The Global Costs of an Influenza Pandemic By Warwick J. McKibbin and Alexandra A. Sidorenko

The deaths of 187 people in 12 countries through May 2007 from Type A (H5N1) avian infl uenza has raised fears that bird fl u could trigger the next infl uenza pandemic – perhaps even Tone on the scale of the Spanish fl u that swept the world in 1918. While we are in some ways better prepared to deal with a pub- lic-health emergency on this scale than we were a century ago, the integrated nature of the global economy – the dependence on global markets for goods, services and capital – may leave us more vulnerable to a pandemic’s accompanying economic shocks. We can estimate the economic consequences of pan- demics, based on computer simulations incorporating what we know about infl uenza transmission and the likely response by governments, as well as by markets. left: iain masterton/alamy; right: john stanmeyer/vii right: iain masterton/alamy; left:

18 The Milken Institute Review tk

Third Quarter 2007 19 influenza pandemic an insightful account of the events of 1918. some grim facts The natural reservoirs of infl uenza A are Infl uenza pandemics have been recorded for humans and waterfowl. Pigs can be infected more than three centuries, a period in which by both human and avian strains, providing there have been 10 probable and three possi- a mixing bowl for virus “reassortment” – cre- ble outbreaks. The fl u is not the deadliest of ating a strain that can be passed easily from all infectious diseases – smallpox, various human to human. The high density of the hemorrhagic fevers and rabies kill a much human population and its proximity to pigs higher proportion of those who contract and waterfowl make Asia a likely source of a them. But it is spread more easily, and hence future pandemic strain. has the capacity to infect a larger proportion But efforts to predict fl u pandemics have of the population. In the parlance of epidemi- not succeeded. The best one can do is look to ologists, its “attack rate” is high. the record, which suggests an interval of 10 to

The emergence of a flu pandemic is virtually as unpredictable as an earthquake. Thus, the most we can do is to be alert and be prepared.

Infl uenza Type A, the category to which 40 years between major outbreaks. Another H5N1 belongs, is both the most common and pandemic, it seems, is overdue. the least stable of the known human infl uen- If a pandemic strain does emerge, its za viruses. Its evolutionary skills include an spread could be unprecedentedly rapid, ability to change rapidly and to use animals as thanks to mass travel by airplane. Border- reservoirs in which to exchange genes. Three control measures might postpone the arrival known infl uenza pandemics in the 20th cen- of infection, but don’t offer much hope of tury – the Spanish infl uenza of 1918-19, the stopping its transmission. An effective vac- Asian infl uenza of 1957-58 and the Hong cine would not be available until months after Kong infl uenza of 1968-69 – were caused by the onset of the pandemic, since a vaccine infl uenza A viruses. The gravity of clinical accurately targeted for Type A virus can’t be symptoms of the Spanish fl u, which dispro- prepared ahead of time. portionally killed young adults, made it the Moreover, fl u-vaccine manufacturing ca- “deadliest plague in history,” according to pacity is concentrated in a handful of devel- John Barry, the author of The Great Infl uenza, oped countries, and it would be diffi cult to gear up to the scale required to check a pan- demic with billions of potential victims. Cur- WARWICK MCKIBBIN is an economist who heads the rent vaccine technology crucially depends on Australian National University’s Center for Applied Macro- the availability of fertilized chicken eggs, economic Analysis and is a senior fellow at The Lowy Insti- tute for International Policy in Sydney and the Brookings whose supply may be limited during the pan- Institution in Washington. ALEXANDRA SIDORENKO is demic. And if the pandemic is caused by a a health economist at the Australian National University’s National Center for Epidemiology and Population Health highly pathogenic virus, like H5N1, the tech- and the Australian Center for Economic Research on Health. nology will be diffi cult to manage safely.

20 The Milken Institute Review Epidemiologists always remind us to ex- used by courts to determine damages, goes pect the unexpected with infl uenza. The emer- like this: using estimates of mortality and gence of a fl u pandemic is virtually as unpre- morbidity, one calculates the loss of future in- dictable as an earthquake. Thus, the most we come during a working lifetime. Then one can do is to be alert and be prepared. adds in the loss of time and income by spous- Since 1970, many countries have invested es or relatives who must care for them, along in the development of national-preparedness with the cost of medical care and supporting plans, along with a global surveillance system services.

