The Impact of Disasters on Inflation

Original Paper

Abstract

This paper studies how disasters affect consumer price inflation. There is a marked heterogeneity in the impact between advanced economies, where the impact is negligible, and developing economies, where the impact can last for several years. There are also differences in the impact by type of disasters, particularly when considering inflation sub-indices. Storms increase food price inflation in the near term, although the effect dissipates within a year. Floods also typically have a short-run impact on inflation. Earthquakes reduce CPI inflation excluding food, housing and energy.

Keywords

Disasters Inflation 

JEL Classification

E31 Q54 

Introduction

Disasters caused by natural hazards have the potential to cause massive economic disruption, and often are accompanied by a significant human toll. Recent examples of disasters include: earthquakes in Japan, Chile, Haiti and New Zealand in 2010 and 2011; the devastation of Vanuatu by Cyclone Pam in 2015; The 2011 floods in Thailand; ongoing drought in California and the eruption of Eyjafjallajökull in Iceland in 2010. With greater concentrations of population and activity in vulnerable regions, the incidence of economically significant disasters is increasing (Cavallo and Noy 2011). Barro (2009) estimates the welfare cost of these rare, but extreme, events at 20 percent of output, far beyond the 1.5 percent estimated welfare cost of normal business cycle fluctuations.

Until recently, our understanding of the economic impact of disasters was limited. Progress has been made over the past decade in investigating the impact of disasters on output, with a number of authors systematically studying the impact across disaster type, level of development and sectors of the economy (see, e.g. Noy 2009; Raddatz 2009; Loayza et al. 2012; Fomby et al. 2013; Felbermayr and Gröschl 2014).

Yet while the impact on output is now better understood, the impact on prices has received far less attention. Indeed, Cavallo and Noy (2011) in their recent survey of the literature on disasters point to the effect on prices as being one of the main remaining gaps in our knowledge. Only a small handful of papers have carried out any investigation of price impacts, and most of them do so only as part as part of a broader case study of the impact of a particular event, rather than a dedicated analysis of the impact on prices. Given the range of factors noted in the literature that affect the impact of disasters, caution should be exercised before extrapolating universal conclusions from single events. Even within countries, disasters of different magnitudes and types may have differing effects on inflation.

Section “How Disasters May Affect Prices” below reports the results from these prior studies. Of them, Heinen et al. (2015) is worthy of mention as the only one that extends the analysis beyond a single, or small number of similar events. They quantify the impact of hurricanes and floods on inflation in 15 Caribbean economies, finding evidence of a positive impact from both in the month the disaster occurs.

This article extends the analysis of Heinen et al. (2015) in a number of important dimensions. First, it also covers earthquakes and droughts, important disaster types that are covered in the literature on output effects.1 Second, it extends coverage to 212 economies. As Heinen et al. (2015) note, the 15 Caribbean economies they study are particularly vulnerable to diasters due to their small physical size, geographic isolation, high population densities, limited natural resources and reliance on imports. This particular vulnerability may result in impact estimates that do not readily translate to other economies.

Third, as described below in Section “How Disasters May Affect Prices”, the impact of disasters on inflation may differ over time, as initial short-term disruption and shortages may give way to rehabilitation and a rebuild-led demand surge. Heinen et al. (2015) study the impact on inflation in the initial three months following the disaster. This article extends that analysis to consider the first three years following the disaster. As results here show, there are prolonged impacts on inflation that can last a number of years, and may differ in sign. For example, Heinen et al. (2015) find positive effects from hurricanes on food prices. The analysis here finds that that influence is relatively short-lived and is reversed beyond the horizon of analysis of that earlier paper.

Finally, the analysis here considers the impact of disasters on major sub-components of CPI, allowing for a finer understanding of likely direct and indirect impacts on various sectors of the economy. While Heinen et al. (2015) also look at subcomponents for the economies and disasters they study, less is known about the impact on sub-components for other disasters and economies.

As a result of these important extensions, this article provides a systematic analysis of the impact of disasters on prices that is comparable to the well-known studies of the impact on output mentioned above. As such, it provides a significant step in closing one of the important remaining gaps in our knowledge of the effects of disasters. This article also makes a further contribution to the literature on the impact of disasters by using higher frequency data. Quarterly GDP data is not widely available outside of the most developed countries. Studies of the effects on output have focused on annual (or five-year average) GDP growth. Consumer prices are available at more frequent intervals, permitting the analysis here to study at a more granular level the evolution through time of the impact of disasters.

There are a number of benefits in knowing the likely path for inflation following a disaster. Understanding the effect on prices provides monetary policy makers with greater guidance on how to set policy in the immediate aftermath of the disaster. It can help with estimating the insurance costs for rebuild or cash settlement; it provides aid donors with a metric for determining the value of cash donations or gifts in kind; it assists fiscal authorities with calculating the future costs of the rebuild programme. Finally, the path for inflation has implications for the exchange rate and capital account policies.

We combine two sets of data to undertake the analysis here. The first is the EM-DAT database on disasters widely used in the literature. This contains information on a wide range of disasters, including number of people killed, number of people affected and (less frequently) damage caused. This data set is widely used in the literature and is the only one with widespread coverage that is publicly available. As a robustness check, we also undertake the analysis using the Ifo GAME database of geological and meteorological hazards, presented in Felbermayr and Gröschl (2014).

The second data set is the consumer price data from Parker (2016). These data cover consumer prices for 223 countries and territories over the period 1980-2012. We restrict our sample to those countries with at least 40 quarterly observations, resulting in 212 included in the analysis here. The data include information on headline consumer prices, as well as sub-indices for food, housing, energy and the remainder of the index. The panel is not balanced, with coverage for the sub-indices less complete for less developed countries. Nonetheless, coverage of sub-indices far exceeds any other database for consumer prices.

Previous studies have highlighted a large heterogeneity in the impact of disasters on output, particularly between advanced and developing countries. The impact on inflation is similarly diverse. Disasters on average have negligible impact on inflation in advanced countries, but typically increase inflation in developing countries. That said, the impact for severe disasters (those in the upper quartile) is larger, and significant even in high income countries.

The impact of diasters on inflation differs by sub-index. The impact on food price inflation is in general positive, if short lived. The impact on housing and other sub-indices is in general negative. Differences in expenditure weights on these sub-indices will in part explain the differences witnessed in headline inflation numbers by level of development.

Earthquakes do not significantly affect headline inflation, but do significantly reduce CPI inflation excluding food, housing and energy. Storms cause an immediate increase in food price inflation for the first six months, although this impact is reversed in the subsequent two quarters, resulting in no significant impact over the entire first year, or beyond. Floods increase headline inflation in the quarter that the flooding occurs in middle and low income countries, but have no significant impacts in subsequent quarters. In high income countries, the impact on headline inflation is negative, although insignificant. Droughts increase headline inflation for a number of years.

How Disasters May Affect Prices

As noted above, there has yet to be a systematic review of the impact of disasters on prices. Nonetheless, evidence from the literature on the impact on economic activity and a small number of case studies provide some guide to the potential channels of impact.

Short-run Impacts

Disasters affect economic activity via a number of channels in the short run. The immediate direct impact of diasters can cause death and injury to people, and cause damage to buildings, transport infrastructure and livestock. The destruction of harvests or housing can create shortages, pushing up the price of remaining food or houses. The size of the increase in prices may depend on market power of firms and perceptions of customers – it may not be in the long-run interest of a firm to be seen to be profiteering from customers’ misery. Rotemberg (2005), p 835 notes examples of customer protests at prices increases following the 1994 earthquake in the Los Angeles area.2

Beyond the direct impact, other businesses and households may be indirectly affected, such as being unable to bring goods to market due to lack of transport infrastructure. For example, farmers may react to the shortage of feed caused by a drought by slaughtering livestock. This could potentially reduce meat prices in the near term, but increase them in the medium term as farmers act to rebuild livestock numbers once the drought has ended. If the disruption to economic activity is sufficiently large it may reduce demand for goods and services from sectors not directly affected. This lower demand could reduce the prices in these other sectors.

There are a number of papers that aim to quantify the impact of disasters on economic activity. Noy (2009) examines the impact of 507 disasters over the period 1970-2003, finding a significant impact on GDP. The effect is greater for smaller and for less developed countries. Higher per capita income, literacy rates and institutional capacity help to mitigate the impact. The impact of disasters appears to differ by type of disaster. Raddatz (2009) finds that climatic disasters (storms, floods, droughts and extreme temperatures) have a significant negative impact on GDP, mostly in the year of the disaster. Other disasters are not found to have a significant impact. Felbermayr and Gröschl (2014) similarly find differential impacts by type of disaster and level of development, using a database of geological and meteorological events.

A number of authors also consider the impact of disasters on differing sectors of the economy. Loayza et al. (2012) find no significant effect on overall GDP using five-year growth averages over the period 1961-2005, although droughts are negative and storms and floods are positive. Droughts and storms negatively affect agricultural output, whereas floods are positive. The authors suggest that this positive effect may derive from plentiful rainfall providing benefit to crops that outweighs the localised damage from flooding, and the additional nutrients that aid the following season. Furthermore, cheaper electricity from more abundant hydropower aids industry. Nonetheless, this positive effect disappears in the presence of more severe flooding. Fomby et al. (2013) find that earthquakes affect agricultural production in developing countries, potentially a result of damaged infrastructure. Fomby et al. (2013) also find differing impacts of disasters depending on their severity.

