A quick method for assessing economic damage caused by natural disasters: an epidemiological approach.
Keywords Direct economic damage * Natural disaster * Epidemiological model
JEL 3 * 15 * 29 * 31
Japan is a natural disaster prone country. Figure 1 compares the top 10 countries in terms of the occurrences of natural disaster per 100,000 square kilometers of land territory for the period 1990-2010. Japan ranks number two after the Philippines. Among the high income countries, however, Japan stands at the top by the same metric.
[FIGURE 1 OMITTED]
In any occurrence of a natural disaster, it is imperative, as well as academically challenging, to assess the value of economic damage involved. From a practical point of view, assessment needs to be done in the midst of an on-going emergency, where time is the most scarce resource of all. In retrospect, it must be conducted in order to search for possible policy measures for future mitigation.
However, as Cochrane (2004, pp.290-291) says in relation to the 9/11 incident in New York, 2001, "Loss accounting, as currently practiced, is problematic for several reasons." He goes on to say that" ... most problems stem from double counting, failure to identify clearly an accounting stance, ignoring non-market losses, confusion as to whether post-disaster economic trends are a product of the event or some other unrelated factor, and the employment of too limited a time frame."
Thus, the purpose of this paper is to present a simple epidemiological model for assessing economic damage, which is applicable to any event or natural disaster. The method comprises an alternative to the common accounting practice of aggregation by parts.
In the next section, we will explain what is commonly done in Japan regarding the economic damage measurement. In the section following that, we will present a social epidemiological model of hazard and vulnerability leading to economic damage. We will then examine the data availability and select variables to be used for estimation. Finally, we will present the fixed effect estimation result.
In the section after that, we will propose yet another model of a simple equation, which is convenient for a quick estimation. By applying this method to calculate the damage of the North East Japan earthquakes and tsunami of March 11, 2011 (the Tohoku earthquake), we conclude that the true direct economic damage may well be the double of government's estimation.
Measurement of Post-Disaster Economic Damage: The Common Practice
The Fire and Disaster Management Agency (FDMA) of the Japanese government has been compiling statistics regarding natural disasters based on reports filed by 47 prefectures. (1) The FDMA classification of disaster damage is shown in Table 1. All types of damage are reported in physical units except economic damage.
Table 1 FDMA classification of Natural disasters damage Damage Type of damage Unit classification Human damage dead person missing person injured seriously person injured injured person affected population affected household household affected person population Building damage housing damage totally building destroyed half building destroyed partially building destroyed inundated building above floor inundated building below floor- non-housing public building buildings other building Others rice field destroyed ha rice field flooded ha fruit and vegetable ha patch destroyed fruit and vegetable ha patch flooded Public public school campus infrastructure infrastructure bridge bridge river site landslide site Transportation railway blocked site ship ship Economic damage thousand yen
Data is compiled on a yearly basis, and it covers the period 1995-2007. Hence, we have a cross section of 47 prefectures and 24 types of damage each year, and the data runs over 13 years now, which is enough to configure a panel data.
However, there is a critical flaw in the economic damage data. (2) In order to calculate this number, FDMA adds up numbers compiled by prefectures. Prefectures are interested only in damage inflicted on infrastructure, public buildings, and agriculture and fishery facilities, (3) and they report just that. Hence, the government data on economic damage excludes damage done to an important segment of the economy: the private sector. This is a serious omission, since disasters that hit urban areas bring about more damage to residential houses, condominiums, commercial buildings, and office buildings than to public infrastnicture. (4) This means that we must augment the FDMA data to cover all direct economic damage prior to our analysis.
An Epidemiological Analysis of Economic Damage
We start from a conceptual framework: economic damage is a product of hazard and the community's vulnerability factors. (5) Measurement of the magnitude of hazard is usually done in terms of some physical unit. For example, a number in Richter magnitude scale is commonly used to represent the energy released in the case of earthquakes. In case of meteorological disasters, precipitation, or wind velocity, the Saffir-Simpson category scale is used.
However, when dealing with all natural disasters as we are doing here, we need to find some social index that represents the scale and frequency of disasters that hit the area in a year. This is an empirical question to which we will try to give our answer in the next section.
