拉丁美洲的新冠疫情：對生命和生計造成的巨大損失

1、WP/21/168COVID-19 in Latin America: A High Toll on Lives andLivelihoodsby Bas B. Bakker and Carlos GoncalvesIMF Working Papers describe research in progress by the author(s) and are published to elicit comments and to encourage debate. The views expressed in IMF Working Papers are those of the autho

2、r(s) and do not necessarily represent the views of the IMF, its Executive Board, or IMF management. 2021 International Monetary FundWP/21/168IMF Working PaperWestern Hemisphere DepartmentCOVID-19 in Latin America: A High Toll on Lives and Livelihoods Prepared by Bas B. Bakker and Carlos Goncalves1Ju

3、ne 2021IMF Working Papers describe research in progress by the author(s) and are published to elicit comments and to encourage debate. The views expressed in IMF Working Papers are those of the author(s) and do not necessarily represent the views of the IMF, its Executive Board, or IMF management.Ab

4、stractLatin America was hit hard by Covid-19, both in terms of lives and livelihoods. Early lockdowns in the second quarter of 2020 prevented an explosion of deaths at the time but did not stop the pandemic from later wreaking havoc in the region. This paper investigates the dynamics of pandemics in

5、 Latin America and how it differed from elsewhere. We probe the role of non-pharmaceutical interventions; the effectiveness (or lack of thereof) of lock-downs in Latin America; which structural factors contributed to the high death toll in Latin America; and the extent to which the epidemic harmed t

6、he economy. Finally, we briefly analyze the roots of the second-waves that started in the fourth quarter of 2020.JEL Classification Numbers: I12, I15, I18Keywords: Latin America, COVID19, Economic growth, Reproduction rates Authors E-Mail Address: bbakker; HYPERLINK mailto:cgoncalves cgoncalves1 Thi

7、s paper is an extended and updated version of Chapter 2 of the IMFs October 2020 Regional Economic Outlook for Western Hemisphere “COVID-19 in Latin America and the Caribbean: A High Toll on Lives and Livelihoods (Bakker et al., 2020). The paper has benefited from comments from Alejandro Werner, Fra

8、ncesca Caselli, Joshua Felman, Anna Ivanova, Leslie Lipschitz, Pedro Rodriguez, Frederik Toscani and the Argentinean authorities. All remaining errors are the sole responsibility of the authors.Contents HYPERLINK l _bookmark0 Introduction and Conclusions 4 HYPERLINK l _bookmark4 Slow Burns, Forest F

9、ires and Put-outs the Dynamics of Covid-19 Epidemics 8 HYPERLINK l _bookmark5 SEIR Model 8 HYPERLINK l _bookmark12 Reduced Form Equation 13 HYPERLINK l _bookmark25 The Dynamics of Covid-19 Epidemics: Some Empirics 17 HYPERLINK l _bookmark26 Higher Stringency, Lower Mobility and Higher Temperatures a

10、re Associated HYPERLINK l _bookmark26 with a Lower Growth Rate of New Deaths 17 HYPERLINK l _bookmark30 Later Lockdowns Lead to Higher Deaths 20 HYPERLINK l _bookmark33 Randomness 22 HYPERLINK l _bookmark34 Geographical Spread 22 HYPERLINK l _bookmark35 Why did Lockdowns in Latin America not Stop th

11、e Pandemic? 25 HYPERLINK l _bookmark36 Early Lockdowns Require a Sharper Reduction of R0t 25 HYPERLINK l _bookmark37 Mobility Rebounded as Cases Increased 27 HYPERLINK l _bookmark38 Weak Institutional Capacity may Have Hampered the E ectiveness of Lock- HYPERLINK l _bookmark38 downs 30 HYPERLINK l _

12、bookmark42 Second Waves 34 HYPERLINK l _bookmark43 What Factors made Latin America Vulnerable? 36 HYPERLINK l _bookmark47 The Impact of Lockdowns and Fear on Economic Activity and Mobility 41 HYPERLINK l _bookmark48 The Impact of Covid-19 on Economic Growth 41 HYPERLINK l _bookmark54 What Explains t

