00:00 OK,
00:01 great.
00:01 Um,
00:02 in that case,
00:03 uh,
00:03 let me,
00:04 um,
00:04 welcome everyone to the first in this series
00:07 of e-seminars on topics related to COVID-19,
00:11 uh,
00:12 that we have organized here in DC.
00:14 Um,
00:14 this is the first one in a series of 6 or 7
00:17 that we have planned and we'll be adding more as we,
00:20 uh,
00:20 as we go over time.
00:22 Um,
00:22 but without advertising too much about the,
00:25 uh,
00:25 the,
00:26 the,
00:26 the future,
00:27 let me just say a couple of,
00:28 um,
00:29 other introductory words.
00:30 The one is,
00:31 um,
00:31 I just want to thank everybody for your patience with the new technology.
00:35 This,
00:35 uh,
00:35 we don't have a lot of experience running seminars in this format.
00:39 Um,
00:41 most of you will be watching on the YouTube live stream.
00:44 Um,
00:45 if you would like to send questions during the Q&A,
00:47 you can do so using the chat function there,
00:50 which is being monitored,
00:51 and the questions will come to me,
00:52 and I'll try to,
00:53 uh,
00:53 convey them
00:54 as best possible.
00:56 Apologies also to those of you who are on YouTube.
01:00 For having,
01:01 uh,
01:02 um,
01:02 for having uh restricted Webex access.
01:05 Unfortunately,
01:05 this is something we had to do given the limitations of Webex.
01:09 In subsequent events,
01:10 we'll be trying to,
01:11 uh,
01:12 see if we can get a few more people hooked up through Webex,
01:14 so it's a little bit more interactive.
01:17 Um,
01:17 I also just wanted to note for the benefit of both
01:20 the presenters and anybody who intervenes in the Q&A that,
01:23 uh,
01:23 this session is being recorded for subsequent
01:26 posting in order to accommodate our colleagues in
01:28 time zones that aren't very friendly to this time slot.
01:32 Um,
01:33 so then without further ado,
01:35 let me just quickly introduce the,
01:37 uh,
01:37 the,
01:38 the,
01:39 the 3 papers that we have.
01:40 Uh,
01:40 we have a series of 3 presentations and we'll just
01:43 go through them,
01:44 uh,
01:44 in sequence with about 15 minutes per presentation for each paper.
01:49 Uh,
01:49 the first paper is,
01:51 uh,
01:51 a Simple Planning Problem of COVID-19 lockdown,
01:54 which is co-authored by Fernando Alvarez,
01:56 David Argente,
01:57 and Francesco Lippi,
01:59 and I believe Francesco will be presenting.
02:01 Our second paper is the Macroeconomics of Epidemics,
02:05 um,
02:05 co-authored by Marty Eichenbaum,
02:07 Sergio Rebelo,
02:08 and Matthias Trabant,
02:09 and I believe that Marty will be presenting this one.
02:12 And then the third paper is The Economic Ripple Effects of COVID-19,
02:17 uh,
02:17 which is co-authored by Paco Buira,
02:19 Andres Nomeyer,
02:20 Roberto,
02:21 uh,
02:22 our own Roberto Fatal,
02:24 and,
02:24 uh,
02:24 Yung-sukhin.
02:26 So,
02:26 uh,
02:27 and then
02:27 we'll follow the three presentations with a short discussion from,
02:32 um,
02:32 our other World Bank colleague,
02:34 uh,
02:34 Ana Paolo Casalito,
02:36 uh,
02:36 who's a senior economist in EFI.
02:39 Uh,
02:39 so without further ado,
02:40 let me turn over to,
02:41 um,
02:42 uh,
02:42 to I guess,
02:43 Francesco to begin.
02:44 Thank you very much,
02:44 Francesco.
02:54 Marty has to stop sharing his screen so that I can share mine.
03:02 Hello.
03:05 Martin,
03:05 you on your screen,
03:07 Martin.
03:13 Martin,
03:13 can you hear us?
03:16 Could you stop sharing your screen?
03:22 I don't think he's hearing it.
03:24 OK,
03:24 um,
03:25 OK,
03:25 Francesca,
03:26 could you try just to share the screen and see what happens?
03:31 It doesn't let me because it says that Marty has to stop sharing his screen.
03:41 Uh,
03:42 you see my screen now?
03:44 Now,
03:44 now we can see it.
03:45 Yes,
03:46 perfect,
03:46 great.
03:48 Great.
03:49 So shall I start?
03:52 Yes,
03:53 please go ahead.
03:55 OK.
03:56 So this is joint work with Fernando Alvarez and David Argente,
04:00 and it's a,
04:02 it's a simple uh first attempt to analyze
04:06 an optimal planning problem.
04:08 So
04:09 these days it's hard for us as anybody to think about anything else than the COVID,
04:15 and one of the measures that's been implemented in many
04:17 places around the world is this idea of the lockdown,
04:20 say,
04:21 say to the people to stay home,
04:23 and what we want to do here
04:25 is to
04:26 balance this idea,
04:28 you know,
04:28 Of locking down people in their homes,
04:31 which is obviously good from an epidemiology point of view as a first response
04:35 to the diffusion of the virus
04:37 with the costs
04:39 that such a policy implies for an economy,
04:42 the costs of people not being able to produce,
04:44 so ultimately
04:46 the loss of GDP and income and eventually.
04:48 You know,
04:49 those things are important,
04:50 so we're trying
04:51 to essentially use
04:53 an off the shelf
04:55 epidemiology model
04:56 and lay on top of it an economist's view
04:59 of how you would optimally balance that problem,
05:02 which is a serious and tragic one,
05:04 with another equally dire problem,
05:06 the one of not producing.
05:09 So the framework,
05:10 it's a pretty standard
05:13 SIR model.
05:15 So these are models that are used in epidemiology.
05:18 There are 3 types in the population.
05:20 The S
05:21 are the citizens who are susceptible of being infected.
05:25 The I is a fraction of citizens who are infected,
05:28 and the R are the citizens who are recovered.
05:32 And in this model,
05:32 it's assumed that once you recover,
05:34 you're not going to get the,
05:36 the disease again.
05:37 In the recovery there's also a fraction of dead people,
05:41 and the idea,
05:42 as I said,
05:42 is the planner will trade off,
05:44 will try to choose the lockdown
05:47 policy,
05:47 trading off 22 objects.
05:51 One is the discounted value of economic activity,
05:54 the present value.
05:55 So obviously
05:57 having an economy that produces value to the planner
06:01 and on the other hand,
06:02 trading off the pecuniary value of life.
06:05 The planner doesn't want to have too many dead people,
06:08 and you know,
06:09 obviously there's no need to explain
06:11 that.
06:13 An important ingredient in our framework.
06:15 is
06:17 the congestion
06:19 of the healthcare system,
06:20 so these models will have
06:22 something that is called the fatality rate.
06:24 This is the rate at which
06:26 infected people die,
06:30 and we will assume that this rate depends on Level of
06:33 the infection itself so that when you have more infected people,
06:37 the fatality rate increases.
06:39 The idea behind it is that,
06:41 you know,
06:41 as you get congested,
06:43 intensive care units,
06:44 it's harder to cure everybody so that more people
06:48 will die.
06:50 As I mentioned,
06:51 we assume that those recovered can be identified,
06:55 so they will not be in lockdown and will not be subject to being infected again.
07:00 Instead we'll be assuming that infected and
07:03 susceptible people will be in lockdown.
07:05 The planner is not able to distinguish between these two types in the population.
07:10 So when a lockdown is announced,
07:12 it affects everybody.
07:14 There's a maximum lockdown that we assume it can be
07:17 done because the economy has to work at a minimum level
07:20 to provide basic services,
07:22 and moreover,
07:23 lockdown is.
07:23 Going to be imperfect.
07:24 Not everybody in lockdown is having zero contacts.
07:26 There will be
07:27 some contacts.
07:28 Finally,
07:29 we'll be assuming that
07:31 as the economy is in lockdown or as time goes by,
07:34 really,
07:35 a cure may arrive with a stochastic probability
07:39 that has an expected duration of 1.5 years.
07:42 So once the cure arrives,
07:44 the problem will not be with us anymore.
07:47 So formally
07:48 we're using this off the shelf
07:51 SIR model where the total population N
07:54 is divided into three groups the infected,
07:57 I,
07:57 the susceptible S,
07:59 and the recovered.
08:00 And really if you go to this line here,
08:04 the S dot,
08:05 this is the change in time,
08:06 the fraction of susceptible people.
08:09 How come you move from being susceptible to being infected?
08:12 Well,
08:12 a susceptible person meets an infected one,
08:16 so the product.
08:17 Of these two variables is what epidemiologists use to
08:21 capture the idea that,
08:22 you know,
08:22 you need to have some infected to get
08:24 other people infected,
08:25 and the more
08:27 you have,
08:28 the more the bigger the change in the number of
08:30 susceptible people who move from being susceptible to being infected,
08:34 and beta
08:36 is the rate at which this happens.
08:37 Beta is the contact rate of a susceptible
08:40 of an infected person which is susceptible.
08:42 It's basically the number of people that each infected person can infect.
08:46 And,
08:47 and likewise,
08:48 there is a dynamic law of motion for the fraction of infected people who are,
08:52 uh,
08:53 you know,
08:53 fed
08:54 by those who move from being susceptible to being infected the first term
08:59 and exit from the state of being infected at a rate gamma,
09:03 which is measured the average duration of the disease,
09:07 and,
09:07 and you exit the infection state by either.
09:11 Recovering in good health or dying,
09:14 and I'll say more about that later on.
09:16 Now we modify this little model where you have the Sxi product by injecting a control
09:23 for the planner.
09:24 It's this term over here I'm trying to highlight it.
09:27 I hope you see it.
09:28 It is,
09:29 so the fraction of the populations susceptible
09:33 and the fraction of the population I.
09:35 It can be subject to a lockdown.
09:37 What is a lockdown?
09:38 Well,
09:38 there is not,
09:40 not all of the people are around,
09:42 but just a fraction of 1 minus 1.
09:45 So the planner commands the citizens to stay home,
09:48 a fraction of the citizens to stay home,
09:51 so that they,
09:52 you know,
09:52 the contact rate naturally will be smaller because there are less people around.
09:57 Now if TIA was one,
09:58 this lockdown is fully.
10:00 Infected.
10:00 So if Tita is 1 and L is 1,
10:02 these contacts fall to zero and you have a perfect control of the infection,
10:07 that is unrealistic,
10:09 first,
10:09 because,
10:10 as I mentioned,
10:10 you cannot shut down the whole economy.
10:12 Second,
10:13 because even if you tell people to stay home,
10:15 they're still seeing the other people in their building and
10:17 they have to go out to buy some groceries.
10:20 So TIA is our measure of how effective the lockdown is.
10:25 Uh,
10:26 so,
10:26 you know,
10:26 once you,
10:27 you introduce this control,
10:28 essentially the problem for the planner becomes
10:31 that,
10:32 the one of controlling LT.
10:34 LT is uh what kind of lockdown to have at each,
10:37 at each period T,
10:38 um,
10:39 and also let me mention the fatality.
10:41 Rate here at the bottom because that's another objective
10:45 that the policy that the planner wants to control.
10:48 So the N dot is the number of fatalities,
10:50 how many people die per unit of time,
10:53 and it's given by,
10:54 you know,
10:54 what fraction of the infected IT die.
10:57 And we'll assume this function.
10:59 So as I mentioned,
11:00 this fatality rate is increasingly high,
11:02 so it is a constant fatality rate of gammaF
11:05 that you can think of as a fatality under normal conditions.
11:09 Plus when you have a lot of infected people,
11:12 this fatality rate may increase.
11:15 So what is the problem that the policymaker is trying to,
11:18 the planner is trying to solve?
11:20 He's trying to minimize the present discounted value,
11:23 this object here on the top
11:25 of the welfare losses coming from the two terms,
11:29 the blue term and the red term.
11:31 The blue term are the losses
11:33 coming from
11:34 foregone GDP.
11:36 If you put people in lockdown,
11:38 so think of this as a number between 0 and a maximum,
11:42 say 0.7.
11:42 25,
11:43 then you're not enjoying the 0.75% of output
11:48 and the agents that will not be producing,
11:51 so they will be out of the production process are both the S and the I types.
11:55 So the blue term is the foregone is the cost of GDP losses,
11:59 and the red term is the cause of the deaths,
12:02 as I mentioned.
