intro: markus brunnermeier...markus’ intro previous/future webinars ramanan laxminarayan:...
TRANSCRIPT
![Page 1: Intro: MARKUS BRUNNERMEIER...Markus’ intro Previous/future webinars Ramanan Laxminarayan: Epidemology models Daron Acemoglu: On the benefits of targeted policies Jeremy Stein Fed-Treasury](https://reader034.vdocument.in/reader034/viewer/2022042411/5f2a63c5c4f00278d4430ec9/html5/thumbnails/1.jpg)
Twitter: @MarkusEconomistIntro: MARKUS BRUNNERMEIER
![Page 2: Intro: MARKUS BRUNNERMEIER...Markus’ intro Previous/future webinars Ramanan Laxminarayan: Epidemology models Daron Acemoglu: On the benefits of targeted policies Jeremy Stein Fed-Treasury](https://reader034.vdocument.in/reader034/viewer/2022042411/5f2a63c5c4f00278d4430ec9/html5/thumbnails/2.jpg)
Markus’ intro Previous/future webinars Ramanan Laxminarayan: Epidemology models Daron Acemoglu: On the benefits of targeted policies Jeremy Stein Fed-Treasury credit programs
Speakers
![Page 3: Intro: MARKUS BRUNNERMEIER...Markus’ intro Previous/future webinars Ramanan Laxminarayan: Epidemology models Daron Acemoglu: On the benefits of targeted policies Jeremy Stein Fed-Treasury](https://reader034.vdocument.in/reader034/viewer/2022042411/5f2a63c5c4f00278d4430ec9/html5/thumbnails/3.jpg)
Ramanan Laxminarayan https://www.youtube.com/watch?v=z1yHjM7szBk
Agent based models to capture behavioral response Expectations play limited role
5/8/2020 3
Anderson & May
cddep.org
Leading textbook
![Page 4: Intro: MARKUS BRUNNERMEIER...Markus’ intro Previous/future webinars Ramanan Laxminarayan: Epidemology models Daron Acemoglu: On the benefits of targeted policies Jeremy Stein Fed-Treasury](https://reader034.vdocument.in/reader034/viewer/2022042411/5f2a63c5c4f00278d4430ec9/html5/thumbnails/4.jpg)
3blue1brown: Grant Sanderson
Random Corporation
5/8/2020 5
Nice simulations
https://www.rand.org/pubs/tools/TLA173-1/tool.htmlhttps://www.youtube.com/watch?v=gxAaO2rsdIs
![Page 5: Intro: MARKUS BRUNNERMEIER...Markus’ intro Previous/future webinars Ramanan Laxminarayan: Epidemology models Daron Acemoglu: On the benefits of targeted policies Jeremy Stein Fed-Treasury](https://reader034.vdocument.in/reader034/viewer/2022042411/5f2a63c5c4f00278d4430ec9/html5/thumbnails/5.jpg)
Ex-ante vs. ex-post targeting
Trade-off: Ex-ante vs. ex-post targeting (flexibility)
German “emergency break” “Regional targeting” in Germany Lockdown region if
More than 50 out of 100,000 inhabitants Infected within a week
Enforceability? Incentive to monitor your neighbor?
5/8/2020 6
![Page 6: Intro: MARKUS BRUNNERMEIER...Markus’ intro Previous/future webinars Ramanan Laxminarayan: Epidemology models Daron Acemoglu: On the benefits of targeted policies Jeremy Stein Fed-Treasury](https://reader034.vdocument.in/reader034/viewer/2022042411/5f2a63c5c4f00278d4430ec9/html5/thumbnails/6.jpg)
Ethics: Statistical Value of Life
Infer from wage difference between hazardous and non hazardous occupations Impacted by risk aversion
Should targeted policy includes statistical value of life
Angus Deaton is critical of this concept (see earlier webinar) Expected life expectancy
Q: Doesn’t matter since only externality (spreading virus) should be taken into account.