13、he Sharp Drop in Mobility in Latin America Policies or HYPERLINK l _bookmark54 Fear? 46 HYPERLINK l _bookmark57 The Future of the Pandemic in Latin America 50 HYPERLINK l _bookmark62 Bibliography 53 HYPERLINK l _bookmark77 Data Issues 55 HYPERLINK l _bookmark78 Covid in Individual Countries 59 HYPER

14、LINK l _bookmark79 Data Sources 60Introduction and ConclusionsLatin America has been hit hard by the Covid-19 pandemic. As of early June 2021, the death toll in the region is similar to that in Western Europe and the United States (Figure 1.1) despite earlier and longer lockdowns, and a much younger

15、 population. HYPERLINK l _bookmark1 1 Further, it is likely that the o cial death count is far below the real one, since testing has been low and excess-over-normal deaths in some countries have been far higher than Covid-related deaths (Annex A). HYPERLINK l _bookmark2 2At the same time, the patter

16、n of the pandemic has been strikingly di erent. Western Europe saw sharp explosions in the Spring (Q2) of 2020, very modest infections during the Summer of 2020, and renewed explosions in late 2020 and early 2021. Latin America mostly avoided these explosions (which we here dub as forest- res ) but

17、also did not see periods of very modest infections (Figure 1.1 and 1.2). HYPERLINK l _bookmark3 3 Instead, its daily death toll gradually increased ( slow-burn ) and, in many countries, only peaked in the fall of 2020 or the rst half of 2021.Here we shed light on some important questions, such as wh

18、y did the lockdowns not reduce the pandemic in the region as they did in Europe at the time? In addition, to what extent were formal i.e., regulatory or legal measures correlated with e ective reductions in contact and mobility? What was the impact of the epidemic on the economy of Latin America? To

19、 what extent was the impact the result of stringency (i.e., government policies) and what role did fear (as proxied by daily deaths) play?The paper addresses these questions using a reduced form of a SEIR (susceptibility, ex- posed, infectious, recovered) model. An epidemic explodes when the e ectiv

20、e reproduction number (R) is greater than 1; it is contained when R is below 1. The latter can best be achieved with e cacious vaccines, but until these were generally available there was great uncertainty and debate on how to proceed. For most countries the stated strategy was to bridge the time un

21、til vaccines became available through non-pharmaceutical measures i.e., lockdowns and mask mandates. Limiting infections would prolong the period before herd immunity could be achieved, but it would also reduce deaths and prevent medical facilities being overwhelmed in the pre-vaccine period.The dat

22、a show that lockdowns and other non-pharmaceutical interventions helped. Higher stringency reduced mobility and slowed the spread of the pandemic. But, not surprisingly, the e cacy of early lockdowns was greater to the extent that the population susceptibility had already fallen. Moreover, we nd tha

23、t measures of government e ec- tiveness exerted an independent, signi cant e ect: lockdowns delivered better results in countries with higher measures of government e ectiveness. Finally, the impact of higher stringency on mobility and thus on infections declined over time.1 With the exception of Ur

24、uguay.2 In early June 2021, just before this paper was published, Peru almost tripled its o cial death toll after a scienti c review of medical records ordered by the government.3 For data on individual countries, see Annex B.But stringent measures also had costs. Our regression analysis suggests th

25、at both stringency and daily deaths a ected economic activity, but the impact of the former was quantitatively more important. The sharp downturn in Latin America in the second quarter of 2020 was chie y associated with tough and binding lockdowns; and the recovery in the second half with both an ea

26、sing of formal measures and a reduced impact of stringency on mobility and activity.Explanations based on only a few major policy-sensitive variables cannot tell the whole story, of course. Other variables demographic, economic, and sociological characteris- tics that are not amenable to policy in u

27、ence in the relevant time intervals are important in explaining substantial di erences between countries.? Countries with older populations and/or higher levels of obesity fared signi cantly worse.? The pandemic has been more di cult to contain in countries and municipalities with low income levels

28、not surprisingly, the imperative of earning living incomes makes it more di cult for lower-income workers to reduce their mobility.? Areas with high population density whether large cities like New York, or areas of signi cant poverty like the favelas of Rio are disadvantaged.? The medical services

29、infrastructure is important: countries with more hospital beds saw a lower death toll.? Temperature and the caprice of seasonality are important. Temperature has a nega- tive in uence on the spread and morbidity of the pandemic, hence di erences when the northern hemisphere is moving into Spring and