Regarding the vulnerability factors, we note Weichselgartner (2001, p.89) who indicated that the concept of vulnerability includes pre-existing characteristics of the community, such as preparedness and prevention, and the post-disaster response and recoverability. Again, we will give our empirical answer to the question: what are the proper indices for preparedness, prevention, and response?
Epidemiological Model and Data
We employ the following epidemiological model:
ln([Edamage.sub.it]) = [[alpha].sub.0] + [[alpha].sub.1] ln([Dead.sub.it]) + [[alpha].sub.2] ln([Hdamage.sub.it])+ [[alpha].sub.3]ln([HQ.sub.it]) + [[beta].sub.1] ln(Under [15.sub.it]) + [[beta].sub.2] ln([Stocks.sub.it])+ [[beta].sub.3] ln([Recovery.sub.it-1]) + [[beta].sub.4]1n([Recovery.sub.it-2]) + [[beta].sub.5] ln([Reeovery.sub.it-3]) + [[beta].sub.6] ln([Recovery.sub.it-4]) + [[beta].sub.7] ln([Recovery.sub.it-5]) + [u.sub.i] + [[theta].sub.t] + [[epsilon].sub.it],
where In stands for the natural logarithm, [theta], is a time fixed effect for year t (t=19952007), [u.sub.i] is a fixed effect for prefecture i (1=1,2,...,47) and [[epsilon].sub.it] the error tem. We will apply this model to the fixed effect estimation by panel data consisting of 47 prefectures and 13 years.
All variables are scaled to the size of prefectures as is summarized in Table 2. Edamage, for example, is the direct economic damage caused by natural disasters divided by the gross regional product of the prefecture involved. As noted above, the FDMA numbers for economic damage do not include damages done to the private sector. We, therefore, augment the FDMA figures by adding the economic damage of buildings caused by natural disasters taken from the Building Destruction Statistics compiled by the Ministry of Land, Infrastructure, Transport and Tourism.
Table 2 Variables and statistics Factors Index Variables Description Dependent damage Edamagc direct damage/gross variable regional product Hazard magnitude of Dead number of dead or major disaster missing persons/1000 prefecture population magnitude of Hdamage number of houses minor disaster partially destroyed/total number of housing units in prefecture frequency of HQ number of times disasters disaster head quarters are set up/number of cities and townships Vulnerability preparedness Stocks sum of social and private stocks per capita prevention Under 15 number of people under 15 years of age/prefecture population response Recovery 5 year average of ratios of recovery expenditures to total public investments in prefecture
Another amendment we made to the FDMA data concerned the Niigata prefecture in 2004 and 2007. In these years, Niigata was hit by middle-sized earthquakes. However, we found that economic damages from the earthquakes are not reflected in the FDMA numbers. We added the publicly known damage numbers to the statistics.
Hazard and Vulnerability Variables
As for the data for hazard factor, we chose to use: 1) the number of dead or missing persons per prefectural populations (Dead); 2) the ratio of partially destroyed houses to the total number of housing units in prefecture (Hdamage); and 3) the number of times local emergency headquarters are set up divided by the number of cities and townships in the prefecture (HQ).
Clearly, the variable Dead represents the extent to which disasters are deadly. The Hdamage variable is to complement Dead in measuring the magnitude and frequency of less severe disasters. The ratios of buildings totally destroyed or half destroyed have a strong correlation with the Dead variable. This indicates that when a disaster kills people, the disaster also brings about building devastation. (6)
Each time a sizeable disaster hits a locality, its government is mandated to set up a local emergency headquarter. Hence, the number of such headquarters indicates how many significant disasters hit the prefecture in a year.
We also decided to use three variables as indices of vulnerability: I) the sum of social and private capital stocks per capita compiled by the Cabinet Office (Stocks), 2) the ratio of population 15 years old and under in the total local population (Under15), and 3) the ratio of post disaster recovery expenditures to the total public investments in the prefecture (Recovety).
The variable Stocks is used as a preparedness variable, since it is expected that the local community in which capital stocks are heavily accumulated is relatively well prepared for disaster, and inversely, perceived infrequency of disaster may be the reason for such accumulation.