12:03 How many people die is this first term,
12:05 the product of the fatality rate times infected,
12:09 and the cost of Each death is given by this term in the square bracket,
12:13 that is the present value of lifetime earnings w over r
12:17 plus,
12:18 you know,
12:18 the sky term which allows us to eventually
12:21 consider additional pecuniary motives of dead people.
12:26 OK,
12:26 so the optimal control problem is the control of this value functions and I
12:32 at each time t from t going running into the infinite future.
12:38 OK,
12:39 so,
12:39 uh,
12:40 and there's going to be an initial condition for this problem,
12:43 how many infected people we have at the moment that will start
12:48 taking this control problem into account.
12:49 So there's going to be a fraction epsilon of people infected,
12:53 and we assume a fraction 1 minus epsilon of persons susceptible.
12:58 We need to parameterize this model to,
13:01 to look at some,
13:02 uh,
13:02 uh,
13:03 to look at a solution,
13:04 and we parameterize the model,
13:06 you know,
13:06 it's hard to,
13:07 to know exactly what these parameters are.
13:11 They are serious measurement issues,
13:12 but we try to take,
13:15 you know,
13:15 see what the consensus
13:17 values are out there.
13:19 Let me
13:20 discuss very briefly three key parameters for us in this simulation.
13:23 Those are the ones highlighted in red.
13:26 The gamma F is the Francesco,
13:29 as you do so,
13:29 I'll just give you your 5 minute warning,
13:31 please.
13:32 Thank you.
13:32 OK,
13:33 OK,
13:34 so the gamma F is the fatality rate out of the population of the infected.
13:38 This kappa is this additional
13:40 extra effect coming from the congestion,
13:43 and chi is the extra cost of death which in this benchmark case we set to zero.
13:49 So
13:50 you know,
13:50 the problem has a solution.
13:51 It has a value function that it lives in a two dimensional space.
13:55 The state of the problem is the point in this r square plane.
13:58 How many infect,
13:59 how many is acceptable.
14:00 The values for this problem,
14:02 you can read them on the
14:03 on the vertical axis in terms of
14:06 per capita
14:07 GDP loss.
14:08 So this is a 3% GDP loss.
14:10 The policy function.
14:11 What we're interested in is here,
14:14 you know,
14:14 this is a phase diagram,
14:15 and it's a heat map.
14:16 When it's yellow,
14:17 you want to exercise a lot of control.
14:19 When it's blue,
14:19 you don't exercise control.
14:21 So endowed with these instruments,
14:23 what you can do,
14:24 you know,
14:24 this is a phase diagram.
14:25 If the economy starts here
14:27 and you don't exercise any control,
14:30 the economy will follow the black path.
14:32 This is how the state of the economy.
14:34 Move until eventually
14:35 there are no more infected here.
14:37 There's going to be a lot of dead people,
14:39 as you're going to see if
14:41 control is exercised,
14:42 that's going to be the right path
14:44 as you start with no control,
14:46 but as the state moves into the yellow region,
14:48 the planner will exercise control and then we get out.
14:51 So to give you a concrete sense of what we get
14:54 in a benchmark scenario.
14:56 The top panel shows you the lockdown policy.
14:59 You start with,
15:00 you know,
15:00 gradually climbing up to a serious lockdown of 80%
15:04 for about
15:06 15 weeks and then you unlock it.
15:09 Let me just briefly mention that this
15:10 parameterization is kind of a pessimistic scenario.
15:13 We have a lot of
15:15 death
15:15 cases,
15:16 but notice here in terms of the number of deaths
15:19 that the blue line without control,
15:22 you would have you would end up having many
15:24 more dead people than under the control scenario.
15:27 We analyzed.
15:28 more benign scenario here where the convexity of the
15:32 fatality rate is less
15:35 extreme,
15:36 which gives us a somewhat shorter period of lockdown.
15:39 Population in lockdown is about 20% of the
15:42 whole population after 15 weeks of lockdown,
15:45 and the death rate,
15:46 you know,
15:47 eventually is around
15:49 1% of the population.
15:51 We
15:52 compute several comparative statistics with respect to the
15:56 key parameters of interest.
15:58 These tables at the end of the paper are also very useful to kind of
16:02 get a handle on how costly is this lockdown in terms of GDP per capita,
16:07 say,
16:08 so our benchmark evaluation here is that if you were to do no lockdown,
16:14 the consequences of these debts
16:16 would amount to a permanent loss.
16:18 of something like 3.8% of GDP per capita forever.
16:22 That's a huge number,
16:24 which is reduced to a big number,
16:26 2%,
16:27 but much smaller than 3.8%
16:29 once you exercise control.
16:31 I would say that,
16:32 you know,
16:32 if I have no more time,
16:34 this is my bottom line.
16:35 Perhaps I can put up my concluding slides.
16:38 What we do,
16:39 we present a simple framework
16:41 to analyze this trade-off of the tradeoffs involved with lockdown.
16:46 We highlight what are the main forces between
16:50 that,
16:51 you know,
16:51 motivate
16:52 why and for how long and for how much you have to use the lockdown,
16:57 and of course there are several open questions from Richard parameterization
17:03 to,
17:03 you know,
17:04 some details of the specifications of the model,
17:06 and that's what we're working on right now.
17:16 Hello.
17:17 Excellent.
17:18 Thanks a lot,
17:19 Francesco.
17:19 Shall we just go straight to uh Marty if you can get your slides up,
17:22 please.
17:45 And Marty,
17:45 just a reminder,
17:46 you have to unmute too.
17:50 Damn it.
17:52 You're unmuted now,
17:53 Martin.
17:54 Right,
17:54 I'm trying to figure out how to
17:56 unmute that.
17:57 Sorry,
17:57 I will do that.
17:59 Um,
18:00 sorry about that.
18:03 Oh,
18:04 you're good now.
18:04 We hear you.
18:05 Oh,
18:05 you do.
18:05 OK,
18:06 great.
18:06 OK,
18:06 wonderful.
18:07 OK,
18:08 so here we go to
18:10 full screen mode.
18:11 Thank you very much,
18:12 everyone.
18:13 Um,
18:14 I'm getting used to this new world.
18:16 Um,
18:16 with the,
18:17 I'm talking about some joint work,
18:18 uh,
18:19 with,
18:19 uh,
18:19 Sergio Rebelo
18:21 and,
18:21 uh,
18:21 Mathias Trabant.
18:23 Um,
18:25 brief introduction.
18:26 Um,
18:27 you know,
18:27 as COVID-19 is spreading throughout the world,
18:30 uh,
18:31 governments are struggling with understanding,
18:33 um,
18:34 how and
18:35 precisely how long to manage the epidemic.
18:39 As we all know,
18:40 epidemiology models are very useful and widely used
18:43 to predict the course of the epidemic,
18:46 but
18:47 one important shortcoming
18:49 of those models
18:50 is they really don't allow for the interaction
18:54 between economic decisions
18:56 and rates of infection.
18:58 Of course,
18:59 we think that's very important in the present context.
19:03 Um,
19:04 at the big picture,
19:05 uh,
19:05 we know epidemics have aggregate demand and aggregate supply effects.
19:09 What are those supply effects?
19:11 Well,
19:11 the epidemic exposes people who are working to the virus.
19:15 And people naturally react by reducing their labor supply.
19:19 The demand effect is that
19:22 these epidemics expose people who are purchasing consumption goods to the virus,
19:26 and so they react by reducing their consumption.
19:29 So together those supply and demand effects work together to generate
19:33 a large persistent recession.
19:37 There's a,
19:38 there's a critical externality,
19:39 of course,
19:40 is that people infected with the virus don't really
19:43 internalize the effect of their own consumption and work decisions
19:47 on the spread of the virus.
19:49 So a really important question for economists to ask is,
19:52 you know,
19:52 what policy should a government
19:54 pursue to deal with that infection externality.
19:58 And what Sergio Matias and I have done so far
20:01 is study containment policies that reduce market activity.
20:06 And as a natural consequence,
20:08 they exacerbate the recession,
20:10 but in fact,
20:11 they raise welfare overall,
20:14 by substantially reducing the death toll
20:17 caused by the by by the epidemic.
20:19 So that is the unfortunate trade-off
20:21 that we are faced with.
20:24 Our point of departure is the classic SER model proposed by Kermack and McKendrick.
20:30 Uh,
20:30 and in their model,
20:31 as you recall,
20:32 we all know,
20:33 there's this exogenous transition between health states.
20:37 We imagine that there's a continuum
20:39 of people with measure 1.
20:42 And that very simple model prior to the epidemic,
20:45 all the agents are identical,
20:47 and they maximize utility function in which they care about discounted flow
20:52 from the log of consumption,
20:54 and they also are not happy working,
20:56 so there's a quadratic term in,
20:58 in,
20:58 in,
20:58 in work.
21:01 The household budget constraint,
21:02 of course,
21:02 is very simple.
21:04 Consumption on the left hand side.
21:07 We've allowed for this crucial object called UCT,
21:11 which we're going to formally as a tax rate and consumption,
21:14 but really it's just a proxy
21:16 for all the containment containment measures
21:19 that governments can undertake to reduce social interactions.
21:23 So I'm going to call that mCT containment rate.
21:27 On the right hand side we just have the wage earnings of individuals
21:31 and cap gamma
21:33 which are just lump sum transfers,
21:35 so the government's going to make some revenues from the containment taxes
21:38 and then just rebate them.
21:41 There's a continuum of competitive representative firms,
21:43 very simple unit measure,
21:45 and consumption is produced using a linear technology
21:49 a times nt,
21:51 and there's a government budget constraint
21:52 which just says their containment revenues,
21:55 tax revenues
21:56 are just rebated and equal to the lump sum transfers.
22:01 Now,
22:01 the population dynamics,
22:03 and,
22:03 and this is important,
22:04 cap T on the left-hand side of the newly infected people.
22:08 And so what is this first term on the right hand side?
22:12 Well,
22:12 the idea is that we have the number
22:14 of people that are infected buying consumption goods.
22:18 Right,
22:18 so ITCIT are the total number of infected
22:21 folks that are out there buying consumption goods.
22:24 ST are the number of um
22:27 Uh,
22:27 of a susceptible people who are out buying consumption goods.
22:31 So that's what the superscripts is.
22:33 The parameter pi,
22:34 and we'll come back to this in the parameterization,
22:37 reflects both the amount of time that people spend shopping.
22:40 And the probability of becoming infected as a result of that activity.
22:44 So that's the first uh term.
22:46 The second term represents the number of people infected while working.
22:52 And so ITNT are going to be the total number of infected folks that are working,
22:56 that's this term over here.
22:58 And then the term right to the left of
23:00 it are all these susceptible folks that are working,
23:02 and so PS2 represents the probability
23:05 of becoming infected as a result of work interactions.
23:08 I should point out that we're working with
23:11 or collaborating with a team of epidemiologists in Pittsburgh
23:16 who have very,
23:16 very detailed data
23:18 on breakdown of,
23:19 of the kind of people work due and,
23:21 and,
23:21 and,
23:21 and infection rates
23:24 that there's no time for that today.
23:26 Finally,
23:26 the last term just represents other ways of getting infected,
23:30 which is exactly what the epidemiologists emphasized this exogenous object STIT
23:36 and PS3 is,
23:37 is uh the the probability associated with
23:40 things like interacting with a neighbor,
23:42 etc.
23:44 OK,
23:44 so if you think about the total number of susceptible people at time T,
23:47 it's the number that there were in the beginning of the period,
23:50 or the end of last period,
23:51 minus new infected folks.
23:54 The number of infected people at time T + 1 is whatever it was initially,
23:59 plus new infections,
24:01 and unfortunately
24:02 we have some of these folks that are dying.
24:04 PiD is going to be a rate at which
24:06 there's a mortality rate,
24:08 and P
24:09 PR represents people that fortunately are recovered from becoming infected.
24:15 What is the number of recovered people at Time T?
24:17 Well,
24:18 it's the number that were recovered yesterday plus
24:20 inflows from infected people who were recovered.
24:23 So that's the good news.
24:24 And very sadly,
24:26 the number of deceased people are the number that were infected yesterday,
24:29 plus new people that are entering into that state unfortunately
24:33 as a result of being infected and then passing away.
24:39 OK,
24:40 so the idea is that at time 0,
24:41 there's a fraction epsilon,
24:42 a very small fraction of folks that were infected,
24:45 and so the susceptible at that initial point are 1
24:48 minus the number of the people that were initially infected.
