Macroeconomics of the Great Influenza Pandemic, 1918–1920 (2024)

Abstract

Data for 48 countries during the Great Influenza Pandemic imply flu-related deaths in 1918–1920 of 40 million, 2.1 percent of world population, implying 160 million deaths when applied to current population. Regressions with annual information on flu deaths 1918–1920 and war deaths during WWI imply flu-generated economic declines for GDP and consumption in the typical country of 6 and 8 percent, respectively. Higher flu death rates also decreased realized real returns on stocks and, especially, on short-term government bills.

I. The Great Influenza Pandemic

A reasonable upper bound for mortality and economic effects from the ongoing COVID-19 pandemic can be derived from the world’s experience with the Great Influenza Pandemic (popularly and unfairly known as the Spanish Flu¹), which began and peaked in 1918 and persisted through 1920. Our estimate, based on data discussed later on flu-related death rates for 48 countries, is that this pandemic killed around 40 million people worldwide, corresponding to 2.1 percent of the world’s population at the time. These numbers are likely the peak of worldwide mortality from a “natural disaster” in modern times, though the impact of the plague during the black death in the 14^th century was much greater as a share of the population.²

The Great Influenza Pandemic arose in three main waves, the first in spring 1918, the second and most deadly from September 1918 to February 1919, and the third for the remainder of 1919. (A fourth wave applies in some countries in 1920.) This airborne infection was based on the Influenza A virus subtype H1N1. The coincidence of the two initial waves with the final year of World War I (1918) encouraged the spread of the infection, due to crowding of troops in transport, including large-scale movements across countries. An unusual feature was the high mortality among young adults without pre-existing medical conditions. This pattern implies greater economic effects than for a disease with comparable mortality that applied mostly to the old and very young.

The pandemic killed a number of famous people, including the sociologist Max Weber, the artists Gustav Klimt and Egon Schiele, the child saints Francisco and Jacinta Marto, and Frederick Trump, the grandfather of Donald Trump. Many more famous people were survivors, including Franz Kafka, Friedrich Hayek, General Pershing, Walt Disney, the Spanish King Alfonso XIII, the actress Mary Pickford, and the leaders of France and the United Kingdom at the end of World War I, Georges Clemenceau and David Lloyd George. Of particular note, the disease severely impacted U.S. President Woodrow Wilson, whose impairment likely had a negative impact on the negotiations of the Versailles Treaty in 1919.

Table 1 shows our estimates of excess mortality rates from the Great Influenza Pandemic. These rates are expressed relative to the total population for 48 countries for each year from 1918 to 1920.³ These data come from an array of sources, detailed in Ursúa (2009) and Weng (2016). Important references are Johnson and Mueller (2002), Murray, et al. (2006), Mitchell (2007), and Human Mortality Database. Notably, the Murray, et al. (2006) study used all vital registration data available worldwide from 1915 to 1923. For countries with annual statistics on death tolls from the flu and flu-related deaths such as pneumonia, these direct numbers are used to measure excess mortality rates for 1918–1920. For some other countries, we followed their methodology to calculate the annual all-cause excess mortality rate for 1918–1920, measuring deaths that were above the average mortality rate from three years before and after the 1918–1920 period. Comparisons of direct yearly estimates of death rates from influenza/pneumonia with all-cause excess mortality rates for countries with both types of data indicate a close correspondence for the two methods. For the few countries for which there is little or no detail on the annual flu breakdown, we used the time distribution of deaths in neighboring countries as an approximation.

The 48 countries covered (42 of which have GDP data for the relevant timeframe) constitute 92 percent of estimated world population in 1918.⁴ These covered places would represent a larger share of world GDP at the time.

The numbers in Table 1, combined with information on country population, correspond to total flu deaths for the 48 countries of 25.0 million in 1918, 9.2 million in 1919, and 2.8 million in 1920, for a total of 36.9 million. When inflated to the world’s population (assuming comparable flu death rates in the uncovered places), the numbers are 27.1 million in 1918, 9.9 million in 1919, and 3.1 million in 1920, for a world total of flu deaths of 40.1 million cumulated over 1918–1920. The estimated aggregate flu death rates for the 48 countries were 1.42 percent for 1918, 0.52 percent for 1919, and 0.16 percent for 1920; the sum of these death rates is 2.10 percent.