The prevention factor is reflected in the variable Under15 which is an indication that the community is inhabited by younger families. It is known that communities with relatively young population are resilient against natural disasters. Researches point to several factors for this observation. It is easier for younger generations to take quick actions in the wake of disaster than older generations. Moreover, community events like religious festivals and school events of sports and culture promoted by younger families solidify local social capital, which contributes to sharing information, emergency supplies, and shelters in emergency.
The variable Recovery corresponds to response. Toya (2009) has shown that the post-disaster public investment made in the previous year contributes to reducing disaster damage in the current year. Thus, the volume of expenditures which local government invested in the post-disaster recovery program in the past is expected to reduce economic damage in the current year.
We must hasten to add, however, that we have tried other variables as candidates for the prevention variable. Data is available for gross prefectural product, population 65 year and over, savings per household, unemployment rate, the ratio of flood and forest control investments in the total local public expenditures, and population density. However, none of these variables showed statistical significance in our estimate.
We also rejected the ratio of houses constructed before 1981, when the building code was strengthened against disaster risks, because our estimate suggested the existence of of unit root problem. (7)
The result of our epidemiological estimation is summarized in Table 3. The three hazard variables, Dead, Hdamage, and HQ, are all statistically significant at the 1 % level. The result shows clearly that heavier economic damage is brought about when there are more human casualties, more house destructions, and more emergency headquarter setups. We may take it that these three variables are fair indices for the hazard factor.
Table 3 Fixed effect estimation Dependent variable: Edamage R-sq: within 0.495 prefecture 47 Between 0.234 year 13 Overall 0.002 sample: 611 Hazard Dead 0.107 *** Hdamage 0.170 *** HQ 0.126 *** Vulnerability Under 15 -9.731 *** Stocks -6.543 * Recovery L1 -0.164 * L2 0.081 L3 -0.183 L4 0.101 L5 -0.285 *** Year 1995 dropped 1996 -0.583 ** 1997 -0.430 1998 -0.140 1999 0.389 2000 -0.489 2001 0.890 2002 -0.789 2003 -0.824 2004 -0.424 2005 -1.426 2006 -1.004 2007 -1.081 Cons 0.341 Significance level: ***:1 %, **:5 %, * 10% cocf. Prob F-test 9.87 0.000 Hausman 245.5 0.000 BP-LM 83.5 0.000 Hcttcst 3387 0.000 Serial 0.014 0.905
Among the vulnerability variables, Under 15 shows the same significance level as hazard variables and its effect is negative on economic damage. Stocks is significant at the 10 % level, and it contributes to the mitigation of economic damage. The Recovery variables show an intriguing pattern of influence; recovery expenditures of 1 and 5 year vintage work negatively on economic damage.
Also, some tests results are shown in Table 3. Our F test, Hausman test, and Breush Pagan Lagrange multiplier (LM) test support the fixed effect model. Since heteroskedasticity is suspected by Wald test, this estimation is conducted by hetero-skedasticity consistent estimation by White. As for possible serial correlation, we performed the Lagrange multiplier (LM) test, which rejected the existence of First order Autoregressive Process, AR(1), serial correlation.
Our results support the basic hypothesis that economic damage caused by natural disaster can be explained by the magnitude and frequency of hazard occurrences and social vulnerability factors. We found that the proportion of the population under 15 years, the existence of per capita public and private capital stocks, and the level of past post-disaster recovery expenditures all contribute positively to the reduction of vulnerability.
In particular, the elasticity of damage reduction with respect to the young generation ratio and per capita private and public capital stocks are noteworthy: a 1 % increase in the young generation ratio will lead to a reduction of damage per gross regional product by 9.7 %, and a 1 % increase in the per capita stocks reduces the damage by 6.5 %.
It is commonly believed that a reinforcement of social infrastructure, an installation of a stricter building code, and a stronger system of command and control in an emergency are effective measures for damage reduction from the engineering perspective. However, our results show that attracting a young population, promoting economic growth, and accumulating private, as well as public capital stocks, should also be a crucial strategy for disaster mitigation. Post-disaster recovery efforts in previous years work also to reduce economic damage in the current year.