Small-scale studies of individual disasters, or small groups, point to differing inflation impacts by type of disaster. The most comprehensive study to date, Heinen et al. (2015), considers the impact of hurricanes and floods on the inflation rates of 15 Caribbean islands. Damaging hurricanes increased monthly headline CPI inflation by 0.05 percentage points, with a greater effect on impact on food prices. More damaging hurricanes have a proportionately higher impact on inflation, with the implied inflationary impact of the largest hurricane in their sample being 1.4 percentage points on monthly headline CPI inflation. Flooding had an average 0.083 percentage point impact on inflation, with the implied largest effect 0.604 percentage points. The impact of both hurricanes and floods takes place in the month of the event, with no significant effects in subsequent months.

In terms of case studies of individual events or countries, Laframboise and Loko (2012) estimated that headline inflation increased by an additional 2 percent in Pakistan following the severe floods of 2010. Abe et al. (2014) find little increase in prices following the Great East Japan earthquake of 2011. Reinsdorf et al. (2014) compare this earthquake with the Chilean earthquake of 2010 using online data for supermarkets. Their data point to a sharp fall in product availability in the immediate aftermath of both earthquakes, without concurrent increases in price.

Doyle and Noy (2015) find no significant aggregate impact on New Zealand consumer prices from the Canterbury earthquakes of 2010 and 2011. At a disaggregated level, Parker and Steenkamp (2012) and Wood et al. (2016) find large increases in rents and construction costs within Canterbury, consistent with restricted housing supply following the widespread destruction of the housing stock. Munoz and Pistelli (2010) investigate the impact on inflation of a small number of large earthquakes, by comparing inflation outturns with a forecast based on information prior to the event. While they find that some earthquakes resulted in higher inflation, it was by no means universal. Given their small sample of events they were unable to explain the causes of this different response.

Kamber et al. (2013) study the impact of droughts on New Zealand, using measures of rainfall and soil moisture deficit in a VAR framework. Their findings suggest a drought of the magnitude of that of early 2013 raises CPI food prices by around 1.0 - 1.5 percent. In particular, milk cheese and eggs prices increase by 3 percent, reflecting the importance of dairy in domestic agriculture. Wholesale electricity prices increase by as much as 8 percent following such a drought, as lower lake levels increase the cost of hydroelectricity, although this cost increase does not appear to pass through to retail. Conversely, depressed economic activity results in falling prices for other non-tradable sectors, resulting in no significant impact on overall CPI. Buckle et al. (2007) similarly found no significant overall impact on consumer prices from droughts in New Zealand.

Medium-run Impacts

There may be some longer-lasting impacts on prices beyond the immediate destruction and disruption. The destruction of ports and infrastructure may disrupt imports, driving up the price for those goods which are imported. Conversely, the lack of ports for export may lead to a domestic oversupply and price falls in goods normally exported. International investors may also choose to withdraw capital from a country recently hit by a disaster, pushing down on the exchange rate and increasing the cost of imports. Ramcharan (2007) finds that in flexible exchange rate regimes, the real exchange rate depreciates by 10.25 percent in the year following a windstorm. The exchange rate effect is uncertain, however, since domestic investors repatriating foreign investments could lead to an exchange rate appreciation; the yen appreciated sharply in the immediate aftermath of the 2011 Tōhoku earthquake (Neely 2011).

Over the medium term, as resources are allocated to damage and reconstruct destroyed buildings and infrastructure there may be a ‘demand surge’, placing upward pressure on prices. Keen and Pakko (2011) calibrated a DSGE model to simulate the impact of Hurricane Katrina. In their simulation, the destruction of capital stock and temporary fall in productivity causes firms to raise prices, resulting in higher inflation

However, this demand surge is not certain. The incidence of a disaster may cause revisions of people’s perception of disaster risk and cause outward migration. Boustan et al. (2012) find outward migration from areas affected by tornadoes, Hornbeck (2012) from heavily eroded counties in the Dust Bowl era, and Hornbeck and Naidu (2014) document substantial outward migration following the 1927 Mississippi floods. Coffman and Noy (2012) use synthetic control methods to estimate a 12 percent drop in population on the island of Kauai in Hawaii, following Hurricane Iniki. The population of New Orleans fell sharply following Hurricane Katrina (Vigdor 2008), although the destruction of housing stock was far greater, resulting in higher house prices and rents. The destruction of disasters may also create poverty traps where households are unable to regain previous wealth and income (Carter et al. 2007). Such scenarios would put downward pressure on prices over the medium term in areas affected by disasters, although it is less certain the extent to which this affects the overall national price level.

Taking the above factors into consideration, the overall impact of disasters on inflation is ambiguous. The prior research on activity suggests there may be at the very least differences in the impact of disasters on inflation: by type of disaster; between the short and medium term; by different sub-component of the inflation basket; by level of development, and; by severity of the disaster. The analysis that follows accounts for these differing potential effects in turn.

Data and Method

Disasters

The most widely used source for disasters is the EM-DAT database collected by the Centre for Research on the Epidemiology of Disasters at the Université catholique de Louvain. The database covers disaster events which meet one of the following criteria: ten or more people killed; 100 or more people affected (defined as people requiring immediate assistance during a period of emergency, i.e. requiring basic survival needs such as food, water, shelter, sanitation and immediate medical assistance); declaration of a state of emergency; or call for international assistance. Alongside the date of the disaster, the EM-DAT database also includes information on the number of people killed and the number of people affected. For a smaller set of disasters the database includes an estimate of the damage caused.

It is worth noting that the EM-DAT database measures the ex post effects of disasters, which as shown in Noy (2009) and elsewhere depend on a number of country specific factors such as institutions. The relevant institutional factors, for example good economic governance, may also affect inflation dynamics which can lead to endogeneity, and potentially biased coefficient estimates. Furthermore, it is known that coverage of disasters within EM-DAT has increased over time (Cavallo and Noy 2011), although that applies more to the coverage of during the 1960s and 1970s, which are not studied here. Moreover, according to Cavallo and Noy (2011), the increase in coverage appears to mostly relate to small disasters, which are less likely to have pronounced macroeconomic effects.

More recently, studies have attempted to proxy for disasters using data on the underlying natural hazard to determine the impact of disasters. For example, Heinen et al. (2015) use a dataset of windspeed and precipitation for their study of windstorms and floods for a select group of Caribbean islands. Felbermayr and Gröschl (2014) use a large dataset of earthquake magnitude, windspeed, temperature and precipitation. The aim of such datasets is to attempt to avoid the potential measurement error in the EM-DAT database described above.

But the use of data for underlying natural hazards is not itself without problems for economic studies seeking to estimate the impact of diasters on the wider economy. It is the interaction of natural hazards with existent economic and demographic factors that determines the ultimate severity. Take as an example two recent large earthquakes in New Zealand: the magnitude 7.8 Dusky Sound earthquake on 15 July 2009 and the magnitude 6.2 Christchurch earthquake on 22 February 2011. Both occurred in the same country – in fact on the the same island – and were close to each other in time. Clearly institutions did not change much between the two earthquakes. Yet the former, much more powerful, earthquake took place in an unpopulated area and caused little economic destruction. The latter took place right on top of the country’s second largest city and caused damage estimated to be 17 percent of national GDP (Wood et al. 2016).

Similarly, the concentration of ex ante hazard risks by country may mean exposed countries put in place institutions to mitigate damage. The extent that the building code ensures resilience to earthquakes – and the extent to which that code is enforced – may well determine the impact on economic activity and inflation. A magnitude 6 earthquake is likely to have less impact in Los Angeles, where the risk is known, than London, where earthquakes are rare and generally very weak. In the presence of such endogeneity, the estimated coefficients may suffer from attenuation bias, which is to say smaller in magnitude than their true value, and less likely to be found significant.

Seen from the view point of economic impact, using country-level hazard data is not, therefore, itself without risk of measurement error. A second problem with ex ante hazard data is the difficulty in assessing the relative impact across disaster types. It is not clear, for example, that a two standard deviation wind speed event is of the same order of potential impact magnitude as, say, a two standard deviation temperature event.

Recognising that both measures have their shortcomings, we favour here the use of ex post measures of disaster impact that can translate across disaster types, since the purpose of this paper it to consider the impact of disasters on inflation, rather than assessing the magnitude of disasters given ex ante hazards. For completeness, we re-estimate our equations in Section “Alternative Measure – Underlying Natural Hazards”, using the ex ante hazards data from Felbermayr and Gröschl (2014).

Only disasters with likely macroeconomic effects are considered here, namely: earthquakes, storms, floods, droughts and other disasters (mass movements, insect infestations, extreme temperatures, volcanoes and wildfires). In order to estimate the effect of disasters on inflation, we require the quarter in which the disaster took place. The EM-DAT database does not always have precise start dates for droughts (even to the three-month period required) so as a consequence many droughts included in EM-DAT have been dropped from the analysis.