that includes a network of laboratories capa- This approach has always been limited in ble of identifying the virus. Challenges sur- the sense that it doesn’t include the intangible rounding pandemic-preparedness planning value people place on their own lives. But it is include increasing the emergency capacity of becoming increasingly evident that conven- critical-care facilities, building antiviral drug tional analysis is inappropriate even on its stockpiles and, of course, improving the abil- own terms for infectious diseases that have ity to produce massive quantities of vaccine large numbers of victims and that lack a vac- very quickly. All these measures are costly and cine. Examples include the HIV/AIDS, Severe need to be justifi ed on the grounds of the ex- Acute Respiratory Syndrome (SARS) and in- pected benefi ts. Which is where economics fl uenza pandemics. The impact of these three fi ts in. diseases provide insight into the wider eco- nomic consequences of their spread. how much does a disease cost? HIV infection mostly affects the young Conventional analysis of the tangible eco- and has a long-term demographic effect

chine nouvelle/sipa nomic costs of death and disability, the sort through decreased fertility and healthy-life

Third Quarter 2007 21 PREDICTED DEATHS IN EACH REGION (THOUSANDS)

MILD MODERATE SEVERE ULTRA NUMBER %POPULATION NUMBER %POPULATION NUMBER %POPULATION NUMBER %POPULATION US 20.2 0.007% 201.9 0.07% 1,009.3 0.35% 2018.6 0.70% Japan 21.5 0.017 214.6 0.17 1,073.1 0.84 2146.2 1.68 UK 7.6 0.013 76.0 0.13 380.0 0.64 759.9 1.28 Europe 56.5 0.010 565.5 0.10 2,827.4 0.50 5654.9 1.00 Canada 3.1 0.010 30.9 0.10 154.5 0.49 309.1 0.99 Australia 2.1 0.011 21.4 0.11 107.1 0.54 214.2 1.09 New Zealand 0.5 0.013 5.2 0.13 25.8 0.65 51.5 1.31 Indonesia 114.3 0.054 1,142.5 0.54 5,712.6 2.70 11,425.1 5.39 Malaysia 10.9 0.045 108.9 0.45 544.5 2.24 1,089.1 4.48 Philippines 41.5 0.052 415.5 0.52 2,077.5 2.60 4,155.0 5.20 Singapore 1.4 0.035 14.4 0.35 72.0 1.73 144.1 3.46 Thailand 16.2 0.026 162.1 0.26 810.3 1.32 1,620.5 2.63 China 284.9 0.022 2,848.6 0.22 14,242.8 1.11 28,485.6 2.22 India 242.4 0.023 2,423.6 0.23 12,118.1 1.16 24,236.1 2.31 Taiwan 5.6 0.025 55.9 0.25 279.4 1.24 558.8 2.48 Korea 11.8 0.025 117.5 0.25 587.6 1.23 1,175.2 2.47 Hong Kong 1.6 0.024 16.4 0.24 82.0 1.21 163.9 2.42 LDCs 330.9 0.022 3,308.6 0.22 16,543.1 1.08 33,086.2 2.15 EEFSU 67.1 0.013 670.7 0.13 3,353.7 0.66 6,707.3 1.32 OPEC 181.6 0.035 1,816.3 0.35 9,081.5 1.77 18,163.1 3.54 TOTAL 1,421.6 0.022 14,216.5 0.22 71,082.3 1.10 142,164.5 2.21 source: The authors

expectancy. The virus affects households, costs of hypothetical infl uenza pandemics. businesses and governments through lower Martin Meltzer, Nancy Cox and Keiji Fukuda household incomes, loss of labor supply and of the Centers for Disease Control and Pre- increased government spending for health vention put the fi gure for the U.S. economy at care and support for orphans. $73-167 billion. The Asian Development The fl u virus is far more contagious than Bank staff produced an estimate of a 2.6 to HIV, and the onset of an epidemic is sudden 6.5 percent loss for Asia’s GDP and 0.6 per- and unexpected. Hence, if the world’s re- cent for global GDP. Yet another study, by the sponse to SARS is any indication, the spread U.S. Congressional Budget Offi ce, estimated of a strain similar to Spanish infl uenza in its that a mild-to-severe pandemic would reduce virulence and the severity of symptoms would U.S. GDP by between 1.5 and 5 percent. almost certainly induce panic. We used our own technical study, pub- Indeed, the SARS epidemic in 2003 dem- lished last year by the Lowy Institute for In- onstrated that even in an event with a rela- ternational Policy and the Australian Nation- tively small number of cases and deaths, the al University’s Center for Applied Macroeco- global costs can be large and reach beyond the nomic Analysis, to outline the likely global countries directly affected. Epidemics have economic consequences of an infl uenza pan- signifi cant effects on economies through large demic under four possible scenarios: reductions in consumption of various goods • A mild pandemic, modeled on the 1968-69 and services (like tourism and group recre- Hong Kong Flu. ation), increases in business operating costs, • A moderate pandemic, modeled on the and the fl ight of capital. Asian Flu of 1957-58. Several earlier studies have estimated the • A severe pandemic, with mortality and