A Damage Estimation for the Tohoku Earthquake
Choice of Variables
Finally, we would like to present our estimation of economic damage caused by the earthquake and tsunami which hit the Tohoku region on March 11, 2011. We notice first that the model in the previous section is not convenient for a forecasting purpose, since the result is too sensitive to a small perturbation in the estimated coefficients of log variables. (8)
Thus, we tested a model without log transformation. We found that some of the variables whose statistical significance was verified in the log linear model lost their significance. HQ and Stocks and Recovery and year dummies turned out to be not statistically significant at the 10 % confidence level. We are thus led to a model consisting only of hazard variables and fixed effect of prefectures. This model is useful in on-going emergency situations, because it only requires the basic and minimum amount of information about variables whose values are publicized on a daily basis in many cases.
The result is shown in Table 4. In order to adjust for the effect of prefecture dummies, we did an Least Square Dummy Variable (LSDV) estimate. In Table 4, only the coefficients of dummies for the affected prefectures are shown.
Table 4 LSDV estimation Dependent variable: Edarnage LSDV estimation R-sq 0.7977 prefecture 47 year 13 sample: 611 Hazard Dead 298468.600 *** H damage 1360.474 ** Prefecture dummies Hokkaido 0.910 Aomori 2.853 ** Iwatc 5049 ** Miyagi (dropped) Yamagata 1.603 Fukushima 2.027 * ibaragi 0.010 Tochigi 1.703 Gunma 0.915 Chiba 0.009 Tokyo 0.103 cons Kanagawa -0.041 -0.126 significance level: ***:1%, **:5%, * 10%
Estimated Economic Damage
We then performed out-of-sample predictions. We applied the result above to the current Tohoku disaster. Data for the Dead variable was taken from the Police Agency's daily report as of September 9, 2011 and divided by the prefecture population in 2009. The Hdamage data was also taken from the Policy Agency's daily report and divided by the number of housing units in prefectures reported in the House and Land Survey 2008.
Substituting these values to the variables in the estimated equation in Table 4, we obtained the economic damage as a ratio to gross prefectural product. Finally, we multiplied the number by gross prefectural product 2007 to obtain our estimate of economic damage for each affected prefecture. Table 5 summarizes the fatalities and housing destructions for 12 prefectures and estimated total economic damage for the affected prefectures.
Table 5 Estimated damage for Tohoku Prefecture Dead or missing Buildings partially Estimated damage (persons) destroyed (buildings) (million yen) Hokkaido 1 7 1,134 Aomori 4 107 18,805 lwatc 6,348 6,974 7,409,008 Miyagi 11,640 149,856 14,482,443 Yamagata 2 0 2,328 Fukushima 1,844 134,905 4,311,176 Tokyo 7 257 13,016 lbaraki 25 154,839 2,094,745 Tochigi 4 62,803 939,688 Gunma 1 16,150 202,756 Chiba 22 30,099 326,387 Kanagawa 4 279 6,844 Total 19,902 556,276 29,808,330 As of September 6, 2011
Many organizations and research groups rushed to give their estimates of economic damage after 3/11. The Cabinet Office said their latest estimate came down to 16.9 trillion yen from 25 trillion yen, including damage on buildings, lifeline facilities, social infrastructure and agriculture, and forestry and fishery facilities.9 However, how they came up with this number is in the black box.
Inada et al. (2011, pp.1-9) estimated the direct damage done to housing units, social infrastructure, plant and equipment, automobiles, and ships and inventory at 17.8 trillion yen, and the indirect loss on Japan's GDP at 6 trillion yen. They applied the same damage rate to all kinds of stocks in the affected regions as a convention.
The Development Bank of Japan (2011) published yet another estimate. They estimated damage in the seaboard and inland areas separately and came up with the direct damage at 16 trillion yen for the four heaviest damaged prefectures combined. Hayashida et. al. (2011, pp.1-25) noted that almost all preceding estimates on direct damage fall in the range of 14 to 18 trillion yen and presented simulations as to the future course of regional economies.
However, when it comes to assessment of direct damage, all studies utilized one form of aggregation by parts or other. There has been no effort made so far to employ epidemiological models as we reported above. Furthermore, our estimate lies on the higher end of the scenarios these estimates draw. This result alone suggests the need for further studies that should lead to a widely accepted method of estimation.