5/8/2020 7
![Page 7: Intro: MARKUS BRUNNERMEIER...Markus’ intro Previous/future webinars Ramanan Laxminarayan: Epidemology models Daron Acemoglu: On the benefits of targeted policies Jeremy Stein Fed-Treasury](https://reader034.vdocument.in/reader034/viewer/2022042411/5f2a63c5c4f00278d4430ec9/html5/thumbnails/7.jpg)
Epidemics & Growth
Interact epidemic model with (endogenous) growth model Across different exit strategies/testing Paul Romer’s webinar
5/8/2020 8Brunnermeier, Merkel, Payne, Sannikov (2020) (Tuesday at VMACS)
𝑡𝑡
ln𝐺𝐺𝐺𝐺𝐺𝐺
New chapter for
![Page 8: Intro: MARKUS BRUNNERMEIER...Markus’ intro Previous/future webinars Ramanan Laxminarayan: Epidemology models Daron Acemoglu: On the benefits of targeted policies Jeremy Stein Fed-Treasury](https://reader034.vdocument.in/reader034/viewer/2022042411/5f2a63c5c4f00278d4430ec9/html5/thumbnails/8.jpg)
End of MARKUS’ INTRODUCTORY REMARKS
Now Please ask questions in Q&A box
![Page 9: Intro: MARKUS BRUNNERMEIER...Markus’ intro Previous/future webinars Ramanan Laxminarayan: Epidemology models Daron Acemoglu: On the benefits of targeted policies Jeremy Stein Fed-Treasury](https://reader034.vdocument.in/reader034/viewer/2022042411/5f2a63c5c4f00278d4430ec9/html5/thumbnails/9.jpg)
A Multi-Risk SIR Model withOptimally Targeted Lockdown
Daron Acemoglu Victor Chernozhukov Ivan Werning Michael D. Whinston
April 2020.
![Page 10: Intro: MARKUS BRUNNERMEIER...Markus’ intro Previous/future webinars Ramanan Laxminarayan: Epidemology models Daron Acemoglu: On the benefits of targeted policies Jeremy Stein Fed-Treasury](https://reader034.vdocument.in/reader034/viewer/2022042411/5f2a63c5c4f00278d4430ec9/html5/thumbnails/10.jpg)
Introduction
I SIR models are playing an increasingly central role in understanding andpolicy-making in the context of the COVID-19 pandemic.
I One of the key trade-offs is between economic and public health outcomes.
I But “optimal policy” coming from baseline models assuming homogeneousvulnerability and economic participation may be misleading.
![Page 11: Intro: MARKUS BRUNNERMEIER...Markus’ intro Previous/future webinars Ramanan Laxminarayan: Epidemology models Daron Acemoglu: On the benefits of targeted policies Jeremy Stein Fed-Treasury](https://reader034.vdocument.in/reader034/viewer/2022042411/5f2a63c5c4f00278d4430ec9/html5/thumbnails/11.jpg)
Varieties of Heterogeneity
I Many dimensions of heterogeneity—occupation, productivity, issues related to jointlabor supply.
I Risk factors are particularly important (COVID-19 characterized as two separatediseases by some medical professionals, a deadly one for older populations or thosewith comorbidities, and similar to seasonal flu for younger, healthier groups).
Age Group Mortality rate
20-49 0.00150-64 0.0165+ 0.06
Table: Mortality rates (conditional on infection) from COVID-19.
![Page 12: Intro: MARKUS BRUNNERMEIER...Markus’ intro Previous/future webinars Ramanan Laxminarayan: Epidemology models Daron Acemoglu: On the benefits of targeted policies Jeremy Stein Fed-Treasury](https://reader034.vdocument.in/reader034/viewer/2022042411/5f2a63c5c4f00278d4430ec9/html5/thumbnails/12.jpg)
What We Do
I Develop a multi-risk SIR model and characterize dynamics of infection in thiscontext.I In practice, apply this to a setting with three groups, young, middle-aged and old
(65+).
I Set up an optimal control problem for this environment—allowing targeting by group.
I Characterize and contrast optimal uniform and optimal targeted policies usingparameter values from the literature for COVID-19.
I Clarify how the trade-offs change with targeted policies.
![Page 13: Intro: MARKUS BRUNNERMEIER...Markus’ intro Previous/future webinars Ramanan Laxminarayan: Epidemology models Daron Acemoglu: On the benefits of targeted policies Jeremy Stein Fed-Treasury](https://reader034.vdocument.in/reader034/viewer/2022042411/5f2a63c5c4f00278d4430ec9/html5/thumbnails/13.jpg)
Summary of Main FindingsI Big gains and significantly improved trade-offs from targeted policy.I Most of the gains can be realized by very simple semi-targeted policies that just
treat the 65+ group differentially.
