1. Introduction
Today, as well as historically, hurricanes have caused widespread destruction, and disruption to economic activity, and death.1 As a matter of fact, one can find references to their existence and the accompanying fear as far back as in the early civilization of the Mayans of Central America and the Tainos of the Caribbean. More explicit accounts of the actual extent of destruction exist since colonialization of the region. For example, after the great hurricane of 1831, the St. Vincent Gazette noted that on many plantation estates, all buildings were destroyed (see Smith 2012). Even today, when a hurricane strikes, the destruction and its aftermath can dominate much of the international media for several days. It is not surprising then that both natural and social scientists have long been fascinated with studying these storms and their impact.
Nevertheless, the study of hurricanes as scientific phenomena and as potentially severe disruptions to economies has only relatively recently taken quantitative routes. In this regard economists lag much behind meteorologists, who first started seriously studying a hurricane’s structure in the early part of the nineteenth century when, for example, in 1821 William Redfield was able to make conclusions in this regard by studying the damages caused by a storm in Connecticut (see Emanuel 2005). Further understanding of their workings then came with the increased development of the laws of physics in the 1930s and ultimately with the advent of reconnaissance of aircraft and radar in the 1940s and of satellites in the 1960s. In contrast, beyond descriptive statistics, the academic economic literature only addressed the issue of the quantitative economic impact of hurricanes in any more comprehensive manner with the commencement of the twenty-first century, and our understanding of the quantitative effects is still limited to the modern period—that is, no earlier than the 1950s, when statistical data on economic activity across countries over time are readily available.2,3
Arguably, however, a better understanding of the effects of hurricanes beyond modern times may fill an important gap in our knowledge of the role of these storms for a number of reasons. Most obviously, it will allow us insight into an aspect of history that for civilizations in the region, hurricanes constituted an important part of life. For example, Mulcahy (2006) argues that “hurricanes shaped the mental and physical world of colonists” (p. 3) in that they “routinely swept across the colonies, destroying fields and crops, leveling plantations, cities, and towns, disrupting shipping and trade and causing widespread deprivation and death among colonists and slaves” (p. 2) and thus “raised questions about how permanent settlements would develop” (p. 4). In this regard, it is noteworthy that the dates of storms were listed on almanacs alongside public holidays and the birthdays of monarchs, and voyages into the region were generally scheduled outside the hurricane season. Moreover, these storms did not just have large impacts on the region but, because their economies served primarily to export, they also affected the royal empires through loss of income and driving up prices. Additionally, damages caused by the storms increased the demand for imports into the colonies, such as food, building supplies, and furniture, from the colonial powers.
More generally, recently among economic academic circles, a school of thought has emerged that has argued that historic events are important determinants of economic development today. For instance, Acemoglu et al. (2001) provide evidence that the disease environment across former colonies determined whether colonial power invested in institutions to protect property rights and thus played an important role in the investment in good institutions and economic development since their independence.4 Relatedly, Johnson (2011) argues that some of the signal events of the Age of Revolution in the Caribbean were a consequence of natural disasters, such as hurricanes. Finally, as we will show, combining historical economic and meteorological data together in order to quantify the economic impact of hurricanes over time may allow one to indirectly assess the quality of the latter in terms of capturing the extent of hurricane activity in the past.
In this paper we provide what we believe to be the first statistical study of the economic impact of hurricanes prior to the mid-twentieth century. More specifically, we estimate the quantitative impact of hurricane strikes on the colonies/countries of the Caribbean during the 1700–1960 period. To this end we gather data on both historical exports of the sugar colonies and hurricane activity from a variety of sources that allow us to determine to what extent hurricane strikes may have affected sugar production in the Caribbean. Our choice of focus on sugar, apart from data availability issues, is that it constituted the Caribbean’s main generator of income during the colonial period after the “sugar revolution” (Pons 2007). Furthermore, some authors contend that Caribbean sugar production yielded an exceptional “economic surplus” that contributed to the Industrial Revolution and growth in the metropolitan economies in colonial Europe (Sheridan 1961; Williams 1944).
