Slides Covid PDF

Estimating and Simulating
a SIRD Model of COVID-19
Jesús Fernández-Villaverde and Chad Jones
April 13, 2020

(Very preliminary and incomplete)
0 / 46
Outline
• Basic model
◦ Social distancing via a time-varying β
• Estimation and simulation
◦ Different countries and states

◦ Robustness to parameters
◦ “Forecasts” from each of the last 7 days
◦ Standard errors coming soon
• How much can we relax social distancing?
1 / 46
Basic Model
2 / 46
Notation
• Number of people who are (stocks):
S = Susceptible
I = Infective
R = Recovered
D = Dead
• Constant population size is N
St + It + Rt + Dt = N
3 / 46
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 / 46
SIRD Model: Laws of Motion
∆St+1 = −βSt It /N
| {z }
new infections
∆It+1 = βSt It /N − γIt

| {z } |{z}
new infections resolving infections
∆Rt+1 = (1 − δ)γIt
| {z }
recover
∆Dt+1 = δγIt
|{z}
die
R0 = D0 = 0
5 / 46
New notation (terrible) R0 : Initial infection rate
• Let s0 ≡ S0 /N = fraction susceptible
• Terrible recycled notation: “Reproduction number” R0 ≡ β/γ
◦ β = rate at which you get the virus from one infective person
◦ 1/γ = average time infected
◦ So R0 = expected number of infections generated by one
sick person when no herd immunity (s0 ≈ 1)
6 / 46
Basic Properties of Differential System (Hethcote 2000)
• If R0 s0 > 1, the disease spreads, otherwise declines immediately
• Initial exponential growth rate is β − γ
• As t → ∞, the total fraction of people ever infected, e∗ , solves

(assuming s0 ≈ 1)
1
e∗ = − log(1 − e∗ )
R0
Long run is pinned down by R0 (and death rate),

γ affects timing
7 / 46
Social Distancing
• What about a time-varying infection rate βt ?
◦ Disease characteristics – fixed, homogeneous

◦ Regional characteristics (NYC vs Montana) – fixed,
heterogeneous
◦ Social distancing – varies over time and space
• Model: assume two key parameters β0 and β ∗
• Economy decays exponentially from β0 to β ∗ at rate λ:
βt = β0 e−λt + β ∗ (1 − e−λt )
⇒ can think about initial R0 = β0 /γ

and final R∗0 = β ∗ /γ
8 / 46
Estimates and Simulations
9 / 46
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 (Heinsberg, Germany random sampling)
◦ λ = 7%: βt falls halfway to new value each 10 days
• Parameters that vary across countries/states
◦ β0 and β ∗
◦ I0 : initial number of infections (gets timing right)
• Objective function:
◦ Equally weighted SSR for Cumulative deaths (logs) and

Daily deaths (logs)
10 / 46
Guide to Graphs
• 7 days of forecasts: 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 = .001)
◦ Green = highest parameter value (e.g. δ = .005)
11 / 46
New York: Cumulative Deaths per Million People (δ = .003/.001/.005)
New York
R0=4.2,0.9 = 0.003, =0.07, =0.2, %Infected(today)=21
450
THRU APR 10
400
Cumulative deaths per million people
350
Three estimates (δ = .003/.001/.005)
300 fit equally well in levels
250
200
150
100
50
0
Mar 14 Mar 17 Mar 20 Mar 23 Mar 26 Mar 29 Apr 01 Apr 04 Apr 07 Apr 10
2020
12 / 46
New York: Daily Deaths per Million People (δ = .003/.001/.005)
New York
R0=4.2,0.9 = 0.003, =0.07, =0.2, %Infected(t+30)=42
50
THRU APR 10
45
40
Different futures for daily