22 The Milken Institute Review morbidity similar to that of the Spanish Flu An infl uenza pandemic would be expected of 1918-1919. to lead to: • An ultra pandemic, modeled on high-end • A decline in the size of the labor force due estimates of casualties from the Spanish Flu. to a rise in mortality and disabling illness. • An increase in the cost of doing business, the model especially in the service sectors in which To simulate the economics of a fl u pandemic, human interactions are largest. we used a computer model developed by • A shift in consumer preferences away from Warwick McKibbin and Peter Wilcoxen (Syr- services that require exposure to other peo- acuse University). It includes 20 countries or ple, which is independent of the effects of regions (the United States, Japan, the United changes in incomes and prices. Kingdom, Europe, Canada, Australia, New • A re-evaluation of investment risk in light Zealand, Korea, China, India, Indonesia, Thai- of the responses of governments and their land, Taiwan, Hong Kong, Philippines, Singa- health systems to the infl uenza pandemic. pore, Malaysia, Eastern Europe and the for- Epidemic shocks. We used two indicators mer Soviet Union, OPEC, and other develop- to build the mortality shocks for each sce- ing economies) with six sectors of production nario in each country or region: an index of in each economy (energy, mining, agricul- geographic susceptibility to an infl uenza pan- ture, durable-goods manufacturing, nondu- demic and an index of health policy. rable-goods manufacturing, and services). The index of geography refl ects the ease The equations capture both trade and fi nan- with which the infl uenza virus can enter a cial market linkages between and within country through air travel, and its capacity to economies. A more detailed description is spread within the country once introduced. available at http://www.brookings.edu/views/ The fi rst (international) component of the papers/mckibbin/200602.htm. geography index is based on air transport The prediction of shocks associated with data. The domestic component is a measure the spread of the fl u follows the methods de- of population density. The higher the density veloped by McKibbin and Jong-Wha Lee for and frequency of contacts, the faster the analyzing the economic costs of SARS, adapt- spread of the epidemic. ed to model the effect of a pandemic on each The health policy index measures the like- country. The model captures economic rela- ly effi ciency of a country’s health care system tionships among fi rms, consumers and gov- to deal with the pandemic, and is constructed ernments. The regions in the model are from data on per capita health expenditure and linked by fl ows of goods and assets. Money infl uenza-specifi c policies in each country. supply and demand are also explicitly includ- These two indexes combined create an in- ed in the equations. The impact of the spread- dex of mortality for each country. This trans- ing infl uenza pandemic is fed into the equa- lates into an aggregate world scenario defi ned tions through a series of shocks constructed in terms of total world deaths, with the mild to match each of the four scenarios. The scenario generating 1.4 million deaths (0.022 shocks are constructed for the benchmark percent mortality), the moderate scenario 14 economy, the United States, then scaled and million deaths (0.22 percent mortality), the adjusted by various factors for the other severe scenario 71 million deaths (1.1 percent countries. mortality), and the ultra scenario 142 million

Third Quarter 2007 23 influenza pandemic in the second year. The mortality shocks are deaths (2.2 percent mortality). much larger for developing countries. There are substantial differences in mor- Sector exposure. Some productive sectors tality rates among countries. The hardest hit would be particularly vulnerable to a pan- are Indonesia and the Philippines: in the demic. We calculate the share of these vulner- “ultra” scenario, more than 5 percent of their able sectors in each country’s GDP. Note that population dies. The smallest shocks occur in there are major differences in the structure of the United States, Europe and Canada. These service industries across economies, from less differences, together with the underlying then 2 percent of exposed services in China structures of economies and the linkages to more than 35 percent in Hong Kong. These through global trade and capital markets, sector-exposure indexes are used to weight drive the economic consequences. the effect of mortality shocks on both pro- duction costs and demand. GDP LOSS BY REGION Risk-premium calculations. The initial “risk (PERCENTAGE CHANGE, 2006) premium” added to the cost of capital is