24:51 This is a rational expectations model,
24:53 so everyone knows the initial infection and they understand the laws of motion
24:57 governing the population health dynamics.
25:00 Significantly,
25:01 everybody,
25:02 because they're atomistic,
25:02 they take as given these aggregate variables IT CIT,
25:07 infected people consuming,
25:09 and the number of infected people that are working.
25:14 Uh,
25:14 what's the value function,
25:15 if you like,
25:16 uh,
25:16 of,
25:17 of,
25:17 of,
25:17 uh,
25:17 the lifetime utility of a susceptible person?
25:20 Well,
25:21 they start off with a period utility function
25:23 in which they care about their consumption,
25:26 how much they work.
25:27 They discount the future,
25:29 concentrate beta
25:31 with 1 with probability 1 minus tau,
25:34 and that's the probability that a susceptible person becomes infected.
25:38 Uh,
25:39 they will,
25:40 um,
25:41 uh,
25:42 uh,
25:42 they will become infected.
25:43 So we have ta T times UIT.
25:45 They're gonna be an infected person at T + 1,
25:48 with probability 1 minus T,
25:50 uh,
25:50 t,
25:50 they don't get infected,
25:51 and so they're just susceptible again.
25:53 And then they have a straightforward budget constraint.
25:56 Now,
25:56 it's important,
25:57 their perceived law of motion of becoming infected,
26:00 well,
26:00 they take
26:01 the total number of infected people times
26:04 how much of those infected folks are shopping,
26:06 they take that as given.
26:08 But they do understand that the more they shop,
26:11 the higher is their probability of getting infected.
26:15 Similarly,
26:16 they take the total number of people that are working
26:19 that are infected as given,
26:20 but they understand that they,
26:21 as individuals,
26:22 if they go to work,
26:23 they'll increase their probability of infected.
26:25 And then there's this exogenous
26:27 standard SER type model.
26:31 If you're an infected person,
26:33 well,
26:34 this is your period utility,
26:35 C I T N I T.
26:37 Next period,
26:38 um,
26:39 you may stay,
26:41 uh,
26:42 you will be infected if you don't pass away or you,
26:45 or you don't recover.
26:47 If you do recover,
26:48 you're just going to become a recovered person.
26:51 OK,
26:51 now it is important to emphasize
26:53 that this expression for you.
26:57 Uh,
26:57 embodies the common assumption
26:59 in macro and health economics that the cost of death
27:03 is simply the foregone utility of life.
27:05 So that,
27:06 that's an important assumption in the literature,
27:08 and we pursue it.
27:10 Finally,
27:11 if you're a recovered person,
27:12 well,
27:13 that's a very nice state because you're gonna get utility from your consumption
27:17 and work,
27:17 and we are assuming that you will not get infected.
27:20 Uh,
27:20 that's a little bit controversial,
27:23 uh,
27:23 but the mainstream view is that once you're recovered,
27:25 you are recovered.
27:28 Government budget constraint is quite standard,
27:31 so I'm just going to skip that.
27:32 Basically,
27:32 the government's revenues have to be equal to their lump sum taxes,
27:37 which are rebated on a per capita basis.
27:41 OK,
27:41 the model has other features which I won't
27:43 go into the math because of time constraints,
27:45 but we all know
27:46 that limits to healthcare capacity are very,
27:48 very important.
27:50 And so we want to take that into account
27:52 by assuming that the mortality rate is some constant,
27:55 but it's also increasing in the number of people
27:59 that are infected in the population.
28:02 So that's very,
28:02 very important,
28:03 that the the the constraints on the medical system.
28:06 We also assume that
28:09 effective treatment arrives with an exogenous probability delta C.
28:14 Um,
28:15 and so too do vaccines that make allow people to be recovered with some probability,
28:21 Delta
28:21 V.
28:23 We could,
28:24 we could numerically allow for alternative paths,
28:26 sort of not these constant probability objects,
28:29 and,
28:29 and that's something that is straightforward and probably something we should do.
28:34 OK.
28:35 Parameterizations,
28:36 uh,
28:36 uh,
28:37 Francesco mentioned difficulties.
28:38 We worked pretty hard
28:40 to get some reasonable numbers here.
28:42 So the first,
28:43 based on the medical evidence,
28:44 um,
28:45 we're gonna assume it takes about 18 days for
28:47 people to either recover or die from an infection.
28:51 Um,
28:52 and then we're going to assume a mortality rate of 0.5%.
28:58 Now,
28:58 that's based on the average of mortality rates by age in South Korea
29:04 that was computed using US population weights for
29:07 people that are younger than 70 years old.
29:10 So that,
29:10 that's something that
29:14 Obviously one could disagree about the mortality rates.
29:16 We use South Korea
29:17 because they have the highest testing rates,
29:19 as far as we know in the world.
29:22 The parameters A and theta,
29:25 which are basically coming the utility function,
29:28 come from
29:29 a parameterization which we've calibrated so that the pre-epidemic
29:32 steady state representative person works about 28 hours a week
29:36 and earns a weekly weekly income of 58,000 divided by 52.
29:41 A discount rate 0.961/52. That's important.
29:46 We want the present value of life here.
29:49 To be roughly what it is by the government.
29:51 Obviously this is controversial.
29:53 We didn't want to come up with independent estimates,
29:55 and so we're just tying ourselves
29:57 to what the government,
29:57 for example,
29:58 the Environmental Protection Agency,
29:59 says,
30:00 and that turns out to be a number like $9.3 million
30:03 2019 dollars.
30:08 OK.
30:08 Uh,
30:08 the transmission function,
30:09 which of course is very,
30:10 very important,
30:11 um,
30:12 so if you look at the epidemiology literature,
30:15 um,
30:16 they,
30:18 this is one particular study well known,
30:20 assumed that 30% of,
30:22 um,
30:23 these,
30:23 uh,
30:23 uh,
30:23 uh,
30:24 um,
30:25 analogous diseases
30:26 occur in the household,
30:28 um,
30:29 38 or 38% occur in the general community,
30:33 and 37% in schools and workplaces.
30:37 So excuse me,
30:38 Marty,
30:38 I just want to jump in and give you a 3 minute warning.
30:42 Three-minute warning.
30:42 OK.
30:43 Um,
30:44 all right.
30:45 So,
30:45 uh,
30:45 I will be careful here.
30:46 Uh,
30:47 we use the Bureau of Labor Statistics
30:49 time use survey to estimate the percentage of time spent
30:52 on something called general community activities.
30:55 And then we do a,
30:57 uh,
30:58 a way of translating,
30:59 um,
31:00 workplace,
31:01 um,
31:02 using weighted average daily contacts from the epidemiologists,
31:05 uh,
31:05 to come up with the,
31:07 the calibration.
31:09 All right,
31:09 because of time constraints,
31:10 let me get to the,
31:11 the basic
31:12 point about
31:14 the uh
31:15 standard
31:16 epidemiology model
31:18 is that
31:19 economic activity does not
31:21 affect
31:22 the inflation rates or agents certainly don't take that into account.
31:26 If in fact you take that into account,
31:29 and here you would look at the blue lines versus the dotted lines,
31:33 the crucial point is
31:35 that
31:36 the
31:37 the you get a much bigger recession
31:40 once you take into account
31:42 economic agents' activities.
31:44 So that's really what figure 1 is.
31:47 Figure 3,
31:50 I'll come back to when I talk about optimal policy,
31:53 but it's in the simplest model,
31:55 what you want to do is
31:56 basically induce a larger recession
32:00 than economic agents would do on their own,
32:03 precisely because they're not internalizing this
32:05 externality associated with economic activity.
32:08 So you can see very severe recession,
32:10 aggregate consumption falling by 25%.
32:14 Now
32:15 in our benchmark model,
32:16 which has vaccines,
32:17 treatment,
32:18 and medical preparation,
32:19 the same basic message occurs,
32:22 but these dotted lines over here as opposed to the blue lines basically show you
32:28 that
32:28 in this model it's optimal to immediately induce sharp containment,
32:34 but then
32:35 As the infections rise,
32:38 you want to keep the tax rising with it,
32:40 the containment with it,
32:42 because that's when the externalities are,
32:43 are greatest.
32:44 And then slowly remove them,
32:47 um,
32:47 as the infection starts to wane.
32:49 So,
32:49 you're basically
32:51 balancing optimal containment,
32:53 um,
32:54 so to you get herd immunity in a socially optimal way.
33:00 Now,
33:00 we end the paper right now with this question of
33:05 the politicians are under a lot of pressure for various reasons to
33:10 end containment prematurely.
33:12 So what we did was an exercise and said,
33:14 look,
33:15 suppose you gave in to that urge at the peak of the infection.
33:18 What would you accomplish?
33:20 That's what the red line is.
33:22 The dotted line is the optimal thing to do.
33:25 And
33:26 So,
33:27 what you can see is if you leave too early,
33:30 let's go to this bottom one,
33:31 this is when you leave at the,
33:32 at the,
33:33 um,
33:34 at the peak of the infection,
33:36 um,
33:37 you would land up getting a temporary boost in economic activity,
33:40 of course,
33:41 uh,
33:42 even larger if you,
33:43 uh,
33:44 left earlier,
33:44 but you would get a huge surge in infections right after.
33:48 And so you wouldn't get a V-shaped recovery.
33:50 You would get
33:51 a big recovery,
33:52 and then you would go right back into recession
33:54 as the infection took hold and agents understood that.
33:58 And then finally,
33:59 if you start early,
34:00 if you start too late rather than starting at time 0,
34:04 the longer you wait,
34:05 the bigger you have to contract once you get underway.
34:09 So there's a penalty for starting late
34:11 and there's a huge penalty for starting early,
34:14 which we can use the model to quantify.
34:16 So let me wrap up because I know I'm almost done.
34:18 The basic idea is to take the canonical models that epidemiologists are using
34:24 and
34:24 internalize,
34:25 take into account.
34:27 The economic decisions that agents make,
34:30 use that to quantify how to exacerbate recessions,
34:33 and then ask what does optimal policy do,
34:36 and optimal policy makes the recession
34:38 even worse
34:39 to buy time for the vaccines
34:41 and treatments to arrive.
34:43 So as usual,
34:44 the scientists will be the ultimate heroes in this story,
34:46 but we can buy them time,
34:48 hopefully in a socially optimal way.
34:51 Thank you.
34:54 Thanks very much,
34:55 Marty.
34:55 Well,
34:55 uh,
34:56 if you don't mind,
34:56 uh,
34:57 unsharing your screen,
34:58 please,
34:58 and we'll go over to Roberto.
35:00 Great,
35:01 I'm gonna try and do that.
35:02 Um,
35:03 so I'm gonna go here.
35:17 I'm trying to get back to the screen.
35:18 I,
35:18 I'm really very sorry about that.
35:20 Um,
35:32 For some reason,
35:32 it's not.
35:35 Letting me get back to them.
35:38 Um,
35:39 you just need again the third button.
35:43 Well,
35:43 I'm just where it says share content,
35:45 I think there's an option that says like unshare.
35:49 No,
35:49 that's what I'm sharing my screen,
35:50 so let me hit,
35:51 ah,
35:51 there we go,
35:52 and now if I hit share.
35:56 Stop sharing.
35:57 Got it.
35:57 Exactly.
35:58 Sorry about that.
36:00 Thank you so much.
36:07 Sure.
36:12 Um,
36:18 Oh,
36:18 good morning,
36:19 everyone.
36:19 I'm
36:20 Roberto Felhaev.
36:21 I'm with the World Bank and the Research Group.
36:24 I'm presenting preliminary and still ongoing work.
36:27 It's joined with Pacoeira and John Shin from
36:31 Washington University in St.
36:33 Louis and I'm the new mayor from Lila University.
36:38 So the,
36:39 the starting point in our work is to notice in
36:41 the previous two presentations are an example of this that
36:45 the first response among economists to the
36:47 pandemic was to very quickly try to bridge
36:50 epidemiological models and surrounded with an economic model
36:55 so as to understand
36:56 uh
36:57 feedback effects between the economy and the diffusion
37:00 of the uh the pandemic and vice versa.
37:04 Naturally,
37:05 we notice that there is a compromise uh that
37:08 being
37:09 able to capture these features in a tractable function implies.
37:13 So the rich epidemiological structure comes at the expense of a rather stylized,
37:17 uh,
37:18 modeling of the production side of the economy and therefore,
37:21 uh,
37:21 implications for the short to medium term are harder to make sense of.