The remainder of the paper is organized as follows. Section 2 outlines the data sources used, provides descriptive statistics, and describes our statistical methodology. Section 3 presents the results of our analysis. Finally, section 4 concludes.
2. Data and methods
a. Hurricane data
Tropical cyclone events affecting Caribbean colonies/countries for the period 1700–1850 are identified using information from Chenoweth (2006). More specifically, Chenoweth (2006) meticulously went through Poey’s (1855) well-known and widely used list of known historical tropical cyclones in the region, verifying and validating these with other sources, to then construct an updated and corrected chronology of 383 unique tropical storms striking the Caribbean over the period 1700 until 1855.5 Importantly for the purposes here, he identifies for each storm the localities of colonies/countries affected and estimates according to their description whether the cyclone was likely to have been of tropical storm or hurricane strength. We thus use this list to identify the number of tropical storms and their strength affecting colonies/countries in each year for the period 1700–1850. One may note that the quality of this list as well as all accepted tropical storms in the Chenoweth data, in terms of providing an accurate picture of all potentially damaging cyclones affecting localities in the Caribbean region, will ultimately depend on the quality and availability of the primary sources on which they are based.
For the period 1851–1960, our source of tropical cyclone events is the National Hurricane Center’s (NHC) “best track” dataset (HURDAT), which provides 6-hourly reports of storm positions and maximum wind speeds of all known tropical cyclones in the North Atlantic basin.6 Ideally, we would like to use these data to identify all potentially damaging storms that may have affected localities in the Caribbean region, in a similar manner to the historical data described above. More precisely, for each storm we would like to know if it approached or made landfall in the colonies/countries in the region with sufficient strength to potentially cause damage. To do so we first identified all populated areas in the region from the 1950 Latin American and the Caribbean Population Database, which provides population estimates at the 1-km gridcell level. One should note that implicitly our assumption is that these cells may have also been populated or used by local residents in the years prior to 1950. We then calculated the wind speed experienced by each populated 1-km grid cell for each 6-hourly position of each tropical cyclone within 500 km in a similar manner as Strobl (2012), that is, employing the wind field model proposed by Boose et al. (2004).7 This provided us with a maximum wind speed for each grid cell for each tropical storm in the HURDAT dataset during the period 1851–1960. If any cell within a colony/country experienced wind speeds in excess of 119 km h−1 (63 km h−1) for any tropical cyclone, we then identified the relevant cyclone as a potentially damaging hurricane (tropical storm). We provide the number of events for each data source for each economic entity in Table 1.
b. Sugar data
The sugar export data by country/colony are taken from a number of sources. For data prior to 1900 the main source is Deerr (1950), who provides annual quantities of sugar exported for most English, French, Spanish, and Danish colonies, although there are some missing values for some years. For the period 1850–1900, we supplemented a few of the missing observations of the English colonies with data taken from various issues of statistical tables relating to colonial and other possessions of the United Kingdom, whereas for the period from 1831 to 1850, Martin (1843) served to complete some missing values. Data for the period 1900–60 were taken from Bulmer-Thomas (2012). Since there are some sugar-producing countries for which we do not have data prior to 1900, we reduce our sample to those represented in both periods. Moreover, for some of the English colonies, data were reported after a certain year jointly, and we thus combine their exports in the years prior as well.8 Overall, this leaves us with a sample of 21 colonies/countries for which we have potentially data from 1700 to 1960, but for which in reality there are some missing values for some years. All data refer to exports within the given year. A complete list of countries with the earliest year reported, the last year reported, and the number of years with nonmissing/nonzero values is given in Table 2.
Sugar data by country/colony, earliest year reported, last year reported, and the total number of observations for 1700–1850 and 1851–1960.