Even with these selection criteria, there are a large number of disasters in the EM-DAT database which are small relative to the overall size of the country and are unlikely to have any discernable macroeconomic effects. To aid estimation, only disasters with at least major impact are considered in the analysis below. To estimate the severity of the impact of the disasters we construct an impact variable for each disaster, calculated in a similar fashion to Fomby et al. (2013).

The impact variable used in this paper is:
$$ IMP_{i,t}^{\prime} = (EQIMP_{i,t}, STIMP_{i,t}, FLIMP_{i,t}, DRIMP_{i,t}, OTIMP_{i,t})' $$
(1)
where EQIMP, STIMP, FLIMP, DRIMP and OTIMP represent the respective total impact of earthquakes, wind storms, floods, droughts and other disasters. I M P i,t is calculated as:
$$ IMP_{i,t}(k)= {\sum}_{j = 1}^{J} intensity_{i,t,j}^{k} $$
(2)
where
$$\begin{array}{@{}rcl@{}} intensity_{i,t,j}^{k} & = & 100*\frac{fatalities_{i,t,j}^{k} + 0.3*total \ af\!fected_{i,t,j}^{k}}{population_{i,t,j}}, \ \text{if} \ intensity > 0.1 \\ & = & 0 \quad \text{otherwise} \end{array} $$
(3)
and J is the total number of each type-k events (k = 1,2,3,4,5 and responds to earthquakes, wind storms, floods, droughts and other disasters respectively) that took place in each country i in quarter t. The creation of I M P i,t can be described by the following steps. First, for each disaster the intensity was calculated by dividing the number of fatalities and 30 percent of the total people affected by the population. Where this intensity is smaller than 0.1 percent, the impact is set to zero (3). Then for each country, the total impact for each type of disaster is calculated as the sum of the intensities of each such disaster that occurred in each country for each quarter (2).
The criteria on disasters discussed above, together with the availability of consumer price data (see Section “Consumer Prices”), result in a total of 1349 disasters in 163 countries. Table 1 shows the incidence of disasters by type and by country development. We take the World Bank’s classification of High, Middle and Low income countries. We further split High income countries into advanced and other high income countries. Following (Noy 2009) we take advanced countries to be high income members of the OECD in 1990. Indeed all of these countries were members of the OECD before the period analysed here. Floods and storms are the most frequently occurring disasters that meet the criteria for inclusion. Measured droughts are rare in advanced countries and other high income countries, with only three in the sample, compared with 124 in middle income countries.
Table 1

Incidence of disasters

 

Earthquakes

Storms

Floods

Droughts

Other

Total

Number

      

Advanced

17

14

9

2

5

47

Other high income

3

39

21

1

3

67

Middle income

47

288

433

124

29

921

Low income

6

57

155

90

6

314

Total

73

398

618

217

43

1349

75th percentile

      

Advanced

1.19

0.66

0.40

 

1.68

1.09

Other high income

 

1.26

0.28

  

0.94

Middle income

0.93

1.56

0.89

5.47

1.20

1.46

Low income

0.53

1.26

0.92

6.68

8.25

2.75

Total

0.95

1.33

0.87

5.79

1.95

1.64

90th percentile

      

Advanced

2.07

1.69

1.09

 

5.10

4.63

Other high income

 

3.89

0.53

  

3.67

Middle income

2.34

6.12

2.43

10.75

2.38

4.53

Low income

13.47

3.83

3.28

14.34

12.24

7.00

Total

2.63

4.37

2.48

12.00

5.10

4.99

Notes: countries within advanced and other high income groups set out in Table 7 in the Appendix. 75th and 90th percentile impact as calculated per (2) and (3). Measured in percent of population. Impact omitted where there are fewer than 5 events

The impact on inflation is likely to depend on the size of the disaster. We follow Cavallo et al. (2013) and focus here on large disasters in the 75th and 90th percentiles. The 75th percentile disaster is approximately the impact of Hurricane Earl on Antigua and Barbuda in 2010. The hurricane affected around 6 percent of the population and did damage estimated to be around 1 percent of GDP. The 90th percentile is approximately the impact of the 2010 earthquake in Chile, which killed 562 people, affected 2.7 million (16 percent of the population) and had estimated damages of 17 percent of GDP.

Consumer Prices

As noted in Section “How Disasters May Affect Prices” above, different types of disasters may affect different prices, with the prices for food, housing (including rent) and energy being the most commonly cited in the literature. Commonly used international databases, such as the International Financial Statistics of the International Monetary Fund and the World Development Indicators of the World Bank, typically contain information on just the overall, headline CPI index. Information on the sub-indices is normally only available from national sources.

The consumer price data used here are taken from the dataset in Parker (2016). This dataset contains CPI for 223 countries and territories on a quarterly basis for the period 1980-2012. The series contained are the overall index (CPI) the sub-indices for food (CPIF), housing (CPIH), energy (CPIE), and all remaining items in the index (CPIxFHE). Coverage for CPIH and CPIE is relatively sparse relative to the other indices, so a combined housing and energy index is also included (CPIHE) which has observations for a greater number of countries.

We drop countries for which there are fewer than 40 quarters of CPI data. This results in 212 countries with observations for headline CPI. The average number of quarters of headline CPI data per country is 105. Fewer countries have data for the sub-indices, and the length of coverage is also typically shorter, particularly for less developed countries.

Method

To estimate the impact of disasters on inflation, we run a panel regression of the form:
$$ \pi_{i,t}={\sum}_{j = 0}^{p}\beta_{j}D_{i,t-j}+\mu_{i}+\lambda_{t}+\nu_{it} $$
(4)
where π i,t is quarterly log difference in CPI in country i in quarter t. We multiply the inflation rate by 100 to give coefficients that are in units of percentage points for ease of reading. D i,t is a vector of variables capturing the impact of disasters. The analysis that follows also considers the impact on the inflation rate for food, housing, energy and cpi excluding food housing and energy, respectively \(\pi _{i,t}^{f}\), \(\pi _{i,t}^{h}\), \(\pi _{i,t}^{e}\), \(\pi _{i,t}^{xfhe}\). We consider both the impact of all disasters combined, and the five types of disasters (earthquakes, wind storms, floods, droughts and other) individually as described in Section “Disasters” above.

The parameters μ i and λ t are fixed effects for country and time respectively. The country fixed effects capture the time invariant characteristics of each country that explain differences in average inflation rates between countries. The time fixed effects capture global factors that affect all countries, such as global developments in output growth and commodity prices or the Great Moderation. The occurrence of disasters is assumed to be exogenous, and unaffected by current or previous values of CPI.

Previous studies of the impact of disasters on growth have tended to include a large number of controls, such as the saving rate, institutional quality, etc. to attempt to eliminate endogeneity. Yet as Dell et al. (2014) note in their review of the literature, this can lead to a problem of “overcontrolling” if part of the impact of disasters works precisely through these controls, which would result in an underestimate of the impact of disasters. Dell et al. (2014) recommend using the system of country and time fixed effects used here and dispensing with other controls.

One potential problem with this estimation is that CPI data – particularly CPI food – is typically seasonal, which increases the variance of the underlying series. There are a number of approaches to eliminate this seasonality. The first is to use a seasonal adjustment process, the most widely used of which is the Census Bureau’s X12. However, X12 uses both forward and backward looking filters, which violates the exogeneity assumption over CPI and disasters.

The use of country seasonal dummies for each quarter is also unsatisfactory for our purposes if disasters do have an impact on CPI, but are concentrated in particular quarters. Consider windstorms, whose incidence is for the most part concentrated to certain times of the year. In such cases, the seasonal dummy will absorb some of the true impact of disasters. Such quarterly dummies are also unsatisfactory if the seasonal pattern changes over time.

Given these problems, we use quarterly regional seasonal dummies. We separate countries into Northern Hemisphere (defined as those with capital cities north of the Tropic of Cancer), Southern Hemisphere (those with capitals south of the Tropic of Capricorn) and tropical countries (those whose capitals lie between the tropics). These seasonal dummies permit varying seasonal patterns for those countries with pronounced (and opposite) seasons, as well as recognising the existence of weather patterns in the tropics where seasonal temperature variations are reduced and there is often more than one growing season.3

Standard panel estimation assumes that the errors, ν i t , are not correlated cross-sectionally, i.e.:
$$ \rho_{ij}=\rho_{ji}=corr(\nu_{it},\nu_{jt})= 0 \qquad \text{for} \qquad i \neq j $$
(5)

However, such an assumption may not be valid when macroeconomic time series are used. Close trade ties and other economic interactions between spatially grouped countries are likely to result in positive cross-correlations. To test the null hypothesis of cross-sectional independence we use the Pesaran (2004) test, which is the most appropriate given the unbalanced nature of the CPI dataset, and the large N relative to T. We obtain a test statistic of 205, which is significant evidence against the null of cross-sectional independence. The average absolute pairwise cross-sectional correlation is 0.302. A positive cross-sectional correlation results in substantial downward bias to the standard errors calculated using standard panel estimation techniques. To account for this large cross-sectional correlation, and any potential serial correlation, we use Driscoll and Kraay (1998) adjusted standard errors in the estimations that follow.4

Results

This section describes the results from the regression described above in Eq. 4. We initially consider the aggregate impact of all disasters combined on inflation. Given the potential for heterogeneity of impact, as discussed in Section “How Disasters May Affect Prices”, we then analyse in turn the effects on inflation by type of disaster, by level of development and by severity of disaster. To verify the robustness of our findings, we also consider two alternative specifications of impact – damage relative to GDP and considering just the number of disasters rather than differentiating by impact.