37:26 So where we come in is exactly to follow up on this approach
37:30 and sort of reversing
37:32 the compromise.
37:34 We're actually going to dispense from epidemics altogether.
37:38 We're going to start thinking about a lockdown shock.
37:41 We're going to be very explicit about how we define this,
37:44 and we're going to gain
37:46 a rich and heterogeneous set of preparation mechanisms to get
37:50 a better handle on the medium to short run implications
37:53 of the lockdown.
37:56 Now,
37:56 these probation mechanisms are going to hinge upon two critical,
38:00 and a few sets of
38:02 frictions that have been
38:05 mentioned as important in this context,
38:08 credit frictions to account for the fact that
38:11 firms and individuals might have a hard time
38:14 accessing to external finance to smooth this event.
38:17 And labor market frictions to account for
38:19 large unemployment spells that might emerge from from the layoffs
38:25 involved in the lockdown.
38:27 So that's a realistic feature that we want to bring in.
38:30 Secondly,
38:30 we want to do this in the context of imperfect insurance.
38:33 By this we mean that individuals we have only some risk-free pool of savings to
38:40 appeal to to finance and smooth their consumption,
38:43 but they won't be able to fully insure
38:45 against the idiosyncratic risk in the economy.
38:48 And lastly,
38:49 we want to
38:51 consider um uh an individual business owner.
38:53 So the production uh side of the economy is one that's
38:56 not looking at how the corporate sector reacts to a lockdown,
39:00 but more of like
39:01 how individual entrepreneurs react to a lockdown.
39:05 So then in that setup,
39:06 what are,
39:07 what are our questions?
39:08 We,
39:08 we want to characterize
39:10 how does the economy respond to a notion of a lockdown.
39:14 Uh,
39:15 today we're going to focus on aggregate variables of GDP,
39:18 productivity,
39:18 and employment.
39:20 Of course,
39:20 the richness in the model has a lot of uh distribution and implications to,
39:24 to also explore and we,
39:26 we delayed that for down the road.
39:28 And we want to
39:30 characterize these,
39:31 these responses both in a developed credit market setup like the US,
39:35 but also,
39:36 and we will show this today
39:38 uh um characterize them in a calibration of the credit
39:42 market that might be more suitable for a developing country.
39:46 And then we want to emphasize that besides
39:48 whatever magnitudes we bring up to the table,
39:51 we want to think that
39:52 this is a framework for further modeling
39:55 and characterizing palliative policies that governments are entertaining
40:00 to mitigate the recession.
40:02 So more completely was the moral environment.
40:04 There's gonna be heterogeneous households in their ability to run
40:07 firms and to become entrepreneurs and in their wealth.
40:11 So as a result of this,
40:12 they might become business owners or workers in the labor market,
40:17 and they make these decisions,
40:18 as I said before,
40:19 in the context of imperfect insurance.
40:21 It's just a risk-free asset that
40:23 uh that uh individuals can appeal to,
40:26 to smooth consumption.
40:29 Factor markets are going to be frictional.
40:31 The financial markets frictions take the form of a collateral constraint,
40:35 and the labor market take the form of a matching friction.
40:38 So up until here,
40:39 this is a setup that we've used before to study
40:43 the US economy for other types of shocks like the Great Recession.
40:47 Today we are going to engineer a lockdown shock.
40:50 And the form that the lockdown
40:52 shock is going to take is basically
40:55 at the beginning very conservatively assuming that
40:58 a uniform fraction fee of all the firms in the economy are deemed nonessential.
41:04 And that means that they are forced to shut down for the duration of the lockdown.
41:08 We're going to be generous in the sense that these nonessential entrepreneurs
41:12 will go to the labor market and search for a job,
41:15 and there might be unemployment,
41:18 so we're going to insure everybody against this rise of
41:21 unemployment by providing a full replacement of the wage rate.
41:25 But the economy is still going to be hurt by the fact that
41:28 these would-be entrepreneurs are going to take a big hit on their earnings,
41:32 therefore affecting the ability to save and eventually
41:37 manifesting in the aggregate investment rate in the economy.
41:39 That's the mechanism we have in mind.
41:42 So a little bit of notation to formalize this idea.
41:45 So
41:46 an essential business is an individual that
41:49 uh maximizes um
41:52 lifetime utility making consumption and investment decisions.
41:56 The noteworthy property of the essential business is
41:59 that if you look at the budget constraint,
42:01 there's consumption and savings here on the left hand side,
42:04 but the earnings
42:05 show
42:06 the possibility of this individual to actually become an entrepreneur
42:09 if the profits uh merit so.
42:12 So if you're a restaurant owner and you are still doing takeout,
42:16 as opposed to just shutting down and go to search for a job,
42:20 then you can do that,
42:21 and that's going to be part of your earning.
42:23 However,
42:24 and here's where the financial friction comes into play.
42:27 If you are an entrepreneur,
42:29 then the amount of capital that you might be able to attract to your plant
42:33 is going to be limited to a proportion of your wealth.
42:37 And the,
42:37 the,
42:38 the reminder of your earnings are
42:39 the returns to your savings and your savings account,
42:42 and then everyone is going to be contributing to in lump sum fashion
42:46 to the financing of the unemployment insurance that I will talk about in a second.
42:50 Now if you are a non-essential uh business owner,
42:53 I say
42:54 your problem is the same,
42:55 you're still optimizing lifetime utility
42:58 uh uh consumption and,
42:59 and savings,
43:01 but your earnings are just given by your wage.
43:04 So this is the way in which you are forced to shut
43:06 down and go to the labor market that we are fully insuring you
43:09 against unemployment.
43:14 And the other piece of notation we want
43:16 to reduce corresponds to the labor market friction,
43:19 and the way we think about the labor market is as follows.
43:22 So
43:23 imagine at the beginning of the period there will be a mass of unemployed workers,
43:27 a bunch of other workers are going to be fired at the beginning of the period.
43:31 The matching friction.
43:33 Takes the form of only a fraction of that pool
43:36 being matched into a hiring marker,
43:38 a hiring market where the demand side will be waiting for them.
43:42 So as a result,
43:43 the unemployment rate tomorrow is going to be whoever
43:46 was unemployed today like the nearly destroyed jobs,
43:49 net of these
43:50 matches that were created in the period.
43:53 Now,
43:54 at this stage,
43:54 we want to emphasize an extension we're working on,
43:57 which is
43:58 to bring in the notion of,
43:59 of rest and employment.
44:01 Uh,
44:01 I'm not sure where exactly
44:03 uh defining it as Albert Scheimer have introduced the concept to the literature,
44:07 but what we have in mind is the possibility that
44:10 these non-essential firms
44:12 might not have to go through the matching
44:14 function to hire the workers they just fired.
44:17 So this will be like in the rest and employment status,
44:19 and then
44:20 when the lockdown is uh finished,
44:23 the,
44:23 the,
44:23 the non-essential firms can go back and fire them much quicker.
44:27 Like,
44:27 as you can imagine,
44:28 this will be important for
44:29 the duration of the unemployment spells.
44:33 Very quickly over calibration of the
44:36 parameter values.
44:37 I mean we are targeting micro level and aggregate statistics
44:41 of the firm size distribution,
44:42 the unemployment rate which we use to calibrate the
44:45 the level friction parameter
44:47 and external finance to capital ratios to discipline
44:51 the financial friction.
44:52 Today we're going to have two calibrations of this lambda parameter,
44:55 one of a very flexible one for the US
44:58 and a tighter one for developing countries.
45:02 Now,
45:02 again,
45:02 to be more specific,
45:04 what's the lockdown shock?
45:06 We're going to start the economy at the stationary allocation.
45:09 Everything is going according to normal
45:11 and then unexpectedly,
45:13 a fraction fee of businesses are deemed non-essential.
45:17 We are still figuring out what the magnitude and the persistence
45:20 of fee is that's still ongoing and it's an ongoing event,
45:24 so we are going to be conservative.
45:25 I don't know how our number maps,
45:27 for example,
45:27 to what Marty showed over the 75% of the economy.
45:31 Here we are assuming that 30% of businesses
45:34 become nonessential
45:35 and a period in the model is a quarter,
45:38 so
45:39 it's going to be a quarter,
45:40 so this is going to be a one period shock.
45:43 And besides being realistic,
45:44 it allows us to emphasize
45:46 preparation channels through the model.
45:51 So let us show some results.
45:53 Let us begin with the dynamics of GDP.
45:56 We are showing these are normalized on the vertical axis to GDP in the steady state.
46:01 We have two lines.
46:03 The black solid line is the calibration to the US economy,
46:06 and the gray line
46:08 is the calibration of credit markets to a developing economy.
46:13 So we want to make the point that on impact the lockdown
46:17 translates into in this conservative.
46:20 Implementation of the shock into a 12 to 13% reduction in
46:25 quarterly GDP for the US
46:27 and a recovery that takes over roughly a year to undo,
46:31 to do.
46:33 For the developing economies,
46:34 we are finding a stronger decline and impact of more than 15,
46:39 17% of GDP.
46:42 Now what's driving this
46:44 here we show the TFP and the unemployment rate.
46:47 So TFP shows that it helps to account for a large chunk of the initial decline.
46:53 Importantly,
46:54 we note that the magnitude of the decline in TFP.
46:59 is larger in the developing economy as in the US.
47:03 You can think of the US declined TFP as
47:06 more of a mechanical effect coming from the lockdown.
47:10 So the lowdown
47:11 puts a bunch of entrepreneurs out of business
47:14 with decreasing returns.
47:15 You're loading more production into fewer firms,
47:17 and that's bad for aggregate efficiency,
47:19 and that's exactly the decline in entrepreneurship is mostly what drives
47:24 the decline in productivity for the US
47:26 in developing economies,
47:27 despite the shock being the same,
47:29 the decline is stronger,
47:31 and here we see the richness of the model starting to become more prevalent.
47:36 What's driving the stronger decline in developing economies is the fact that
47:39 we see a strengthening of the misallocation due to tighter credit strengths.
47:44 Among the essential firms,
47:47 those that are going to be expanding and,
47:48 and sort of enjoying the lockdown are
47:51 those that can do so.
47:53 And who can do so in a world with tight credit?
47:55 The wealthy entrepreneurs.
47:57 And the wealthy entrepreneurs need not be the most talented ones,
48:00 and that's why
48:01 misallocation is magnified
48:03 an impact
48:05 in the developing country.
48:07 Turning to unemployment rate,
48:10 we see that this is your 5 minute warning.
48:11 Sure,
48:12 thank you.
48:13 Turning to unemployment rate,
48:14 we see that it peaks at roughly 18% for the US and it is lower
48:20 in
48:20 developing countries.
48:22 The reason why the peak is lower in developing countries is because
48:25 this matching friction,
48:26 the way we model it.
48:28 The less traffic there is in the labor market,
48:31 then the less is the unemployment,
48:33 and with tighter credit reallocation is hindered.
48:36 Therefore,
48:37 the traffic in
48:38 that friction is mitigated.
48:43 Turning to more of the demand side,
48:44 I mean the the investment rate in the economy,
48:47 what we are finding is a strong decline in
48:49 investment and obviously a stronger decline for developing economy
48:54 that transmit to a attractive decline of uh the capital stock that lasts for a while,
48:59 but that's not a very substantial one.
49:02 When it comes to investment dynamics,
49:04 we're going to stress that there are two forces at play here.
49:06 One is the fact that GDP went down,
49:08 so investment is going to fall.
49:11 And there are other forces that in our specification
49:14 of the shock kind of mute each other,
49:16 but I want to emphasize because they have to do with the richness
49:19 of the model and they may play a different strengths in other calibrations,
49:22 which is
49:23 the reallocation of net worth.
49:26 Here you're going to have a bunch of agents,
49:28 particularly the,
49:29 the non-essential ones
49:31 that are going to be eating their savings to live through the lockdown.
49:35 And therefore that'll be saving much less and,
49:37 and the economy is going to have less capital through that channel.
49:41 But on the other hand,
49:42 you're going to have the essential businesses for whom
49:45 profits went up,
49:46 you know,
49:46 factor prices are lower now,
49:48 so their earnings have gone up,
49:49 and those guys are expanding
49:51 their net worth.
49:53 So the investment rates sort of
49:55 balances those forces and just showing the fact that GDP went down.
49:59 But in other calibrations we think that this
50:02 distribution of net worth across individuals might,
50:05 might
50:05 have an effect on the shape of the
50:07 of the conversions of investment.