30、 the southern hemisphere into Fall.? BCG vaccines for infants (against tuberculosis) are common in some countries but not in those with a longer history of much lower incidences of tuberculosis. These vaccines appear to reduce somewhat both rates of Covid-19 infection and morbidity.? Finally, new va

31、riants that are more contagious and/or more deadly are another seem- ingly capricious di erence between countries (although, notably, they have been more characteristic of southern hemisphere countries).The lessons from this close analysis of the data can be summarized as follows:? There may be a ne

32、 line in when to lock down. Locking down too late will lead to an explosion of deaths. But locking down very early may not be sustainable and ultimately may not succeed in stopping the pandemic (with the notable exception of small islands).? Lowering stringency and increasing mobility will help the

33、economy, but if done too rapidly can lead to second waves.? The safe level of stringency and mobility depends on the share of the still susceptible population. The higher the total death toll (or the higher the share of the population that has been vaccinated), the more stringency can be reduced. Co

34、untries most at risk of an explosion in new deaths are not the ones where the total death toll is already very high, but those where the total death toll is still low and few have been vaccinated.? The safe level of stringency and mobility also depends on the season. A high level of mobility during

35、the summer months may be consistent with a low level of infections. But keeping the same mobility level when winter approaches may lead to a sharp increase in infections.? Latin America avoided the second wave explosions in late 2020 and early 2021 that Europe and the United States saw because mobil

36、ity had not picked up as much as in Europe, the share of the still susceptible population was lower and because it was summer in the Southern Hemisphere. But with winter approaching, and more virulent new variants, we are seeing new waves in many countries. Much more rapid vaccination will be key to

37、 stop the pandemic.The broad conclusions of this section are substantiated by a careful analysis of the available data in the sections that follow.Slow Burns, Forest Fires and Put-outs the Dynamics of Covid-19 Epi- demicsSEIR ModelAt each moment of time, the population (N ) is divided into ve mutual

38、ly exclusive cate- gories. HYPERLINK l _bookmark6 4 These are susceptible (no immunity) S, exposed (infected but not yet infectious) E, infectious I, recovered R, and dead D. We use the following SEIR model: HYPERLINK l _bookmark7 5dSSdt = Itt N(1)dESdt = ttI N E(2)dIdt = E I(3)dD= I(4)dtdRdt = (1 )

39、I(5)We start with a brief description of the workings of a SEIR model. Every day, an infectious person bumps into t persons. The probability that he will infect a personduring a contact is S t, where Sthe likelihood that the person will be susceptible and tNNis the probability that a susceptible per

40、son will be infected. is the rate at which people5.2that have been exposed to the virus become infectious. Following HYPERLINK l _bookmark75 Wang et al. ( HYPERLINK l _bookmark75 2020) and HYPERLINK l _bookmark63 Atkeson ( HYPERLINK l _bookmark63 2020) it is set to = 1 re ecting an estimated incubat

41、ion period of the disease of 5.2 days. The parameter is the rate at which infectious people either recover or die.18Following HYPERLINK l _bookmark75 Wang et al. ( HYPERLINK l _bookmark75 2020) and HYPERLINK l _bookmark63 Atkeson ( HYPERLINK l _bookmark63 2020) we set = 1re ecting an estimateddurati

42、on of illness of 18 days. The share of those that die is ; the share that recovers1 . We assume = 0.01.4 This section is based on HYPERLINK l _bookmark63 Atkeson ( HYPERLINK l _bookmark63 2020).5 The model is similar to HYPERLINK l _bookmark63 Atkeson ( HYPERLINK l _bookmark63 2020), with the di ere

43、nce that we distinguish between recovered and dead persons.DynamicsAs an infected person is infectious for 1 days, he will infect tt Spersons while infectious. NThis ratio is also known as Rt, the e ective reproduction number:R = tt S(6)t NIf Rt is greater than 1, each infectious person will infect

44、more than one other persons, which means that the number of currently infected people will continue to increase. If it is smaller than 1, the number of currently infected persons will decline, and the epidemic will die out. De ning R0t as the expected number of secondary cases produced by a single (