One may want to note that here we are using sugar export data given the lack of information on actual sugar production over enough countries and years over our sample period to conduct a reliable statistical analysis. In this regard, it is important to point out that the measured effect of hurricanes on exports will not only include the direct impact on reduced production but also the indirect negative impact due to the destruction of transport infrastructure and hence a lower ability to export.
c. Methodology
Another concern, particularly with agricultural data, is that of serial correlation. As a matter of fact, a serial correlation test allowing for fixed effects could decisively reject the null hypothesis of no first-order serial correlation for both the 1700–1850 and 1851–1900 periods11, and we thus implemented a fixed effects model with first-order autoregressive [AR(1)] disturbances to estimate (1).
3. Results
We report the estimated coefficient on H under the fixed effects AR(1) model for the 1700–1850 sample in the first column of Table 3. As can be seen, the null hypothesis β = 0 cannot be rejected under standard significance levels, and hence it indicates that hurricanes do not affect sugar exports in the year that they strike. We next experimented with including further lags of H in columns 2 and 3. Accordingly, hurricane strikes have a 1-yr lagged significant negative effect; however, no further impact is discernible thereafter. In other words, our results suggest that any destructive impact from a hurricane can only be identified in export data the year after the documented strike. This may not be surprising, given that sugar is usually harvested in the Caribbean during January–May, so that the damage due to hurricanes may only appear via a reduction in sugar exports in the following year. One may want to note that this implies that any indirect effect through damage to transport infrastructure, which one would expect to see more strongly in the actual year of the strike, is not dominant in the data.
Regression results for 1700–1850. Standard errors in parentheses. Symbols * and ** identify 5% and 1% significance levels, respectively. In the first column, H: number of hurricanes/tropical storms, H[acc]: hurricanes accepted in Chenoweth (2006), TS[acc]: tropical cyclones accepted in Chenoweth (2006), and H(all): all cyclones in Chenoweth (2006); H(TS) are the tropical storms in the accepted list of Chenoweth (2006).
While for these first estimations we only used the count of hurricanes, Chenoweth’s (2006) verified list also contains a large number of cyclones that were likely to have been tropical storms. In this regard, one might expect tropical storms to have a smaller—if any—impact on sugar exports compared to hurricane strikes. To verify this we first included the total number of tropical cyclones by country and year from Chenoweth’s (2006) list, including their lagged up to t − 2 values in the fourth column of Table 3. As can be seen, this does not change our earlier estimates qualitatively; however, it does produce a somewhat lower coefficient on the Ht−1 variable, suggesting that tropical storms may indeed have a lower negative impact on sugar exports. To further investigate this, we then separately included the total number of hurricanes as well as the number of tropical storms that affected a locality each year in our specification, as shown in the fifth column. The estimates here indicate that unlike hurricanes, tropical storms have no discernible effect on sugar exports in the Caribbean.
In his verified list of relevant events, Chenoweth (2006) rejects a total number of 149 tropical cyclones listed in other sources. If these are truly errors, then one might expect their inclusion as tropical cyclones in H to increase the measurement error of this proxy and thus may result in further attenuation bias, potentially biasing the estimated effect toward zero. To verify this we included in the sixth column of Table 3 all tropical cyclones—that is, those accepted as well as those rejected by Chenoweth (2006)—in H.12 As can be seen, the use of all tropical cyclones as a measure does indeed produce a coefficient on H that is statistically not significantly different from zero. This likely unreliability of rejected tropical cyclones is further supported, at least in terms of their measured impact on sugar exports, by the estimated coefficients in column 7, where we separately included all accepted tropical cyclones by Chenoweth (2006) as well as the ones that he rejected. As can be seen, only the lagged t − 1 measure of accepted cyclones is significant.