After all, what is more pertinent is not who wins in the guessing game, but to learn what factors, natural and social, give rise to economic damage of the magnitude that comes to 6 % of GDP. A disaster-prone country must be equipped with as quick a method of disaster damage assessment as a preparedness measure for post-disaster reconstruction. We hope our research has made a modest contribution.
Acknowledgment This work was supported by 2011 Research Grant from Kampo Financial Group.
Cabinet Office (2011). Press release on damage estimate of the Great East Japan Earthquake (in Japanese), June 24. (http://www.bousai.go.jp/oshirase/h23/110624-1kisya.pdf).
Cavallo, E., Powell, A., & Becera, 0. (2010). Estimating the direct economic damage of the earthquake in Haiti. Economic Journal, 120(546), F298-F312.
Cochrane, H. (2004). Economic loss: myth and measurement. Disaster Prevention and Management, 1.3(4), 290-296.
Development Bank of Japan (2011). Assessment of economic damage on private and public stocks caused by the Tohoku Earthquake (in Japanese), News Release, April 28.
Hayashida, H., Hamagata, S., Nakano, K., Hitomi, K., & Hoshino, Y. (2011). Macroeconomic impacts of the Tohoku earthquake: an assessment by means of the CRIEPI macrocconometric model (in Japanese). Discussion Paper 11024, Socio-Economic Research Center, Central Research Institute of Electric Power Industry, 1-25.
Inada, Y., Irie, Y., Shima, A., &. Toizumi, T. (2011). The Tohoku earthquake's impacts on the Japanese macrocconomy (in Japanese), KISER Report No.6, April 12,1-9.
Toya, H. (2009). The mitigation effect on disaster damage by preventive policy measures: a panel analysis based on prefectural data (in Japanese). In Report on a system of anti-disaster policies with economics perspective, Cabinet Office, No.44, June, 67-89.
United Nations Development Programme (2004). Reducing disaster risk: a challenge for development. i-xii+ 146.
Weichselgartner, J. (2001). Disaster migigation: the concept of vulnerability revisited. Disaster Prevention and Management, 10(4), 85-94.
(1.) Prefectures are subnational jurisdictions governed by publicly elected governors, which are comparable to the states in the United States.
(2.) It docs not include the indirect damage of lost business and employment opportunities either, which are outside the scope of our investigation here.
(3.) The reason for their bias can be found in two laws; "Law of National Government Finance on the Disaster Recovery of Public Facilities" and "Disaster Recovery Expenditures on Agriculture, Forestry, and Fishery Facilities." In order to obtain national government funding, local governments, including prefectures, need to present damage estimates of the relevant facilities as soon as possible.
(4.) In case of the Kobe earthquake of 1995, the private sector damage came to 60% of the total damage.
(5.) This formulation or its variations are frequently used in the United Nations publications. See, for example, UNDP (2004, p.100).
(6.) The Fukushima nuclear power plant accidents pose a serious problem, for there is no reported casualty among the inhabitants and no physical destruction in the vicinity, yet the inconvenience and economic losses incurred by the dislocated people are enormous.
(7.) Houses that were built before 1981 were subject to a more lenient building code than afterwards.
(8.) Cavallo et al. (2010) also focused on the relationship between the human damage and the direct economic damage in disasters. They used a log linear model to predict the direct economic damage of the earthquake in Haiti. However, they came up with a wide range for prediction, which is, basically, an unavoidable consequence of log linear models.
(9.) Cabinet Office (2011).
An earlier version of the paper was presented at the 72nd International Atlantic Economic Conference in Washington D.C. The author wishes to thank the chair of the session and other participants for helpful discussions.
Int Adv Econ Res (2012) 18:417-427
Published online: 13 July 2012
[c] International Atlantic Economic Society 2012
M. Hayashi ([??])
Asia Pacific Institute of Research, Osaka, Japan
DOI 10.1007/s1 1294-012-9367-y
|Printer friendly Cite/link Email Feedback|
|Publication:||International Advances in Economic Research|
|Date:||Nov 1, 2012|
|Previous Article:||An investigation into the positive effect of an educated wife on her husband's earnings: the case of Japan in the period between 2000 and 2003.|
|Next Article:||A time-series investigation of the U.S. real health expenditure: evidence from nonlinear unit root tests.|