Daily deaths per million people
35
deaths
30
25
20
15
10
0
Mar 15 Mar 29 Apr 12 Apr 26 May 10 May 24 Jun 07
2020
13 / 46
New York: Cumulative Deaths per Million (Future, δ = .003/.001/.005)
New York
R0=4.2,0.9 = 0.003, =0.07, =0.2, %Infected(end)=44 = 0.005
1500
THRU APR 10
= 0.003
1000
= 0.001
500
But very different futures!
0
2020
14 / 46
Italy: Cumulative Deaths per Million People (λ = .07/.05/.10)
Italy
R0=5.3,0.7 = 0.003, =0.07, =0.2, %Infected(today)=12
350
THRU APR 10
300
250
Three estimates (λ = .07/.05/.10) fit
pretty well in levels
200
150
100
50
0
Feb 21 Feb 28 Mar 06 Mar 13 Mar 20 Mar 27 Apr 03 Apr 10
2020
15 / 46
Italy: Daily Deaths per Million People (λ = .07/.05/.10)
Italy
R0=5.3,0.7 = 0.003, =0.07, =0.2, %Infected(t+30)=15
16
THRU APR 10
14
12 λ = .07/.05 better fit for daily

deaths
10
0
Feb 22 Mar 07 Mar 21 Apr 04 Apr 18 May 02 May 16
2020
16 / 46
Italy: Cumulative Deaths per Million (Future, λ = .07/.05/.10)
Italy
R0=5.3,0.7 = 0.003, =0.07, =0.2, %Infected(end)=15
600
THRU APR 10
= 0.1
500
= 0.07
= 0.05
400
300
200
Slightly different futures!

100
0
2020
17 / 46
Spain: Cumulative Deaths per Million People (γ = .2/.1)
Spain
R0=5.9,0.3 = 0.003, =0.07, =0.2, %Infected(today)=14
400
THRU APR 10
350
300
γ = .2 fits slightly better in levels
250
200
150
100
50
0
Mar 03 Mar 10 Mar 17 Mar 24 Mar 31 Apr 07 Apr 14
2020
18 / 46
Spain: Daily Deaths per Million People (γ = .2/.1)
Spain
R0=5.9,0.3 = 0.003, =0.07, =0.2, %Infected(t+30)=16
25
THRU APR 10
20
γ = .2 better fit for daily deaths

15
10
0
Mar 04 Mar 18 Apr 01 Apr 15 Apr 29 May 13 May 27
2020
19 / 46
Spain: Cumulative Deaths per Million (Future, γ = .2/.1)
Spain
R0=5.9,0.3 = 0.003, =0.07, =0.2, %Infected(end)=16
800 = 0.1
THRU APR 10
700
600
500 = 0.2
400
300
200
Very different futures!
100
0
Mar 03 Mar 17 Mar 31 Apr 14 Apr 28 May 12 May 26
2020
20 / 46
Repeated “Forecasts” from the
past 7 days of data
– After peak, forecasts settle down.

– Before that, very noisy!
21 / 46
Spain (7 days): Daily Deaths per Million People
Spain
R0=5.9,0.3 = 0.003, =0.07, =0.2, %Infected(t+30)=16
25
THRU APR 10
20
15
10
0
2020
22 / 46
Italy (7 days): Daily Deaths per Million People
Italy
R0=5.3,0.7 = 0.003, =0.07, =0.2, %Infected(t+30)=15
16
THRU APR 10
14
12
10
0
2020
23 / 46
New York (7 days): Daily Deaths per Million People
New York
R0=4.2,0.9 = 0.003, =0.07, =0.2, %Infected(t+30)=42
60
THRU APR 10
50
40
30
20
10
0
2020
24 / 46
New York (7 days): Cumulative Deaths per Million (Future)
New York
R0=4.2,0.9 = 0.003, =0.07, =0.2, %Infected(end)=44
1600
THRU APR 10
1400
1200
1000
800
600
400
200
0
2020
25 / 46
California (7 days): Cumulative Deaths per Million
California
R0=4.0,1.0 = 0.003, =0.07, =0.2, %Infected(today)= 1
18
THRU APR 10
16
14
12
10
0
Mar 04 Mar 11 Mar 18 Mar 25 Apr 01 Apr 08 Apr 15
2020
26 / 46
California (7 days): Daily Deaths per Million People
California
R0=4.0,1.0 = 0.003, =0.07, =0.2, %Infected(t+30)= 3
20
THRU APR 10
18
(noisy and unreliable)
16
14
12
10
0
2020
27 / 46
California (7 days): Cumulative Deaths per Million (Future)
California
R0=4.0,1.0 = 0.003, =0.07, =0.2, %Infected(end)= 5
450
THRU APR 10
400 (noisy and unreliable)