Since the results thus far indicated that only hurricanes, as identified in Chenoweth’s (2006) list, had an effect; moreover, in the year after their strike, we finally reran (1), including this variable as depicted in the last column of Table 3. This produced a significant negative coefficient of 1.626, and we can use this to provide some quantitative feel for the effect of striking hurricanes on sugar exports in the Caribbean. More specifically, it implies that a hurricane strike decreased exports of sugar in the region by 1.626 million pounds in absolute terms. We can take this figure to derive the economic significance of these events for the colonies/countries affected. More specifically, we take sugar and total export figures as well as sugar prices for the English colonies in 1850 and calculate the percentage by which the value of sugar and total exports was reduced by a hurricane strike.13 As the figures in Table 4 show, the impact was substantial for most colonies. On average sugar exports would have fallen by 68.1% across the English Caribbean, with sugar production having been completely desiccated for many of the smaller sugar producers, such as the Bahamas, Bermuda, Montserrat, St. Vincent, and the Virgin Islands, and ranging widely for the other islands. If one considers the impact on the overall wealth of these small economies as proxied by their total exports, then again the experience would have ranged widely, from the 1.2% reduction in Jamaica to the 76.6% reduction in the value of exports in St. Vincent. Overall, the reduction in exports would have been a little over one-quarter of its value. One should, however, note that these rough calculations simply capture the quantity effect and abstract from any price effects of hurricanes by assuming that prices remain constant, since our data do not allow us to directly measure the latter.
Estimated economic effect of hurricane strikes using the estimated coefficient. Upper-threshold effect is 100% (i.e., no sugar exported).
We next proceeded to estimate Eq. (1) for the period 1851–1960 with our constructed proxy of hurricane events, as in section 2. We first start with including up to two lags of our hurricane incident index. As can be seen from the first column from Table 5, unlike the data prior to 1851, we now find that both the t − 1 and the t − 2 lags are statistically significant. We thus included a further lag, shown in the third column, but this indicated no significant impact from a hurricane strike beyond 2 yr. One may want to note that overall, the estimated coefficients suggest a slightly larger effect in t − 2 than in t − 1, although a test of equality of the coefficients cannot be rejected. In considering these results relative to what we found for the earlier period, apart from the more long-term effect, the most striking difference is that the coefficients are nearly 10 times larger than that for the latter period. The most plausible explanation for this is the attenuation bias produced by measurement error.14 More precisely, given that the proxies of hurricane events come from two different sources and two different methods of construction, they are likely to have very different levels of measurement error in terms of serving as a proxy for potentially destructive hurricane events. In particular, one might suspect that the earlier data might be missing relatively more potentially damaging tropical cyclone occurrences or misclassifying some events in terms of localities affected and likely wind speed, given that they are based on a much lower number and more heterogeneous set of documents [see Chenoweth (2006) for a more complete description of documents]. Even if the earlier data constitute an exhaustive list of hurricane events, one could argue that using actual wind speed calculated from a wind field model in economically potentially important areas, as one is able to do with the more recent tropical cyclone data, is likely to provide a more accurate proxy of the potential destruction of these events. Either way, as is well known, large amounts of measurement error in an explanatory variable will lead to biasing its estimated coefficient toward zero (Hyslop and Imbens 2001).
Regression results for 1851–1960. Standard errors in parentheses. Symbols * and ** identify 1% and 5% significance levels, respectively. In the first column, H: number of hurricane/tropical storms and H(W): average wind speed of hurricanes. Fourth column uses hurricane incidence dummies for hurricanes with a minimum strength of 178 km h−1. Fifth column contains results for population-weighted wind speed hurricane proxies.
One should note that we, in order to be congruent with the analysis of the earlier period, simply tried to identify potentially significant hurricane events from the HURDAT best-track data. However, these data and the calculations underlying their identification also allow us to distinguish the strikes in terms of the actual wind speed experienced and the total area affected within each colony/country. To investigate whether ignoring such heterogeneity may affect the estimated impact on sugar exports, we thus calculated the mean wind experienced across all cells within a colony/country in any year and included its current value and up to three lags in our specification in (1), the results of which are depicted in the third column of Table 5. Accordingly, qualitatively the results of using wind speed are similar to those using the hurricane event counts, although here the t − 1 coefficient appears larger than that at t − 2; however, again, a statistical test indicates that there is statistically no difference between the two.
We also redefined the hurricane incidence numbers but are now only considering hurricane wind speeds over 178 km h−1, which correspond to a category 3 or higher on the Saffir–Simpson scale, and are believed to be particularly damaging. The results of using the t − 1 and t − 2 lagged incidence numbers for these more destructive hurricanes are shown in the fourth column of Table 5 and indicate, as would be expected, that the impact on sugar export losses is a good bit larger than for all levels of hurricanes.