50:11 Now,
50:11 prior to concluding,
50:12 we wanted to
50:14 show a case that might sound more realistic to many
50:18 in terms of what's the timing of the shocks.
50:21 And in particular when you consider the situation where
50:25 prior to learning that there is a lockdown,
50:28 you have already made some commitments of capital rental.
50:32 So think of a barbershop or a nail salon that they still have to pay for their lease
50:37 and then they are told that they cannot open.
50:39 So we implement this by forcing
50:43 both the essential and the non-essential firms to have
50:46 to pay for their capital in the first period.
50:49 As you can imagine,
50:50 this is going to be a tremendous
50:51 shock for the nonessential firms because their earnings
50:54 not only go down because of the foregone profits,
50:57 but also whatever wage they get,
50:58 they have to use partly to pay for this capital.
51:02 So the figures below show.
51:05 The the same dynamics as before.
51:07 So the solid black is the US,
51:10 the solid gray is the developing country,
51:12 but now we have a dashed line for the US
51:16 under this timing.
51:18 And the point I want to make is that this timing is a lot more damaging to the economy,
51:23 bringing the decline in impact to roughly 20% of GDP for the US
51:28 And what we can see from the TFP dynamics is that now TFP is massively affected,
51:34 mostly because we are measuring TFP as in national
51:36 income accounts and essentially that's capturing underutilized capital.
51:41 So if you were more,
51:43 uh if you are accounting for underutil capital utilization,
51:45 that will be captured by capital stock,
51:48 but um otherwise it's TFP that taking the toll of the underutilized capital.
51:52 So
51:53 we want to give a sense that even a very conservative calibration of the shock,
51:57 a very generous social insurance program of the unemployed,
52:01 the recession can be very big depending on
52:04 how we assume the,
52:06 the timing of the lockdown uh enters the,
52:09 enters the economy.
52:11 So what I wanted to conclude is to
52:13 advertise some extensions.
52:14 So
52:15 we think that the framework can still be enriched prior to thinking about policies.
52:21 Uh,
52:21 one direction is we want to make the exit a little
52:23 bit more endogenous rather than just a uniform exit shock.
52:27 And one way to do that is to have a more interesting field of entrepreneurship where
52:32 creating a firm entails a setup cost and a fixed cost of operation.
52:36 That is going to bring some hysteresis into entrepreneurship,
52:39 inducing
52:41 some of the non-essential firms to still want to be
52:44 an entrepreneur despite earning zero profits
52:46 for the sake of avoiding to pay the setup of costs again in the future.
52:50 So that
52:51 that might be a better way to handle the shock.
52:53 And finally,
52:54 as I alluded earlier,
52:56 um,
52:57 we want to account for the rest unemployment
52:59 and the possibility of a quicker rehiring of the
53:02 of the recently fired workers.
53:05 Once we have converged to the framework we think is the most reasonable,
53:09 then
53:10 We have a set of palliative policies in mind and the list can go on,
53:13 but so far,
53:15 so for example,
53:15 Argentina just announced yesterday labor market
53:18 policy where they basically forbid any firm
53:22 They forbid nonessential businesses from firing the labor force.
53:26 So you can think of this as a model as not only forcing
53:28 firms to pay for their capital as in the last experiment I showed,
53:32 but in addition,
53:32 forcing them to pay for the labor force.
53:34 That's,
53:34 that's a policy we can,
53:35 we can study.
53:37 Peru has announced something very creative on the
53:39 credit market stand uh branch of the,
53:41 of the economy,
53:42 um,
53:43 issuing,
53:44 uh,
53:44 like free loans
53:46 from the central bank implemented through commercial banks that will give
53:50 uh financing for working capital to Peruvian businesses
53:54 so we can implement those in the model by relaxation,
53:57 simultaneous relaxation of the credit constraints
54:00 at the same time we hit with the lockdown,
54:02 the lockdown shock.
54:03 And also we have the flexibility to experiment with various
54:06 types of lump sum transfers that could be targeted,
54:08 non-targeted,
54:09 so um.
54:11 Yeah,
54:11 we think that we have the flexibility to,
54:13 to study all,
54:13 all these points.
54:15 So let me stop that.
54:29 I just want to kick off the discussion with some observations,
54:31 but please go ahead.
54:35 OK.
54:35 Um,
54:36 so,
54:37 first of all,
54:37 I would like to um thank the organizers for inviting me to discuss uh
54:42 the three papers which I read them with really great interest and pleasure.
54:47 Uh,
54:47 and I want to congratulate the speakers,
54:49 uh,
54:50 for their timely contributions,
54:51 which definitely help us understand much better
54:54 the policy trade-off that governments are currently
54:57 facing,
54:58 uh,
54:58 when dealing
54:59 with the COVID-19 crisis.
55:02 So I will start,
55:03 I have 4 questions for Martin,
55:05 um,
55:06 a question,
55:07 uh,
55:07 for,
55:07 um,
55:08 Francesco,
55:09 and then
55:09 a question and a request,
55:11 uh,
55:11 for,
55:12 for Roberto.
55:13 So Martin,
55:14 as you mentioned in your paper,
55:16 uh,
55:17 the classic model is a special case of your model in the sense that the propagation
55:22 of the disease
55:23 is unrelated to economic activity,
55:26 and this is precisely the contribution of,
55:28 uh,
55:28 your paper.
55:29 So if we compare the dynamics of aggregated
55:31 consumption and work hours between both models,
55:34 there is a huge difference between them with your model
55:36 actually displaying a much more severe recession but saving more
55:40 lives.
55:41 So given that there There has been a lot of debate on what
55:44 type of government interventions could be
55:46 effective for dealing with the pandemic,
55:48 and your paper focused only
55:50 on containment measures which are applied without discrimination to everybody,
55:54 which I think it's a little bit
55:56 extreme in the sense that you are taxing susceptible,
55:58 infective and recovered people,
56:01 but you don't actually explore the possibility of testing,
56:05 let's say universally,
56:06 and then tracing and targeting the containment
56:09 measures only to those that are infected.
56:12 I am curious
56:13 and would like uh your thoughts on this.
56:15 If you can confirm if the results of the classic model would reflect actually the
56:19 aggregate dynamics of consumption and what are in the extreme case that we can have
56:24 actually extensive testing and target containment measures only to those
56:28 that are affected,
56:29 because if that were the case,
56:30 then I think
56:31 Uh,
56:32 even beyond actually thinking about the,
56:34 the role of externalities here,
56:36 uh,
56:36 it will be nice to highlight that there could be less extreme
56:40 policy solutions that can deliver better economic outcomes,
56:43 uh,
56:44 than the ones that we will see,
56:45 uh,
56:46 right now.
56:46 And of course,
56:47 I
56:48 understand that I,
56:49 I want to acknowledge that there will be a trade-off
56:51 because we will not be able to build up a large
56:54 fraction of immune people and as a result,
56:56 we will end up with more susceptible
56:59 people to infection later.
57:01 My second question is related to the dynamics of the optimal policy.
57:06 So,
57:07 in your model,
57:08 the dynamics of the optimal policy really
57:10 mimics the dynamics of the infection rate,
57:12 and this is super intuitive.
57:13 I mean,
57:14 containment measures
57:15 internalize the externality caused by the behavior of infected people,
57:19 so as the number of infected people actually rise,
57:22 it is optimal to intensify containment measures and vice versa.
57:27 What I am really struggling is um that once you achieve the
57:32 in the peak of infection rate and basically start
57:37 um relaxing the containment measures,
57:39 the infection rate also goes down.
57:42 But if you basically look to what happened,
57:45 Um,
57:46 in the US across cities,
57:48 with the pandemic flu in 1818,
57:50 you will observe that most of the cities went through a second long wave of infection
57:56 as governments start basically relaxing the initial containment measures.
58:00 And if you actually look at the evolution of death over time,
58:04 um,
58:05 we see that it's like the distribution is bimodal with a second peak that was
58:10 uh lower than the first one,
58:12 but still very high.
58:13 So when thinking about the lessons learned from the previous uh pandemic episode,
58:18 uh,
58:18 there has been a lot of debate on what caused the second long wave of infections,
58:22 and some argue that the timing,
58:24 the intensity,
58:25 and the duration
58:26 of the containment measures were not optimal.
58:28 So,
58:28 what I really want to know is your thoughts.
58:31 Uh,
58:31 to what extent your,
58:32 uh,
58:32 optimal policies will definitely rule
58:35 Out,
58:35 um,
58:36 the possibility of a second long
58:38 um wave of infection
58:41 and to what extent the current policy responses imposed by countries in the US,
58:46 in China,
58:47 and Italy are optimal,
58:48 so we are sure that we will not go through
58:50 a second long wave of infections that these governments actually
58:53 start relaxing the containment measures.
58:57 My third question is related to the dynamics of aggregate consumption.
59:01 So in your model,
59:02 the aggregate,
59:02 the dynamics of aggregate consumption really mimics
59:05 the U-shaped pattern of the awards work,
59:07 which indeed basically reflects
59:09 labor supply decisions of susceptible agents.
59:12 And I assume because the focus of the paper
59:14 is on highlighting the role of externalities and how
59:17 actually governments can internalize that externality,
59:20 you abstract from other actually
59:22 uh key factors that can influence the consumption
59:24 patterns of the susceptible growth like uncertainty,
59:28 or the patient's expectations about the timing,
59:31 the intensity and the duration of the commitment.
59:33 So we know from recent research that indeed has been actually conducted in
59:37 Italy that compliance with COVID-19 social distancing
59:41 measures is a function of people's expectations
59:45 about how governments will react
59:47 with the containment measures.
59:49 And I wouldn't be surprised
59:50 that the
59:51 Consumption pattern also,
59:52 some part of the fluctuation
59:54 uh reflect those expectations.
59:56 So have you thought,
59:58 uh,
59:58 how the consumption trajectory could change as you introduce these factors
1:00:02 and give us a little bit more realistic approach
1:00:05 to uh model consumption behavior of the susceptible population.
1:00:09 And my last question because I think that there are a lot of World Bank people,
1:00:13 uh,
1:00:13 connected,
1:00:14 um,
1:00:15 to,
1:00:15 to this seminar is like you mentioned that you're
1:00:18 basically trying to understand.
1:00:20 Um,
1:00:21 and explore the role of fiscal transfers to people
1:00:24 and loans to keep actually firms going from bankrupt.
1:00:27 And I think that
1:00:28 given that these policies are currently widely discussed at the bank,
1:00:31 it will be super useful for us to hear from you,
1:00:35 uh,
1:00:36 if you have actually preliminary findings.
1:00:39 So,
1:00:40 Um,
1:00:41 I,
1:00:41 I have one question,
1:00:43 uh,
1:00:44 for Francesco,
1:00:45 and indeed,
1:00:46 um,
1:00:47 Francesco,
1:00:47 your paper actually and Martin's built on the same epidemiological,
1:00:51 uh,
1:00:52 setup.
1:00:52 So it's like,
1:00:53 the first question that I did for Martin also applied.
1:00:56 Uh,
1:00:57 for,
1:00:57 for you too.
1:00:58 But when I am comparing,
1:01:00 I mean,
1:01:00 I have been actually trying to compare
1:01:02 the dynamics of the optimal policy across both papers.
1:01:06 I mean,
1:01:07 uh Martin's model,
1:01:08 the dynamic
1:01:09 really mimics the dynamics of the infection rate and once
1:01:12 you achieve the peak it start actually going down fast.
1:01:16 But in your model,
1:01:17 Uh,
1:01:18 the infection actually rate,
1:01:20 um,
1:01:20 doesn't go so fast and even when you start actually declining,
1:01:24 you have,
1:01:25 um,
1:01:26 the lockdown at the highest level for a really long period,
1:01:31 which is more or less
1:01:32 like
1:01:33 I don't know,
1:01:34 9 weeks.
1:01:35 So,
1:01:36 if,
1:01:36 if you actually consult uh infectious disease specialist,
1:01:40 they will tell you,
1:01:41 well,
1:01:41 you have to keep the lockdown
1:01:43 very high,
1:01:44 uh,
1:01:45 even when they actually infection rates start going down
1:01:48 to avoid actually going to the second longest.
1:01:51 wave of infections,
1:01:52 but I don't think that this is the case.