Aggregate Impact of Disasters on Inflation

We first estimate (4) on the aggregate impact of all disasters, which is to say D i,t is the sum by country and by quarter of the impact across all types of disaster. We include up to 11 lags, since we find joint significance up to three years following the incidence of the disaster. Further lags are not individually or jointly significant. The individual coefficients from the estimation are included in Supplementary Table A1. To aid assessment of the impact of a typical disaster, we multiply the coefficients by the impact value of the 75th and 90th percentile disasters (see Table 1) to give the estimated effect on inflation of these disasters. These estimated impact results are shown in Table 2.
Table 2

Estimated inflation impact of disasters

 

Headline

Food

Housing

Energy

CPIxFHE

75th percentile

Quarter 0

0.266**

0.170 **

− 0.062

− 0.138

0.015

Quarter 1

0.190*

0.170*

− 0.051

− 0.156

− 0.097*

Quarters 2-3

0.151

-0.235*

-0.245**

-0.122

− 0.125

Year 1

0.607*

0.104

-0.358**

-0.415

− 0.207*

Year 2

0.911

− 0.030

− 0.348*

0.153

− 0.142

Year 3

0.770*

0.276

− 0.132

0.203

0.034

90th percentile

Quarter 0

0.806**

0.514**

− 0.189

− 0.418

0.045

Quarter 1

0.577*

0.515*

− 0.155

− 0.472

− 0.294

Quarters 2-3

0.457

-0.713*

− 0.743**

− 0.369

− 0.378

Year 1

1.839 *

0.316

− 1.087**

− 1.259

− 0.627*

Year 2

2.763

− 0.092

− 1.057*

0.463

− 0.432

Year 3

2.336*

0.838

− 0.399

0.616

0.104

Observations

22471

18933

8191

9167

12639

R2

0.051

0.051

0.047

0.172

0.056

Notes: *, ** significant at 5 and 1 percent level respectively. CPIxFHE is consumer prices excluding food, housing and energy. Shows estimated impact for 75th and 90th percentile disaster. Underlying regression coefficients in Supplementary Table A1. Quarter 0 is the quarter the disaster takes place. Year 1 is quarters 0 through 3 combined, year 2 is quarters 4 through 7, year 3 is quarters 8 through 11

Our results estimate that a disaster in the 75th percentile would have a contemporaneous (i.e. quarter 0) impact on headline inflation of 0.27 percentage points (pp). There is a further significant impact of 0.19pp on headline inflation in the quarter immediately following the disaster (quarter 1). Since exact timing of effects may differ between individual disasters, we combine the coefficients for quarters 2 and 3. The impact at this horizon is positive, but insignificant. The combined impact on inflation of the 75th percentile disaster for the first year (quarters 0 through 3) is estimated to be 0.61pp. The impact over the second year (quarters 4 through 7) is estimated to be 0.91pp, although this is not significant. Finally, the impact over the third year (quarters 8 through 11) is significant, and estimated to be 0.77pp.

Turning to the sub-indices, there is a positive and significant contemporaneous impact on food prices of 0.17pp, and a similar impact in the first quarter following the disaster. However, in the subsequent two quarters there is a negative and significant impact on inflation, such that the overall impact on food prices over the first year is insignificant and close to zero. There is no significant impact on food prices beyond the first year.

Housing inflation is significantly reduced in the aftermath of disasters, by 0.36pp and 0.35pp in the first two years following the disaster. There is no significant impact on energy prices. CPI inflation excluding food, housing and energy is significantly lower in the aftermath of disasters, by an estimated 0.21pp in the first year for the 75th percentile disaster. There is no significant impact beyond the first year. Table 2 also includes the estimated figures for the 90th percentile disaster. Since this involves multiplying the underlying coefficients by a larger impact coefficient, the overall pattern of effects and significance are unchanged from the 75th percentile case.

Strictly speaking, it is not possible to draw conclusions from the individual sub-indices for the overall impact on headline inflation. The samples differ for each sub-index because of the lack of availability of some sub-indices. In particular, the sub-indices for housing and energy are frequently unavailable outside of high income countries. There is also a noticeable difference in relative weights in the sub-indices between countries. For example, the weight of food in the index is around 10-15 percent in advanced countries, but frequently exceeds 50 percent in low income countries (Parker 2016). For the purposes of robustness, we include the estimation results on a balanced panel of 78 countries over the period 1996-2012 for the sub-indices for food, the combined housing and energy sub-index and CPIxFHE (see Supplementary Table A2). The sample in this balanced panel is heavily biased towards high income countries, and the estimates are similar in nature to those for this group of countries (see Section “Impact by Level of Development”).

Impact by Type of Disasters

As noted above in Section “How Disasters May Affect Prices”, disasters have heterogeneous impacts on activity, dependent on type. To test whether this finding also holds for inflation, we re-estimate (4) with separate impact variables of each type of disaster. The coefficients from this estimation are shown in Supplementary Tables A3, A4, A5, A6 and A7. The results are summarised in Table 3. We again multiply the coefficients by the impact of the 75th percentile disaster of the relevant type. The ‘other’ category of disasters has almost no significant coefficients, perhaps unsurprising given the diversity of disasters within the category, so these disasters are unreported in Table 3.
Table 3

Impact on inflation by type of disaster

 

Headline

Food

Housing

Energy

CPIxFHE

Earthquakes

     

Quarter 0

0.247*

0.282*

0.051

− 0.233

− 0.093

Quarter 1

0.055

− 0.075

0.194*

0.704*

− 0.223**

Quarters 2-3

− 0.052

− 0.219

− 0.296*

− 0.446

− 0.312**

Year 1

0.250

− 0.012

− 0.051

0.025

− 0.628**

Year 2

0.364

0.273

0.032

1.928

− 0.449**

Year 3

0.218

0.389

− 0.300

− 0.725

− 0.357*

Storms

     

Quarter 0

0.099

0.152*

− 0.088

− 0.522

0.020

Quarter 1

0.060

0.220**

− 0.037

− 0.316

− 0.095

Quarters 2-3

− 0.158

− 0.334**

− 0.259**

0.000

− 0.067

Year 1

0.001

0.038

− 0.384

− 0.839

− 0.141

Year 2

− 0.125

− 0.009

− 0.641**

0.075

− 0.113

Year 3

− 0.066

0.058

− 0.386*

0.160

0.201

Floods

     

Quarter 0

0.382*

0.164

− 0.075

− 0.244

− 0.147

Quarter 1

0.185

0.092

− 0.072

− 0.335

− 0.153**

Quarters 2-3

0.347

− 0.208

− 0.006

− 0.234

− 0.033

Year 1

0.914

0.048

− 0.154

-0.814

− 0.333*

Year 2

1.650

-0.031

− 0.332

1.211*

-0.063

Year 3

1.464

0.255

− 0.116

− 0.047

0.045

Droughts

     

Quarter 0

1.369*

0.535

− 0.079

0.278

0.139

Quarter 1

1.313*

0.441

− 0.251

− 0.272

− 0.171

Quarters 2-3

1.951

− 0.126

− 0.762**

− 0.579

− 0.647**

Year 1

4.634*

0.850

− 1.092

− 0.573

− 0.679

Year 2

7.079

0.121

− 0.682

− 0.708

− 0.207

Year 3

5.217**

1.753

− 0.379

1.168

− 0.134

Observations

22471

18933

8191

9167

12639

R2

0.057

0.054

0.051

0.175

0.062

Notes: *, ** significant at 5 and 1 percent level respectively. CPIxFHE is consumer prices excluding food, housing and energy. Shows estimated impact for 75th percentile disaster. Underlying regression coefficients in Supplementary Tables A3 through A7. Quarter 0 is the quarter the disaster takes place. Year 1 is quarters 0 through 3 combined, year 2 is quarters 4 through 7, year 3 is quarters 8 through 11

Earthquakes do not have a significant impact on headline or food inflation at any horizon. An earthquake in the 75th percentile is estimated to increase housing inflation in the first quarter after it takes place by 0.18pp and energy inflation in that quarter by 0.79pp. These increases appear to be unwound in subsequent quarters, with the estimated impact over the first year combined not significantly different from zero. CPI inflation excluding food, housing and energy is significantly reduced by earthquakes in each of the three years following the disaster, by 0.63pp, 0.45pp and 0.36pp respectively.

Storms are estimated to have a contemporaneous positive impact on headline inflation, and a positive impact the following quarter, although insignificant in both cases. The second quarter following the storm is negative and significant. Overall, the estimated impact for the 75th percentile storm over the first year is 0.00pp. There is no significant impact on headline inflation in subsequent years. There is a significant impact on food price inflation during the first year. A 75th percentile storm significantly increases food price inflation by 0.16pp contemporaneously and by a further 0.22pp in the first quarter following the storm. These increases are unwound in the subsequent two quarters, leaving the total estimated impact over the first year to be insignificantly different from zero. Storms reduce housing price inflation in the three years that follow, by 0.38pp, 0.64pp and 0.39pp respectively, although only the second year is significant. The impact on other sub-indices is insignificant.

