Learning from natural disasters: Evidence from enterprise property insurance take-up in China

Abstract

This paper examines the causal impact of natural disasters on property insurance take-up of firms. Using the data of industrial firms in China, we find that a one-standard-deviation increase in typhoon damage leads to a 2.6% increase in the purchase of property insurance the following year. This increase gradually declines and returns to the previous level three years later. Our results demonstrate that this impact is driven not by changes in risk preferences or supply-side variation but by the updated risk beliefs learned from the typhoon experience. Unlike household insurance decisions, the learning effect of firms demonstrates an indirect but positive effect of typhoons on the firm’s insurance decisions from its related firms in the upstream or downstream sectors, but not from competitors.

Appendices

Appendix 1. Calculation process for typhoon damage

The wind speeds a firm experiences during a typhoon are essential in measuring the damage to an individual firm. It will depend on the firm’s exact location relative to the storm’s movement and features. We apply the well-known wind field model in physics (Boose et al., 2004; Holland, 2008) to model the wind speed at any point $i$, for storm k, in time t:

$${v}_{ikt}=GF\left[{v}_{m,kt}-S(1-Sin({T}_{ikt}\left)\right)\frac{{V}_{h,kt}}{2}\right]{\left[{\left(\frac{{R}_{m,k,t}}{{R}_{it}}\right)}^{{B}_{kt}}\text{e}\text{x}\text{p}\left(1-{\left(\frac{{R}_{m,k,t}}{{R}_{it}}\right)}^{{B}_{kt}}\right)\right]}^{0.5},$$

(6)

where ${v}_{max,kt}$ is the maximum sustained wind velocity anywhere in the typhoon; ${T}_{ikt}$ is the clockwise angle between the forward path of the typhoon and a radial line from the typhoon center to the point of interest i; ${V}_{h,kt}$ is the forward velocity of the tropical storm; ${R}_{max,k,t}$ is the radius of maximum winds and ${R}_{it}$ is the radial distance from the center of the tropical storm to point i. G is the gust factor, and F, S, and B are the scaling factors for surface friction, asymmetry due to the storm’s forward motion, and the shape of the wind profile curve, respectively.

In terms of implementing Eq. (1), the maximum wind speed ${v}_{max,kt}$ is taken from the IBTrACS data, ${V}_{h,kt}$ can be calculated following the movement path of the storm, while R and T can be determined by using the relative position between the eye of the typhoon and our point of interest i. We set G equal to 1.5 and S equal to 1 following Boose et al. (2004), respectively. For the surface friction indicator, Vickery et al. (2009) suggest that in open water, the reduction factor is around 0.7. Following Elliott et al. (2015) and Elliott et al. (2019), we linearize the reduction factor to capture the friction effect of the typhoon as it moves inland. In terms of the pressure profile parameter and the maximum radius winds, we adopt Holland (2008)’s approximation method following the suggestions by (Elliott et al., 2019).

As noted by Emanuel (2011), there are energy dissipation reasons to assume that the relationship between wind speed experienced and damage incurred is to the cubic power. Moreover, there is unlikely to be any damage from winds that fall below 92 km/h. To incorporate these features and to ensure that the percentage of damage varies between 0 and 1, Emanuel (2011) proposes the following damage function for the wind of storm k experienced at location i:

$${TD}_{ik}=\frac{{v}_{ik}^{3}}{1+{v}_{ik}^{3}},$$

(7)

where

$${v}_{ik}=\frac{Max\left[\left({v}_{ik}-{v}_{zero}\right),0\right]}{{v}_{half}-{v}_{zero}},$$

(8)

where ${v}_{ik}$ is the maximum wind speed of storm k experienced at the point of interest i; ${v}_{zero}$ is the threshold below which no damage occurs, and ${v}_{half}$ is the threshold at which half of the property is damaged. Following Emanuel (2011), we assume values of 92 km/h and 204 km/h for ${v}_{zero}$ and ${v}_{half}$, respectively. Note that Eq. (1) is defined in terms of the percentage of damage caused per storm. Given that our firm-level data is annual, a firm could be attacked by several storms in a year (as shown in Panel B of Appendix Fig. 4). Hence, to account for the possibility of multiple strikes, we assume that damages can accumulate over a year but restrict the impact of additional storms to have less than a simple cumulative effect. Instead, under the assumption that fragile structures get destroyed first, leaving more resistant structures, we allow damage to be multiplicative for storms k = 1, …, K within a year t as follows:

$${TD}_{it}=\sum _{k=1}^{{K}^{i}}\prod _{i=1}^{k}{TD}_{ik}.$$

(9)

For robustness, we also define the total typhoon damage for firm i in year t as follows:

$${TDT}_{it}=\sum _{k=1}^{{K}^{i}}\sum _{i=1}^{k}{TD}_{ik}.$$

(10)

Appendix 2. Supplementary figures

Fig. 4

Distribution of typhoon damage above zero and number of typhoon attack across years

Fig. 5

The mean wind speed during various relative periods. Note: The figure shows the mean wind speed during various relative periods to landfall. Typhoon data comes from the International Best Track Archive for Climate Stewardship (IBTrACS) database. It contains typhoons’ geographic coordinates and wind speed for each three hours. Each bin of the periods represents three hours (a day before and after the typhoon’s landfall).

Fig. 6

The distribution of average typhoon damage across provinces (1998–2007). Note: The average typhoon damage is obtained by calculating the average of typhoon damage to firms in each province. See Section  2.2 in the paper for the data sources and calculation of typhoon damage to firms.

Fig. 7

The impact of typhoon at different periods on firm EPI purchase. Note: This figure presents the regressions of $EPID_t$  (A) and $Log\;(EPIE)_t>0$  (B) on $TD_{i,t-\tau}$, where $\tau$ = 3, 2, 1 (baseline), 0, -1 (placebo). All regressions include climatic controls, year fixed effects, firm fixed effects, and region-specific trends. Standard errors are clustered at 2-digit industry, geographic region, and year level.

Fig. 8

Spillover impact of typhoon on EPI (extensive effect). Note: This figure presents the estimates of the impact of own typhoon damage (A), typhoon damage in the same industry (B) and typhoon damage in related (upstream and downstream) industries (C) on premium expenditure of EPI (log) (> 0) within different cut-off distances. All regressions include climatic controls, year fixed effects, firm fixed effects, and region-specific trends. Standard errors are clustered at 2-digit industry, geographic region, and year level.

Fig. 9

Spillover impact of the typhoon on EPI (intensive effect). Note: This figure presents the estimates of the impact of own typhoon damage (A), typhoon damage in the same industry (B) and typhoon damage in related (upstream and downstream) industries (C) on premium expenditure of EPI (log) (> 0) within different cut-off distances. All regressions include climatic controls, year fixed effects, firm fixed effects, and region-specific trends. Standard errors are clustered at 2-digit industry, geographic region, and year level.

Appendix 3. Supplementary table

Table 8 Regressions using alternative samples

Table 9 The results with two-year cutoff or four-year cutoff

Table 10 Regressions considering supply changes (extensive and intensive effect)