One problem with our analysis, for both the older and modern periods, is that we do not control for vulnerability within an island and how this might affect the extent of the negative impact of hurricanes through exposure. For instance, Boruff and Cutter (2007) highlight that only small parts of the population of Barbados and St. Vincent live in the extremely hazardous parts of the islands. Since vulnerability is likely to have changed over time in order to account of this, we would ideally like to have annual information on the spatial distribution of sugar plantations within countries and how these spatial areas were affected by the striking hurricanes. This is clearly not possible with the hurricane incidence data for the older period. Even for the more modern period, when the hurricane track data allow us to calculate wind speeds experienced for spatial areas within economies, we do not have time-varying information on the spatial economic activity related sugar within countries. As a fairly rough check, however, neglecting this might affect our results; we used the 1950 Latin American and the Caribbean Population database, which provides population estimates for 1950 of 1-km grid cells within the region, and we assume that the distribution of the population was the same as far back as 1850. These population distributions are then used to calculate a population-weighted estimate of wind speed experienced, the results of which are shown in the fifth column of Table 5. Comparing the coefficient on the population-weighted wind speed to the unweighted one in column 1 finds a substantially larger 1-yr lagged effect for the former, while there is now no 2-yr lagged impact. Although assuming the population distribution in the mid-nineteenth century was the same as the mid-twentieth century is unlikely to be very realistic and thus one may not necessarily want to read too much into these results, it does at the very least suggests that one would ideally like to control for time-varying vulnerability if the data allowed one to do so.15
We can now proceed to assessing the quantitative impact of our estimates for this latter period and to reestimating the incidence- and wind-speed-based proxies only for those lags that were significant in the sixth and seventh columns of Table 5, respectively. One should note that taken at face value, our coefficients on the events proxy suggests that a hurricane strike will cause a reduction in sugar exports in a colony/country in the region of 33.1 million pounds over the 2 yr subsequent to its occurrence. If we consider the actual average wind speed experienced, then the average hurricane strike (i.e., the average nonzero value) suggests a 25.9 fall in millions of pounds of sugar exported, while using its largest observed value of our proxy—which corresponded to the destruction in Puerto Rico due to Hurricane San Ciriaco in 1899, believed to be the longest-lived and eleventh deadliest hurricane since the start of the HURDAT data—suggests a loss of 113.6 million pounds of sugar exports. Actual accounts from the time confirm that losses from this storm were indeed large. For example, Schwartz (1992) notes that estimates of damages amounted to roughly $20 million (U.S. dollars).
We can again employ our estimates to assess the economic impact on local economies in 1850, as shown in Table 4. Using our incidence estimates, the implied impact on sugar production by a hurricane strike would have, ceteris paribus, been devastating, wiping out all sugar exported over 2 yr, except for Jamaica and Trinidad and Tobago, with an average loss of 90.6%. In terms of overall loss of wealth, as measured by total exports, this would have resulted in an average fall of nearly half. Again, however, the experience differs widely, with some countries, such as Antigua, Barbados, Dominica, St. Kitts and Nevis, St. Lucia, and St. Vincent, experiencing a loss of around three-quarters, while other countries, such Jamaica, for whom sugar was a relatively small part of their exports, having losses below 10%.
Finally, we graph the total losses by year due to hurricane incidence across the Caribbean as suggested by our modern-period estimates in Fig. 1. Accordingly, there were a number of loss-heavy years prior to the nineteenth century. For example, in 1766 there was a total of four destructive hurricanes, where one of these hurricanes affected 17 countries, killing between 600 and 1600 people in Martinique alone (see Rappaport and Fernández-Partagás 1995). While the years from the late eighteenth century to the mid-nineteenth century were relatively calm, the number of losses increased substantially since then, particularly around the turn of the twentieth century.
This figure depicts the estimated sugar export losses due to hurricanes.