The 75th percentile flood is estimated to have a positive and significant contemporaneous impact on headline inflation of 0.38pp. There is estimated to be a positive impact on headline inflation throughout the first three years following the flood, although this is not significant. Energy price inflation is estimated to be lower for the first year, before rebounding in the following year. This would be consistent with plentiful rainfall lowering hydroelectric generation costs in the near term. The 75th percentile flood is estimated to have no significant impact on food or housing price inflation. Inflation in the remainder of the index is estimated to be lower by 0.33pp in the first year following the flood.

The 75th percentile drought is estimated to increase headline inflation by 1.36pp in the start quarter, and by 1.30pp in the subsequent quarter.5 The impact on food price inflation is typically positive, although insignificant. The impact on housing and CPIxFHE price inflation is negative, significantly so in the second and third quarters following the start of the drought.

Impact by Level of Development

Previous research has highlighted that disasters have greater impact on activity in developing economies than in advanced economies (Noy 2009; Raddatz 2009; Fomby et al. 2013). We investigate whether this finding holds for the impact on inflation by estimating (4) separately for advanced countries, other high income countries and for the remaining countries. There are insufficient observations for low income countries, particularly for the sub-indices, to merit estimating these countries separately. The estimated impact for the 75th percentile disaster in each country group is shown in Table 4. The underlying coefficient estimates are provided in Supplementary Tables A8, A9 and A11. Given the relative lack of individual sub-indices for housing and energy in middle and low income countries, we use the combined housing and energy sub-index that is more widely available in these countries.
Table 4

Impact of disasters by level of development

Advanced countries

 

Headline

Food

Housing

Energy

CPIxFHE

Quarter 0

0.005

0.054

− 0.126

− 0.050

− 0.007

Quarter 1

− 0.017

0.097*

− 0.018

− 0.143

− 0.034

Quarters 2-3

− 0.076*

− 0.074

− 0.080

− 0.061

− 0.049

Year 1

− 0.088

0.076

− 0.224

− 0.254

− 0.091*

Year 2

− 0.054

− 0.162*

− 0.251*

− 0.221

− 0.017

Year 3

0.124

0.121

0.229

0.143

0.067

Observations

2783

2741

2167

2715

2591

R2

0.302

0.247

0.172

0.507

0.361

Other high income countries

 

Headline

Food

Housing

Energy

CPIxFHE

Quarter 0

1.067

0.226*

− 0.007

− 2.427**

0.029

Quarter 1

1.152

0.296*

0.041

− 0.514**

− 0.177

Quarters 2-3

0.745*

0.267

0.048

0.906*

− 0.055

Year 1

2.965

0.789*

0.082

− 2.035*

− 0.204

Year 2

0.951*

0.920**

0.086

0.201

− 0.189

Year 3

0.690*

0.072

0.077

− 0.201

0.117

Observations

4887

4106

2486

2465

2852

R2

0.095

0.181

0.121

0.302

0.151

Middle and low income countries

 

Headline

Food

Housing & Energy

CPIxFHE

 

Quarter 0

0.273**

0.184**

0.061

0.025

 

Quarter 1

0.183*

0.173*

− 0.048

− 0.081

 

Quarters 2-3

0.164

− 0.250*

0.053

− 0.144

 

Year 1

0.620*

0.108

0.065

− 0.200

 

Year 2

1.007

− 0.025

0.000

− 0.121

 

Year 3

0.832*

0.307

− 0.080

0.053

 

Observations

14801

12086

7301

7196

 

R2

0.059

0.051

0.068

0.062

 

Notes: *, ** significant at 5 and 1 percent level respectively. CPIxFHE is consumer prices excluding food, housing and energy. Shows estimated impact for disaster in 75th percentile. Underlying regression coefficients in Supplementary Tables A8, A9 and A11. Quarter 0 is the quarter the disaster takes place. Year 1 is quarters 0 through 3 combined, year 2 is quarters 4 through 7, year 3 is quarters 8 through 11

Disasters do not have significant impact on either headline, food or energy price inflation in advanced countries. Housing price inflation is significantly lower in the second year after the disaster, by 0.25pp for the 75th percentile disaster. CPIxFHE inflation is significantly lower in the first year following the disaster, by 0.09pp.

In other high income countries, the 75th percentile disaster is estimated to increase headline inflation 2.97pp over the first year, but only the increase in quarters 2 and 3 is significant. There are significant increases in the second and third year after the disaster, by 0.95pp and 0.69pp respectively. Food price inflation is significantly increased in the first two years following the disaster, conversely energy prices fall. There are no significant impacts on the other sub-indices.

There are insufficient events to consider the impact by disaster separately for advanced and other high income countries, so Supplementary Table A10 considers the impact by type of disaster for all high income countries. The 75th percentile earthquake in high income countries has no significant effect on headline inflation, but significantly reduces CPIxFHE inflation by 1.43pp in the first year, by 1.62 in the second year and by 1.61pp in the third year. The 75th percentile storm has a positive impact on headline inflation. Over the first year, the estimated impact is 4.59pp, although this is insignificant. For the second and third year the impact is positive and significant at 1.50pp and 0.96pp respectively. Food price inflation is higher by 0.96pp in the first year and by 1.27pp in the second year.

For middle and low income countries, the 75th percentile disaster is estimated to increase headline inflation by 0.62pp in the first year, by (an insignificant) 1.01pp in the second year and by 0.83pp in the third year. By sub-component, food price inflation is significantly higher in the quarter that the disaster takes place and in quarter 1. But in the subsequent two quarters, this higher inflation is partly reversed, such that the combined impact for the first year is insignificant. Disasters do not have significant impact on the other sub-components in middle and low income countries. Split by type of disaster (see Supplementary Table A12), earthquakes lower CPIxFHE inflation, and the 75th percentile drought has a large and positive impact on headline inflation: 4.99pp in the first year, 7.34pp (insignificant) in the second year and 5.37pp in the third year.

Impact by Severity of Disaster

Given the heavily skewed distribution of disaster impacts, it is possible that there are non-linearities in their effect on inflation. We construct a series for the impact of severe disasters, S E V I M P t,i in an analogous fashion to Eqs. 2 and 3, but set the cutoff threshold to be the 75th percentile of the distribution. Thus the upper quartile of disasters – 337 disasters in total – are classified as ‘severe’. We then estimate (4) including both I M P t,i and S E V I M P t,i in the vector of impact variables, D i,t . The estimated coefficients on the I M P t,i variables represent the impact of major disasters (those in the first three quartiles of disaster impact) on inflation. The coefficients on S E V I M P t,i capture any additional effect on inflation from severe disasters.

Given that the effect of disasters differs by level of development (Section “Impact by Level of Development”), we estimate high income countries separate from middle and low income countries. Table 5 shows the estimated impact on inflation of a (severe) disaster in the 90th percentile, split by level of development.
Table 5

Impact of severe disasters on inflation

High income countries

 

Headline

Food

Housing

Energy

CPIxFHE

Quarter 0

1.645

0.251

− 0.476*

− 3.640

0.086

Quarter 1

2.029

0.717**

− 0.059

− 0.331

− 0.470

Quarters 2-3

0.322

− 0.437

− 0.600*

0.435

− 0.235

Year 1

3.996*

0.530

− 1.136**

− 3.536

− 0.619

Year 2

0.888*

0.749

− 0.744*

− 0.633

− 0.662*

Year 3

0.395

− 0.786

0.236

− 0.133

− 0.171

Observations

7670

6847

4653

5180

5443

R2

0.079

0.166

0.107

0.341

0.169

Middle and low income countries

 

Headline

Food

Housing & Energy

CPIxFHE

 

Quarter 0

0.632*

0.563**

0.305

0.044

 

Quarter 1

0.491

0.453*

− 0.117

− 0.277

 

Quarters 2-3

0.591

− 0.517

0.193

− 0.350

 

Year 1

1.714

0.499

0.381

− 0.582

 

Year 2

3.232

− 0.067

− 0.004

− 0.338

 

Year 3

2.675*

1.128

− 0.041

0.263

 

Observations

14801

12086

7301

7196

 

R2

0.060

0.053

0.070

0.064

 

Notes: *, ** significant at 5 and 1 percent level respectively. CPIxFHE is consumer prices excluding food, housing and energy. Shows estimated impact for disaster in 90th percentile. Underlying regression coefficients in Supplementary Tables A13 through A16. Quarter 0 is the quarter the disaster takes place. Year 1 is quarters 0 through 3 combined, year 2 is quarters 4 through 7, year 3 is quarters 8 through 11

For high income countries, there is a significant positive impact on inflation in the first two years following a severe disaster (Supplementary Table A14). The impact on housing price inflation is negative for the first two years. A split by disaster type is not worthwhile for high income countries. There are only 23 severe disasters, and individual types are concentrated in certain countries. For example, the four earthquakes are split evenly between Chile and New Zealand, with the two New Zealand earthquakes taking place less than six months apart.