1:01:54 I mean,
1:01:54 the
1:01:55 the mechanism that it's actually working
1:01:59 uh in,
1:01:59 in your model actually
1:02:01 nice if you actually make a comparison
1:02:03 of the optimal policy,
1:02:05 uh,
1:02:06 dynamics in your paper visa vis um
1:02:09 Martin's paper and actually explain why you have
1:02:12 to keep in your model actually the lockdown,
1:02:14 uh,
1:02:15 for a really long period of time.
1:02:17 And then my last
1:02:19 question or request uh to Roberto,
1:02:22 um,
1:02:23 I mean,
1:02:23 I felt obliged actually to bring the discussion of
1:02:26 digital technology here specifically when we are actually talking about
1:02:31 developing countries.
1:02:32 So there has been a lot of interest at the banking pushing with the digital agenda
1:02:36 and the goal,
1:02:37 basically,
1:02:38 the,
1:02:38 the world is actually going virtual,
1:02:40 no,
1:02:40 so we are learning that,
1:02:41 I mean,
1:02:41 the effects will not be the same for those countries that manage actually
1:02:46 Uh,
1:02:46 to connect their people and their firms,
1:02:49 uh,
1:02:49 in how actually they respond
1:02:51 to the,
1:02:52 um,
1:02:53 to the,
1:02:53 to the pandemic crisis.
1:02:55 So it will be super nice because your paper talks about distortions and also brings
1:03:01 uh explicitly the firm size.
1:03:03 If we can actually work together and incorporate an additional extension
1:03:09 uh
1:03:09 to bring the role of the Digital,
1:03:12 um,
1:03:13 economy in,
1:03:14 into this discussion.
1:03:16 It will be super helpful if we can actually conduct
1:03:19 counterfactual exercise of how much
1:03:21 actually developing countries would have benefit
1:03:24 if they were,
1:03:25 I mean,
1:03:26 uh,
1:03:26 their people and their firms are already connected.
1:03:29 So I will actually stop here.
1:03:32 Um,
1:03:32 thanks a lot for giving me the opportunity,
1:03:34 uh,
1:03:35 to participate in this seminar.
1:03:38 Uh,
1:03:38 thank you,
1:03:39 Anna.
1:03:39 Um,
1:03:39 I'd like to make sure that we leave enough room for the,
1:03:42 uh,
1:03:43 3 sets of authors to respond,
1:03:45 uh,
1:03:45 to questions.
1:03:46 So what I'd like to suggest is that we go for 10 minutes with,
1:03:49 uh,
1:03:50 with Q and A.
1:03:51 We'll just accumulate them all together and then we'll go
1:03:53 back with 5 minutes to each group of presenters.
1:03:56 For those of you
1:03:57 Who are on,
1:03:58 um,
1:03:58 uh,
1:03:59 Webex,
1:03:59 may I ask you to use your raise your hand function.
1:04:02 It's a little circle beside your name at the top right with a hand in it
1:04:05 and raise it.
1:04:06 And I'll ask,
1:04:07 uh,
1:04:07 Ale because I think she can see them to acknowledged speakers.
1:04:11 Um,
1:04:11 and then if you're watching on YouTube,
1:04:13 you can send in through the chat function a question.
1:04:16 While,
1:04:17 uh,
1:04:17 while we're waiting for people to do both of those,
1:04:19 let me just exercise my prerogative as chair to ask two quick questions.
1:04:23 Um,
1:04:24 the one,
1:04:25 which I think applies for the first two presentations
1:04:27 is just to get a sense of the sensitivity of the findings to the,
1:04:31 um,
1:04:31 the calibrated evaluation of life
1:04:34 and to the,
1:04:35 um,
1:04:35 I know I had a sort of side conversation
1:04:37 last week already with Marty and Sergio about this,
1:04:39 but it It'd be great
1:04:40 to get a sense of that.
1:04:41 This is a non-trivial question because
1:04:43 we might think that that valuation is very different and
1:04:46 particularly when we think about doing this in a developing country setting,
1:04:49 you know,
1:04:49 should we be thinking about valuation of life in terms of
1:04:52 calibrating the model relative to income or something like that?
1:04:56 The second quick thing that I'd like to ask that I
1:04:58 think is probably most germane to Marty and Sergio and Mattis's paper
1:05:03 is to think a little bit more about the incentives to defy containment measures.
1:05:07 I was thinking about this particularly in the context of poor countries where
1:05:11 you've got people who,
1:05:12 you know,
1:05:13 Have to work to eat,
1:05:14 right?
1:05:15 And in a situation like that,
1:05:16 the incentives to defy are huge and are not
1:05:18 well captured by the framework you have here.
1:05:21 I think one easy fix might be to sort of put in a
1:05:23 little sort of stone geary term to allow some people to be very close to,
1:05:27 uh,
1:05:28 infinite marginal utility of consumption.
1:05:31 Um.
1:05:31 On the flip side of that though,
1:05:33 I think another interesting thing to look into is
1:05:35 to think a little bit more about the penalties
1:05:37 for,
1:05:38 um,
1:05:38 defying containment measures.
1:05:40 So,
1:05:41 in,
1:05:41 uh,
1:05:41 in the second paper presented,
1:05:42 the penalty is kind of a smooth thing and at the margin,
1:05:45 you can
1:05:45 sort of comply a little more or a little less.
1:05:48 But what if we thought about them as being sort of pretty discrete ones?
1:05:51 I mean,
1:05:51 this is
1:05:52 germane right now because in Virginia,
1:05:53 if I venture down the street without a valid excuse now,
1:05:56 I face a $2500
1:05:58 fine,
1:05:59 which is sort of a non-marginal thing.
1:06:01 Um,
1:06:01 let me stop there.
1:06:02 Um,
1:06:03 Ala,
1:06:04 since I think you can see who has hands up,
1:06:06 do you want to acknowledge a couple of
1:06:11 Hands from Webex.
1:06:13 Anybody,
1:06:13 but,
1:06:13 um,
1:06:14 please let me know and I I'll also let you know art
1:06:17 as soon as I see.
1:06:19 OK,
1:06:20 great.
1:06:20 Then we do have one question from the YouTube already,
1:06:24 uh,
1:06:25 question sent by Luis Fernandez,
1:06:26 uh,
1:06:27 question for the first presentation.
1:06:29 How sensitive are the results to the dependence of the death rate
1:06:33 on the proportion infected?
1:06:35 Uh,
1:06:35 I believe a linear function was assumed,
1:06:37 but a highly nonlinear one is more likely.
1:06:45 And um let me just pause here to see
1:06:47 if there's some other questions coming from uh the,
1:06:49 the folks who we invited to participate by Webex because you can speak for yourself.
1:07:01 OK,
1:07:01 while,
1:07:01 while we're waiting again,
1:07:02 another question from YouTube,
1:07:04 from our colleague Alvaro,
1:07:05 uh,
1:07:06 in,
1:07:06 uh,
1:07:06 who didn't get a link.
1:07:07 Uh,
1:07:08 there is still a lot of uncertainty about
1:07:10 the pandemic parameters,
1:07:12 mortality rates,
1:07:13 transmission,
1:07:14 etc.
1:07:15 How does optimal policy look under uncertainty?
1:07:20 And I think that probably applies to everybody.
1:07:23 Uh,
1:07:24 I mean,
1:07:24 all,
1:07:24 all three papers that is.
1:07:27 Ala,
1:07:28 do you have any hands?
1:07:30 Um,
1:07:31 not yet,
1:07:32 um,
1:07:32 and nothing in the chat.
1:07:36 OK,
1:07:36 well,
1:07:36 um,
1:07:37 but there's plenty in,
1:07:38 uh,
1:07:39 in Anna's questions to go on.
1:07:41 So why don't I suggest that we just go back to the three sets of authors in turn,
1:07:45 and if I got a couple more questions,
1:07:46 uh,
1:07:47 coming in from,
1:07:48 uh,
1:07:48 from YouTube,
1:07:49 I will,
1:07:49 uh,
1:07:50 uh,
1:07:50 I'll just jump in and let you know.
1:07:52 So,
1:07:52 uh,
1:07:52 going back to Francesco and whichever combination of you would like to respond,
1:07:56 uh,
1:07:57 Laura's yours for
1:07:58 5 minutes,
1:07:59 please.
1:08:04 OK.
1:08:05 So,
1:08:06 I can take the floor and take my 5 minutes to respond?
1:08:10 Yeah,
1:08:10 go for it.
1:08:11 Thanks.
1:08:13 Thank you.
1:08:14 Let me share a screen.
1:08:15 I think that will be useful,
1:08:18 um.
1:08:20 So
1:08:21 First,
1:08:22 uh,
1:08:22 the first question by Ana
1:08:25 about,
1:08:25 you know,
1:08:26 the dangers of a second cycle
1:08:28 and why,
1:08:29 uh,
1:08:30 it may,
1:08:30 why in our model the policy stays put for so long,
1:08:34 even though you see the infected decline.
1:08:36 Well,
1:08:36 in these models,
1:08:37 in the SAR model,
1:08:38 you have to understand,
1:08:39 first of all,
1:08:40 that there is at most one cycle,
1:08:43 so that is that the I curve has at most one peak,
1:08:46 and the state is two dimensional.
1:08:48 What does it mean?
1:08:49 It means that it's not enough to know what is the level of the infected today,
1:08:53 to know whether it's a good time or not to lift the lockdown.
1:08:57 You have to know INS
1:09:00 because if you have a small I but you still have a very bigs,
1:09:04 like in this region of the state space.
1:09:07 It's like,
1:09:07 you know,
1:09:07 you're walking around with a match and there is a lot of gasoline around.
1:09:11 There is a big risk that you set off a big fire.
1:09:14 It's a very different situation if I is small,
1:09:18 but also S is small,
1:09:19 so you're more like in this region of the state space.
1:09:22 There,
1:09:23 the potential risk that those infected can do is much smaller.
1:09:26 That's why the policy function depends on two objects,
1:09:30 not one.
1:09:31 And when you ask yourself the question,
1:09:32 you know,
1:09:33 is it a good time or not to lift the lockdown,
1:09:36 you don't have.
1:09:36 To look just at I.
1:09:38 You have to look at both I and S.
1:09:41 Now in the model,
1:09:42 the reason why the policy stays locked
1:09:44 for so long,
1:09:45 as you can see here,
1:09:46 this is the white line.
1:09:47 We start here in this red dots.
1:09:49 There is a little bit of infected,
1:09:51 but a lot of potential infected.
1:09:54 That's why it takes a long time for the policy to be relaxed.
1:09:57 Until it's yellow,
1:09:58 it means you're keeping it
1:09:59 high,
1:10:00 high down,
1:10:01 because if you lift it here,
1:10:03 actually any point you lift it before.
1:10:05 25,
1:10:06 this is a Change of face,
1:10:07 you will have like the,
1:10:08 the number of infected will start growing again.
1:10:10 That's,
1:10:11 you know,
1:10:11 that's my quick answer to,
1:10:13 uh,
1:10:13 to your
1:10:15 question and sorry,
1:10:15 you had to,
1:10:16 to look at a very preliminary paper.
1:10:18 Now,
1:10:18 the second question by art,
1:10:20 you know,
1:10:20 the sensitivity to the statistical value of statistical life,
1:10:23 it's very easy to analyze in the model.
1:10:25 In fact,
1:10:25 our model
1:10:27 uses the value of a statistical life in between 6.5 million and 13 million,
1:10:33 which is what is consistent with what other people have done.
1:10:36 Uh,
1:10:37 in the literature.
1:10:38 It's,
1:10:38 it's just one parameter in the model.
1:10:40 So if you want to explore,
1:10:43 uh,
1:10:43 different,
1:10:44 uh,
1:10:44 cases,
1:10:45 say for developing countries with the higher or lower values,
1:10:49 that's,
1:10:49 that's
1:10:50 very easy to do within the model.
1:10:53 Uh,
1:10:53 there was a question from the web on the,
1:10:55 I think it was Luis Fernandez
1:10:58 on the sensitivity to the nonlinearities.
1:11:01 It,
1:11:01 it is a crucial,
1:11:03 uh,
1:11:03 feature of the model,
1:11:04 how non-linear the cost function is.
1:11:06 In the presentation,
1:11:07 in the model that I presented,
1:11:10 the,
1:11:10 um,
1:11:11 you know,
1:11:11 here is what,
1:11:12 it's exactly here.
1:11:13 My,
1:11:14 uh,
1:11:14 fatality rate is actually nonlinear,
1:11:16 it is not linear,
1:11:17 it is nonlinear
1:11:19 and quadratic like the one that Martin used.
