Estimating and Simulating [-3pt] a SIRD Model of...

Estimating and Simulating

a SIRD Model of COVID-19

Jesus Fernandez-Villaverde and Chad Jones

April 22, 2020

(Preliminary and incomplete)

0 / 58

Outline

• Basic model

◦ Social distancing via a time-varying β

• Estimation and simulation

◦ Different countries, U.S. states, and New York City

◦ Robustness to parameters

◦ “Forecasts” from each of the last 7 days

• Re-opening and herd immunity

◦ How much can we relax social distancing?

1 / 58

Basic Model

2 / 58

Notation

• Number of people who are (stocks):

S = Susceptible

I = Infectious

R = Recovered

D = Dead

• Constant population size is N

St + It + Rt + Dt = N

3 / 58

SIRD Model: Overview

• Susceptible get infected at rate βIt/N

New infections = βIt/N · St

• Infections resolve at Poisson rate γ, so the average number of

days until resolution is 1/γ so γ = .2 ⇒ 5 days.

• Resolution happens in one of two ways:

◦ Death: fraction δ

◦ Recovery: fraction 1 − δ

4 / 58

SIRD Model: Laws of Motion

∆St+1 = −βStIt/N︸︷︷︸

new infections

∆It+1 = βStIt/N︸︷︷︸

new infections

− γIt︸︷︷︸

resolving infections

∆Rt+1 = (1 − δ)γIt︸︷︷︸

recover

∆Dt+1 = δγIt︸︷︷︸

R0 = D0 = 0

5 / 58

Recyled notation (terrible) R0: Initial infection rate

• Initial reproduction number R0 ≡ β/γ

R0 = β × 1/γ

# of infections

from one sick

person

# of lengthy

contacts per

# of days

contacts are

infectious

• R0 = expected number of infections via the first sick person

◦ R0 > 1 ⇒ disease initially grows

◦ R0 < 1 ⇒ disease dies out: infectious generate less than 1

new infection

• If 1/γ = 5, then easy to have R0 >> 1

6 / 58

Basic Properties of Differential System (Hethcote 2000)

• Let st ≡ St/N = fraction susceptible

• If R0st > 1, the disease spreads, otherwise declines

• Initial exponential growth rate is β − γ

• As t → ∞, the total fraction of people ever infected, e∗, solves

(assuming s0 ≈ 1)

e∗ = −1

R0log(1 − e∗)

Long run is pinned down by R0 (and death rate),

γ affects timing

7 / 58

Social Distancing

• What about a time-varying infection rate βt?

◦ Disease characteristics – fixed, homogeneous

◦ Regional factors (NYC vs Montana) – fixed, heterogeneous

◦ Social distancing – varies over time and space

• Reasons why βt may change over time

◦ Policy changes on social distancing

◦ Individuals voluntarily change behavior to protect

themselves and others

◦ Superspreaders get infected quickly but then recover and

“burn out” early

8 / 58

Model of Social Distancing

• Assume two key parameters β0 and β∗

• Economy decays exponentially from β0 to β∗ at rate λ:

βt = β0e−λt + β∗(1 − e−λt)

⇒ can think about initial R0 = β0/γ

R0(t) = βt/γ and final R∗

0 = β∗/γ

• Interpretation:

◦ λ governs the rate of convergence to R∗

9 / 58

Estimates and Simulations

10 / 58

Estimation: Countries and States

• Parameters that are fixed and homogeneous

◦ γ = 0.2: average duration is 5 days (or γ = 0.1)

◦ δ = 0.003

Apr 1: 15% of mothers giving birth in NYC infected

With δ = .004, model says only 13.6% ever infected on Apr 1

• Parameters that vary across countries/states

◦ β0 and β∗

◦ λ: speed at which you move to β∗

◦ I0: initial number of infections (gets timing right)

• Objective function:

◦ Equally weighted SSR for Cumulative deaths (logs) and

Daily deaths (logs)11 / 58

Estimation based entirely on death data

• Excess death issue

◦ New York City added 3000+ deaths on April 15 ≈ 45% more

◦ The Economist ⇒ increases based on vital records

⇒ We adjust all NYC deaths before April 15 by this 45%

and non-NYC deaths upwards by 33%

• We use 5-day moving averages (centered)

◦ Otherwise, very serious “weekend effects” in which deaths

are underreported

◦ Even zero sometimes, followed by a large spike

12 / 58

Guide to Graphs

• 7 days of forecasts: Rainbow color order!

ROY-G-BIV (old to new, low to high)

◦ Black=current

◦ Red = oldest, Orange = second oldest, Yellow =third oldest...

◦ Violet (purple) = one day earlier

• For robustness graphs, same idea

◦ Black = baseline (e.g. δ = .003)

◦ Red = lowest parameter value (e.g. δ = .002)

◦ Green = highest parameter value (e.g. δ = .004)

13 / 58

Guide to Graphs (continued)

• R0 in subtitle: R0 / R0(today) / R∗

◦ Initial / Today / Final

• “%Infect”

◦ Today / t+30 / Final

◦ This is the percent ever infected

◦ (so fraction δ will eventually be deaths)

14 / 58

New York City: Cumulative Deaths per Million (δ = .003/.002/.004)

Mar 16 Mar 23 Mar 30 Apr 06 Apr 13 Apr 20 Apr 27

leNew York City (only)

R0=3.4/1.4/0.9 = 0.003, =0.05, =0.2, %Infect=64/70/70

DATA THROUGH 21-APR-2020

δ = .003/.004 fit equally well in levels

δ = .002 cannot fit: 1700 deaths per

million people already means nearly

100% infected!

15 / 58

New York City: Daily Deaths per Million People (δ = .003/.002/.004)

Mar 17 Mar 31 Apr 14 Apr 28 May 12 May 26 Jun 09

New York City (only)

R0=3.4/1.4/0.9 = 0.003, =0.05, =0.2, %Infect=64/70/70

δ = .003/.004 fit equally well

16 / 58

New York City: Cumulative Deaths per Million (δ = .003/.002/.004)

leNew York City (only)

R0=3.4/1.4/0.9 = 0.003, =0.05, =0.2, %Infect=64/70/70

= 0.003

= 0.002

= 0.004

But very different futures!

17 / 58

Spain: Cumulative Deaths per Million People (γ = .2/.1)

Mar 05 Mar 12 Mar 19 Mar 26 Apr 02 Apr 09 Apr 16 Apr 23

lion p

R0=4.6/1.4/0.6 = 0.003, =0.07, =0.2, %Infect=21/23/23

γ = .2 fits slightly better in levels

18 / 58

Spain: Daily Deaths per Million People (γ = .2/.1)

Mar 06 Mar 20 Apr 03 Apr 17 May 01 May 15 May 29

lion p

R0=4.6/1.4/0.6 = 0.003, =0.07, =0.2, %Infect=21/23/23

γ = .2 better fit for daily deaths

19 / 58

Spain: Cumulative Deaths per Million (Future, γ = .2/.1)

Mar 05 Mar 19 Apr 02 Apr 16 Apr 30 May 14 May 28

leSpain

R0=4.6/1.4/0.6 = 0.003, =0.07, =0.2, %Infect=21/23/23

Very different futures!

20 / 58

Italy: Daily Deaths per Million People (γ = .2/.1)

Feb 24 Mar 09 Mar 23 Apr 06 Apr 20 May 04 May 18

lion p

R0=4.4/1.8/0.8 = 0.003, =0.07, =0.2, %Infect=19/22/22

γ = .1 gives a flatter right tail of

deaths

21 / 58

Repeated “Forecasts” from the

past 7 days of data

– After peak, forecasts settle down.

– Before that, very noisy!