For middle and low income countries, there are a number of significant coefficients on the severe disaster variables. The reversal in the impact on headline and food price inflation typically seen in quarters 2 and 3 is not as pronounced, and is no longer significant. The impact on the other sub-indices is more positive for severe relative to major disasters, significantly so in the third year following the disaster, although the aggregate impact of severe disasters on these sub-indices remains insignificant. The impact of severe disasters by type of disaster in middle and low income countries is similar in pattern to that when all disasters are combined (Supplementary Table A17).

Alternative Measure – Underlying Natural Hazards

Previous studies of the impact of disasters have used a number of different metrics of impact, each of which has its own disadvantages. For the purposes of robustness, we re-ran the analysis using two other measures of disasters from the EM-DAT dataset that are commonly used in the literature – number of disasters and damage caused as a share of GDP. The results for these measures are not shown here, but are qualitatively similar to the main results shown above.6

As noted earlier, the impact of disasters is complex, and is a function of a number of institutional factors. These institutional factors may in turn also affect the inflationary process, raising concerns around potential estimation bias. Previous authors have also pointed to a number of omissions in the EM-DAT database (e.g. Cavallo and Noy 2011; Felbermayr and Gröschl 2014), and there is the potential for bias in inclusion if developing countries are more likely to declare a state of emergency in the hope of receiving additional aid.

To address these issues, we carry out a final robustness check by using the Ifo GAME database of geological and meteorological hazards, presented in Felbermayr and Gröschl (2014). This database measures the strength of the ex ante hazards – earthquake strength, wind speed, rainfall – rather than ex post measures of disaster impact. This has the advantage of being truly exogenous to the economic process. Conversely, ex ante hazards do not necessarily translate directly to diaster potential - a strong earthquake that takes place in a remote, unpopulated area has smaller disaster potential than one that takes place at the same time, in the same country, but in a more populated and economic active region. As a result, ex post measures can provide a more precise assessment of the impact, which is in turn more useful for determining impacts on economic variables.

We convert the monthly GAME data to quarterly frequency in a manner equivalent to that taken by Felbermayr and Gröschl (2014) for annual frequency. For earthquakes, we take the maximum strength earthquake that takes place in the quarter. Since smaller earthquakes are unlikely to cause widespread damage, we discard the lower quartile, resulting in a cut-off value of 4 on the Richter scale. For storms, we take the maximum total wind speed in the quarter. Again, the lower quartile is discarded, resulting in a minimum threshold of 35 knots, which is rated 7 on the Beaufort Scale and rarely associated with significant damage or destruction.

For floods, we average across the quarter the monthly percentage deviation of precipitation from the long-run average. Where that quarterly average is below zero (i.e. precipitation is below average), we set the flood index to zero. For droughts, we carry out the same procedure as for floods, but exclude all values above -50, which is to say precipitation must be less than half the long-run average to signal a drought. For ease of comparison of coefficients, we multiply the drought index by -1, so that higher numbers represent a more severe lack of precipitation. As noted above, the disadvantage of EM-DAT is that droughts are not always provided with a precise date, so overall coverage is low for the analysis above. The advantage of the GAME dataset is that it provides much greater coverage, but the impact of sustained, multi-year events are less easily calculated than using the ex post impacts incorporated into EM-DAT.

Using the impact indices derived from the GAME dataset, we re-run (4). The results are presented in Table 6. We standardise for the 75th percentile of the underlying natural hazard, in keeping with the main results.
Table 6

Impact of natural hazards on inflation

 

Headline

Food

Housing

Energy

CPIxFHE

Earthquakes

     

Quarter 0

0.110

0.125

0.053

0.153

− 0.120

Quarter 1

0.007

− 0.077

0.063

0.420*

0.066

Quarters 2-3

0.181

− 0.072

− 0.267

0.142

− 0.117

Year 1

0.298

− 0.025

− 0.151

0.715

− 0.171

Year 2

− 0.660

− 0.074

− 0.139

− 0.158

− 0.528*

Year 3

− 0.115

− 0.170

0.132

0.193

− 0.231

Storms

     

Quarter 0

− 0.173

0.034

− 0.009

− 0.710**

− 0.189

Quarter 1

− 0.208

− 0.028

0.155

− 0.083

0.140

Quarters 2-3

− 0.462

− 0.424*

− 0.078

0.182

0.021

Year 1

− 0.842

− 0.419

0.068

− 0.611

− 0.028

Year 2

0.564

0.224

− 0.333

− 0.146

0.050

Year 3

0.787

0.084

− 0.537

− 0.192

0.361

Floods

     

Quarter 0

− 0.075*

− 0.014

0.036

0.088

0.021

Quarter 1

0.044

0.071

0.041

− 0.054

0.025

Quarters 2-3

0.015

0.088

0.044

− 0.048

− 0.033

Year 1

− 0.015

0.145

0.121

− 0.014

0.014

Year 2

− 0.004

0.000

0.031

0.122

0.078

Year 3

0.089

0.113

0.015

0.091

0.113*

Droughts

     

Quarter 0

− 0.067

− 0.318

0.191

0.353

− 0.034

Quarter 1

0.396

0.294

0.151

− 0.074

0.210

Quarters 2-3

0.652

1.229**

0.009

− 0.806

0.101

Year 1

0.980

1.205**

0.351

− 0.526

0.277

Year 2

− 0.161

− 0.410

− 0.762

0.415

− 0.717

Year 3

0.743

0.466

− 0.237

0.706

0.043

Observations

16875

13852

5806

6799

9090

R2

0.056

0.062

0.085

0.199

0.068

Notes: *, ** significant at 5 and 1 percent level respectively. CPIxFHE is consumer prices excluding food, housing and energy. Shows estimated impact for hazards in 75th percentile of distribution. Underlying regression coefficients in Supplementary Tables A18, A19, A20 and A21. Quarter 0 is the quarter the hazard takes place. Year 1 is quarters 0 through 3 combined, year 2 is s quarters 4 through 7, year 3 is quarters 8 through 11

Overall, the impact of disasters on inflation using the GAME database is qualitatively in line with the main results. Earthquakes have a short-term positive impact in Q1 on energy prices and overall put downward pressure on CPI excluding food, housing and energy, including by a significant 0.5pp in year 2. Storms have an initial upward pressure on food prices, but this is unwound in quarters 2-3.

Droughts are found to have a significant, upward impact on food prices in the first year. The contrast with the main results lies with the likely difference in coverage of the two databases. GAME covers far more events, but the index does not pick up the impact of prolonged, severe events to the same extent that EM-DAT does. Nonetheless, the overall consistency of the results between the two measures of disasters – ex ante hazards and ex post impact – provides comfort that the main results presented in Section “Results” are robust.

Discussion of Results

That the impact of disasters on inflation varies so much by type, severity, location, and sector of the economy is not surprising. Previous research has highlighted the varying impact on economic activity across those dimensions. As noted in Section “How Disasters May Affect Prices”, disasters are complex and the impact on CPI is uncertain in both the short run and the medium run given potential impacts on both supply and demand. Since differing types of disasters have varying impacts on crops, buildings and infrastructure, there is an inevitable difference on the effects on the various sub-indices of consumer prices. This section provides some candidate explanations for the diversity of results found here, building on the results found here and on insights garnered from the existing literature.

Earthquakes

Earthquakes cause significant destruction of physical capital, in terms of housing, business premises and infrastructure. There is also evidence of a reduction in agricultural output in developing countries (Fomby et al. 2013). After the initial shock, reconstruction takes place, bringing about an eventual lift in economic activity. Initial impacts therefore include shortages of housing and potentially food.

The findings here suggest significant upward initial impact on overall prices and food prices. This contrasts with the findings of Abe et al. (2014), Reinsdorf et al. (2014) and Doyle and Noy (2015) who do not find significant impacts when studying earthquakes in Japan, Chile and New Zealand. The contrast in findings can be explained by differences in sample. These previous studies all involved high income countries, and the positive impact found here is confined to middle and low income countries.

CPI housing prices (essentially rent) rise in the quarter after the earthquake, likely a result of a reduction in housing stock putting pressure on rents for houses that remain habitable. This accords with the findings of higher rents by Parker and Steenkamp (2012) and Wood et al. (2016). The significant higher energy prices in the short term may reflect a reduction in electricity generating capabilities if power stations are damaged in the earthquake. The damage to the Fukushima Daiichi nuclear power plant following the Great East Japan earthquake in 2011 is a recent example.

The other notable feature of earthquakes is the persistent negative impact on CPI excluding food, housing and energy. This downward impact lasts for three years following the earthquake. The damage to capital stock and wealth appears to depress activity in these sectors. This explanation is corroborated by Fomby et al. (2013), who find that the destruction of capital stock and loss of labour supply cancels out the boost to activity from reconstruction, resulting in no overall significant effect on output from earthquakes.

Storms

Storms similarly have the potential to damage buildings and crops, although horizontal infrastructure such as roads is less prone to damage than from earthquakes. The findings here are certainly consistent with short-term destruction of crops, although within a year the upward impact on food price inflation is unwound. Despite being large, the impact is not persistent.