Citation: Weather, Climate, and Society 5, 1; 10.1175/WCAS-D-12-00029.1
4. Conclusions
Hurricanes are important natural and largely unpredictable phenomena that have shaped the economic history of the Caribbean. This study provides quantitative evidence of their economic impact by combining annual hurricane events constructed using historical accounts of hurricane strikes and hurricane tracks with historical sugar export data in a regression framework over the 1700–1960 period. Overall, our analysis suggests that hurricanes have had large negative economic consequences for Caribbean sugar colonies. Differences across time in the estimated effect are likely due to greater measurement error in the earlier data. More generally, our results confirm historical documents that argued that hurricanes had a major impact on the sugar industry in the Caribbean.
REFERENCES
Acemoglu, D., Johnson S. , and Robinson J. A. , 2001: The colonial origins of comparative development: An empirical investigation. Amer. Econ. Rev., 91, 1369–1401.
Belasen, A. R., and Polachek S. W. , 2009: How hurricanes affect employment and wages in local labor markets. Amer. Econ. Rev., 98, 49–53.
Boose, E. R., Serrano M. I. , and Foster D. R. , 2004: Landscape and regional impacts of hurricanes in Puerto Rico. Ecol. Monogr., 74, 335–352.
Boruff, B. J., and Cutter S. L. , 2007: Environmental vulnerability of Caribbean Island nations. Geogr. Rev., 97, 24–45.
Bulmer-Thomas, V., 2012: The Economic History of the Caribbean since the Napoleonic Wars. Cambridge University Press, 720 pp.
Chenoweth, M., 2006: A reassessment of historical Atlantic basin tropical cyclone activity, 1700-1855. Climatic Change, 76, 169–240.
Deerr, N., 1950: The History of Sugar. Vol. 1. Chapman and Hall, 636 pp.
Emanuel, K., 2005: Increasing destructiveness of tropical cyclones over the past 30 years. Nature, 436, 686–688.
Hallegatte, S., and Przyluski V. , 2010: The economics of natural disasters. CESifo Forum, 11 (2), 14–24.
Hsiang, S. M., 2010: Temperatures and cyclones strongly associated with economic production in the Caribbean and Central America. Proc. Natl. Acad. Sci. USA, 107, 15 367–15 372.
Hyslop, D., and Imbens G. , 2001: Bias from classical and other forms of measurement error. J. Bus. Econ. Stat., 19, 475–481.
Johnson, S., 2011: Climate and Catastrophe in Cuba and the Atlantic World in the Age of Revolution. University of North Carolina Press, 306 pp.
Kahn, M. E., 2005: The death toll from natural disasters: The role of income, geography, and institutions. Rev. Econ. Stat., 87, 271–284.
Martin, M., 1843: The British Colonies: Their History, Extent, Condition, and Resources. London Printing and Publishing Company, 886 pp.
Mulcahy, M., 2006: Hurricanes and Society in the British Greater Caribbean, 1624–1783. Johns Hopkins University Press, 257 pp.
Noy, I., 2009: The macroeconomic consequences of disasters. J. Dev. Econ., 88, 221–231.
Nunn, N., 2009: The importance of history for economic development. Annu. Rev. Econ., 1, 65–92.
Pielke, R., Gratz J. , Landsea C. , and Collings D. , 2008: Normalized hurricane damages in the United States: 1900-2005. Nat. Hazards Rev., 29, 29–42.
Poey, A., 1855: A chronological table, comprising 400 cyclonic hurricanes which have occurred in the West Indies in the North Atlantic within 362 years, from 1493 to 1855. J. Roy. Geogr. Soc. London, 25, 291–328.
Pons, P. M., 2007: History of the Caribbean: Plantations, Trade and War in the Atlantic World. Markus Wiener Publishers, 402 pp.
Rappaport, E., and Fernández-Partagás J. , 1995: The deadliest Atlantic tropical cyclones, 1492–1994. NOAA Tech. Memo. NWS NHC-47, 41 pp.