Deaths
outp
ut lo
ss
0
No Control
Maximal Feasible Control
Maximal Fully Effective Control
Deaths
outp
ut lo
ss
0
No Control
Maximal Feasible Control
Maximal Fully Effective Control
Minimal Economic Loss
Optimal Uniform Policies
Optimal Targeted Policies
![Page 14: Intro: MARKUS BRUNNERMEIER...Markus’ intro Previous/future webinars Ramanan Laxminarayan: Epidemology models Daron Acemoglu: On the benefits of targeted policies Jeremy Stein Fed-Treasury](https://reader034.vdocument.in/reader034/viewer/2022042411/5f2a63c5c4f00278d4430ec9/html5/thumbnails/14.jpg)
Important Caveats
I We are not epidemiologists.
I These are really sensitive topics · · ·I There is huge amount of uncertainty about both the disease parameters and the
relevant economic parameters.
I We welcome comments, suggestions and criticisms.
![Page 15: Intro: MARKUS BRUNNERMEIER...Markus’ intro Previous/future webinars Ramanan Laxminarayan: Epidemology models Daron Acemoglu: On the benefits of targeted policies Jeremy Stein Fed-Treasury](https://reader034.vdocument.in/reader034/viewer/2022042411/5f2a63c5c4f00278d4430ec9/html5/thumbnails/15.jpg)
Outline
I1S1R1
Non-ICU
ICUD1
I2S2R2
Non-ICU
ICUD2
NetworkContacts
Figure: MR-SIR: Multiple-Risk Susceptible Infected Recovered Model. Solid lines show theflows from one state to another. Dashed lines emphasize interactions that take place across riskgroups.
![Page 16: Intro: MARKUS BRUNNERMEIER...Markus’ intro Previous/future webinars Ramanan Laxminarayan: Epidemology models Daron Acemoglu: On the benefits of targeted policies Jeremy Stein Fed-Treasury](https://reader034.vdocument.in/reader034/viewer/2022042411/5f2a63c5c4f00278d4430ec9/html5/thumbnails/16.jpg)
Key Equations
I In classic SIR:new infections = βSI .
I In our model, absent lockdowns and isolations:
new infections in group j = βSj
∑k ρjk Ik
(∑
k ρjk(Sk + Ik + Rk))2−α
where {ρjk} are social contact rates between group j and k .
I Here α ∈ [1, 2] is the returns to scale in matching.
I We normalize:∑
j Nj = 1.
I As usual:Sj(t) + Ij(t) + Rj(t) + Dj(t) = Nj .
![Page 17: Intro: MARKUS BRUNNERMEIER...Markus’ intro Previous/future webinars Ramanan Laxminarayan: Epidemology models Daron Acemoglu: On the benefits of targeted policies Jeremy Stein Fed-Treasury](https://reader034.vdocument.in/reader034/viewer/2022042411/5f2a63c5c4f00278d4430ec9/html5/thumbnails/17.jpg)
Infection Dynamics
I Before the vaccine, t ∈ [0,T ), for group j :
Ij = βMj(1− θjLj)Sj
∑k
ρjk(1− θkLk)Ik − γj Ij ,
Mj =
(∑k
ρjk [(Sk + Ik + Rk)(1− θjLk)]
)α−2
,
where ρjk ≥ 0 are contact coefficients and I have assumed notesting/tracing/isolation for the slides (these are allowed in the paper andquantitative analysis below).
I Employment is then given by
Ej(t) = (1− Lj(t))(Sj(t) + Ij(t) + Rj(t)).
![Page 18: Intro: MARKUS BRUNNERMEIER...Markus’ intro Previous/future webinars Ramanan Laxminarayan: Epidemology models Daron Acemoglu: On the benefits of targeted policies Jeremy Stein Fed-Treasury](https://reader034.vdocument.in/reader034/viewer/2022042411/5f2a63c5c4f00278d4430ec9/html5/thumbnails/18.jpg)
Parameters
I γj = δdj (t) + δrj (t): exit rate from infection due to recovery or death.