Table 1 shows that the flu mortality rate varied greatly across countries and years. Some observations are zero; for example, because of a swift quarantine response, Australia avoided the pandemic during 1918. Moreover, Australia did not suffer unusually high death rates when the flu arrived in 1919; instead, the overall death rate of 0.3 percent is comparatively low. (The presence of Australia in the Southern Hemisphere is not the key factor here because New Zealand’s death rate was more than twice as high and South Africa’s was greater by a factor of more than ten.)

The highest cumulative death rate is for Kenya at 5.8 percent, followed by India at 5.2 percent.⁵ Because of its high population (320 million), India accounted in 1918–1920 for 16.7 million flu deaths out of the world total of 40.1 million; that is, 42 percent of the total. The next highest cumulative death rates were for Guatemala at 3.9 percent, Madagascar 3.5 percent, South Africa 3.4 percent, Mauritius 3.2 percent, and Indonesia 3.0 percent. China’s death rate was not nearly as high, but because of its large population (about 570 million), its 8.1 million deaths (20 percent of the world total) were second highest across the countries. Spain is not special, with a cumulative death rate of 1.4 percent and a corresponding number of deaths of 300 thousand. The United States had a cumulative death rate of 0.5 percent, with an associated number of deaths of 550 thousand.

The mortality rates shown in Table 1 apply to total populations. The underlying data here are numbers of deaths and sizes of total populations. Mortality rates related to numbers infected are much less reliable because they depend on counts of infections, which are less accurately measured than deaths. A commonly quoted figure is that one-third of the world’s population was infected by the H1N1 virus during the Great Influenza Pandemic. If this number were accurate, a mortality rate of 2 percent for the overall population would translate into a mortality rate of 6 percent for the infected population.

The one-third number for the world infection rate seems to come from Taubenberger and Morens (2006, p. 15), who cite Frost (1920) and Burnet and Clark (1942).⁶Frost’s (1920) evidence for the United States derives from surveys of 130,000 people in 11 U.S. cities and rural areas carried out in 1919 by the U.S. Public Health Service. Excluding Louisville, which had a truncated survey, the morbidity rates—based on self-diagnosed recall of flu-like symptoms— ranged from 18.5 percent for New London to 53.5 percent for San Antonio. The overall infection rate was 29.3 percent (computed from the numbers given in Frost [1920, table on p. 588 and map on p. 585]). Frost (pp. 584–586) notes that the underlying canvases were carried out intelligently and on reasonable size samples. But he also observes (p. 597) that the numbers on morbidity are unreliable even for the whole of the United States: “As to the value of the statistics … they represent so few localities and such a small number of observations … that … they contribute little towards giving a picture of the epidemic in the country at large.” Results from Mills, et al. (2004), based on an epidemiological model fit to the time profile of observed excess deaths in U.S. cities in 1918, accord with a roughly one-third infection rate. However, this conclusion comes from the model, not from data on morbidity. For other countries, there seems to be no reliable information on numbers of infections during the Great Influenza Pandemic. Therefore, the estimated infection rate of one-third and the resulting infected mortality rate of 6 percent have to be regarded as speculative. On much firmer ground is the estimated mortality rate of 2.1 percent out of the total population. The regressions implemented below use the estimated mortality rates out of the total population in each country, as shown in Table 1.

The present analysis focuses on the impact of a country’s flu death rate on its economic outcomes, not on reverse effects of economic conditions on the death rate. However, it is worth noting that the cumulative flu death rate for 1918–1920 has a correlation of −0.43 with the log of a country’s real per capita GDP in the prior year 1913 (for the 42 countries with data on GDP). The magnitude of this correlation would likely be larger if we were able to include the countries with missing data on GDP—in our sample, these are mostly places in sub-Saharan Africa and Central America, which have high flu death rates and low levels of economic development. The inverse relation between the death rate and the prior level of per capita GDP likely reflects the negative impact of better health services and better organization more broadly on the probability of death from the disease (reflecting partly risk of infection and partly the mortality rate given infection). Another force—apparently only partly offsetting—is that more advanced economies tend to have greater mobility and interactions, which foster spread of contagious disease.

Applying the flu death rates from the Great Influenza Pandemic to current population levels (about 7.5 billion worldwide in 2020) generates staggering mortality numbers. A death rate of 2.1 percent corresponds in 2020 to around 150 million deaths worldwide, 6.8 million in the United States. However, these numbers likely represent the worst-case scenario today, particularly because public-health care and screening/quarantine procedures are more advanced than they were in 1918–1920. Other factors, such as greater international travel, work in the opposite direction. In addition, those worst-case scenarios do not account for differences in the demographic profiles of the Great Influenza Pandemic compared to the ongoing COVID-19.

II. Effects on Economic Growth

A major objective is to use the cross-country data to estimate the impact of the Great Influenza Pandemic on economic growth. Barro and Ursúa (2008) found that this impact might have been substantial. That research focused on rare macroeconomic disasters, using a definition of a disaster as a cumulative decline over one or more adjacent years by 10 percent or more in real per capita GDP or real per capita consumption (based on data on real personal consumer expenditure). Using this definition, the three most important adverse global events since 1870 were World War II,⁷ the Great Depression of the early 1930s, and World War I. The results further suggested that the Great Influenza Pandemic of 1918–1920 might have been the next most important negative macroeconomic shock for the world. Specifically, 12 countries were found (in Barro and Ursúa [2008, Table C2]) to have macro disasters based on GDP with trough years between 1919 and 1921, and 8 were found (in Table C1 for a smaller sample of countries with data) to have these disasters based on consumption. A complicating factor in this analysis was the difficulty in distinguishing effects of World War I from those of the Great Influenza Pandemic. Therefore, it is important that the present analysis allows for a separation of these two forces.

The long-term annual national-accounts information described in Barro and Ursúa (2008) was subsequently expanded to 42 countries and covers the period of World War I and the Great Influenza Pandemic.⁸ We use these data to study the determinants of growth rates of GDP and private consumption, notably to isolate effects from the Great Influenza Pandemic. This analysis exploits variations in flu intensity from 1918 to 1920 across countries and over time, as shown in Table 1.

To hold fixed the effects of World War I, we gauge the war intensity for each country that participated in the war by the ratio of military combat deaths to total population. The data by country on combat deaths, including missing in action, come mainly from Urlanis (2003, part II). In terms of annual death rates during the war, we found estimates for seven countries (France, Germany, Italy, United Kingdom, United States, China, and Taiwan). For the remaining countries involved in World War I, we use the annual distribution of deaths from countries that either fought alongside or against the given country. For example, British Commonwealth countries and colonies are assigned the time distribution of the United Kingdom, while Austria, Japan, Russia, and Greece follow that of Germany. The resulting data are in Table 2.

Our sample has a total of 6.2 million combat-related military deaths from 1914 to 1918. This number substantially understates the commonly cited total death toll for World War I of around 20 million, but this larger figure includes civilian excess deaths from a variety of causes, as well as deaths of soldiers due to illness and while prisoners of war. The main point is that the deaths of soldiers in combat are measured most accurately and are likely to be a satisfactory proxy for the intensity of the war across countries and over time.⁹

An important point is that the data contained in Tables 1 and ​and22 encompass a lot of independent movements in flu and war death rates in 1918, the peak year of the Great Influenza Pandemic and the final year of World War I. Notably, many countries that experienced the flu were not involved in the war.

Table 3 uses regression analysis to assess effects of the Great Influenza Pandemic and World War I on economic growth, gauged by growth rates of real per capita GDP and real per capita consumption (personal consumer expenditure). The sample periods for annual growth rates are 1901 to 1929. The start year is arbitrary, and results are similar if we go back to 1870. The ending of the sample in 1929 simplifies the analysis by excluding the Great Depression. The cross-section corresponds to the 42 countries for which we have data on real per capita GDP. (The sample for consumption is smaller because of missing data.) The explanatory variables are the flu and war death rates, as shown in Tables 1 and ​and2.2. Values for the flu death rate outside of 1918–1920 are set to zero,¹⁰ and similarly for the war death rate outside of 1914–1918. The regressions include no other explanatory variables aside from constant terms. That is, our focus is on the two disaster shocks—flu and war—which we treat as exogenous shocks. In interpreting the results, we view the associated events—World War I and the Great Influenza Pandemic—as unanticipated and contemporaneously perceived as having some persistence but ultimately being temporary. The results for GDP growth are in the first three columns and those for consumption growth are in the next three columns. Estimation is by panel least squares, with standard errors of estimated coefficients computed by allowing for clustering of the error terms by year.¹¹

Importantly, this analysis does not suffer from problems typically associated with cross-country growth regressions, notably that the included explanatory variables may proxy for an array of excluded variables. The present analysis assesses how two large shocks—flu deaths and war deaths—relate to differential changes across countries in rates of economic growth. Although there is some association of the flu shocks with prior levels of economic development, the flu death-rate variable mainly picks up exogenous and unanticipated variations for individual countries. Moreover, the results change negligibly if measures of prior levels of economic development (such as real per capita GDP in 1913) are held constant.

The regression for GDP growth in column 1 includes only the contemporaneous values of the flu and war death rates. The two estimated coefficients are significantly negative at least at the 5 percent level—indicating that flu and war are both bad for economic growth.¹² The coefficient of −3.0 on the flu death rate means that, at the cumulated aggregate death rate of 0.021 for 1918–1921 (Table 1), the Great Influenza Pandemic is estimated to have reduced real per capita GDP by 6.2 percent in the typical country. Given the cross-country range of experience with flu intensity, this result accords with the observation from before that the pandemic could have caused a substantial number of rare macroeconomic disasters in the sense of declines in real per capita GDP by 10 percent or more.

The coefficient of −17.9 on the war death rate means that, at the cumulated mean death rate of 0.0047 for 1914–1918, World War I is estimated to have reduced real per capita GDP in the typical country by 8.4 percent. This result accords with the large number of macroeconomic disasters associated with World War I, as reported in Barro and Ursúa (2008, Table C2).

The form of the regression in column 1 of Table 3 implies that the negative effects of temporary flu and war on growth rates are temporary and, hence, that the adverse effects on levels of real per capita GDP are permanent. Column 2 tests for these implications by including lags of flu and war death rates in the specification. If the depressing effects of temporarily high flu and war death rates on the level of per capita GDP were only temporary, then lagged values of these death rates should, eventually, have positive coefficients—that is, negative growth-rate effects would be reversed in the long run by recovery in the form of positive growth-rate effects.

Column 2 adds as regressors the average of the flu and war death rates for annual lags 1 through 4 and for annual lags 5 through 8. For flu death rates, the estimated coefficient on these two lagged variables are each positive but insignificantly different from zero at the 5 percent level. The two lags are also jointly insignificantly different from zero (p-value=0.25). However, we do not reject the hypothesis that the sum of the coefficients on the contemporaneous and lagged flu variables add to zero (p-value=0.48). Therefore, the results cannot rule out effects of the flu pandemic on the level of real per capita GDP that are fully permanent (corresponding to a coefficient of zero on the lagged variables) or fully temporary (where the coefficients on the contemporaneous and lagged variables sum to zero) or somewhere in between.

For war death rates, the first lag variable is significantly negative, indicating that the adverse effect of war on GDP growth tends to build up for a while. Then the second lag variable is significantly positive, indicating a systematic tendency for recovery of per capita GDP following a prior war. In this case, the sum of the three coefficients related to the war death rate is significantly negative (p-value=0.012). This result implies that the recovery from wartime economic decline is only partial; that is, part of the negative effect on the level of per capita GDP—roughly half—is permanent. This finding accords with broader results about rare macroeconomic disasters reported in Nakamura, Steinsson, Barro, and Ursúa (2013) and Barro and Jin (2021). Those studies found for a broad panel of countries that about half of disaster-related declines in consumption were permanent.

Columns 3 and 4 repeat the analysis for consumption growth rates. The sample size is smaller than that for GDP mostly because only 30 of the countries have full annual data on consumption going back at least to 1914. The main results are analogous to those for GDP growth rates, although the estimated effects on consumption growth are larger in magnitude. This result is not surprising for wartime effects, because the expansion of government outlays for the war would depress consumption beyond the effect from lower GDP. However, this pattern is surprising for flu effects.

We noted before the substantial number of rare macroeconomic disasters with troughs between 1919 and 1921. One of these events is the sharp U.S. economic decline from 1918 to 1921 (12 percent for GDP, 16 percent for consumption). In the U.S. history since 1870, this event comes just after the Great Depression in terms of the extent of proportionate declines in GDP and consumption.¹³ However, although it likely played a role, the Great Influenza Pandemic is probably not the main source of the large contraction. First, the U.S. cumulated flu death rate of 0.52 percent corresponds to estimated decreases by only 1.5 percent for GDP and 2.1 percent for consumption (using the respective regressions coefficients on the influenza death rate from columns 1 and 3 of Table 3). Second, part of the timing is off—although there were substantial declines in GDP and consumption in 1919 and 1920, the largest decreases were in 1921 (6 percent for GDP, 7 percent for consumption), well after the peak of the U.S. flu death rate in 1918.

In contrast to the United States, the magnitude of expected declines and their timing fit better for other cases in our sample. As an example for GDP, our regression results (column 1 of Table 3) imply a contraction in India by 15.6 percent driven by both death rates. This amount is close to the observed contraction by 14.6 percent in India between 1916 and 1918, troughing in the year when it was most affected by the pandemic. As an example for consumption, our results (column 3 in of Table 3) imply a contraction in Canada by 16.5 percent. The actual figure for 1918–1921 was 19.6 percent, but a large part of that contraction happened between 1918 and 1919 (by 12.3 percent), which at least in part can be attributed to the negative impact of the pandemic in combination with war deaths.

III. Effects on Rates of Return and Inflation Rates

We now turn to exploring the effects of the pandemic- and war-related death rates on asset prices. Table 4 shows regression results for effects of flu and war death rates on realized real rates of return and inflation rates. As noted before, in interpreting the results, we view the associated events—World War I and the Great Influenza Pandemic—as being unanticipated and contemporaneously perceived as having some persistence but ultimately being temporary. We consider returns on two types of assets: stocks (based on broad market indexes) and short-term government bills (analogous to U.S. Treasury Bills). In carrying out this analysis, we excluded observations with the most extreme inflation rates, which included hyperinflationary outcomes for Austria and Germany after World War I—the peak inflation rate was 1.8×10¹⁰ percent per year in Germany in 1923. These observations are sensitive to measurement error for inflation and, therefore, for real assets returns, which are computed from data on nominal returns and inflation rates. The simple linear relationships that we use also would not work for these extreme cases.

Columns 1 and 2 of Table 4 apply to realized real returns on stocks. The contemporaneous effect of the flu death rate is negative but statistically insignificantly different from zero. However, the point estimate, −13.1 (with a p-value of 0.13), is large. At a flu death rate of 2.1 percent (aggregate value from Table 1), this coefficient implies that the real stock return would be lower by 28 percentage points. At the U.S. death rate of 0.52 percent, the impact would be only 7 percentage points. Lagged effects are unimportant; that is, there is no prediction that the short-term negative effect will be reversed.

For the war death rate, the estimated contemporaneous effect is significantly negative. The coefficient, −40.0, implies that, at the mean war death rate of 0.0047 (from Table 2), the real stock return would be depressed by 19 percentage points. In this case, lagged effects are important, particularly the positive coefficient on the second lag. (The p-value for joint significance of the two lagged variables is 0.050.) A test that the coefficients of the contemporaneous and two lagged terms add to zero is accepted with a p-value of 0.27. Therefore, the results predict an eventual recovery from the short-term stock-market decline, and the overall impact of war on real stock-market value might be zero.

Columns 3 and 4 of Table 4 apply to realized real returns on government bills. The estimated coefficient on the contemporaneous flu death rate is significantly negative. The coefficient of −7.0 implies that the real return is depressed by 14 percentage points at a flu death rate of 2.1 percent (or by 3.6 percentage points at a flu death rate of 0.52 percent). This large effect can be interpreted partly as a decline in the “safe” expected real interest rate and partly as an effect of higher inflation (considered next) on the realized real returns on nominal claims (to the extent that bills have non-negligible maturity). The estimated coefficients on the lagged variables are individually and jointly insignificantly different from zero.

For the war death rate, the estimated coefficient on the contemporaneous variable is significantly negative. The coefficient of −27.2 means that, at the mean war death rate of 0.0047, the real return would be depressed by 13 percentage points. Lagged effects are unimportant here.

Columns 5 and 6 of Table 4 apply to the inflation rate. The data refer to reported price levels, which would have been influenced by price controls pursued during World War I in the United States and other countries, including Germany and the United Kingdom. The estimated effect of the Great Influenza Pandemic is significantly positive—the contemporaneous coefficient of 10.1 means that the inflation rate would have been higher by 21 percentage points at a flu death rate of 2.1 percent (or by 5 percentage points at a flu death rate of 0.52 percent). However, the estimated first lag coefficient is significantly negative and about the same magnitude, thereby indicating that the eventual effect on the price level could have been negligible (p-value =0.9 for the hypothesis that the coefficients of the contemporaneous and two lagged values add to zero).

For the war death rate, the contemporaneous coefficient is significantly positive, and the first lag coefficient is also significantly positive. In this case, the results reject the hypothesis (p-value=0.000) that the ultimate effect on the price level is nil.

The results on inflation confirm that the Great Influenza Pandemic and, especially, World War I increased inflation rates at least temporarily. These responses are important in interpreting the effects of these events on realized real rates of return, especially for those on real T-bill returns.

IV. Comparison with the Coronavirus Pandemic

The Great Influenza Pandemic of 1918–1920 represents a plausible worst-case scenario for global disease outbreaks such as the one caused by COVID-19 in 2020–2021. The estimated worldwide deaths of 40 million in 1918–1920, corresponding to an excess death rate of 2.1 percent, translates into 160 million persons when applied to the world’s population of about 7.5 billion in 2020. For the United States, the estimated deaths of 550,000 in 1918–1920, or an excess mortality rate of 0.52 percent, corresponds to 1.7 million deaths when applied to the U.S. population of 331 million in 2020. In contrast, according to Opportunity Insights (2021), the actual U.S. death rate from COVID-19 from April 2020 through September 2021 was 0.20% or 674,000 persons. That is, the U.S. excess mortality rate was about 40% of that in 1918–1920. This shortfall may reflect advances in medical treatment—especially vaccines—and in political measures intended to mitigate propagation. But, of course, COVID-19 is also not the same virus as the one that caused the Great Influenza Pandemic.

According to our regression results, the excess mortality in 1918–1920 caused declines in the typical country’s GDP by about 6 percent. These contractions are comparable to those during the Great Recession of 2008–2009 but fall short of those in 2020. However, the 2020 global recession has a surprising shape; appearing as a sharp decline in 2020 followed by an equally sharp recovery in 2021. This V-shaped pattern differs from the recessions related to the Great Influenza Pandemic; in fact, the 2020–2021 business-cycle pattern is unique. This outcome likely reflects not the nature of the disease caused by COVID-19 but rather the extreme political response. Unlike in 1918–1920, governments shut down large swaths of economic activity during 2020. The rapid recoveries in 2021 were, in large part, re-openings of economies that had been artificially closed down.

Footnotes

Contributor Information

Robert J. Barro, Harvard University.

José F. Ursúa, Dodge & Cox Joanna Weng, EverLife.

References