Schwartz, S. B., 1992: The hurricane of San Ciriaco: Disaster, politics, and society in Puerto Rico, 1899–1901. Hisp. Amer. Hist. Rev., 72, 303–334.
Sheridan, R., 1961: The West Indian sugar crisis and British slave emancipation, 1830–1833. J. Econ. Hist., 21, 539–551.
Smith, S. D., 2012: Storm hazard and slavery: The impact of the 1831 Great Caribbean Hurricane on St Vincent. Environ. Hist., 18, 97–123.
Strobl, E., 2011: The economic growth impact of hurricanes: Evidence from U.S. coastal counties. Rev. Econ. Stat., 93, 575–589.
Strobl, E., 2012: The macroeconomic impact of natural disasters in developing countries: Evidence from hurricane strikes in the Central American and Caribbean region. J. Dev. Econ., 97, 130–141.
Toya, H., and Skidmore M. , 2007: Economic development and the impacts of natural disasters. Econ. Lett., 94, 20–25.
Williams, E. E., 1944: Capitalism and Slavery. University of North Carolina Press, 285 pp.
Wooldridge, J. M., 2002: Econometric Analysis of Cross Section and Panel Data. The MIT Press, 752 pp.
See Emanuel (2005) for an extensive history of hurricanes.
Much of the literature focused on the economic impact of natural disasters in general; see, for instance, Kahn (2005), Toya and Skidmore (2007), and Noy (2009) to name just a few. There are far fewer studies exclusively focused on hurricanes, such as Belasen and Polachek (2009), Hsiang (2010), Strobl (2011), and Strobl (2012).
Pielke et al. (2008) calculate the total direct losses due to hurricanes for the United States for the period 1900–2005. Direct losses arguably may only constitute a small proportion of potential effects in comparison to indirect losses and long-term effects; see Hallegatte and Przyluski (2010). Moreover, they do not investigate the total net economic impact of these losses.
See Nunn (2009) for a comprehensive review of the literature relating historic events to current economic well-being.
Poey’s (1855) constructed his original list by referring to a number of earlier lists as well as using other primary historical sources.
The original list of storms in HURDAT originally started in 1886. This was then extended in 2000 back to 1851 (see http://www.aoml.noaa.gov/hrd/hurdat/Documentation.html#Partagas). One may want to note that prior to 1944, most storm tracks were constructed from information gathered from land stations and ships at sea.
This is a wind field model based on the well-known Holland model and takes, among other things, the position of any point relative to the storm, the maximum wind speed, the traveling speed and direction, and whether the storm has made landfall into account.
These are St. Kitts and Nevis, and Trinidad and Tobago.
1000 lbs is equal to 0.5 tons.
Alternatively, we could have included a set of country dummy variables, which is equivalent to a fixed effects estimator but would lower the degrees of freedom. We also experimented with running a standard regression including time-invariant country characteristics, such as size, average elevation, agroecological zone, and type of sugar colony, and the results were similar to our fixed effects specification.
The test statistics were 23.66 and 9.88, respectively; see Wooldridge (2002) for details about the test employed.
One should note that for the rejected cyclones, no information is provided on whether they are likely to have been tropical storms or hurricanes.
Total sugar and export values were only available to us for the English Caribbean and come from the Statistical Tables Relating to Colonial and Other Possessions of the United Kingdom. For colonies for which data for 1850 were not available, we took the figures from the most recently available year thereafter.
We also experimented with estimating our specification with the more recent data for subperiods, but this always showed a similar—about tenfold larger—effect than what we find for the earlier period.
In addition to the robustness checks reported here, we also experimented with allowing for a dynamic specification by including t − 2 sugar exports; however, this only marginally changed the estimates. Furthermore, rather than using a fixed effects estimator, we instead included variables indicating country size, elevation, agoecological zone classification, and sugar-colony-phase dummies; however, again this only marginally changed results. Finally, for the limited sample for which data were available, we investigated whether hurricane incidence affected rum exports, as suggested by Mulcahy (2006). However, surprisingly we found no significant impact on rum exports, although this may have been due to our rather limited sample.