I δdj (t) = ψj(total infections): probability of death for individual of type j dependingon overcapacity in the hospital system.
I ρij : social contact rate between individuals of group i and j .
I β: infection rate.
I Lj(t): time-varying lockdown.I Lj(t) ≤ Lj ≤ 1, where Lj < 1 allows for “essential” workers.
I θj : effectiveness of lockdown.
I wj : economic contribution of an individual from group j .
I χj : additional non-pecuniary cost of death.
![Page 19: Intro: MARKUS BRUNNERMEIER...Markus’ intro Previous/future webinars Ramanan Laxminarayan: Epidemology models Daron Acemoglu: On the benefits of targeted policies Jeremy Stein Fed-Treasury](https://reader034.vdocument.in/reader034/viewer/2022042411/5f2a63c5c4f00278d4430ec9/html5/thumbnails/19.jpg)
An Aggregation Result
I Our MR-SIR model generalizes the standard SIR model.
I Suppose βjk = β and γj = γ.
I Consider uniform lockdowns Lj(t) = L(t) for all j .
I Suppose further that infection rates are initially identical across groups, so thatSj(0)/Nj , Ij(0)/Nj and Rj(0)/Nj are independent of j .
I Then the model behaves identically to a homogeneous SIR model.I except deaths that naturally vary by group.
![Page 20: Intro: MARKUS BRUNNERMEIER...Markus’ intro Previous/future webinars Ramanan Laxminarayan: Epidemology models Daron Acemoglu: On the benefits of targeted policies Jeremy Stein Fed-Treasury](https://reader034.vdocument.in/reader034/viewer/2022042411/5f2a63c5c4f00278d4430ec9/html5/thumbnails/20.jpg)
Optimal ControlI Social objective:
min
∫ ∞0
e−rt∑j
(wj(Nj − Ej(t)) + χjδdj (t)Ij(t)) dt
I χjδdj (t)Ij(t): non-pecuniary costs of deaths
I With vaccine (and cure) arriving at T , integration-by-parts yields∫ T
0
e−rt∑j
Ψj(t) dt, (1)
where, allowing for partial isolation of the infected and identification of therecovered, the flow cost for group j is given by
Ψj(t) = wjSj(t)Lj(t) + wj Ij(t)(1− ηk(1− Lj(t)))
+wj(1− κj)Rj(t)Lj(t) +
[(1− e−r∆j )wj
r+ χj
]δdj (t)Ij(t).
![Page 21: Intro: MARKUS BRUNNERMEIER...Markus’ intro Previous/future webinars Ramanan Laxminarayan: Epidemology models Daron Acemoglu: On the benefits of targeted policies Jeremy Stein Fed-Treasury](https://reader034.vdocument.in/reader034/viewer/2022042411/5f2a63c5c4f00278d4430ec9/html5/thumbnails/21.jpg)
Parameter Choices
I Fatality rates (conditional on infection) from the table above.
I Ny = 0.53, Nm = 0.26, and No = 0.21.
I wo = 0 and wy = wm = 1.
I Lo = 1 and Lj = 0.7.
I θj = 0.75
I probabilities of detection and isolation in the baseline: φj = τj = 0
I probability of identifying recovered individuals: κj = 1
I γj = 1/18.
I βρij = β = 0.2 (later allow ρij = ρ for i 6= j where ρ = 0.5).
I δdj (t) = δdj · [1 + λ · total infections]. We set λ such that if there is a 30% infectionrate in the overall population, then mortality rates are 5 times the base mortalityrates.
![Page 22: Intro: MARKUS BRUNNERMEIER...Markus’ intro Previous/future webinars Ramanan Laxminarayan: Epidemology models Daron Acemoglu: On the benefits of targeted policies Jeremy Stein Fed-Treasury](https://reader034.vdocument.in/reader034/viewer/2022042411/5f2a63c5c4f00278d4430ec9/html5/thumbnails/22.jpg)
Optimal Uniform Policy
0 200 400
0.00
0.25
0.50
0.75
1.00Lockdown Policy
ymo
Outcomes
Economic Loss 0.2429
Pop. Fatalities 0.0183
Y Fatality Rate 0.0012
M Fatality Rate 0.0116
O Fatality Rate 0.0698
0 200 4000.00
0.05
0.10
0.15Normalized Infection Rates
ymo
Base: Uniform Policy for = 0.75 = 2.0 = 1.0 = 20
I Both high numbers of lives lost (1.8% of adult population) and big economicdamage (24.3% of one year’s GDP).