350
300
250
200
150
100
50
0
2020
28 / 46
U.K. (7 days): Cumulative Deaths per Million
U.K.
R0=3.6,1.5 = 0.003, =0.07, =0.2, %Infected(today)= 8
200
THRU APR 10
180
160
140
120
100
80
60
40
20
0
Mar 05 Mar 12 Mar 19 Mar 26 Apr 02 Apr 09 Apr 16
2020
29 / 46
U.K. (7 days): Daily Deaths per Million People
U.K.
R0=3.6,1.5 = 0.003, =0.07, =0.2, %Infected(t+30)=51
120
THRU APR 10

100
80
60
40
20
0
Mar 06 Mar 20 Apr 03 Apr 17 May 01 May 15 May 29
2020
30 / 46
U.K. (7 days): Cumulative Deaths per Million (Future)
U.K.
R0=3.6,1.5 = 0.003, =0.07, =0.2, %Infected(end)=60
2500
THRU APR 10
2000
1500
1000
500
0
2020
31 / 46
Sweden (7 days): Daily Deaths per Million People
Sweden
R0=3.2,1.3 = 0.003, =0.07, =0.2, %Infected(t+30)=32
40
THRU APR 10
30
25
20
15
10
0
2020
32 / 46
France (7 days): Daily Deaths per Million People
France
R0=4.4,1.7 = 0.003, =0.07, =0.2, %Infected(t+30)=66
100
THRU APR 10

80
70
60
50
40
30
20
10
0
Feb 16 Mar 01 Mar 15 Mar 29 Apr 12 Apr 26 May 10
2020
33 / 46
Cumulative Deaths per Million, Log Scale
CUMULATIVE DEATHS PER MILLION
Spain Italy
256 France DATA THROUGH 12-APR-2020
U.K.
128
3%
35%
64 U.S. Hubei, China
=2
Iran
s, g
s, g
ay
32 day
3d
%
14
ry 2
ery
16
ev , g=
ys
eve
da
les
8 5
bles
ub
ry
ve
Do
Dou
e S. Korea
4 es
ubl
Do , g=
7%
2 ays
d
y 10
ver
1 es e
ubl
Do
0.5
Italy seems a better guide to France/U.K.

0.1
0 10 20 30 40 50 60 70 80 90
DAYS SINCE 1 DEATH PER 10 MILLION PEOPLE
34 / 46
Hubei, China (7 days): Daily Deaths per Million People
Hubei, China
R0=2.5,0.5 = 0.003, =0.07, =0.2, %Infected(t+30)= 2
4.5
THRU APR 10
3.5
2.5
1.5
0.5
0
Jan 23 Feb 06 Feb 20 Mar 05 Mar 19 Apr 02 Apr 16
2020
35 / 46
S. Korea (7 days): Daily Deaths per Million People
S. Korea
R0=1.8,0.9 = 0.003, =0.07, =0.2, %Infected(t+30)= 0
0.25
THRU APR 10
0.2
0.15
0.1
0.05
0
2020
36 / 46
Washington (7 days): Daily Deaths per Million People
Washington
R0=1.4,1.4 = 0.003, =0.07, =0.2, %Infected(t+30)=19
40
THRU APR 10
30
25
20
15
10
0
2020
37 / 46
Louisiana (7 days): Daily Deaths per Million People
Louisiana
R0=3.5,0.6 = 0.003, =0.07, =0.2, %Infected(t+30)=13
16
THRU APR 10
14
12
10
0
2020
38 / 46
Florida (7 days): Daily Deaths per Million People
Florida
R0=4.0,1.3 = 0.003, =0.07, =0.2, %Infected(t+30)=11
80
THRU APR 10
60
50
40
30
20
10
0
Mar 07 Mar 21 Apr 04 Apr 18 May 02 May 16 May 30
2020
39 / 46
Michigan (7 days): Daily Deaths per Million People
Michigan
R0=3.7,0.5 = 0.003, =0.07, =0.2, %Infected(t+30)=13
25
THRU APR 10

20
15
10
0
Mar 19 Apr 02 Apr 16 Apr 30 May 14 May 28 Jun 11
2020
40 / 46
Massachusetts (7 days): Daily Deaths per Million People
Massachusetts
R0=2.4,1.6 = 0.003, =0.07, =0.2, %Infected(t+30)=55
70
THRU APR 10

50
40
30
20
10
0
Mar 21 Apr 04 Apr 18 May 02 May 16 May 30 Jun 13
2020
41 / 46
Assessing Uncertainty
41 / 68
How Do We Evaluate Uncertainty?
• Basic SIRD model is deterministic
◦ There is an implicit appeal to a law of large numbers holding

in populations.
◦ How do we map it into data?
◦ Measurement error (huge undercount of deaths in Italy and
Spain, massive under-reporting of infections). However,
unlikely to be classical measurement error.
◦ Un-modeled shocks (we have not specified them explicitely).
• Thus, assessing uncertainty is not straightforward.
42 / 68
Our Approach
• A simple empirical Bayesian method

◦ Specify a prior on β, γ, and δ centered around our point
estimates and having a variance based on plausibility/our
reading of the medical literature.
◦ You sample from the prior and forecast the future behavior of
the model based on the sampled parameter values.
◦ Informative about the properties of the model and range of
likely outcomes
• We tried to compute standard errors.

◦ We got numerically unstable values.
◦ However, our provisional values are within the range of our
priors (if anything, our computed standard errors feel “too
small”).
43 / 68
R0 is key for dynamics...
U.S., uncertainty regarding R 0

R0=2.4,1.3 = 0.003, =0.07, =0.2, %Infected(today)= 3
1600
R0 today = 1.1
R0 today = 1.3
1400 R0 today = 1.4
R0 today (point estimate) = 1.6
R0 today = 1.7
Cumulate deaths per million people
1200 R0 today = 1.9

R0 today = 2.0 10-Apr-2020
1000
800
600
400
200
0
2020
44 / 68
New York, uncertainty regarding R 0

R0=2.4,0.9 = 0.003, =0.07, =0.2, %Infected(today)=21
1800
1600
1400
10-Apr-2020
1200
1000
800
600
R0 today = 0.9
R0 today = 1.1
400
R0 today = 1.2
R0 today (point estimate) = 1.4
R0 today = 1.5
200
R0 today = 1.7
R0 today = 1.8
0
2020
45 / 68
California, uncertainty regarding R 0

R0=2.4,1.0 = 0.003, =0.07, =0.2, %Infected(today)= 1
300
R0 today = 0.8
R0 today = 1.0
R0 today = 1.1
250 R0 today (point estimate) = 1.3
R0 today = 1.4
R0 today = 1.6
R0 today = 1.7
200
150
10-Apr-2020
100
50
0
2020
46 / 68
Spain, uncertainty regarding R 0

R0=2.4,0.3 = 0.003, =0.07, =0.2, %Infected(today)=14
600
500
10-Apr-2020
400
300
200
R0 today = 0.3
R0 today = 0.4
R0 today = 0.6
100 R0 today (point estimate) = 0.7
R0 today = 0.9
R0 today = 1.0
R0 today = 1.2
0
2020
47 / 68
γ not so much
U.S., uncertainty regarding keeping R 0 constant

R0=2.4,1.3 = 0.003, =0.07, =0.2, %Infected(today)= 3
1400
= 0.170
= 0.180
= 0.190
1200
= 0.200 10-Apr-2020
= 0.210
= 0.220
1000 = 0.230
800
600
400
200
0
2020
48 / 68
γ not so much
New York, uncertainty regarding keeping R 0 constant

R0=2.4,0.9 = 0.003, =0.07, =0.2, %Infected(today)=21
1400
10-Apr-2020
1200
1000
800
600
400 = 0.170
= 0.180
= 0.190
= 0.200
200 = 0.210
= 0.220
= 0.230
0
2020
49 / 68
γ not so much
California, uncertainty regarding keeping R 0 constant