1:11:22 Now how much,
1:11:24 how big is this quadratic effect depends on this parameter kappa.
1:11:27 So if you make it smaller,
1:11:29 you will have less of a lockdown because essentially
1:11:32 you have a lesser problem to deal with.
1:11:34 I mean the peak will not be as high.
1:11:37 The congestion effect will not be.
1:11:38 As severe,
1:11:40 but even with the kappa zero,
1:11:41 even with the linear effect,
1:11:42 there may be parameterization
1:11:44 under which you want to go to a complete lockdown.
1:11:47 I don't have time here to get into the details,
1:11:49 but that is indeed
1:11:50 one of the important parameters and uh it's easy to
1:11:54 analyze in the paper.
1:11:56 And finally,
1:11:57 uh,
1:11:58 uh,
1:11:59 your question about the uncertain,
1:12:01 uh,
1:12:01 uncertainty of the parameters,
1:12:03 uh,
1:12:03 uh,
1:12:04 I think it was you are,
1:12:05 uh.
1:12:06 Uh,
1:12:07 you know,
1:12:08 uh,
1:12:08 we,
1:12:08 all of us,
1:12:09 uh,
1:12:09 acknowledge that,
1:12:10 uh,
1:12:11 first of all,
1:12:11 we are not experts in this literature.
1:12:13 Second,
1:12:13 even when you talk to the experts,
1:12:15 we don't really know what is the fatality rate,
1:12:17 how many are infected,
1:12:19 so,
1:12:20 uh,
1:12:21 What we did
1:12:22 for now is to experiment across a variety of parameterization and check
1:12:27 the robustness of the results.
1:12:30 Perhaps a more interesting experiment would be to try to incorporate this
1:12:35 parameter or model uncertainty
1:12:37 ex ante
1:12:38 and to think of an exercise in robust.
1:12:41 Control
1:12:42 where you are actually trying to control a system
1:12:45 uh about which you kind of know something of the law of motions,
1:12:50 but you know that you're not sure
1:12:52 about many of its details and,
1:12:53 and that's what,
1:12:54 you know,
1:12:55 there are tools in the literature
1:12:57 to develop these exercises like Sargeant and Hansen
1:13:01 and
1:13:01 are two economists who have done a great deal of contributions to this
1:13:06 line of inquiry and,
1:13:07 uh,
1:13:07 and perhaps that might be a next interesting step.
1:13:14 Great,
1:13:14 thanks very much,
1:13:14 Francesco.
1:13:15 And just since I'm interrupting now to pass the floor on,
1:13:18 I'll,
1:13:19 I'll pass on one other question that came in online.
1:13:23 Um,
1:13:23 I'm not 100% sure I'm going to do it justice,
1:13:25 but I think it's intriguing enough just to toss out there,
1:13:28 and that is that
1:13:30 Uh,
1:13:31 essentially the question is,
1:13:32 so these are,
1:13:33 these are models that governments can use to
1:13:34 sort of make decisions about optimal containment policies.
1:13:38 Um,
1:13:38 what are the political risks that we should think about if
1:13:41 these models turn out to be incorrect and you end up,
1:13:43 say,
1:13:44 with much higher mortality or a much deeper recession than otherwise.
1:13:47 It's a little bit of a general question.
1:13:49 Um,
1:13:50 I'll leave that in,
1:13:51 uh,
1:13:51 in the pot for,
1:13:52 uh,
1:13:53 for Marty's group or,
1:13:54 uh,
1:13:54 Roberto's group to,
1:13:55 uh,
1:13:56 to,
1:13:56 to respond to.
1:13:57 So over to you next,
1:13:58 Marty.
1:14:07 Um,
1:14:08 let,
1:14:08 let me start with Anna's first question about,
1:14:11 uh,
1:14:11 containment.
1:14:13 Uh,
1:14:13 she's absolutely correct that we have a particular instrument,
1:14:16 you know,
1:14:17 as always with RAMSI problems.
1:14:19 We are
1:14:20 solve have solved for
1:14:23 the social planning problem,
1:14:25 which allows one to discriminate between the different types of agents.
1:14:30 And those results are not quite ready,
1:14:32 which is why they're not included,
1:14:34 uh,
1:14:34 but,
1:14:34 um.
1:14:36 I was personally surprised tentatively that
1:14:40 the results are not all that different than one would expect,
1:14:43 although,
1:14:43 of course,
1:14:44 you would send
1:14:46 recovered agents to work and infected would stay at home,
1:14:49 but they're sort of offsetting effects so that the aggregates,
1:14:51 at least so far,
1:14:52 don't look very different,
1:14:54 but we,
1:14:54 we,
1:14:55 we are definitely going to
1:14:57 write a sequel to our paper on smart containment.
1:15:00 And there I think you really have to distinguish also between old and young,
1:15:04 and the demographics become important,
1:15:05 so I completely
1:15:08 agree that
1:15:09 that
1:15:10 could change some interesting things.
1:15:12 Very good question about the dynamics of optimal
1:15:16 policy and what happened during the Spanish flu.
1:15:20 If you look at the last
1:15:22 2nd to last figure in our paper.
1:15:25 What you'll notice is
1:15:27 Giving up too early,
1:15:29 you would get exactly what Anna was alluding to,
1:15:32 that you get a second wave of infections.
1:15:34 So if you give in to either by mistakenly or political pressure
1:15:39 and end prematurely that policy
1:15:42 of containment.
1:15:44 The model,
1:15:45 this little model says,
1:15:46 yeah,
1:15:47 you're gonna get a boom for a little bit,
1:15:49 but then you're just gonna slide back into
1:15:51 that second infection because you haven't gotten the herd immunity.
1:15:55 So,
1:15:55 yes,
1:15:55 I,
1:15:56 I completely agree
1:15:57 that uh giving up too early,
1:15:59 and we've already seen some indications of this from various politicians,
1:16:02 there are intense pressures to give up too early.
1:16:05 So,
1:16:05 uh,
1:16:06 I think
1:16:07 the historical evidence from,
1:16:08 from the Spanish flu and our model
1:16:10 are consistent with,
1:16:11 with deep concern about doing that.
1:16:14 Um,
1:16:15 dynamics of aggregate consumption,
1:16:18 um,
1:16:19 well,
1:16:20 that's a great question.
1:16:21 Uh,
1:16:21 obviously,
1:16:23 for example,
1:16:23 we don't have investment in the model,
1:16:25 right?
1:16:26 And so agents,
1:16:27 um,
1:16:28 are,
1:16:28 their consumption is mimicking their labor supply decisions very quickly.
1:16:32 Um,
1:16:33 we would love to extend the,
1:16:34 the model.
1:16:36 Including uncertainty about what the government's going to do,
1:16:39 that obviously is a major complication,
1:16:41 but,
1:16:41 but it's something that in principle is good.
1:16:44 On the issue of fiscal transfers,
1:16:46 very interesting question.
1:16:48 Um,
1:16:50 One of the
1:16:52 big differences between this crisis in 2008,
1:16:55 2008,
1:16:56 and I was actually speaking to some bank executives yesterday,
1:16:59 nobody had any clue what the problem was.
1:17:03 We all in this crisis,
1:17:04 really do know what the problem is.
1:17:07 And so one bank executive told me,
1:17:08 he said,
1:17:09 optimal policy in terms of
1:17:11 fiscal transfers,
1:17:12 etc.
1:17:13 is building a bridge to the other side.
1:17:16 What would that bridge mean?
1:17:18 Well,
1:17:18 it means,
1:17:19 for example,
1:17:20 uh,
1:17:20 that we do want to help people,
1:17:22 um,
1:17:23 because we don't think there's a big moral hazard
1:17:25 problem.
1:17:26 I don't.
1:17:27 Uh,
1:17:27 we don't want people becoming
1:17:30 permanently disaffected from the labor force,
1:17:32 and we don't want good firms being broken up.
1:17:34 And none of that's in our model,
1:17:35 of course,
1:17:36 but I think
1:17:37 it is central.
1:17:38 These fiscal transfers are partly,
1:17:40 you want to arrange them in a way so that people continue
1:17:43 a relationship with
1:17:45 their firms
1:17:47 and don't suffer sort of from hysteresis effect
1:17:50 of becoming disaffected
1:17:52 or broken from,
1:17:53 from,
1:17:54 from their employment relationships.
1:17:56 That's obviously way beyond what we can do now,
1:17:58 but I think it's a real world issue is extremely important.
1:18:02 Um,
1:18:04 Art's question about the sensitivity value of life,
1:18:07 really,
1:18:08 and first,
1:18:09 thank you,
1:18:09 Art,
1:18:09 for prodding us to be clear about that in our
1:18:12 private correspondence.
1:18:14 We now have a table in the paper,
1:18:16 which documents sensitivity to various parameters,
1:18:19 because,
1:18:20 you know,
1:18:20 I,
1:18:20 I think there obviously is great uncertainty.
1:18:22 We should add
1:18:24 uh sensitivity to beta,
1:18:26 which actually controls the present value of life.
1:18:29 Uh,
1:18:29 in our model,
1:18:30 and I promise you we will do that.
1:18:31 That's an easy thing for us to do,
1:18:33 and,
1:18:33 uh,
1:18:34 I think we'll,
1:18:35 we'll go some way to at least giving you the information
1:18:38 to,
1:18:38 to assess that sensitivity.
1:18:41 On the incentives to divide containment measures,
1:18:44 um,
1:18:45 Again,
1:18:45 a very,
1:18:46 very good question.
1:18:47 I would say in this model,
1:18:49 people are hand to mouth,
1:18:50 so it is quite painful for them,
1:18:53 right?
1:18:53 These,
1:18:53 these,
1:18:53 these containment measures.
1:18:54 If they're working less,
1:18:56 they're consuming less right away.
1:18:58 So,
1:18:58 so I don't think.
1:19:01 They have a lot of,
1:19:02 it is difficult for them to,
1:19:04 to do things.
1:19:04 Now,
1:19:05 in the real world,
1:19:06 there are lots of people that,
1:19:07 you know,
1:19:08 do really crazy things,
1:19:09 um,
1:19:10 and,
1:19:10 um,
1:19:11 uh,
1:19:12 we're not capturing that,
1:19:13 of course,
1:19:14 because we have rationality.
1:19:16 Uh,
1:19:16 you could go with Stone Geary,
1:19:17 but I,
1:19:17 but I do think it's important,
1:19:18 these people don't,
1:19:19 they really are suffering
1:19:21 from,
1:19:21 from their containment measures instantly.
1:19:24 Um,
1:19:26 Have I gotten everything?
1:19:27 Oh,
1:19:27 on the,
1:19:28 um,
1:19:28 the politics,
1:19:30 the risks of the models.
1:19:31 Oh,
1:19:31 I,
1:19:31 I wanna say,
1:19:33 uh,
1:19:33 I want to endorse Francesco's remarks about
1:19:35 robust control.
1:19:36 I think that's very interesting.
1:19:38 Uh,
1:19:38 I also have a colleague,
1:19:39 Chuck Mansky,
1:19:40 who,
1:19:40 of course,
1:19:40 has made a very important contribution thinking about Bayesian decision making.
1:19:45 Um,
1:19:46 for right now,
1:19:47 I think
1:19:48 robustness,
1:19:49 both in our calculations,
1:19:51 um,
1:19:51 and in thinking of robustness in the Hanson Sergeant way,
1:19:55 would go a long way.
1:19:56 I think policymakers,
1:19:57 and I'm only speaking for myself now,
1:20:00 ought to be erring on the side of saving lives when there's uncertainty,
1:20:03 and,
1:20:04 uh,
1:20:04 that,
1:20:04 that,
1:20:05 that to me seems like the right default concern.
1:20:08 Thank you.
1:20:11 Great,
1:20:11 thanks very much,
1:20:11 Marty.
1:20:12 Um,
1:20:13 uh,
1:20:13 just before I go to Roberto for a moment,
1:20:15 I'll read out one other question that we got from,
1:20:17 uh,
1:20:18 uh,
1:20:18 from the web,
1:20:19 uh,
1:20:20 which is also I think a very interesting real-world one,
1:20:22 and that is how the,
1:20:23 um,
1:20:23 uh,
1:20:24 dynamics of the SI SIR model
1:20:26 are affected by migration
1:20:28 and how do we think about
1:20:30 perturbations in the susceptible population
1:20:32 that,
1:20:33 um,
1:20:34 uh,
1:20:34 that are driven by people migrating in or migrating out of a particular area.