![Page 23: Intro: MARKUS BRUNNERMEIER...Markus’ intro Previous/future webinars Ramanan Laxminarayan: Epidemology models Daron Acemoglu: On the benefits of targeted policies Jeremy Stein Fed-Treasury](https://reader034.vdocument.in/reader034/viewer/2022042411/5f2a63c5c4f00278d4430ec9/html5/thumbnails/23.jpg)
Optimal Semi-Targeted PolicyI Just applying a more strict lockdown on the oldest age group improves things
significantly.I Overall mortality rate declines from 1.8% to 1%, and economic losses decline from
24% of one year’s GDP to under 13%.
0 200 400
0.00
0.25
0.50
0.75
1.00Lockdown Policy
ymo
Outcomes
Economic Loss 0.1281
Pop. Fatalities 0.0102
Y Fatality Rate 0.0011
M Fatality Rate 0.0113
O Fatality Rate 0.0317
0 200 4000.00
0.05
0.10
0.15Normalized Infection Rates
ymo
Base: SemiTargeted Policy for = 0.75 = 2.0 = 1.0 = 20
![Page 24: Intro: MARKUS BRUNNERMEIER...Markus’ intro Previous/future webinars Ramanan Laxminarayan: Epidemology models Daron Acemoglu: On the benefits of targeted policies Jeremy Stein Fed-Treasury](https://reader034.vdocument.in/reader034/viewer/2022042411/5f2a63c5c4f00278d4430ec9/html5/thumbnails/24.jpg)
Optimal Fully-Targeted PolicyI The young and the middle-age treated differently, but small gains relative to optimal
semi-targeted policy.
0 200 400
0.00
0.25
0.50
0.75
1.00Lockdown Policy
ymo
Outcomes
Economic Loss 0.1268
Pop. Fatalities 0.01
Y Fatality Rate 0.0012
M Fatality Rate 0.01
O Fatality Rate 0.0321
0 200 4000.00
0.05
0.10
0.15Normalized Infection Rates
ymo
Base: Targeted Policy for = 0.75 = 2.0 = 1.0 = 20
![Page 25: Intro: MARKUS BRUNNERMEIER...Markus’ intro Previous/future webinars Ramanan Laxminarayan: Epidemology models Daron Acemoglu: On the benefits of targeted policies Jeremy Stein Fed-Treasury](https://reader034.vdocument.in/reader034/viewer/2022042411/5f2a63c5c4f00278d4430ec9/html5/thumbnails/25.jpg)
The “Pareto” FrontierI The advantage of semi-targeted policies true for different values of χ.
0.00 0.02 0.04Fatalities as Fraction of Population
0.0
0.2
0.4
0.6
0.8
PD
V o
f Eco
nom
ic L
oss
in F
ract
ions
of G
DP
PDV of Economic Loss vs Fatalities
Uniform PolicySemiTargeted PolicyTargeted Policy
Base: Outcomes for = 1, = 2, = 1, T = 546; {0, 5, 10, 20, . . . , 80}
![Page 26: Intro: MARKUS BRUNNERMEIER...Markus’ intro Previous/future webinars Ramanan Laxminarayan: Epidemology models Daron Acemoglu: On the benefits of targeted policies Jeremy Stein Fed-Treasury](https://reader034.vdocument.in/reader034/viewer/2022042411/5f2a63c5c4f00278d4430ec9/html5/thumbnails/26.jpg)
With Greater Value of Life, Another Interesting Contrast
0 200 400
0.00
0.25
0.50
0.75
1.00Lockdown Policy
ymo
Outcomes
Economic Loss 0.3972
Pop. Fatalities 0.0122
Y Fatality Rate 0.0008
M Fatality Rate 0.0077
O Fatality Rate 0.0464
0 200 4000.00
0.05
0.10
0.15Normalized Infection Rates
ymo
Base: Uniform Policy for = 0.75 = 2.0 = 1 = 1.0 = 30
0 200 400
0.00
0.25
0.50
0.75
1.00Lockdown Policy
ymo
Outcomes