MILD MODERATE SEVERE ULTRA calculated for individual countries. In the US -0.6% -1.4% -3.0% -5.5% model, large differences in these risk premi- Japan -1.0 -3.3 -8.3 -15.8 ums among countries lead to rapid move- UK -0.7 -2.4 -5.8 -11.1 Europe -0.7 -1.9 -4.3 -8.0 ment of capital – capital fl ight – and atten- Canada -0.7 -1.5 -3.1 -5.7 dant dislocation in fi nancial markets. -0.8 -2.4 -5.6 -10.6 Australia Shocks to costs of production. The shocks New Zealand -1.4 -4.0 -9.4 -17.7 Indonesia -0.9 -3.6 -9.2 -18.0 to costs of doing business are assumed to Malaysia -0.8 -3.4 -8.4 -16.3 vary across sectors, countries and scenar- Philippines -1.5 -7.3 -19.3 -37.8 Singapore -0.9 -4.4 -11.1 -21.7 ios, and are based on the Lee and McKib- Thailand -0.4 -2.1 -5.3 -10.3 bin SARS study. The shocks to service indus- China -0.7 -2.1 -4.8 -9.1 India -0.6 -2.1 -4.9 -9.3 tries are adjusted by the service sector expo- Taiwan -0.8 -2.9 -7.1 -13.8 sure index. This difference shows up stark- Korea -0.8 -3.2 -7.8 -15.1 Hong Kong -1.2 -9.3 -26.8 -53.5 ly in China, where it is small relative to the LDCs -0.6 -2.4 -6.3 -12.2 other countries. EEFSU -0.6 -1.4 -2.9 -5.4 OPEC -0.7 -2.8 -7.0 -13.6 Shocks to demand. The model generates shifts in spending patterns as a result of the source: The authors changes in tastes, incomes, wealth and relative prices – cable television would presumably Impact on labor supply. The morbidity rate fare far better than live concerts. We model captures losses due to absenteeism, including this behavior by imposing shocks on the de- workers who stay out to care for family mem- mand for various products, following the bers. In the mild scenario, this effect dom- spending shifts assumed by Lee and McKib- inates the impact on the labor force in the bin for the SARS observations in Hong Kong fi rst year, but has no long-term consequenc- and Singapore. es. Deaths, of course, lead to a permanent loss Shocks to risk. The risk premium weights of labor. Under the severe and ultra scenari- discussed earlier are combined with the mor- os, the morbidity shocks are assumed to halve tality shocks to produce a risk-premium shock within a year and then disappear altogether for all countries. This indicator of the degree

24 The Milken Institute Review of fi nancial panic suggests that the countries ing from 0.6 percent for the United States to most prone to panic are those in East Asia 1.5 percent in the Philippines. The contribu- and other developing areas. tion of capital fl ight and demand shifts are The numbers. As the labor supply contracts small relative to the supply-side shocks. because of mortality and morbidity, the re- In the severe scenario, the world economy turn on capital falls in all affected countries is truly hammered by the pandemic. In parts – but more so in those countries experienc- of Asia, GDP contracts by as much one- ing larger shocks. Growth slows everywhere fourth – a major shock by any measure. The as output falls. But the differences among relatively large contraction in Asia compared countries imply that fi nancial capital will fl ow to the United States and Europe partly refl ects from the developing countries to the United the much larger shocks in these economies as States and Europe, where investors feel safer. well as the fl ight of capital to North America In the model, Japan experiences a larger mor- and Europe. The cost shocks also play a much tality shock than North America and Europe, larger role on the GDP losses in the severe and takes an extra hit as well from the fact scenario. Financial markets close down as the that its economy is more closely tied to East shocks intensify. Asia and is thus less of a safe haven than the As the pandemic grows more damaging in other rich economies. increasingly severe scenarios, the largest GDP In the mild scenario, the labor-force shock losses are linked to rising production costs. The is the largest driver of the GDP contraction in panic component captured by capital fl ight most countries. Next most important is the appears to have much less effect on GDP than increase in the costs of production. The fall in the fundamentals driven by supply shocks. In

associated press associated GDP is signifi cant even in this scenario, rang- addition, the effects on wealth and income

Third Quarter 2007 25 influenza pandemic from stocks to bonds. This is not true in are larger in developing countries, and the Hong Kong and the developing-country block contraction of demand is therefore much which, by pegging their currencies to the U.S. larger than in Europe and North America. dollar, are forced to raise interest rates in Japan is an intermediate case, since it order to prevent currency depreciation. This would experience a larger shock than other tightening of domestic monetary policy in- industrialized economies but a smaller one creases the output losses from the pandemic, than the rest of East Asia. Japan is also more but reduces its infl ationary potential. The U.S. integrated with the collapsing East Asian dollar appreciates against all currencies ex- economies, and bears a secondary burden as- cept those tightly pegged to it. sociated with a decline in regional trade. Changes in countries’ external fi nancial The loss of labor through deaths and sick- accounts refl ect changes in capital fl ows. As ness reduces output in all countries and can the scenarios worsen, there is a larger fall in be expected to raise infl ation in the short run equity prices in the more-affected economies;