1:20:38 I think,
1:20:39 um,
1:20:40 I'm,
1:20:40 I'm not going beyond the question as it was sent,
1:20:42 but I think this is something that's particularly relevant when you think
1:20:45 about sort of big within-country flows of people that you see,
1:20:48 for example,
1:20:48 in India in response to the
1:20:50 containment measures where you have sort of huge flows from one area to the next as,
1:20:55 uh,
1:20:55 as uh within-country migrant workers have to,
1:20:58 um,
1:20:58 have to move.
1:20:59 Um,
1:21:00 but
1:21:00 just to give you a moment to think about it,
1:21:02 we'll go to Roberto for a minute and then give uh
1:21:05 a couple of minutes each to Francesco and Marty to,
1:21:08 to close us out.
1:21:09 Go ahead,
1:21:10 Roberto.
1:21:10 Thank you.
1:21:11 Thank you,
1:21:11 Art and Ana,
1:21:12 thank you so much.
1:21:14 Um,
1:21:15 so the,
1:21:15 the question that you bring us as usual when we,
1:21:18 we talk about cross support is a difficult one.
1:21:22 so
1:21:23 one way to think about bringing in the digital economy into
1:21:27 our quantitative exercise,
1:21:30 the most sophisticated approach will be like once we get to the,
1:21:35 to the entrepreneurial model with set up and fixed code of operation.
1:21:39 You could think that a digital economy,
1:21:41 a digital economy.
1:21:43 It gives the opportunity of firms to incur in bigger in higher setup costs.
1:21:48 You have to install all this IT and remote connectivity and so on,
1:21:52 but then it would allow you to have a higher priority of survival when,
1:21:56 when the lockdown hits.
1:21:58 So that will be like the sophisticated way to capture that force
1:22:02 and then there will be some parameter combination in the
1:22:06 setup on fixed costs that we can vary across countries to capture how countries are
1:22:10 heterogeneously equipped
1:22:12 to,
1:22:13 to,
1:22:13 you know,
1:22:14 let firms
1:22:16 operate remotely.
1:22:18 Now a more reviews form and quicker response that we
1:22:21 could provide is to add a little bit more,
1:22:24 and this is a more general response to
1:22:26 the sort of naive way in which we are introducing the lockdown shock
1:22:30 as being uniform across the spectrum of firms.
1:22:34 Uh,
1:22:34 one dimension we could look at
1:22:37 some firms are.
1:22:39 As you say,
1:22:40 more equipped to handle the,
1:22:41 the,
1:22:41 the work from home.
1:22:42 So,
1:22:43 uh,
1:22:44 not every firm will suffer the,
1:22:45 the,
1:22:45 the,
1:22:46 the shock in the same manner.
1:22:47 Um,
1:22:48 I,
1:22:48 I just received,
1:22:49 uh,
1:22:49 some,
1:22:50 some data that uh one of my co-authors,
1:22:52 Andy Mayer sent.
1:22:54 Uh,
1:22:54 not every firm is responding equally.
1:22:57 In terms of the labor force to the lockdown,
1:23:00 it seems to me that the smallest firms are
1:23:03 the biggest contributors to job destruction,
1:23:06 so our implementation of the lockdown shock could be more targeted
1:23:10 to certain firms in the
1:23:12 in
1:23:12 the in the population of firms in the economy.
1:23:14 So
1:23:16 given that flexibility in how we bring in the shock
1:23:18 by making it heterogeneous,
1:23:20 that's another way
1:23:21 in which we can bring in.
1:23:23 Again,
1:23:24 in more reduced form a notion of a digital economy to
1:23:28 a model and
1:23:29 to a question about the political implications,
1:23:32 uh,
1:23:32 ours because we don't,
1:23:34 we don't say anything about optimal policy here,
1:23:38 what I do want to advertise is that
1:23:40 at least our framework will allow us,
1:23:43 will allow us to derive the
1:23:45 economic effects of
1:23:47 Various forms of action and inaction.
1:23:49 Of course we have nothing to say about how inaction translates into
1:23:53 death and disease and so on,
1:23:56 but,
1:23:56 you know.
1:23:57 If we don't do nothing with respect to unemployment insurance,
1:24:00 what will happen if we don't do nothing with
1:24:02 respect to alleviating credit constraints in developing countries?
1:24:05 What will happen?
1:24:06 Those are questions that
1:24:08 eventually a policymaker could
1:24:09 draw on the model
1:24:11 to get a sense of what's at stake
1:24:14 in terms of the economic costs of different policies.
1:24:18 Thanks.
1:24:20 Thanks,
1:24:21 Roberto.
1:24:21 Um,
1:24:22 I don't know if,
1:24:22 uh,
1:24:23 Francesco or uh Marty you'd like to say anything on the question about
1:24:26 migration,
1:24:27 which by the way,
1:24:28 came from Gabriela Cugat.
1:24:33 Um,
1:24:33 you're muted,
1:24:34 Marty.
1:24:37 Marty,
1:24:38 you'll have to unmute.
1:24:42 I'm trying to do.
1:24:43 You,
1:24:43 you're,
1:24:44 yeah.
1:24:45 No.
1:24:50 Marty,
1:24:51 Marty,
1:24:51 you're,
1:24:51 you're unmuted.
1:24:53 Oh,
1:24:53 OK,
1:24:53 great.
1:24:54 Um,
1:24:54 I was just going to say one brief thing about migration.
1:24:58 Um,
1:24:58 I am struck,
1:25:00 as usual,
1:25:00 by
1:25:02 how disproportionately poor people are affected
1:25:05 in this,
1:25:06 uh,
1:25:06 pandemic.
1:25:07 Um,
1:25:07 there's migrations,
1:25:09 uh,
1:25:09 across cities,
1:25:10 but
1:25:11 even within cities,
1:25:12 um,
1:25:13 uh,
1:25:13 poor people do have to go to work.
1:25:16 They have to take off in the subway.
1:25:18 Um,
1:25:20 To the extent that we don't alleviate their economic situation in the short run.
1:25:26 Um,
1:25:27 you know,
1:25:27 they will suffer even more than they already are.
1:25:29 The problem is if they're not doing the jobs that they do,
1:25:32 the recessions will be even larger than it is.
1:25:36 So,
1:25:37 I would say migration is a special case of poor people
1:25:40 uh within cities,
1:25:42 across regions,
1:25:43 across countries,
1:25:44 and so I'm sympathetic,
1:25:46 and it goes back to this issue of how do we help people in the short run,
1:25:50 get through this horrible situation
1:25:53 that will inevitably result in a larger recession in the short run.
1:25:57 That,
1:25:57 that,
1:25:57 that's really all I have to say.
1:26:02 Uh,
1:26:03 great,
1:26:03 thanks.
1:26:03 Uh,
1:26:03 uh,
1:26:04 Francesco,
1:26:04 do you want to
1:26:05 chime in on that?
1:26:07 Yeah,
1:26:08 so,
1:26:09 thank you,
1:26:09 uh,
1:26:10 Art.
1:26:11 I just,
1:26:11 I wanna make one remark about,
1:26:13 um,
1:26:14 the issue we discussed before,
1:26:16 the sensitivity to the,
1:26:17 the value of life.
1:26:19 Um,
1:26:20 as I was mentioning,
1:26:21 I think this is,
1:26:22 uh,
1:26:23 really a great question and one of the
1:26:25 really important points if a policymaker has to assess
1:26:29 what is the good policy,
1:26:30 uh,
1:26:31 it has to take a stand on how you balance this,
1:26:34 you know,
1:26:34 lost lives vis a vis.
1:26:36 The,
1:26:36 uh,
1:26:37 the equally difficult to measure economic consequences of,
1:26:40 of the lockdown.
1:26:41 And,
1:26:41 uh,
1:26:42 so in the paper,
1:26:43 we have,
1:26:43 uh,
1:26:44 a preliminary attempt to do that.
1:26:46 If you focus on this middle line in my table,
1:26:49 the ones that we call medium effectiveness,
1:26:51 this is the effectiveness of the lockdown,
1:26:54 how many,
1:26:55 by how much you reduce the contacts between,
1:26:58 uh,
1:26:58 the citizens,
1:26:59 the people,
1:27:00 once you Once they are in lockdown,
1:27:03 so we assume in this medium case that the contacts are cut by 50%
1:27:08 and you can see in these two columns the output cost of the,
1:27:12 of the lockdown policies.
1:27:14 Of course,
1:27:15 with no lockdown,
1:27:16 with no policy whatsoever,
1:27:17 which I think,
1:27:18 by the way,
1:27:18 is a situation
1:27:19 that is not a bad description of the pandemic
1:27:22 of the Spanish flu from the 20s.
1:27:24 It doesn't seem to me that they implemented.
1:27:26 Uh,
1:27:27 big restrictions measures back then.
1:27:28 Rather,
1:27:29 they actually tried to keep it,
1:27:30 uh,
1:27:31 hidden
1:27:31 to the broader public,
1:27:33 uh,
1:27:33 and,
1:27:33 and,
1:27:34 you know,
1:27:34 if you value the life more,
1:27:36 of course,
1:27:36 the cost is bigger,
1:27:37 379 versus
1:27:39 1.89.
1:27:40 And,
1:27:40 and when I say high and low,
1:27:42 I'm referring to estimates that I found in the
1:27:45 statis in the literature on the value of a statistical
1:27:48 life that ranged between 6.5 million to 13 million.
1:27:51 So when we move to a low-end valuation,
1:27:55 the cost of the pandemic,
1:27:57 if unchecked,
1:27:58 is in the order of 1.9% of GDP,
1:28:01 and checking that allows you to reduce it by 60 basis points,
1:28:06 essentially.
1:28:06 The gain is bigger
1:28:08 if you want to attribute,
1:28:09 if you attribute lives.
1:28:11 A bigger value like in the order of 13 million,
1:28:13 then again it's much bigger and you can see it up here.
1:28:16 So again this is
1:28:18 to clarify that your previous question that
1:28:22 it's important to explore the sensitivity of the model results to
1:28:26 key parameters and indeed the value of lives is one of them.
1:28:31 Thank you.
1:28:33 Great,
1:28:33 thanks very much,
1:28:34 Francesco.
1:28:35 You know,
1:28:35 just to wrap up,
1:28:36 there was one other question that I wanted to ask earlier to all of you,
1:28:40 which concerned,
1:28:41 you know,
1:28:41 really thinking about heterogeneous agents in the model,
1:28:44 which,
1:28:44 you know,
1:28:44 Roberto has without the epidemiology,
1:28:47 but the first two papers don't,
1:28:48 and thinking about sort of the distributional consequences.
1:28:52 This is particularly sparked by your Comment,
1:28:54 uh,
1:28:55 just now,
1:28:55 Marty,
1:28:55 about poor people being hurt worse.
1:28:57 Um,
1:28:58 but,
1:28:58 you know,
1:28:59 just to show how fast research is evolving in real time,
1:29:02 just 5 minutes ago,
1:29:03 I was sent a paper,
1:29:05 um,
1:29:06 by Andrew Glover,
1:29:07 Jonathan Heathcote,
1:29:08 Dirk Kruger,
1:29:09 and Jose Victor Rio Ruh that,
1:29:11 um,
1:29:12 uh,
1:29:12 does exactly this.
1:29:13 It's an April 1st paper.
1:29:14 So
1:29:15 tough to keep up these days with how fast things are moving,
1:29:18 uh,
1:29:18 but I guess I'll recommend that to everybody to have a look at.
1:29:21 Um,
1:29:22 so let me wrap up by,
1:29:23 um,
1:29:24 thanking all of you who joined us.
1:29:26 We had about 350 people joining us from YouTube at the peak,
1:29:29 so that was a great audience to reach.
1:29:31 Um,
1:29:32 thanks very much to all of the,
1:29:34 uh,
1:29:34 speakers,
1:29:35 and,
1:29:36 um,
1:29:36 I'm inviting all of you who remain on YouTube.
1:29:39 To come and join us for the next in this uh
1:29:42 uh series of events which is going to be
1:29:44 on Tuesday and focusing on macro policy responses to,
1:29:48 uh,
1:29:48 to COVID-19.
1:29:50 That's led by our colleague,
1:29:51 Norman Loeza.
1:29:52 So hope to see many of you online again
1:29:54 soon and thank you very much for participating today.
1:29:58 Thank you.
1:29:59 Bye.
1:29:59 Thank you.
1:30:00 Thank you.
1:30:00 Goodbye everyone.
- add-style
- lp-body-content