45、typical) infection in a completely susceptible population:we can further rewrite this asR0t= ttS(7)Rt = R0t N(8)The e ective reproduction rate Rt depends both on R0t and the share of the still susceptible population.It should be noted that R0t is not necessarily constant, as it depends on the number

46、 of contacts t and the transmission probability t. If the number of contacts an infectious person has falls (for example, because of a drop in mobility), or if the transmission prob- ability declines (for example, because people start to wear masks and wash their hands frequently), R0t will drop.The

47、 Epidemic in the Absence of Sanitary measures and Behavior ChangesRIn the absence of any sanitary measures and behavior changes (i.e., with unchanged R0t), the number of currently infected persons will continue to explode until the share of the susceptible population has dropped below 1 . At that st

48、age, each infected person will0tinfect less than one other person, and the disease will start to die out.S 1NR0t Rt 1(9)NThe lower R0t, the higher the level of Sat which Rt falls below 1. For example, if R0t = 3,Rt will fall below 1 if the share of the still susceptible population is less than one t

49、hird (i.e., two thirds of the population has been infected), while if R0t is 1.25, Rt falls below one if the share of the still suceptible population is less than 80 percent (i.e., 20 percent of the population has been infected).The Impact of Lockdowns: TheoryLockdowns and other sanitary measures re

50、duce R0t by reducing the number of contacts and the transmission probability. We will distinguish between fully e ective lockdowns and partially e ective lockdowns, and between early and late lockdowns:? A fully e ective lockdown is a lockdown which reduces R0t to below 1 and manages to keep it ther

51、e. A partially e ective lockdown is a lockdown which reduces R0t but to a level above 1? An early lockdown is a lockdown that occurs when few people have been infected (i.e, the share of the still suceptible population is high), while a late lockdown is a lockdown that occurs when many people have a

52、lready been infected.The dynamics of an epidemic after a lockdown will depend on both the timing and e ec- tiveness of the lockdown (Figure 2.1):? An early and e ective lockdown will resemble a put-out . The disease soon disap- pears.? An early and partially e ective lockdown will resemble a slow-bu

53、rn. The number of infected persons will continue to rise after the lockdown (albeit at a slower rate) until the share of the still susceptible population has fallen enough. For example, with an early lockdown that reduces R0t to 1.25, the number of currently infected people will continue to rise unt

54、il the the share of the still susceptible population has fallen to 80 percent.? A late and e ective lockdown will resemble a forest- re similar to the no-intervention scenario, but peaking at a lower level.? A late and partially e ective lockdown will also resemble a forest- re , but peak at a highe

55、r level than in the late and e ective scenario.It should be noted that while an early, partially e ective lockdown will not stop a pandemic, a late, partially e ective lockdown may. To see this note that whether a lockdown reduces Rt to below 1 depends not only on by how much R0t falls, but also on

56、the share of the still susceptible population. For example, reducing R0t to 1.25 will not stop an epidemic when 10 percent of the population has been infected, but will do so when 30 percent has been infected.It should be noted that in an SEIR model there is an inverse U relationship between the tot

57、al number of deaths and new deaths (Figure 2.2). The number of daily deaths increases until total deaths has reached a certain threshold; thereafter it declines. The growth rate of new deaths declines steadily as the total number of deaths rises (Figure 2.3).The patterns in Figure 2.1 and 2.2 can be

58、 observed in practice. New York City in the Spring of 2020 (which locked down very late) had a severe forest re and looked almost likethe no-intervention pattern (Figure 2.4). Spain in the Spring of 2020 also had a forest re (Figure 2.5). Argentina, by contrast, had a slowburn daily deaths only peak

59、ed in October (Figure 2.6). HYPERLINK l _bookmark11 66 Note the di erent y-axes in Figures 2.4, 2.5 and 2.6.Reduced Form EquationThe key variable in a pandemic is the e ective reproduction rate Rt. If it is above 1, the epidemic will continue to explode, and when it is below 1, the epidemic will sta

60、rt to die out. Recall from equation ( HYPERLINK l _bookmark9 8) (which we repeat here for convenience) that the e ective reproduction rate Rt depends both on R0t (the expected number of secondary cases produced by a single (typical) infection in a completely susceptible population) and the share of