Even a “small” pandemic would generate hundreds of billions in losses — and a disproportionate amount of those losses would be borne by developing countries.

to the extent that output falls by more than in the safe-haven economies like North Amer- demand falls because of contraction of in- ica and Europe, stock prices fall less. This re- come and wealth. Because the shock is ex- fl ects capital infl ow into these rich, stable pected to be temporary, households would economies, which ends up being invested in respond to falling income by reducing savings part in stocks. – and thus consumption would remain rela- Temporary rises in government spending tively strong. The rise in the cost of doing tend to reduce the GDP loss in the most-af- business also tends to push up prices. On the fected economies, and to raise infl ation (as other side of the ledger, the shift in demand well as short-term interest rates) in those away from the most vulnerable productive economies. The fi scal response could be a sectors tends to lower the relative price of critical determinant of the impact of the pan- those products and services, while the im- demic on bond markets. posed fall in aggregate spending (investment, in particular) also tends to lower prices. what it all means These opposing factors act to create infl a- Even a mild pandemic would likely make a tion in most economies. The standout excep- noticeable dent in global economic output. tion in the model is Hong Kong, but there is The mild scenario, estimated to cost the also mild defl ation in the Philippines, Malay- world 1.4 million lives, would reduce total sia and New Zealand. output by nearly one percent, or approxi- The model’s assumptions imply that, in mately $330 billion in the fi rst year. As the as- most countries, stock prices fall and interest sumed scale of the pandemic increases, so, of rates fall along with them as investors shift course, do the economic costs. A massive

26 The Milken Institute Review global economic slowdown occurs in the tract signifi cantly. “ultra” scenario in which more than 142 mil- Whether a pandemic causes infl ation or lion people die and output in some develop- defl ation depends on the relative size of de- ing economies shrinks by half in the year of clines in demand and supply across sectors. the pandemic. The loss in output in this sce- Consumption-smoothing – the tendency of nario is $4.4 trillion in the fi rst year – 12.6 households to try to maintain living stan- percent of global GDP. The composition of dards in the teeth of temporary distress – im- the slowdown differs sharply across countries, plies that aggregate demand would decline by with a major shift of global capital from the less than supply. This, together with increases affected economies to the less affected safe in the costs of doing business, suggests that havens of North America and Europe. infl ation is a more Some robust results – that is, results that likely consequence are not terribly sensitive to the model’s un- of pandemic than derlying assumptions – do emerge. One is defl ation. But a suf- that stock markets fall and bond markets fi ciently strong shift rally, although to differing degrees in differ- in spending prefer- ent countries. The stock price reactions ap- ences that caused se- pear to be reasonably small because the eco- rious economic dis- nomic hit associated with a pandemic takes location could lead place over a brief period and recoveries are to defl ation. Coun- quick. In addition, the fall in interest rates tries that try to pre- also buoys equity prices. vent exchange-rate The second robust result is that monetary changes are more policy responses play a key role in determin- likely to experience ing economic consequences. Countries ex- a defl ationary shock pected to focus on preventing exchange-rate as tight monetary policy compounds the eco- depreciation end up with very tight money, nomic contraction. which raises the costs of the pandemic. This is There is plainly a great deal of uncertainty particularly true of Hong Kong, which re- about how individuals and markets would re- ceives the largest shock of any country thanks spond to mass illness and signifi cant loss of to its density and openness, yet has the most life. Some things are clear, nonetheless. A rigid exchange-rate regime. 1918-style pandemic would destroy trillions We also fi nd that the more severe the pan- of dollars of potential output along with mil- demic, the more that developing countries are lions of lives. Even a “small” pandemic would hurt relative to North America and Europe; generate hundreds of billions in losses – and Japan is caught in the middle. The differences a disproportionate amount of those losses in the epidemiological outcome generate would be borne by developing countries, fl ows of capital from the most affected devel- which could least afford the blow. At least one oping countries to industrialized economies, conclusion, then, is crystal clear: it would pay worsening the current account positions of to invest considerable resources now to pre- the receiving countries and putting down- vent a fl u pandemic and to prepare for the ward pressure on developing-countries’ ex- consequences of massive economic jolts if the M john stanmeyer/vii change rates. World trade would likely con- pandemic does come.

Third Quarter 2007 27