Economic Loss 0.1511
Pop. Fatalities 0.0093
Y Fatality Rate 0.001
M Fatality Rate 0.0103
O Fatality Rate 0.0287
0 200 4000.00
0.05
0.10
0.15Normalized Infection Rates
ymo
Base: SemiTargeted Policy for = 0.75 = 2.0 = 1 = 1.0 = 30
![Page 27: Intro: MARKUS BRUNNERMEIER...Markus’ intro Previous/future webinars Ramanan Laxminarayan: Epidemology models Daron Acemoglu: On the benefits of targeted policies Jeremy Stein Fed-Treasury](https://reader034.vdocument.in/reader034/viewer/2022042411/5f2a63c5c4f00278d4430ec9/html5/thumbnails/27.jpg)
Bigger Gains with Between-Group Distancing
0 200 400
0.00
0.25
0.50
0.75
1.00Lockdown Policy
ymo
Outcomes
Economic Loss 0.1759
Pop. Fatalities 0.0134
Y Fatality Rate 0.001
M Fatality Rate 0.0086
O Fatality Rate 0.0505
0 200 4000.00
0.05
0.10
0.15Normalized Infection Rates
ymo
CS1: Uniform Policy for = 0.75 = 2.0 = 0.5 = 20
0 200 400
0.00
0.25
0.50
0.75
1.00Lockdown Policy
ymo
Outcomes
Economic Loss 0.0796
Pop. Fatalities 0.0063
Y Fatality Rate 0.001
M Fatality Rate 0.0086
O Fatality Rate 0.0168
0 200 4000.00
0.05
0.10
0.15Normalized Infection Rates
ymo
CS1: SemiTargeted Policy for = 0.75 = 2.0 = 0.5 = 20
![Page 28: Intro: MARKUS BRUNNERMEIER...Markus’ intro Previous/future webinars Ramanan Laxminarayan: Epidemology models Daron Acemoglu: On the benefits of targeted policies Jeremy Stein Fed-Treasury](https://reader034.vdocument.in/reader034/viewer/2022042411/5f2a63c5c4f00278d4430ec9/html5/thumbnails/28.jpg)
Even Bigger Gains with Testing-Tracing
0 200 400
0.00
0.25
0.50
0.75
1.00Lockdown Policy
ymo
Outcomes
Economic Loss 0.0587
Pop. Fatalities 0.0057
Y Fatality Rate 0.0007
M Fatality Rate 0.0074
O Fatality Rate 0.0161
0 200 4000.00
0.05
0.10
0.15Normalized Infection Rates
ymo
Base with Testing: SemiTargeted Policy for = 0.75 = 2.0 = 1.0 = 20
![Page 29: Intro: MARKUS BRUNNERMEIER...Markus’ intro Previous/future webinars Ramanan Laxminarayan: Epidemology models Daron Acemoglu: On the benefits of targeted policies Jeremy Stein Fed-Treasury](https://reader034.vdocument.in/reader034/viewer/2022042411/5f2a63c5c4f00278d4430ec9/html5/thumbnails/29.jpg)
Combining Between-Group Distancing and Testing-Tracing
0 200 400
0.00
0.25
0.50
0.75
1.00Lockdown Policy
ymo
Outcomes
Economic Loss 0.0203
Pop. Fatalities 0.003
Y Fatality Rate 0.0006
M Fatality Rate 0.0045
O Fatality Rate 0.007
0 200 4000.00
0.05
0.10
0.15Normalized Infection Rates
ymo
Group Dist and Testing: SemiTargeted Policy for = 0.75 = 2.0 = 0.5 = 20
![Page 30: Intro: MARKUS BRUNNERMEIER...Markus’ intro Previous/future webinars Ramanan Laxminarayan: Epidemology models Daron Acemoglu: On the benefits of targeted policies Jeremy Stein Fed-Treasury](https://reader034.vdocument.in/reader034/viewer/2022042411/5f2a63c5c4f00278d4430ec9/html5/thumbnails/30.jpg)
ConclusionI Still much to be done. But our research suggests targeted (semi-targeted) policies
can do much better, especially with some between-group social distancing andtesting-tracing: