixels. Figures 1 and 2 present the night-time light images of Greater Christchurch from the 2013 cloud-free composite DMSP/OLS and VIIRS DNB satellites, respectively. The brightly lit area in the figures corresponds to Christchurch City. It is noticeable that the DMSP data have saturation centered on the city area while the VIIRS shows much more detail. The latter has a better spatial resolution, about 750 m to the 2.7 km resolution of the former (NOAA, 2013). Due to the difference in time coverage of the two datasets, both are used in this paper, but for the main results on insurance and recovery, we use the higher-resolution VIIRS data.6 Specifically, the DMSP data of satellite F16 and F18, from 2009 to 2012, are used to capture the reduction in nightlights as the indicator of short-run disaster impact. The DMSP data are publicly available at annual frequency only. We then use quarterly VIIRS data for the period from 2012 to 2016 for each AU for the recovery trajectory.7 Figure 1 Open in new tabDownload slide Raw image of average DMSP/OLS night-time light in 2013. Figure 1 Open in new tabDownload slide Raw image of average DMSP/OLS night-time light in 2013. Figure 2 Open in new tabDownload slide Raw image of average VIIRS DNB night-time light in 2013. Figure 2 Open in new tabDownload slide Raw image of average VIIRS DNB night-time light in 2013. Figure 3 Open in new tabDownload slide Average VIIRS DNB night-time light in 2013 aggregated to the area unit level. Figure 3 Open in new tabDownload slide Average VIIRS DNB night-time light in 2013 aggregated to the area unit level. 5.2. Insurance claim data To measure the payments provided by the insurance sector during the recovery, we use the geo-coded payment data from the EQC. The dataset includes individual claims for earthquake events during 2010–2011. For each insured event, the data provide the actual amount that EQC has spent on each property and the estimated total damage cost as it was apportioned for each earthquake.8 We use records of approximately 220,000 claims for nearly 100,000 properties in Greater Christchurch. More than 85% of these claims came from Christchurch city. Three-fourths of the claims are for building structure and the rest are for land and content exposure.9 Table 1 provides summary statistics of quarterly claim payment data at the AU level. In Greater Christchurch, the average total of quarterly claim payments for each exposure per AU are USD 462,696, USD 17,347 and USD 60,240 for structure, content and land, respectively. The standard deviation of land claim payments is high, relative to its mean, as there are claims with high land remediation cost due to land movement and cliff collapse.10 Table 1 Summary statistics of claim payment data, by asset damage type Variables Building Content Land Total (N = 143,545) (N = 68,324) (N = 73,123) (N = 220,898) Mean Standard deviation Mean Standard deviation Mean Standard deviation Mean Standard deviation Total claim payment (USD) 462,695 696,423 17,347 43,840 60,240 1,424,564 540,284 1,642,358 Total exposed value of the assets (USD) 6,680,840 7,645,051 274,319 520,982 694,532 2,844,193 7,651,877 9,406,143 Proportion of cash paid/total settlement 0.73 0.26 1.00 0.00 0.55 0.41 0.71 0.35 Time to settlement (days) 845 538 489 439 688 514 984 542 Variables Building Content Land Total (N = 143,545) (N = 68,324) (N = 73,123) (N = 220,898) Mean Standard deviation Mean Standard deviation Mean Standard deviation Mean Standard deviation Total claim payment (USD) 462,695 696,423 17,347 43,840 60,240 1,424,564 540,284 1,642,358 Total exposed value of the assets (USD) 6,680,840 7,645,051 274,319 520,982 694,532 2,844,193 7,651,877 9,406,143 Proportion of cash paid/total settlement 0.73 0.26 1.00 0.00 0.55 0.41 0.71 0.35 Time to settlement (days) 845 538 489 439 688 514 984 542 Notes: Claim payments are aggregated at AU level and quarterly frequency. Summary statistics of variable ‘time to settlement’ are first calculated at the individual claim level. Open in new tab Table 1 Summary statistics of claim payment data, by asset damage type Variables Building Content Land Total (N = 143,545) (N = 68,324) (N = 73,123) (N = 220,898) Mean Standard deviation Mean Standard deviation Mean Standard deviation Mean Standard deviation Total claim payment (USD) 462,695 696,423 17,347 43,840 60,240 1,424,564 540,284 1,642,358 Total exposed value of the assets (USD) 6,680,840 7,645,051 274,319 520,982 694,532 2,844,193 7,651,877 9,406,143 Proportion of cash paid/total settlement 0.73 0.26 1.00 0.00 0.55 0.41 0.71 0.35 Time to settlement (days) 845 538 489 439 688 514 984 542 Variables Building Content Land Total (N = 143,545) (N = 68,324) (N = 73,123) (N = 220,898) Mean Standard deviation Mean Standard deviation Mean Standard deviation Mean Standard deviation Total claim payment (USD) 462,695 696,423 17,347 43,840 60,240 1,424,564 540,284 1,642,358 Total exposed value of the assets (USD) 6,680,840 7,645,051 274,319 520,982 694,532 2,844,193 7,651,877 9,406,143 Proportion of cash paid/total settlement 0.73 0.26 1.00 0.00 0.55 0.41 0.71 0.35 Time to settlement (days) 845 538 489 439 688 514 984 542 Notes: Claim payments are aggregated at AU level and quarterly frequency. Summary statistics of variable ‘time to settlement’ are first calculated at the individual claim level. Open in new tab We also exploit other information in the claim data; in particular, we focus on two variables: time it took to settle the claim and proportion of cash in settlement. The first is the average number of days to claim settlement, since the day the claim was launched, for each quarter in each AU. Online Appendix Figure 7 presents the temporal distribution of payments (per AU) for Q2 of each of the years following the earthquakes.11 There is no apparent spatial pattern in the temporal distribution of the claim payments, suggesting there is most likely no selection problem in the time-to-payment variable. It took 1–4 years for most claims to be resolved (average is nearly 3 years). The second variable is the proportion of cash payment amount relative to the total claim settlement in each AU. Table 1 illustrates that 71% of the amount the insurer paid to claimants was in cash, while the number of cash-paid claims was 60% (EQC, 2017). Figures 4 and 5 describe the association between the insurance claim data and the measured nightlights, both for the immediate impact phase in Figure 4, and for the longer-term recovery period (in Figure 5). Figure 4 Open in new tabDownload slide Change in nightlight from 2009 to 2011 and damage ratio. Note: The y-axis variable is described in Equation (1) and the x-axis variable is defined in Equation (2). Figure 4 Open in new tabDownload slide Change in nightlight from 2009 to 2011 and damage ratio. Note: The y-axis variable is described in Equation (1) and the x-axis variable is defined in Equation (2). Figure 5 Open in new tabDownload slide Quarterly change in NTL and same quarter insurance payments. Notes: The y-axis variable is described in Equation (4) and the x-axis variable is defined in Equation (5). Figure 5 Open in new tabDownload slide Quarterly change in NTL and same quarter insurance payments. Notes: The y-axis variable is described in Equation (4) and the x-axis variable is defined in Equation (5). 