The impact on headline CPI in the initial quarter of 0.10 for a 75th percentile storm is insignificantly different from the average value of 0.05 found by Heinen et al. (2015). Referencing to just the 15 economies covered by that paper gives a somewhat closer figure of 0.087. The finding by Heinen et al. (2015) that food prices are more heavily affected than headline inflation is confirmed by the results here. Where the findings here diverge from Heinen et al. (2015) is over the horizon that these positive impacts take place. In that paper, the effect on food prices is only positive. Yet as shown here, the duration of effects on food prices is short-lived, and is reversed beyond the horizon of analysis in Heinen et al. (2015).

CPI housing falls in the years following a storm. This appears at odds with a shortage of housing caused by destruction of the housing stock. The answer may lie in an overall drop in household spending power from lower economic activity and lower wealth. It may also arise from outward migration from the affected area, as noted by Coffman and Noy (2012) and Vigdor (2008) in their studies of the impacts of respectively Hurricane Iniki and Hurricane Katrina.

Floods

The damage from floods differs from earthquakes and storms in that it may be more localised, for example riverine floods are restricted to the immediate proximity of rivers. Depending on severity, flood damage to housing may be superficial rather than structural, although there is likely to be loss incurred to contents. Initial damage to crops may be offset in future years by increases to fertility with some beneficial effects to agriculture in later years (Fomby et al. 2013).

There is evidence of an initial positive impact to overall CPI. In middle and low income countries, there is also an initial positive impact on food, likely a result of crop destruction. In high income countries, the potential benefits from greater agricultural production are reflected in a negative (albeit insignificant) impact on food prices in the years following the flood. This beneficial impact does not appear to occur with severe floods. Here, the damage to agricultural land, perhaps even including washing away topsoil, appears to outweigh any nutrient gain. Fomby et al. (2013) also find no beneficial effect from severe floods.

In terms of case studies for where the inflation impact of floods has been previously estimated, Laframboise and Loko (2012) estimate that the 2010 floods in Pakistan increased headline CPI by 2 percentage points. The model estimates here (noting that this is a severe flood in a middle income country) suggest a positive headline CPI impact of 1.6 percentage points for this event, around the order of magnitude experienced.

Droughts

Droughts differ from the other types of diaster studied here insofar as they inflict little by way of physical damage to buildings or infrastructure. Nonetheless, they can have significant impact on agricultural production, and can reduce reduce productive capital through killing livestock. Their impact can be particularly marked in developing countries where a greater share of the population is in poverty and the share of food expenditure in total consumption is much greater.

As noted in Section “Disasters”, the EM-DAT dataset does not always have precise timings for droughts, so the results from the GAME dataset provide a worthwhile addition. The positive effects on food prices and negative effects on other components of CPI found here are consistent with droughts creating shortages in food and consumers therefore constraining non-food expenditure. The results here are in line with the findings of Kamber et al. (2013) for New Zealand.

A further difference between droughts and the other main disaster types is the duration of impacts. Droughts maintain their impact on CPI into the third year, whereas the impact of disasters tends to be insignificant by that point. There are a range of potential explanations for this increased duration. First, droughts themselves may last over a number of years, with impact typically worsening as successive crops are devastated. Second, the impact of droughts can be more widespread: widespread rainfall may result in localised flooding, but droughts can affect a large area. Finally, droughts are more prevalent in middle and low income countries, where the impact on inflation tends to be higher.

Level of Development and Severity of Disaster

The findings here point to a greater inflationary impact from disasters in developing countries. This result is to be expected as a number of authors (e.g. Noy 2009; Raddatz 2009; Fomby et al. 2013) find that the impact of disasters on output is greater in these countries. It follows that the impact on inflation is likely to be greater.

Institutional quality was found by Noy (2009) to be one of the determining factors in the severity of output loss from a disaster. Institutional quality – particularly central bank independence – is also known to have a bearing on inflation outcomes, including inflation volatility (See, among others, Cukierman et al. 1992; Alesina and Summers 1993; Klomp and de Haan 2010). Where monetary policy is more credible, inflation expectations are more likely to remain anchored, reducing the impact of the disaster on inflation. Since central bank credibility tends to be positively correlated with development, a lower impact of inflation from a given disaster impact is likely on average in more developed economies.

The proportionately greater impact of larger disasters on inflation is also in line with the findings of the literature on the effects on economic activity. Fomby et al. (2013) and Loayza et al. (2012) both find that while some types of disasters (such as floods) can have overall positive impacts on output, this positive effect disappears once disasters cross a certain severity threshold. Larger disasters exceed an economy’s capacity to remain resilient.

Conclusion

This paper has analysed the impact of disasters on inflation, using a panel of consumer price indices for 212 countries. It extends the literature by being the first to jointly model earthquakes, storms, floods and droughts together, by expanding the coverage from single events or limited number of countries to as close as possible to the universe of countries, and also extends the timeframe of analysis beyond the very near term. As such, it provides a comprehensive analysis of the impact of disasters on inflation to match the existing literature on the impact on output (e.g. Noy 2009; Raddatz 2009; Loayza et al. 2012; Fomby et al. 2013; Felbermayr and Gröschl 2014). The findings point to a considerable heterogeneity in the impact of disasters by type of disaster, by sub-index of CPI, by level of development and by timing.

There is a clear differentiation in the inflation impact of disasters by level of development. The impact of disasters in advanced countries is for the most part insignificant, and even where there is a significant impact on a sub-index, its magnitude is negligible. Conversely, the impact for less developed countries is more marked, with significant effects on headline inflation persisting even three years post-disaster. That said, there is a significant impact in high income countries from severe disasters (those in the upper quartile).

In terms of sub-indices, the impact on food price inflation is in general positive, if short lived. The impact on housing and other sub-indices is in general negative. Differences in expenditure weights on these sub-indices will in part explain the differences witnessed in headline inflation numbers by level of development.

Earthquakes do not significantly affect headline inflation, but do significantly reduce CPI inflation excluding food, housing and energy. Storms cause an immediate increase in food price inflation for the first six months, although this impact is reversed in the subsequent two quarters, resulting in no significant impact over the entire first year, or beyond. Floods increase headline inflation in the quarter that the flooding occurs in middle and low income countries, but have no significant impacts in subsequent quarters. In high income countries, the impact on headline inflation is negative, although insignificant. Droughts increase headline inflation for a number of years.

The results here provide a starting place for future research into the impact on prices. Despite the overall uncertainty, they provide a number of lessons for policymakers when considering the impact of disasters on inflation. First, movements in inflation, even if short lived, can be substantial. In the context of developing countries where food expenditures often exceed half the household’s budget, marked shifts in food prices can put notable pressure on poorer households nationwide, beyond just those located in the directly affected area.

Second, it is clearly important to focus on the entire general equilibrium effects of income, wealth and economic activity on prices, rather than focusing narrowly on disaster-induced shortages of goods. Indeed, beyond shortages in the most directly affected sectors (food and housing), disasters appear on balance to be deflationary. It follows that – particularly over the medium term – the impact of indirect disruption and lower consumption from reduced wealth tend to dominate any inflationary forces arising from rebuild activity. This is the reverse of the findings in Keen and Pakko (2011) using a calibrated DSGE model, and as such the recommended tightening of monetary policy does not appear supported by real-world data. The results here instead suggest that forward-looking central banks reacting to medium-term inflationary pressures should, if anything, loosen policy.

Footnotes

  1. 1.

    The estimation here additionally includes mass movements, insect infestations, extreme temperatures, volcanoes and wildfires, but since those disasters combined are not found to have significant impacts we do not discuss these disaster types in any detail.

  2. 2.

    In some jurisdictions it is illegal to increase prices of certain goods, termed ‘price gouging’, in the immediate aftermath of a disaster (Gerena 2004).

  3. 3.

    For robustness, we also estimate using country seasonal dummies and using data seasonally adjusted using X12 (not reported). The results using the seasonally adjusted data are qualitatively similar to those presented here, although the impact of windstorms on food is no longer significant. As noted above, this lack of significance is unsurprising.

  4. 4.

    The Pesaran (2004) test is carried out in Stata using the xtcsd command of Hoyos and Sarafidis (2006). The estimation using Driscoll and Kraay (1998) adjusted standard errors is implemented using the xtscc command developed by Hoechle (2007).

  5. 5.

    Note that unlike the other disasters considered here, droughts may continue for several quarters, indeed even years. The 75th percentile drought is also much greater in impact than the 75th percentile of the other disasters.

  6. 6.

    Regression results are available on request.

Notes

Acknowledgements

The author thanks Ilan Noy, Eric Strobl and Paul Raschky for useful comments.