22 / 58

Spain (7 days): Daily Deaths per Million People

lion p

R0=4.6/0.7/0.6 = 0.003, =0.07, =0.2, %Infect=21/23/23

23 / 58

Italy (7 days): Daily Deaths per Million People

lion p

R0=4.4/0.9/0.8 = 0.003, =0.07, =0.2, %Infect=19/22/22

24 / 58

New York City (7 days): Daily Deaths per Million People

New York City (plus)

R0=3.5/1.2/0.5 = 0.003, =0.04, =0.2, %Infect=53/59/60

25 / 58

New York City (7 days): Cumulative Deaths per Million (Future)

leNew York City (plus)

R0=3.5/1.2/0.5 = 0.003, =0.04, =0.2, %Infect=53/59/60

26 / 58

California (7 days): Cumulative Deaths per Million

lion p

California

R0=4.0/1.2/1.1 = 0.003, =0.08, =0.2, %Infect= 2/ 6/ 8

27 / 58

California (7 days): Daily Deaths per Million People

lion p

California

R0=4.0/1.2/1.1 = 0.003, =0.08, =0.2, %Infect= 2/ 6/ 8

(noisy and unreliable)

28 / 58

California (7 days): Cumulative Deaths per Million (Future)

leCalifornia

R0=4.0/1.2/1.1 = 0.003, =0.08, =0.2, %Infect= 2/ 6/ 8

29 / 58

U.K. (7 days): Cumulative Deaths per Million

leUnited Kingdom

R0=3.8/1.0/0.5 = 0.003, =0.04, =0.2, %Infect=13/20/20

30 / 58

U.K. (7 days): Daily Deaths per Million People

lion p

United Kingdom

R0=3.8/1.0/0.5 = 0.003, =0.04, =0.2, %Infect=13/20/20

31 / 58

U.K. (7 days): Cumulative Deaths per Million (Future)

leUnited Kingdom

R0=3.8/1.0/0.5 = 0.003, =0.04, =0.2, %Infect=13/20/20

32 / 58

Sweden (7 days): Daily Deaths per Million People

Sweden

R0=3.1/1.0/0.5 = 0.003, =0.05, =0.2, %Infect= 9/13/13

33 / 58

France (7 days): Daily Deaths per Million People

Feb 18 Mar 03 Mar 17 Mar 31 Apr 14 Apr 28 May 12

lion p

France

R0=7.4/0.9/0.5 = 0.003, =0.05, =0.2, %Infect=16/20/20

34 / 58

Cumulative Deaths per Million, Log Scale

0 10 20 30 40 50 60 0.1

Doubles every 10 days, g=7%Doubles e

very 5 days, g=14%

Korea, South

SwedenUnited Kingdom

FranceItalySpain

United States

DAYS SINCE 1 DEATH PER 10 MILLION PEOPLE

CUMULATIVE DEATHS PER MILLION

Italy seems a better guide to France/U.K.