5.3. Other variables We also use data from Statistics New Zealand on households, at the AU level, from the most recent census conducted before the earthquakes (in 2006)—this data are summarized in Table 2. Online Appendix Table 3 illustrates the correlation between the 2006 nightlight and control variables from the 2006 Census, matched at the AU level. There are positive correlations between light luminosity as a measure of economic activity and most explanatory variables. Online Appendix Table 3 also shows that light intensity captures the density variables better. For instance, the correlation between nightlight and population is 0.59, while the correlation between nightlight and population density is 0.71. There is also positive relationship between light brightness and income density; and nocturnal light is negatively correlated with the distance between the AUs and the city center. Table 2 Summary statistics of area units (AU) in Greater Christchurch Variables Christchurch city Waimakariri and Selwyn Mean Standard deviation Mean Standard deviation Area in squared kilometer 11.3 44.0 148.2 612.4 Night-time population 1755 1183 1643 1255 Night-time population density 2802 1296 514 698 Household income 68,420 18,220 74,171 18,345 Household income density 40,893 28,853 25,097 29,778 Variables Christchurch city Waimakariri and Selwyn Mean Standard deviation Mean Standard deviation Area in squared kilometer 11.3 44.0 148.2 612.4 Night-time population 1755 1183 1643 1255 Night-time population density 2802 1296 514 698 Household income 68,420 18,220 74,171 18,345 Household income density 40,893 28,853 25,097 29,778 Notes: Household income and night-time population are measured using 2006 Census data provided by Statistics New Zealand. The density variables are per km2. Open in new tab Table 2 Summary statistics of area units (AU) in Greater Christchurch Variables Christchurch city Waimakariri and Selwyn Mean Standard deviation Mean Standard deviation Area in squared kilometer 11.3 44.0 148.2 612.4 Night-time population 1755 1183 1643 1255 Night-time population density 2802 1296 514 698 Household income 68,420 18,220 74,171 18,345 Household income density 40,893 28,853 25,097 29,778 Variables Christchurch city Waimakariri and Selwyn Mean Standard deviation Mean Standard deviation Area in squared kilometer 11.3 44.0 148.2 612.4 Night-time population 1755 1183 1643 1255 Night-time population density 2802 1296 514 698 Household income 68,420 18,220 74,171 18,345 Household income density 40,893 28,853 25,097 29,778 Notes: Household income and night-time population are measured using 2006 Census data provided by Statistics New Zealand. The density variables are per km2. Open in new tab We also use Shake-maps for the September 2010 and February 2011 earthquakes, provided by the USGS.12 These maps provide the spatial distribution of the physical intensity for major earthquakes. We aggregate these macro-seismic intensities to the AU level. In all of our estimations, we exclude the CBD, because the area was cordoned off for 2 years, and its redevelopment was subject to a very different, complex and contentious regulatory regime.13 6. Methodology We now turn to the regression analysis where we explore the change in night-time light in the Greater Christchurch region, during and after the CES. The preliminary set of estimations, described in Section 6.1, is intended to establish the short-term impact of earthquake damage on local economic activity in Great Christchurch. The second, the main focus of this work, aims to estimate the effect of insurance payments on the recovery of residential areas in the region (Section 6.2). 6.1. Earthquake damage and the loss in night-time light We begin by using the immediate reduction in luminosity post-earthquakes as an indicator of the loss in economic activity in Greater Christchurch. The variable is calculated as follows: Economic_Lossieq=ΔNTLi2009-2011=ln⁡(NTLi,2009)-ln⁡(NTLi,2011), (1) where NTLi is our proxy for economic development based on the DMSP nightlight value (taken in logarithms) aggregated for each AU (denoted by i). We next aggregate the insurance claim payments over the whole period, to the AU level, to indicate the financial loss experienced by each AU due to earthquake damage. A number of papers in the literature have stressed that earthquake damage is correlated with income per capita (Kahn, 2005; Toya and Skidmore, 2007; Felbermayr and Gröschl, 2014). Hence, even in the spatially confined study at hand, cross-AU heterogeneity in damage may be driven by cross-AU differences in income per capita. To reduce the endogeneity of the financial loss indicator, we create a damage ratio variable from these aggregate figures (in Equation 2). Damagei,k=Claim_paymenti,k∑kAsset_valuei,k. (2) Damagei represents the total earthquake financial loss on all exposures (k = building, content, land or total) as a ratio of the total exposure value for all dwellings for which there were claims.14 The property value data are from the New Zealand Quotable Value (QV). In the first set of results, we use Economic_Lossieq  as a dependent variable indicating the change in economic activity due to the earthquakes. We hypothesize that AUs that have high ΔNTL experienced large economic losses because of the large amount of damage to property (assets), as follows: Economic_Lossieq=α+βkDamagei,k+γXi+εi, (3) where Damagei,k is earthquake damage (building, content, land or total) as a ratio of exposed value for each AU as specified in Equation (2). We use robust standard errors in order to control for the heteroscedasticity in the error terms. In addition, for robustness we include several control variables (⁠ Xi ⁠) that might also affect the measured economic loss in our regressions such as average household income, night-time population, average number of bedrooms in each household and surface area (taken in logarithms). For robustness, because of endogeneity concerns, we also implement a two-stage least squares instrumental variable (2SLS-IV) method. We use the ShakeMap’s earthquake physical intensity measure (Zi) as the instrumental variable for  Damagei,k ⁠. The correlations between damage ratio measured by property damage and the exogenous seismic intensity of February 2011 earthquake are over 50%, except for land damage. Thus, we expect to have strong first stage where the instrument is highly correlated with the endogenous variable.15 6.2. Insurance settlement and Christchurch recovery In the second set of regressions, we estimate the effect of insurance payments on local recovery, following the earthquakes. In this regression, we use the night-time VIIRS light dataset from April 2012 to August 2016 (VIIRS data are not available before that). We convert the nightlight data from monthly to quarter frequency. In order to identify the economic recovery in Greater Christchurch, we take the proportional change in the night-time radiance value for each quarter. In the specification, the variable is used as the dependent variable. Economic_Recovery i,tPost=ΔNTLi,tQ2.2012-Q3.2016=ln⁡(NTLi,t)-ln⁡(NTLi,t-1). (4) The main explanatory (RHS) variable is the insurance payment. It is the total insurance claim pay-out which an AU received at quarter  t ⁠, as described in Equation (5). Insi,t,k=ln⁡(Claim_paymenti,t,k). (5) The regression model is written as follows: Economic_Recovery i,tPost=αi+τt+βkInsi,t,k+γXi,t+εi,t. (6) whereby Insi,t,k is our measure of insurance payments described in Equation (5). Here, we hypothesize that the insurance payments are associated with the quarterly change in nightlight in the years following the earthquakes (Q2 2012–Q3 2016). We included AU and quarter-fixed effects to control for unobserved variations across individual AUs and over time. We also include AU cluster-robust standard errors to control for heteroscedasticity. Other insurance-related variables such as ‘settlement time’ and ‘proportion cash amount’ are also included. In some AUs, there are quarters without claim payments. The value of the insurance-related variables in this case is set as one.16 We also investigate the interaction term between insurance payment and settlement time and cash payments. By all accounts, EQC and private insurers resolved insurance claim on a first come, first serve basis without prioritizing any certain demographic groups or geographic location. Hence, we assume that the claim settlements were processed randomly across locations. In addition, we carry out spatial panel data analysis. Spatial econometric modeling helps us control for spatial effects. They may reduce the unobserved estimation bias which arises from both spatial and time dependence. More importantly, spatial regression methods permit us to identify spillover effects coming from neighboring AUs over time (Anselin et al., 2008; Lee and Yu, 2010; Elhorst, 2014).17 Following this literature, we implement four different spatial specifications including spatial autoregressive model (SAR), spatial error model (SEM), spatial Durbin model (SDM) and spatial autocorrelation model (SAC)—the estimated equations are described Online in Appendix B. We employ spatial panel maximum-likelihood estimation for the set of regression models with AU and quarter-fixed effects as described below. W is a non-negative spatial weighted matrix that describes the spatial structure of dependence between AUs in the sample. These models include three different types of interaction effects among units: (i) endogenous spatial interaction effects for the dependent variable  (WYi,t) ⁠; (ii) exogenous spatial interaction effects among the explanatory variables WXi,t and (iii) spatial interaction effects among the error terms  (Wϑi,t) ⁠. We employ a contiguity weighted matrix. The elements ωij of matrix W equal to 1/the number of neighbors of AU i if AU i and j share a border, otherwise  ωij=0 ⁠. In another version, we also use a matrix where each element ωij is the inverse distance between AU i and j ⁠. 7. Results The first set of regressions results, examining whether earthquake damages explain the reduction in economic activity in Greater Christchurch, is shown in Table 3. The damage ratio variable (for buildings, contents and land damage) is the main explanatory variable in the regressions. In Columns 1–3 of the table, we estimate the effect of residential building damage on the economic activity in the immediate aftermath of the earthquake events. Other columns focus on the damage for content, land and total damage (sum of the three asset classes) in regressions 4–6, 7–9 and 10–12, respectively. In these specifications, the coefficients of damage variables for buildings are always positive and significant (they are mostly positive but insignificant for the content and land damage variables—see Columns 4–9). For instance, in Column 1 of Table 3, the economic loss will be 0.56% higher, when the residential buildings damage over property value increases by 1%. When controlling for other variables (taken in logarithms), the building damage indicators are still statistically significant and of the same order of magnitude. Maybe not surprisingly, the residential building damage appears to explain some of the economic loss immediately after the disaster; and it is the only variable that consistently has explanatory power. Table 3 Short run economic impact of the earthquakes using the damage ratio variable Variables Dependent variable: Change in night-time light between 2010 and 2011 Building Content Land Total (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) Damage ratio 0.559*** 0.416** 0.957** 0.757** 0.379 0.379 0.016 −0.006 0.747 0.474*** 0.343** 0.912** (0.186) (0.171) (0.401) (0.367) (0.389) (0.389) (0.051) (0.068) (0.493) (0.181) (0.162) (0.415) Household income 0.008 0.003 0.007 0.007 0.012 −0.025 0.007 −0.000 (0.037) (0.045) (0.036) (0.036) (0.038) (0.049) (0.037) (0.041) Night-time Population 0.018 0.017 0.019 0.019 0.019 0.021* 0.019 0.017 (0.012) (0.012) (0.012) (0.012) (0.012) (0.012) (0.012) (0.012) Number of bedrooms −0.060 −0.019 −0.071 −0.071 −0.093 0.023 −0.061 −0.011 (0.094) (0.119) (0.092) (0.092) (0.092) (0.144) (0.092) (0.114) Area square km 0.006 0.005 0.006 0.006 0.006 −0.012 0.005 0.003 (0.008) (0.008) (0.008) (0.008) (0.009) (0.018) (0.008) (0.009) Constant −0.086*** −0.237 −0.251 −0.076*** −0.210 −0.210 −0.052*** −0.228 0.011 −0.079*** −0.224 −0.221 (0.016) (0.312) (0.367) (0.016) (0.301) (0.301) (0.008) (0.309) (0.362) (0.0151) (0.305) (0.335) Observation 158 158 158 158 158 158 158 158 158 158 158 158 R-squared 0.045 0.097 0.058 0.022 0.079 0.079 0.000 0.074 0.031 0.037 0.093 0.043 IV 40.349 35.301 3.171 22.328 Variables Dependent variable: Change in night-time light between 2010 and 2011 Building Content Land Total (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) Damage ratio 0.559*** 0.416** 0.957** 0.757** 0.379 0.379 0.016 −0.006 0.747 0.474*** 0.343** 0.912** (0.186) (0.171) (0.401) (0.367) (0.389) (0.389) (0.051) (0.068) (0.493) (0.181) (0.162) (0.415) Household income 0.008 0.003 0.007 0.007 0.012 −0.025 0.007 −0.000 (0.037) (0.045) (0.036) (0.036) (0.038) (0.049) (0.037) (0.041) Night-time Population 0.018 0.017 0.019 0.019 0.019 0.021* 0.019 0.017 (0.012) (0.012) (0.012) (0.012) (0.012) (0.012) (0.012) (0.012) Number of bedrooms −0.060 −0.019 −0.071 −0.071 −0.093 0.023 −0.061 −0.011 (0.094) (0.119) (0.092) (0.092) (0.092) (0.144) (0.092) (0.114) Area square km 0.006 0.005 0.006 0.006 0.006 −0.012 0.005 0.003 (0.008) (0.008) (0.008) (0.008) (0.009) (0.018) (0.008) (0.009) Constant −0.086*** −0.237 −0.251 −0.076*** −0.210 −0.210 −0.052*** −0.228 0.011 −0.079*** −0.224 −0.221 (0.016) (0.312) (0.367) (0.016) (0.301) (0.301) (0.008) (0.309) (0.362) (0.0151) (0.305) (0.335) Observation 158 158 158 158 158 158 158 158 158 158 158 158 R-squared 0.045 0.097 0.058 0.022 0.079 0.079 0.000 0.074 0.031 0.037 0.093 0.043 IV 40.349 35.301 3.171 22.328 Notes: ***, **, * Significant at 1%, 5% and 10% levels, respectively. AU clustered, robust standard errors are shown in parentheses. All regressions are estimated with OLS. Columns (1)–(3) use a damage ratio measure for buildings, Columns (4)–(6) for contents, Columns (7)–(9) for land damage and Columns (10)–(12) aggregate all damages per AU over time. In all columns, the dependent (LHS) variable is the same. The instrumental variable (IV), in the bottom row, is the robust Kleibergen–Paap rk Wald F-statistic for test of weak instruments. IV regressions have overidentification p-values approximately equal to zero, except for land regression. Open in new tab Table 3 Short run economic impact of the earthquakes using the damage ratio variable Variables Dependent variable: Change in night-time light between 2010 and 2011 Building Content Land Total (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) Damage ratio 0.559*** 0.416** 0.957** 0.757** 0.379 0.379 0.016 −0.006 0.747 0.474*** 0.343** 0.912** (0.186) (0.171) (0.401) (0.367) (0.389) (0.389) (0.051) (0.068) (0.493) (0.181) (0.162) (0.415) Household income 0.008 0.003 0.007 0.007 0.012 −0.025 0.007 −0.000 (0.037) (0.045) (0.036) (0.036) (0.038) (0.049) (0.037) (0.041) Night-time Population 0.018 0.017 0.019 0.019 0.019 0.021* 0.019 0.017 (0.012) (0.012) (0.012) (0.012) (0.012) (0.012) (0.012) (0.012) Number of bedrooms −0.060 −0.019 −0.071 −0.071 −0.093 0.023 −0.061 −0.011 (0.094) (0.119) (0.092) (0.092) (0.092) (0.144) (0.092) (0.114) Area square km 0.006 0.005 0.006 0.006 0.006 −0.012 0.005 0.003 (0.008) (0.008) (0.008) (0.008) (0.009) (0.018) (0.008) (0.009) Constant −0.086*** −0.237 −0.251 −0.076*** −0.210 −0.210 −0.052*** −0.228 0.011 −0.079*** −0.224 −0.221 (0.016) (0.312) (0.367) (0.016) (0.301) (0.301) (0.008) (0.309) (0.362) (0.0151) (0.305) (0.335) Observation 158 158 158 158 158 158 158 158 158 158 158 158 R-squared 0.045 0.097 0.058 0.022 0.079 0.079 0.000 0.074 0.031 0.037 0.093 0.043 IV 40.349 35.301 3.171 22.328 Variables Dependent variable: Change in night-time light between 2010 and 2011 Building Content Land Total (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) Damage ratio 0.559*** 0.416** 0.957** 0.757** 0.379 0.379 0.016 −0.006 0.747 0.474*** 0.343** 0.912** (0.186) (0.171) (0.401) (0.367) (0.389) (0.389) (0.051) (0.068) (0.493) (0.181) (0.162) (0.415) Household income 0.008 0.003 0.007 0.007 0.012 −0.025 0.007 −0.000 (0.037) (0.045) (0.036) (0.036) (0.038) (0.049) (0.037) (0.041) Night-time Population 0.018 0.017 0.019 0.019 0.019 0.021* 0.019 0.017 (0.012) (0.012) (0.012) (0.012) (0.012) (0.012) (0.012) (0.012) Number of bedrooms −0.060 −0.019 −0.071 −0.071 −0.093 0.023 −0.061 −0.011 (0.094) (0.119) (0.092) (0.092) (0.092) (0.144) (0.092) (0.114) Area square km 0.006 0.005 0.006 0.006 0.006 −0.012 0.005 0.003 (0.008) (0.008) (0.008) (0.008) (0.009) (0.018) (0.008) (0.009) Constant −0.086*** −0.237 −0.251 −0.076*** −0.210 −0.210 −0.052*** −0.228 0.011 −0.079*** −0.224 −0.221 (0.016) (0.312) (0.367) (0.016) (0.301) (0.301) (0.008) (0.309) (0.362) (0.0151) (0.305) (0.335) Observation 158 158 158 158 158 158 158 158 158 158 158 158 R-squared 0.045 0.097 0.058 0.022 0.079 0.079 0.000 0.074 0.031 0.037 0.093 0.043 IV 40.349 35.301 3.171 22.328 Notes: ***, **, * Significant at 1%, 5% and 10% levels, respectively. AU clustered, robust standard errors are shown in parentheses. All regressions are estimated with OLS. Columns (1)–(3) use a damage ratio measure for buildings, Columns (4)–(6) for contents, Columns (7)–(9) for land damage and Columns (10)–(12) aggregate all damages per AU over time. In all columns, the dependent (LHS) variable is the same. The instrumental variable (IV), in the bottom row, is the robust Kleibergen–Paap rk Wald F-statistic for test of weak instruments. IV regressions have overidentification p-values approximately equal to zero, except for land regression. Open in new tab The control variables ‘household income’ and ‘number of bedrooms’, which indicate the size of a dwelling, have small and insignificant coefficients across regressions of all asset classes. Moreover, when using the earthquakes’ physical intensities as instrumental variables (Columns 3, 6, 9 and 12) we obtain similar results to the OLS regressions (with higher magnitudes for the coefficients). The magnitude of the coefficient is largest for building and total assets regressions. For instance, a 1% increase in the total asset damage over total asset value is associated with 0.91% reduction in residential economic activity. For the second set of regressions, our primary focus in this paper, we examine the effect of insurance payment on local economic recovery post-earthquakes. Table 4 provides the results for the estimations of Equation (6) including AU and quarter-fixed effects. The insurance payment variables are estimated for each exposure separately (Columns 1–6) and total exposure is provided in the last three columns. Table 4 Economic recovery following the earthquakes (Claim payment)—AU and quarter-fixed effect Variables Dependent variable: quarterly change in night-time light Building Land Total (1) (2) (3) (4) (5) (6) (7) (8) (9) Insurance payment 0.004 0.005 0.071* 0.000 −0.000 0.009 0.004 0.004 0.094*** (0.004) (0.004) (0.038) (0.001) (0.001) (0.009) (0.004) (0.004) (0.031) Settlement time 0.017 −0.112* 0.011*** −0.001 0.031 −0.121** (0.033) (0.058) (0.004) (0.012) (0.030) (0.054) Prop. Cash settlement −0.060*** −0.018 0.011 −0.012 −0.056*** −0.110 (0.022) (0.157) (0.010) (0.028) (0.021) (0.125) Ins. payment* settlement time −0.001*** −0.001 −0.001*** (0.000) (0.001) (0.000) Ins. payment* prop. cash settlement 0.007 0.000 0.005 (0.014) (0.003) (0.010) Constant −0.146*** −0.203 0.664 −0.101*** −0.171*** −0.106 −0.145*** −0.283 0.857* (0.052) (0.224) (0.495) (0.013) (0.024) (0.072) (0.053) (0.195) (0.436) Observation 2686 2686 2686 2686 2686 2686 2686 2686 2686 No. area units 158 158 158 158 158 158 158 158 158 R-squared 0.704 0.705 0.706 0.703 0.704 0.704 0.704 0.705 0.706 Total effect of Ins. payment 0.004 0.005 0.070* 0.000 −0.000 0.005 0.004 0.004 0.128*** (0.004) (0.004) (0.039) (0.001) (0.001) (0.009) (0.004) (0.004) (0.032) Variables Dependent variable: quarterly change in night-time light Building Land Total (1) (2) (3) (4) (5) (6) (7) (8) (9) Insurance payment 0.004 0.005 0.071* 0.000 −0.000 0.009 0.004 0.004 0.094*** (0.004) (0.004) (0.038) (0.001) (0.001) (0.009) (0.004) (0.004) (0.031) Settlement time 0.017 −0.112* 0.011*** −0.001 0.031 −0.121** (0.033) (0.058) (0.004) (0.012) (0.030) (0.054) Prop. Cash settlement −0.060*** −0.018 0.011 −0.012 −0.056*** −0.110 (0.022) (0.157) (0.010) (0.028) (0.021) (0.125) Ins. payment* settlement time −0.001*** −0.001 −0.001*** (0.000) (0.001) (0.000) Ins. payment* prop. cash settlement 0.007 0.000 0.005 (0.014) (0.003) (0.010) Constant −0.146*** −0.203 0.664 −0.101*** −0.171*** −0.106 −0.145*** −0.283 0.857* (0.052) (0.224) (0.495) (0.013) (0.024) (0.072) (0.053) (0.195) (0.436) Observation 2686 2686 2686 2686 2686 2686 2686 2686 2686 No. area units 158 158 158 158 158 158 158 158 158 R-squared 0.704 0.705 0.706 0.703 0.704 0.704 0.704 0.705 0.706 Total effect of Ins. payment 0.004 0.005 0.070* 0.000 −0.000 0.005 0.004 0.004 0.128*** (0.004) (0.004) (0.039) (0.001) (0.001) (0.009) (0.004) (0.004) (0.032) Notes: ***, **, * Significant at 1%, 5% and 10% levels, respectively. AU cluster and robust standard errors are shown in parentheses. All regressions are estimated with AU and quarter-fixed effect. Columns (1)–(3) measure the ‘insurance payment’ for building damages (per AU and quarter), Columns (4)–(6) measure the payments for land damage and Columns (7)–(9) sum up the building, contents and land payments together. In all columns, the dependent variable is the same. Open in new tab Table 4 Economic recovery following the earthquakes (Claim payment)—AU and quarter-fixed effect Variables Dependent variable: quarterly change in night-time light Building Land Total (1) (2) (3) (4) (5) (6) (7) (8) (9) Insurance payment 0.004 0.005 0.071* 0.000 −0.000 0.009 0.004 0.004 0.094*** (0.004) (0.004) (0.038) (0.001) (0.001) (0.009) (0.004) (0.004) (0.031) Settlement time 0.017 −0.112* 0.011*** −0.001 0.031 −0.121** (0.033) (0.058) (0.004) (0.012) (0.030) (0.054) Prop. Cash settlement −0.060*** −0.018 0.011 −0.012 −0.056*** −0.110 (0.022) (0.157) (0.010) (0.028) (0.021) (0.125) Ins. payment* settlement time −0.001*** −0.001 −0.001*** (0.000) (0.001) (0.000) Ins. payment* prop. cash settlement 0.007 0.000 0.005 (0.014) (0.003) (0.010) Constant −0.146*** −0.203 0.664 −0.101*** −0.171*** −0.106 −0.145*** −0.283 0.857* (0.052) (0.224) (0.495) (0.013) (0.024) (0.072) (0.053) (0.195) (0.436) Observation 2686 2686 2686 2686 2686 2686 2686 2686 2686 No. area units 158 158 158 158 158 158 158 158 158 R-squared 0.704 0.705 0.706 0.703 0.704 0.704 0.704 0.705 0.706 Total effect of Ins. payment 0.004 0.005 0.070* 0.000 −0.000 0.005 0.004 0.004 0.128*** (0.004) (0.004) (0.039) (0.001) (0.001) (0.009) (0.004) (0.004) (0.032) Variables Dependent variable: quarterly change in night-time light Building Land Total (1) (2) (3) (4) (5) (6) (7) (8) (9) Insurance payment 0.004 0.005 0.071* 0.000 −0.000 0.009 0.004 0.004 0.094*** (0.004) (0.004) (0.038) (0.001) (0.001) (0.009) (0.004) (0.004) (0.031) Settlement time 0.017 −0.112* 0.011*** −0.001 0.031 −0.121** (0.033) (0.058) (0.004) (0.012) (0.030) (0.054) Prop. Cash settlement −0.060*** −0.018 0.011 −0.012 −0.056*** −0.110 (0.022) (0.157) (0.010) (0.028) (0.021) (0.125) Ins. payment* settlement time −0.001*** −0.001 −0.001*** (0.000) (0.001) (0.000) Ins. payment* prop. cash settlement 0.007 0.000 0.005 (0.014) (0.003) (0.010) Constant −0.146*** −0.203 0.664 −0.101*** −0.171*** −0.106 −0.145*** −0.283 0.857* (0.052) (0.224) (0.495) (0.013) (0.024) (0.072) (0.053) (0.195) (0.436) Observation 2686 2686 2686 2686 2686 2686 2686 2686 2686 No. area units 158 158 158 158 158 158 158 158 158 R-squared 0.704 0.705 0.706 0.703 0.704 0.704 0.704 0.705 0.706 Total effect of Ins. payment 0.004 0.005 0.070* 0.000 −0.000 0.005 0.004 0.004 0.128*** (0.004) (0.004) (0.039) (0.001) (0.001) (0.009) (0.004) (0.004) (0.032) Notes: ***, **, * Significant at 1%, 5% and 10% levels, respectively. AU cluster and robust standard errors are shown in parentheses. All regressions are estimated with AU and quarter-fixed effect. Columns (1)–(3) measure the ‘insurance payment’ for building damages (per AU and quarter), Columns (4)–(6) measure the payments for land damage and Columns (7)–(9) sum up the building, contents and land payments together. In all columns, the dependent variable is the same. Open in new tab The estimated coefficients for the main variable are small and insignificant when other insurance variables are not included (Columns 1, 4 and 7), or when we also control for the time of claim settlements, and the proportion of claims paid in cash (Columns 2, 5 and 8). Nevertheless, the insurance payment variables are positive and significant when we control for the interaction between the amount of insurance payment and the other variables characterizing those payments (time to settlement, and proportion in cash—see Columns 3, 6 and 9). The estimated coefficients are positive and are statistically significant especially for the largest exposure (building damage). Not surprisingly, payments for damage to contents (not shown) and to land (Columns 4–6), which are relatively small, are associated with very small coefficient estimates, and do not have any statistically discernible impact on recovery. In the bottom row of each column (each regression) we aggregate the impact of insurance payments, by also accounting for the interaction effects with the time and cash variables. When the insurance payment for building damage increase by 1% in an AU, for that quarter, the economic recovery of residential areas increases by about 0.70%, on average. This finding is important. It is the first time, as far as we know, that detailed post-catastrophe insurance payments are empirically linked with better local economic recovery outcomes.18 The effect of the settlement time variable on the outcome variable is negative, as we hypothesized. Both the variable in levels, and its interaction term with payment size are negative and significant (Columns 3 and 9) for claims for building damage. In other word, the positive impact of the claim amount is reduced when the settlement process is delayed—i.e. delayed payments are less helpful in generating increased economic activity. This might be because with delayed payments the owner of a ‘delayed’ property may have already moved elsewhere or had already fixed her house without insurance funding, but to a lower standard. The coefficient of the proportion of cash settlement variable19 is negative and statistically significant for building structure and total assets (Columns 2 and 8). It was suggested, in New Zealand and elsewhere, that cash payments of insurance claims enable recipients to move away and not rebuild. While we find some supportive evidence of that in these results, once we include the interaction terms (between cash proportions and the aggregate amount of payments) we no longer find this negative effect. If anything, our regression results show evidence that payments in cash did not necessarily hinder (nor assist) in the process of recovery. In the specifications in Table 4, we also control for the variations across time using the quarter dummies. The coefficients of the quarter dummies are large and volatile for the first 2 years after the CES, their coefficient estimates become smaller in absolute term from 2014 onward. Economic recovery in residential areas occurred mainly in 2012 and 2013 and the recovery rate thus declines as time passes. To further test the robustness of our results, we re-ran similar specifications using spatial panel models—this allows us to control for the spatial dependencies in the regression set. Online Appendix Tables 4–6 report the estimation results examining the effect of insurance payment on local recovery for the different spatial econometric models described in the Online Appendix. The finding using the spatial models are quite similar to the results of the non-spatial regressions presented in Table 4. Building and land specifications have significant coefficients, while content regressions do not. The payment * time interaction term is, as was the case in previous specifications, negative and statistically significant. We carry out model selection tests (Anselin et al., 1996; Olivia et al., 2009; Belotti et al., 2016; Noy et al., 2016), these support the SDM model specification. In addition, we also implement the Hausman test for the spatial panel model to test whether random effect models are preferred. The estimated spatial autoregressive ρ and autocorrelation  (λ) coefficients are significant. The economic recovery of an AU is positively influenced by the recovery of other surrounding AUs. Because the estimation coefficients of the specifications cannot be compared with each other, we derive the direct and spillover effects.20 In general, a 1% increase in insurance payment directly leads to 0.4–0.5% increase in residential recovery. However, this direct positive effect would be reduced when the claim settlement was delayed. If the spatial regression models include the endogenous interaction term  WYi,t ⁠, the direct effects contain the feedback effects in their estimates. The feedback effects occur when the impact goes through neighboring AUs and back to the initial AU (LeSage and Pace, 2009). In our result, when taking the difference between direct effect and point estimate, the feedback effect only accounts for about 10–12% of the direct effect. Generally, higher insurance payment received in an AU does not only lead to better economic recovery locally, but it also increases the economic growth in neighboring AUs. The spillover effect of the delay in claim payment is also observable in these spatial models. 8. Conclusion Very few papers have examined economic recovery in the longer-term (beyond the first 2 years), and none have looked at the role of insurance in facilitating recovery at the local level. This lacuna is mainly due to the limited availability of the required data at the appropriate frequency and over the longer term. Our contribution to the empirical literature is 2-fold: First, we show one can measure the immediate economic impact and the economic recovery of local areas after a sequence of earthquakes using the change in night-time luminosity. Second, we used data on insurance claim payments to examine the effectiveness of these payments in facilitating recovery. We found that the earthquake damage significantly reduced the nightlight radiance in the immediate aftermath of these events, and that the amount of lights bounced back and even increased in the years that followed. Using the insurance payment information, we found that building claim payments contributed significantly to local residential recovery in the years following the earthquakes. However, prolonged settlement delays (in cases when these delayed occurred) reduced the benefits of these insurance payments. We also found that settling claims in cash (versus doing the required reconstruction) did not change the dynamics of recovery in any material way. We also quantify the positive spillover effects of insurance payments to the recovery of other neighboring AUs. As far as we are aware, the average time it took to settle claims was remarkably longer in Christchurch as almost every residential property that was damaged (and almost all were) was also insured. Yet, delays are by no means unique. Complaints about the time it takes to settle claims appear after almost every large insured event. As other countries increase the penetration rates for insurance for natural hazards, this problem may further exacerbate in other jurisdictions as well. It is also important to note that while public earthquake insurance is less prevalent, and less often used, there are many publicly funded programs for flood insurance in many different countries (and not only in high-income countries). Flood insurance programs may suffer from the same weaknesses as the risk is correlated across larger spatial areas than earthquake risk is. The recent events associated with the 2017 Atlantic Hurricane season (especially Hurricane Harvey, which was the most heavily insured) have amply demonstrated that risk can be correlated across a large densely populated region. The role of insurance in the recovery of Houston should clearly be of concern to policymakers and the residents there, and unfortunately, in future events that are bound to occur in many places around the world. Acknowledgements We thank QuakeCoRE (publication #0370) and the Resilience National Science Challenge for providing funding that supported this work. Funding Funding was received from QuakeCoRE and from the RNSC. Footnotes 1 Liability for land damage is capped at the market value of the land. The local currency cap amounts were NZD 100,000 for dwelling damage and NZD 20,000 for contents. These nominal caps were set in the 1993 Act, and modified in 2019. We convert all currency figures to USD, based on the 2016 yearly average exchange rate. 2 See the Online Data Appendix for more detail about the luminosity data. 3 Several research papers have used night-time light to capture the immediate economic impact of climate disasters (Tanaka et al., 2000; Bertinelli and Strobl, 2013; Mohan and Strobl, 2017; Del Valle et al., 2018). 4 Hashetera et al. (1999) used the illumination intensity before and after the 1999 Marmara earthquake in Turkey to identify the impacted areas. Kohiyama et al. (2004) assessed the immediate impact of the 2001 Gujarat earthquake using night-time light intensity, and claimed that the estimated loss is consistent with their fieldwork information. Other examples include Escudero et al. (2017) and Elliott et al. (2015). 5 AUs are aggregation of mesh-blocks (the smallest geographical unit used by Statistics New Zealand). In urban areas, AUs are often a collection of several city blocks while in rural areas, AUs may be similar to localities or communities according to Statistics NZ. 6 The DMSP and VIIRS data are not comparable; even after radiometric inter-calibration undertaken by NOAA, comparison is impossible as the images were acquired at different times at night. 7 We chose to aggregate the data to a quarterly AU panel in order to smooth out spatial and temporal volatility (see figure 3). In principle, we could have conducted the analysis per pixel or per mesh-block (the smallest spatial unit for which data are collected), but the existing evidence suggests that nightlight will not be a good proxy for economic activity at such a high resolution. For the monthly model, none of the variables of interest seems to be statistically significant at that high frequency (these results are available in the Online Appendix Table 7). 8 This estimated damage cost is the total payment that EQC and private insurers would have transferred to the claimants (as insurance liability was based on replacement costs rather than the value of damage). 9 The insurance claim payments across exposures (building, land and content) are highly correlated (Online Appendix Table 2). Online Appendix Figure 4 provides the breakdown of EQC claims across districts and the separate earthquakes in 2010–2011. Even though the epicenter of the first event was located further away from Christchurch City, the number of valid claims for the first earthquake is nearly as high as for the latter one. 10 EQC does not only covered for the visible land damage, but the scheme has also been found liable for ground improvement works or long-term reduction of property values due to increased flood and liquefaction vulnerabilities generated by the earthquakes. EQC is the only insurance scheme globally that offers compensation for such risks. 11 A fuller representation of all quarters is available as a short stop-motion clip at: https://www.youtube.com/watch?v=9wdjOcP9XGk. 12 Seismologists have started to produce detailed shake maps for major earthquakes. The maps capture the exact spatial extend of earth surface movements and their decay in magnitude across space (that decay is not linear in distance and depends on surface conditions). 13 Including the few residential CBD observations does not change any of the results reported below. 14 As almost all houses were insured, the deductible was very low, and almost all houses incurred some damage (even if minor), this sum approximates quite closely the total value of all residential assets in each AU. 15 We assume that the effect of earthquakes’ physical intensity (Zi) on Economic_Lossieq only come from our endogenous explanatory variables— Damagei,k ⁠. When we run the tests of endogeneity, the null hypothesis (H0: damage ratio variable is exogenous) is rejected. 16 So that their log value will be equal to zero. 17 Spatial models have been used in economic geography, urban and regional science (Baltagi and Li, 2004; Kelejian and Piras, 2014; Firmino et al., 2016; Noy et al., 2016). 18 Von Peter et al. (2012), in a widely cited paper, found an association between overall insurance coverage and post disaster GDP growth at the national level. 19 The variable is excluded in the content specification because all the content payments were settled in cash. 20 To obtain the direct and spillover effects estimates, we use the variation of 500 simulated parameter combinations drawn from the multivariate normal distribution implied by the maximum-likelihood estimation. This procedure is widely used in spatial statistic inferences (LeSage and Pace, 2009; Vega and Elhorst, 2015). References Anselin L. , Bera A. K. , Florax R. , Yoon M. J. ( 1996 ) Simple diagnostic tests for spatial dependence . Regional Science and Urban Economics , 26 : 77 – 104 . Google Scholar Crossref Search ADS WorldCat Anselin L. , LeGallo J. , Jayet H. ( 2008 ) Spatial panel econometrics. In Matyas L. , Sevestre P. (eds) The Econometrics of Panel Data, Fundamentals and Recent Developments in Theory and Practice , 3rd, pp. 901 – 969 . Dordrecht : Kluwer . Google Preview WorldCat COPAC Baltagi B. H. , Li D. ( 2004 ) Prediction in the panel data model with spatial autocorrelation. In Anselin L. , Florax R. , Rey S. J. (eds) Advances in Spatial Econometrics: Methodology, Tools, and Applications , pp. 283 – 295 . Berlin : Springer . Google Preview WorldCat COPAC Belotti F. , Hughes G. , Mortari A. P. ( 2016 ) Spatial panel data models using Stata . CEIS Tor Vergata - Research Papers Series , 14 . WorldCat Bertinelli L. , Strobl E. ( 2013 ) Quantifying the local economic growth impact of Hurricane Strikes: an analysis from outer space for the Caribbean . Journal of Applied Meteorology and Climatology , 52 : 1688 – 1697 . Google Scholar Crossref Search ADS WorldCat Besley T. , Reynal-Querol M. ( 2014 ) The legacy of historical conflict: evidence from Africa . American Political Science Review , 108 : 319 – 336 . Google Scholar Crossref Search ADS WorldCat Bickenbach F. , Bode E. , Nunnenkamp P. , Söder M. ( 2016 ) Night lights and regional GDP . Review of World Economics , 152 : 425 – 447 . Google Scholar Crossref Search ADS WorldCat Brown C. , Seville E. , Vargo J. ( 2013 ) The Role of Insurance in Organisational Recovery Following the 2010 and 2011 Canterbury Earthquakes. Resilient Organisations Research Report 2013/04, University of Canterbury. Brown C. , Seville E. , Vargo J. ( 2016a ) Efficacy of insurance for organisational disaster recovery: case study of the 2010 and 2011 Canterbury Earthquakes . Disasters , 41 : 388 – 408 . Google Scholar Crossref Search ADS WorldCat Brown M. , Guin B. , Kirschenmann K. ( 2016b ) Microfinance banks and financial inclusion . Review of Finance , 20 : 907 – 946 . Google Scholar Crossref Search ADS WorldCat Bruederle A. , Hodler R. ( 2017 ) Nighttime lights as a proxy for human development at the local level . PLoS ONE , 13 : e0202231 . Google Scholar Crossref Search ADS WorldCat Chang S. E. ( 2010 ) Urban disaster recovery: a measurement framework and its application to the 1995 Kobe Earthquake . Disasters , 34 : 303 – 327 . Google Scholar Crossref Search ADS PubMed WorldCat Chen X. , Nordhaus W. ( 2011 ) Using luminosity data as proxy for economic statistics . Proceedings of the National Academy of Sciences , 108 : 8589 – 8594 . Google Scholar Crossref Search ADS WorldCat Cole M. A. , Elliott R. J. R. , Okubo T. , Strobl E. ( 2017 ) Natural disasters and spatial heterogeneity in damages: the birth, life and death of manufacturing plants . Journal of Economic Geography, 19: 373 – 408 . WorldCat Del Valle A. , Elliott R. J. R. , Strobl E. , Tong M. ( 2018 ) The short-term economic impact of tropical cyclones: satellite evidence from Guangdong Province . Economics of Disasters and Climate Change , 2 : 225 – 235 . Google Scholar Crossref Search ADS WorldCat Doll C. , Muller J. , Morley J. ( 2006 ) Mapping regional economic activity from night-time light satellite imagery . Ecological Economics , 57 : 75 – 92 . Google Scholar Crossref Search ADS WorldCat duPont Iv W. , Noy I. , Okuyama Y. , Sawada Y. ( 2015 ) The long-run socio-economic consequences of a large disaster: the 1995 Earthquake in Kobe . PLoS ONE , 10 : e0138714 . Google Scholar Crossref Search ADS PubMed WorldCat Elhorst J. P. ( 2014 ) Spatial Econometrics—From Cross-Sectional Data to Spatial Panels . Heidelberg : Springer . Google Preview WorldCat COPAC Elliott R. J. R. , Strobl E. , Sun P. ( 2015 ) The local impact of typhoons on economic activity in China: a view from outer space . Journal of Urban Economics , 88 : 50 – 66 . Google Scholar Crossref Search ADS WorldCat Elvidge C. , Sutton P. , Ghosh T. , Tuttle B. , Baugh K. , Bhaduri B. , Bright E. ( 2009 ) A global poverty map derived from satellite data . Computers & Geosciences , 35 : 1652 – 1660 . Google Scholar Crossref Search ADS WorldCat EQC. ( 2016 ) General Reports: Insurance Liability Valuation. Melville Jessup Weaver. EQC. ( 2017 ) Canterbury Earthquakes—Progress and Updates. Wellington, New Zealand: The Earthquake Commission. Escudero I. , Anil O. , Montero M. , Jose J. , Strobl E. ( 2017 ) The impact of Hurricane strikes on short-term local economic activity: evidence from nightlight images in the Dominican Republic. Policy Research Working Paper Series 8275, The World Bank. Felbermayr G. , Gröschl J. ( 2014 ) Naturally negative: the growth effects of natural disasters . Journal of Development Economics , 111 , 92 – 106 . Google Scholar Crossref Search ADS WorldCat Felbermayr G. , Gröschl J. K. , Sanders M. , Schippers V. , Steinwachs T. ( 2018 ) Shedding light on the spatial diffusion of disasters. CESifo Working Paper Series 7146 . WorldCat Firmino D. C. d. S. , Elhorst J. P. , Neto R. d. M. S. ( 2016 ) Urban and rural population growth in a spatial panel of municipalities . Regional Studies , 51 : 894 – 908 . Google Scholar Crossref Search ADS WorldCat GeoNet. ( 2011 ) M6.3, Christchurch, February 22, 2011. Available online at: http://info.geonet.org.nz/display/quake/M+6.3%2C+Christchurch%2C+22+February+2011. Ghosh T. , Anderson S. , Powell R. , Sutton P. , Elvidge C. ( 2009 ) Estimation of Mexico’s informal economy and remittances using night-time imagery . Remote Sensing , 1 : 418 – 444 . Google Scholar Crossref Search ADS WorldCat Ghosh T. , Powell R. , Elvidge C. , Baugh K. , Sutton P. , Anderson S. ( 2010 ) Shedding light on the global distribution of economic activity . The Open Geography Journal , 3 : 148 – 161 . WorldCat Gillespie T. W. , Frankenberg E. , Chum K. F. , Thomas D. ( 2014 ) Night-time lights time series of Tsunami damage, recovery and economics metrics in Sumatra, Indonesia . Remote Sensing Letters , 5 : 286 – 294 . Google Scholar Crossref Search ADS PubMed WorldCat Hashetera S. , Kohiyama M. , Maki N. , Hayashi H. , Matsuoka M. ( 1999 ) Use of DMSP-OLS images for early identification of impacted areas due to the 1999 Marmara Earthquake Disaster. Proceedings of the 20th Asian Conference on Remote Sensing, 2: 1291 – 1296 . Henderson J. V. , Squires T. , Storeygard A. , Weil D. ( 2018 ) The global distribution of economic activity: nature, history, and the role of trade . Quarterly Journal of Economics , 133 : 357 – 406 . Google Scholar Crossref Search ADS WorldCat Hodler R. , Raschky P. ( 2014a ) Economic shocks and civil conflict at the regional level . Economics Letters , 124 : 530 – 533 . Google Scholar Crossref Search ADS WorldCat Hodler R. , Raschky P. ( 2014b ) Regional favoritism . Quarterly Journal of Economics , 129 : 995 – 1033 . Google Scholar Crossref Search ADS WorldCat Kahn E. M. ( 2005 ) The death toll from natural disasters: the role of income, geography, and institutions. The Review of Economics and Statistics , 87 : 271 – 284 . Google Scholar Crossref Search ADS WorldCat Kelejian H. H. , Piras G. ( 2014 ) Estimation of spatial models with endogenous weighting matrices, and an application to a demand model for cigarettes . Regional Science and Urban Economics , 46 : 140 – 149 . Google Scholar Crossref Search ADS WorldCat Klomp J. ( 2016 ) Economic development and natural disasters: a satellite data analysis . Global Environmental Change , 36 : 67 – 88 . Google Scholar Crossref Search ADS WorldCat Kohiyama M. , Hayashi H. , Maki N. , Higashida M. , Kroehl H. , Elvidge C. ( 2004 ) Early damaged area estimation system using DMSP-OLS night-time imagery . International Journal of Remote Sensing , 25 : 2015 – 2036 . Google Scholar Crossref Search ADS WorldCat Kulkarni R. , Haynes K. , Stough R. , James R. ( 2011 ) Revisiting night lights as proxy for economic growth: a multi-year light based growth indicator (LGBI) for China, India and the U.S. GMU School of Public Policy—Research Paper. Kusuma A. , Nguyen C. , Noy I. ( 2019 ) Insurance for catastrophes: why are natural hazards underinsured, and does it matter? In Okuyama Y. , Rose A. (eds) Advances in Spatial and Economic Modeling of Disaster Impacts . Springer Science & Business Media . Google Preview WorldCat COPAC Lee L. F. , Yu J. ( 2010 ) Some recent developments in spatial panel data models . Regional Science and Urban Economics , 40 : 255 – 271 . Google Scholar Crossref Search ADS WorldCat LeSage J. , Pace K. ( 2009 ) Introduction to Spatial Econometrics . Boca Raton (FL ): CRC Press . Google Preview WorldCat COPAC Melecky M. , Raddatz C. ( 2015 ) Fiscal responses after catastrophes and the enabling role of financial development . World Bank Economic Review , 29 : 129 – 149 . Google Scholar Crossref Search ADS WorldCat Michalopoulos S. , Papaioannou E. ( 2013 ) Pre-colonial ethnic institutions and contemporary African development . Econometrica , 81 : 113 – 152 . Google Scholar Crossref Search ADS PubMed WorldCat Mohan P. , Strobl E. ( 2017 ) The short-term economic impact of tropical cyclone Pam: an analysis using VIIRS Nightlight Satellite Imagery . International Journal of Remote Sensing , 38 : 5992 – 6006 . Google Scholar Crossref Search ADS WorldCat Montalvo J. , Reynal-Querol M. ( 2016 ) Ethnic diversity and growth: revisiting the evidence. Economics Working Papers 1585, Department of Economics and Business, Universitat Pompeu Fabra. Nguyen C. , Noy I. ( 2019 ) Insuring earthquakes: how would the Californian and Japanese insurance programs have fared after the 2011 New Zealand earthquake? Disasters , forthcoming. NOAA. ( 2013 ) Visible Infrared Imaging Radiometer Suite (VIIRS) Sensor Data Record (SDR) User’s Guide. NOAA Technical Report NESDIS 142. Noy I. , duPont W. ( 2018 ) The long-term consequences of disasters: what do we know, and what we still don’t . International Review of Environmental and Resource Economics , 12 : 325 – 354 . Google Scholar Crossref Search ADS WorldCat Noy I. , Taupo T. , Cuffe H. ( 2016 ) Household vulnerability on the frontline of climate change in Tuvalu . Environmental Economics and Policy Studies , 20 : 705 – 739 . WorldCat Olivia S. , Gibson J. , Smith A. D. , Rozelle S. , Deng X. ( 2009 ) An empirical evaluation of poverty mapping methodology: explicitly spatial versus implicitly spatial approach. Paper presented at the Australian Agricultural and Resource Economics Society, Cairns, Australia. Pinkovskiy M. , Sala-i-Martin X. ( 2016 ) Lights, camera … income! Illuminating the national accounts-household surveys debate . The Quarterly Journal of Economics , 131 : 579 – 631 . Google Scholar Crossref Search ADS WorldCat Platt S. , Brown D. , Hughes M. ( 2016 ) Measuring resilience and recovery . International Journal of Disaster Risk Reduction , 19 : 447 – 460 . Google Scholar Crossref Search ADS WorldCat Poontirakul P. , Brown C. , Noy I. , Seville E. , Vargo J. ( 2017 ) Insurance as a double-edged sword? Quantitative evidence from the 2011 Christchurch Earthquake . Geneva Papers on Risk and Insurance , 42 : 609 – 632 . Google Scholar Crossref Search ADS WorldCat Rose A. , Benavides J. , Chang S. E. , Szczesniak P. , Lim D. ( 1997 ) The regional economic impact of an earthquake: direct and indirect effects of electricity lifeline disruptions . Journal of Regional Science , 37 : 437 – 458 . Google Scholar Crossref Search ADS WorldCat Sawada Y. , Shimizutani S. ( 2008 ) How do people cope with natural disasters? Evidence from the Great Hanshin-Awaji (Kobe) earthquake in 1995 . Journal of Money, Credit and Banking , 40 : 463 – 488 . Google Scholar Crossref Search ADS WorldCat Seville E. , Stevenson J. R. , Brown C. , Giovinazzi S. , Vargo J. ( 2014 ) Disruption and Resilience: How Organizations Coped with the Canterbury Earthquakes. ERI Research Report. Skoufias E. , Strobl E. , Tveit T. B. ( 2017 ) Natural disaster damage indices based on remotely sensed data: an application to Indonesia. Policy Research Working Paper Series 8188, World Bank. Stevenson J. R. , Seville E. , Kachali H. , Vargo J. , Whitman Z. ( 2011 ) Post-disaster organisational recovery in a central business district context: the 2010 & 2011 Canterbury Earthquakes. Resilient Organisations Research. Storeygard A. ( 2016 ) Farther on down the road: transport costs, trade and urban growth in sub-Saharan Africa . Review of Economic Studies , 83 : 1263 – 1295 . Google Scholar Crossref Search ADS PubMed WorldCat Sutton P. , Costanza R. ( 2002 ) Global estimates of market and non-market values derived from night-time satellite imagery, land cover, and ecosystem service valuation . Ecological Economics , 41 : 509 – 527 . Google Scholar Crossref Search ADS WorldCat Sutton P. , Elvidge C. , Ghosh T. ( 2007 ) Estimation of gross domestic product at sub-national scales using night-time satellite imagery . International Journal of Ecological Economics & Statistics , 8 : 5 – 21 . WorldCat Tanaka M. , Sugimura T. , Tanaka S. ( 2000 ) Monitoring water surface ratio in the Chinese floods of summer 1998 By DMSP-SSM/I . International Journal of Remote Sensing , 21 : 1561 – 1569 . Google Scholar Crossref Search ADS WorldCat Toya H. , Skidmore M. ( 2007 ) Economic development and the impacts of natural disasters . Economics Letters , 94 : 20 – 25 . Google Scholar Crossref Search ADS WorldCat Vega S. H. , Elhorst J. P. ( 2015 ) The SLX model . Journal of Regional Science , 55 : 339 – 363 . Google Scholar Crossref Search ADS WorldCat Von Peter G. , Von Dahlen S. , Saxena S. C. ( 2012 ) Unmitigated disasters? New evidence on the macroeconomic cost of natural catastrophes. BIS Working Paper 394, Basel: Bank for International Settlements. © The Author(s) (2019). Published by Oxford University Press. All rights reserved. For permissions, please email: journals.permissions@oup.com This article is published and distributed under the terms of the Oxford University Press, Standard Journals Publication Model (https://academic.oup.com/journals/pages/open_access/funder_policies/chorus/standard_publication_model)

Journal

Journal of Economic GeographyOxford University Press

Published: May 1, 2020

There are no references for this article.