Supplementary material

41885_2017_17_MOESM1_ESM.pdf (121 kb)
(PDF 121 KB)

References

  1. Abe N, Moriguchi C, Inakura N (2014) The effects of natural disasters on prices and purchasing behaviors: the case of the great East Japan earthquake. RCESR Discussion Paper Series DP14-1 Research Center for Economic and Social Risks. Institute of Economic Research, Hitotsubashi University. http://ideas.repec.org/p/hit/rcesrs/dp14-1.html
  2. Alesina A, Summers LH (1993) Central bank independence and macroeconomic performance: some comparative evidence. J Money Credit Bank 25 (2):151–162. https://ideas.repec.org/a/mcb/jmoncb/v25y1993i2p151-62.html CrossRefGoogle Scholar
  3. Barro RJ (2009) Rare disasters, asset prices, and welfare costs. Amer Econ Rev 99(1):243–64. https://ideas.repec.org/a/aea/aecrev/v99y2009i1p243-64.html CrossRefGoogle Scholar
  4. Boustan LP, Kahn ME, Rhode PW (2012) Moving to higher ground: migration response to natural disasters in the early twentieth century. Amer Econ Rev 102(3):238–44. https://ideas.repec.org/a/aea/aecrev/v102y2012i3p238-44.html CrossRefGoogle Scholar
  5. Buckle RA, Kim K, Kirkham H, McLellan N, Sharma J (2007) A structural VAR business cycle model for a volatile small open economy. Econ Modell 24(6):990–1017. http://ideas.repec.org/a/eee/ecmode/v24y2007i6p990-1017.html CrossRefGoogle Scholar
  6. Carter MR, Little PD, Mogues T, Negatu W (2007) Poverty traps and natural disasters in Ethiopia and Honduras. World Develop 35(5):835–856. https://ideas.repec.org/a/eee/wdevel/v35y2007i5p835-856.html CrossRefGoogle Scholar
  7. Cavallo E, Noy I (2011) Natural disasters and the economy - a survey. Int Rev Environ Resour Econ 5(1):63–102. http://ideas.repec.org/a/now/jirere/101.00000039.html CrossRefGoogle Scholar
  8. Cavallo E, Galiani S, Noy I, Pantano J (2013) Catastrophic natural disasters and economic growth. Rev Econ Stat 95(5):1549–1561. https://ideas.repec.org/a/tpr/restat/v95y2013i5p1549-1561.html CrossRefGoogle Scholar
  9. Coffman M, Noy I (2012) Hurricane Iniki: measuring the long-term economic impact of a natural disaster using synthetic control. Environ Dev Econ 17:187–205CrossRefGoogle Scholar
  10. Cukierman A, Webb SB, Neyapti B (1992) Measuring the independence of central banks and its effect on policy outcomes. World Bank Econ Rev 6(3):353–398. https://ideas.repec.org/a/oup/wbecrv/v6y1992i3p353-98.html CrossRefGoogle Scholar
  11. Dell M, Jones BF, Olken BA (2014) What do we learn from the weather? The new climate- economy literature. J Econ Literat 52 (3):740–798. https://ideas.repec.org/a/aea/jeclit/v52y2014i3p740-98.html CrossRefGoogle Scholar
  12. Doyle L, Noy I (2015) The short-run nationwide macroeconomic effects of the canterbury earthquakes. New Zealand Econ Papers 49(2):143–156CrossRefGoogle Scholar
  13. Driscoll JC, Kraay AC (1998) Consistent covariance matrix estimation with spatially dependent panel data. Rev Econ Stat 80(4):549–560. https://ideas.repec.org/a/tpr/restat/v80y1998i4p549-560.html CrossRefGoogle Scholar
  14. Felbermayr G, Gröschl J (2014) Naturally negative: the growth effects of natural disasters. J Develop Econ 111(C):92–106. https://ideas.repec.org/a/eee/deveco/v111y2014icp92-106.html CrossRefGoogle Scholar
  15. Fomby T, Ikeda Y, Loayza NV (2013) The growth aftermath of natural disasters. J Appl Economet 28(3):412–434. http://ideas.repec.org/a/wly/japmet/v28y2013i3p412-434.html CrossRefGoogle Scholar
  16. Gerena C (2004) States move to prevent post-disaster price gouging. Econ Focus (Fall) 8. https://ideas.repec.org/a/fip/fedrrf/y2004ifallp8nv.8no.4.html
  17. Heinen A, Khadan J, Strobl E (2015) The inflationary costs of extreme weather. mimeo. Ecole PolytechniqueGoogle Scholar
  18. Hoechle D (2007) Robust standard errors for panel regressions with cross-sectional dependence. Stata J 7(3):281–312. https://ideas.repec.org/a/tsj/stataj/v7y2007i3p281-312.html Google Scholar
  19. Hornbeck R (2012) The enduring impact of the american dust bowl: short- and long-run adjustments to environmental catastrophe. Amer Econ Rev 102(4):1477–1507. https://ideas.repec.org/a/aea/aecrev/v102y2012i4p1477-1507.html CrossRefGoogle Scholar
  20. Hornbeck R, Naidu S (2014) When the levee breaks: black migration and economic development in the American South. Amer Econ Rev 104(3):963–90. https://ideas.repec.org/a/aea/aecrev/v104y2014i3p963-90.html CrossRefGoogle Scholar
  21. Hoyos RED, Sarafidis V (2006) Testing for cross-sectional dependence in panel-data models. Stata J 6(4):482–496. https://ideas.repec.org/a/tsj/stataj/v6y2006i4p482-496.html Google Scholar
  22. Kamber G, McDonald C, Price G (2013) Drying out: investigating the economic effects of drought in New Zealand. Reserve Bank of New Zealand Analytical Notes series AN2013/02, Reserve Bank of New Zealand. http://ideas.repec.org/p/nzb/nzbans/2013-02.html
  23. Keen BD, Pakko MR (2011) Monetary policy and natural disasters in a DSGE model. Southern Econ J 77(4):973–990. http://ideas.repec.org/a/sej/ancoec/v774y2011p973-990.html CrossRefGoogle Scholar
  24. Klomp J, de Haan J (2010) Inflation and central bank independence: a meta-regression analysis. J Econ Surv 24(4):593–621. https://ideas.repec.org/a/bla/jecsur/v24y2010i4p593-621.html CrossRefGoogle Scholar
  25. Laframboise N, Loko B (2012) Natural disasters; mitigating impact, managing risks. IMF Working Papers 12/245, International Monetary Fund. http://ideas.repec.org/p/imf/imfwpa/12-245.html
  26. Loayza NV, Olaberra E, Rigolini J, Christiaensen L (2012) Natural disasters and growth: going beyond the averages. World Develop 40(7):1317–1336. http://ideas.repec.org/a/eee/wdevel/v40y2012i7p1317-1336.html CrossRefGoogle Scholar
  27. Munoz E, Pistelli A (2010) Tienen los Terremotos un Impacto Inflacionario en el Corto Plazo? Evidencia para una Muestra de Paises. Notas de Investigacion Journal Economia Chilena (The Chilean Economy) 13(2):113–127. http://ideas.repec.org/a/chb/bcchni/v13y2010i2p113-127.html Google Scholar
  28. Neely CJ (2011) A foreign exchange intervention in an era of restraint. Review (Sep) 303–324. https://ideas.repec.org/a/fip/fedlrv/y2011isepp303-324nv.93no.5.html
  29. Noy I (2009) The macroeconomic consequences of disasters. J Develop Econ 88 (2):221–231. http://ideas.repec.org/a/eee/deveco/v88y2009i2p221-231.html CrossRefGoogle Scholar
  30. Parker M (2016) Global inflation: the role of food, housing and energy prices. Discussion Paper DP2016/05, Reserve Bank of New ZealandGoogle Scholar
  31. Parker M, Steenkamp D (2012) The economic impact of the Canterbury earthquakes. Reserve Bank New Zealand Bull Reserve Bank New Zealand 75:13–25. http://ideas.repec.org/a/nzb/nzbbul/sep201206.html Google Scholar
  32. Pesaran M (2004) General diagnostic tests for cross section dependence in panels. Cambridge Working Papers in Economics 0435. Faculty of Economics, University of Cambridge. https://ideas.repec.org/p/cam/camdae/0435.html Google Scholar
  33. Raddatz C (2009) The wrath of God: macroeconomic costs of natural disasters. Policy Research Working Paper Series 5039, The World Bank. http://ideas.repec.org/p/wbk/wbrwps/5039.html
  34. Ramcharan R (2007) Does the exchange rate regime matter for real shocks? Evidence from windstorms and earthquakes. J Internat Econ 73(1):31–47. http://ideas.repec.org/a/eee/inecon/v73y2007i1p31-47.html CrossRefGoogle Scholar
  35. Reinsdorf M, Hill R, Cavallo A, Cavallo E, Rigobon R (2014) Prices and supply disruptions during natural disasters. Rev Income Wealth 60:S449–S471. http://ideas.repec.org/a/bla/revinw/v60y2014ips449-s471.html CrossRefGoogle Scholar
  36. Rotemberg JJ (2005) Customer anger at price increases, changes in the frequency of price adjustment and monetary policy. J Monetary Econ 52(4):829–852. http://ideas.repec.org/a/eee/moneco/v52y2005i4p829-852.html CrossRefGoogle Scholar
  37. Vigdor J (2008) The economic aftermath of hurricane katrina. J Econ Perspect 22(4):135–54. https://ideas.repec.org/a/aea/jecper/v22y2008i4p135-54.html CrossRefGoogle Scholar
  38. Wood A, Noy I, Parker M (2016) The Canterbury rebuild five years on from the Christchurch earthquake. Reserve Bank New Zealand Bull 79(3):1–16. https://ideas.repec.org/a/nzb/nzbbul/feb201603.html Google Scholar

Copyright information

© Springer International Publishing AG, part of Springer Nature 2017

Authors and Affiliations

  1. 1.Reserve Bank of New ZealandWellingtonNew Zealand
  2. 2.European Central BankFrankfurt am MainGermany

Personalised recommendations