35 / 58

S. Korea (7 days): Daily Deaths per Million People

pleKorea, South

R0=1.5/0.8/0.5 = 0.003, =0.02, =0.2, %Infect= 0/ 0/ 0

36 / 58

Washington (7 days): Daily Deaths per Million People

Mar 03 Mar 17 Mar 31 Apr 14 Apr 28 May 12 May 26

pleWashington

R0=1.8/1.0/0.5 = 0.003, =0.02, =0.2, %Infect= 4/ 6/ 6

37 / 58

Louisiana (7 days): Daily Deaths per Million People

Louisiana

R0=3.1/0.9/0.8 = 0.003, =0.08, =0.2, %Infect=15/20/21

38 / 58

Florida (7 days): Daily Deaths per Million People

lion p

Florida

R0=3.9/1.0/0.5 = 0.003, =0.05, =0.2, %Infect= 2/ 3/ 3

39 / 58

Detriot (Wayne County, 7 days): Daily Deaths per Million People

Mar 21 Apr 04 Apr 18 May 02 May 16 May 30 Jun 13

Detroit

R0=2.9/1.0/0.5 = 0.003, =0.05, =0.2, %Infect=34/41/41

40 / 58

Massachusetts (7 days): Daily Deaths per Million People

Mar 23 Apr 06 Apr 20 May 04 May 18 Jun 01 Jun 15

pleMassachusetts

R0=3.4/1.6/1.6 = 0.003, =0.20, =0.2, %Infect=15/56/63

41 / 58

District of Columbia (7 days): Daily Deaths per Million People

Mar 23 Apr 06 Apr 20 May 04 May 18 Jun 01 Jun 15

District of Columbia

R0=1.8/1.3/0.5 = 0.003, =0.02, =0.2, %Infect= 9/27/30

42 / 58

Reopening and Herd Immunity

43 / 58

Percent Ever Infected would be very informative

— Percent Ever Infected (today) —

δ = .002 δ = .003 δ = .004

New York City (plus) 78 53 41

Detroit 51 34 26

Spain 32 21 16

Italy 28 19 14

Michigan 25 16 12

France 24 16 12

Massachusetts 23 15 12

United Kingdom 20 13 10

Sweden 13 9 6

District of Columbia 13 9 7

New York excluding NYC 8 5 4

Denmark 5 3 2

Germany 4 3 2

Florida 3 2 2

California 3 2 1

Tennessee 2 1 144 / 58

Herd Immunity

• How far can we relax social distancing?

• Let s(t) = S(t)/N = the fraction still susceptible

◦ The disease will die out as long as

R0(t)s(t) < 1

◦ That is, if the “new” R0 is smaller than 1/s(t)

◦ Today’s infected people infect fewer than 1 person on

average

• We can relax social distancing to raise R0(t) to 1/s(t)

45 / 58

Herd Immunity and Opening the Economy?

Percent R0(t+30) PercentSusceptible with no way back

R0 R0(t) t+30 outbreak to normal

New York City 3.5 1.2 40.5 2.5 55.6

Detroit 2.9 1.0 58.7 1.7 34.8

Spain 4.6 0.7 77.4 1.3 15.1

Italy 4.4 0.9 78.4 1.3 10.9

Michigan 3.4 1.0 76.3 1.3 13.3

France 7.4 0.9 79.8 1.3 5.1

Massachusetts 3.4 1.6 43.8 2.3 38.2

United Kingdom 3.8 1.0 80.5 1.2 8.4

Sweden 3.1 1.0 87.0 1.1 8.0

District of Columbia 1.8 1.3 73.4 1.4 4.2

New York excluding NYC 3.5 0.9 92.9 1.1 8.3

Washington 1.8 1.0 93.9 1.1 13.1

Denmark 2.7 0.7 96.4 1.0 16.1

Germany 4.6 0.9 95.3 1.0 3.7

Florida 3.9 1.0 96.6 1.0 2.9

California 4.0 1.2 94.1 1.1 -4.3

Korea, South 1.5 0.8 99.8 1.0 33.646 / 58

Simulations of Re-Opening

• Begin with the basic estimates shown already

• Different policies are then adopted starting around May 20

◦ Black: assumes R0(today) remains in place forever

◦ Red: assumes R0(suppress)= 1/s(today)

◦ Green: we move 25% of the way from R0(today) back to

initial R0 = “normal”

◦ Purple: we move 50% of the way from R0(today) back to

initial R0 = “normal”

• We assume these R0 values stay in place forever

◦ In practice, over course, β would likely start to fall as

mortality rises

47 / 58

Spain: Re-Opening

Apr May Jun Jul Aug Sep

lion p

R0(t)=0.7, R

0(suppress)=1.3, R

0(25/50)=1.7/2.6, = 0.003, =0.07

48 / 58

Italy: Re-Opening

Mar Apr May Jun Jul Aug Sep

lion p

R0(t)=0.9, R

0(suppress)=1.3, R

0(25/50)=1.8/2.6, = 0.003, =0.07

49 / 58

New York City: Re-Opening

lion p

New York City (plus)

R0(t)=1.2, R

0(suppress)=2.5, R

0(25/50)=1.8/2.3, = 0.003, =0.04

NYC has herd immunity and can

re-open significantly in a month!

50 / 58

New York excluding NYC: Re-Opening

Apr May Jun Jul Aug Sep Oct

lion p

New York excluding NYC

R0(t)=0.9, R

0(suppress)=1.1, R

0(25/50)=1.5/2.2, = 0.003, =0.07

But not the rest of NY state!

51 / 58

California: Re-Opening

California

R0(t)=1.2, R

0(suppress)=1.1, R

0(25/50)=1.8/2.4, = 0.003, =0.07

52 / 58

Sweden: Re-Opening

lion p

Sweden

R0(t)=1.0, R

0(suppress)=1.2, R

0(25/50)=1.5/2.0, = 0.003, =0.05

53 / 58

Detroit: Re-Opening

lion p

Detroit

R0(t)=1.0, R

0(suppress)=1.7, R

0(25/50)=1.5/2.0, = 0.003, =0.05

54 / 58

Massachusetts: Re-Opening

lion p

eopleMassachusetts

R0(t)=1.6, R

0(suppress)=2.7, R

0(25/50)=2.0/2.5, = 0.003, =0.20

55 / 58

Washington, DC: Re-Opening

lion p

District of Columbia

R0(t)=1.3, R

0(suppress)=1.4, R

0(25/50)=1.5/1.6, = 0.003, =0.02

56 / 58

Conclusions

Speculations based on model, we are not epidemiologists

• Time-varying β (or R0) needed to capture social distancing, by

individuals or via policy

• New York City already has around 1700 deaths per million

people

◦ So δ > .002 is required by model

◦ 100% are not already infected and resolved

(continued...)

57 / 58

Conclusions (continued)

• Random sampling in NYC would be very informative about δ and

percent ever infected (for herd immunity effects)

◦ NEJM study of 215 mothers giving birth in NYC found a 15

percent infection rate from Mar 22 to Apr 4

◦ Suggests a high current ever-infected rate (infections

doubling every 3-4 days)

◦ Model suggests δ < .004 to match this

◦ California serology testing also suggests δ ≈ .003 (but

problems?)

• “One size fits all” will not work for re-opening

◦ NYC could potentially re-open soon w/ basic precautions

◦ Rest of NY state could not

58 / 58

Estimating and Simulating [-3pt] a SIRD Model of...

Documents

Transcript of Estimating and Simulating [-3pt] a SIRD Model of...

Estimating and Simulating a [-3pt] SIRD Model of COVID-19 ...chadj/Covid/ITA-Extended...(Light bars = New York City, for comparison) 22/39 Italy: Re-Opening (α =0) Mar Apr May Jun

Estimating and Simulating a [-3pt] SIRD Model of COVID-19 ...chadj/Covid/NH-ExtendedResults2.pdfEstimating and Simulating a SIRD Model of COVID-19 for Many Countries, States, and Cities

Estimating and Simulating a [-3pt] SIRD Model of COVID-19 ...chadj/Covid/PER-ExtendedResults.pdf · Estimating and Simulating a [-3pt] SIRD Model of COVID-19 for [-3pt] Many Countries,

Estimating and Simulating a [-3pt] SIRD Model of COVID-19 for [ …chadj/Covid/PER-Extended... · 2020. 10. 11. · Apr May Jun Jul Aug Sep Oct 2020 0 5 10 15 20 25 Daily deaths per

Estimating and Simulating [-3pt] a SIRD Model of COVID-19chadj/slides-covid.pdfEstimating and Simulating a SIRD Model of COVID-19 Jesu´s Fernandez-Villaverde and Chad Jones´ April

Estimating and Simulating a [-3pt] SIRD Model of COVID-19 ...chadj/Covid/CHL-ExtendedResults2.pdfJun 2020 Jul 2020 Aug 2020 Sep 2020 Oct 2020 Nov 2020 Dec 2020 Jan 2021 Feb 2021Mar