R0=2.4,1.0 = 0.003, =0.07, =0.2, %Infected(today)= 1
150
= 0.170
= 0.180
= 0.190
= 0.200 10-Apr-2020
= 0.210
= 0.220
= 0.230
100
50
0
2020
50 / 68
γ not so much
Spain, uncertainty regarding keeping R 0 constant

R0=2.4,0.3 = 0.003, =0.07, =0.2, %Infected(today)=14
500
10-Apr-2020
450
400
350
300
250
200
150
= 0.170
= 0.180
100 = 0.190
= 0.200
= 0.210
50 = 0.220
= 0.230
0
2020
51 / 68
δ matters for deaths, not for dynamics
U.S., uncertainty regarding

R0=2.4,1.3 = 0.003, =0.07, =0.2, %Infected(today)= 3
2500
= 0.000
= 0.001
= 0.002
= 0.003
2000 = 0.004
= 0.005
= 0.006
1500
10-Apr-2020
1000
500
0
2020
52 / 68
New York, uncertainty regarding

R0=2.4,0.9 = 0.003, =0.07, =0.2, %Infected(today)=21
2500
= 0.000
= 0.001
= 0.002
= 0.003
2000 = 0.004
= 0.005
= 0.006
1500
10-Apr-2020
1000
500
0
2020
53 / 68
California, uncertainty regarding

R0=2.4,1.0 = 0.003, =0.07, =0.2, %Infected(today)= 1
250
= 0.000
= 0.001
= 0.002
= 0.003
200 = 0.004
= 0.005
= 0.006
150
10-Apr-2020
100
50
0
2020
54 / 68
Spain, uncertainty regarding

R0=2.4,0.3 = 0.003, =0.07, =0.2, %Infected(today)=14
700
600
500
10-Apr-2020
400
300
200 = 0.000
= 0.001
= 0.002
= 0.003
100 = 0.004
= 0.005
= 0.006
0
2020
55 / 68
We can plot joint draws
U.S., uncertainty regarding and

R0=2.4,1.3 = 0.003, =0.07, =0.2, %Infected(today)= 3
2000
1800
1600
1400
1200
10-Apr-2020
1000
800
600
400
200
0
2020
56 / 68
Policy Counterfactuals
57 / 68
Reducing R0 has a huge impact...
U.S., infected reducing R 0 at the end of observations

R0=2.4,1.3 = 0.003, =0.07, =0.2, %Infected(today)= 3
6
R0 estimated on 10-Apr-2020 is 1.6
R0 lowered to 1.3
R0 lowered to 0.9
5 R0 lowered to 0.6
R0 lowered to 0.3
4
Infected, percentage
10-Apr-2020
0
2020
58 / 68
New York, infected reducing R 0 at the end of observations

R0=2.4,0.9 = 0.003, =0.07, =0.2, %Infected(today)=21
8
R0 lowered to 1.1
7 R0 lowered to 0.8
R0 lowered to 0.5
R0 lowered to 0.3
6
10-Apr-2020
0
2020
59 / 68
California, infected reducing R 0 at the end of observations

R0=2.4,1.0 = 0.003, =0.07, =0.2, %Infected(today)= 1
0.45 10-Apr-2020

R0 lowered to 1.0
0.4 R0 lowered to 0.8
R0 lowered to 0.5
R0 lowered to 0.3
0.35
0.3
0.25
0.2
0.15
0.1
0.05
0
2020
60 / 68
...but it has a decreasing marginal effects on cumulate deaths
U.S., ever infected reducing R 0 at the end of observations

R0=2.4,1.3 = 0.003, =0.07, =0.2, %Infected(today)= 3
1200
R0 estimated on 10-Apr-2020 is 1.6 10-Apr-2020
R0 lowered to 1.3
R0 lowered to 0.9
1000 R0 lowered to 0.6
R0 lowered to 0.3
800
600
400
200
0
2020
61 / 68
New York, ever infected reducing R 0 at the end of observations

R0=2.4,0.9 = 0.003, =0.07, =0.2, %Infected(today)=21
1400
10-Apr-2020
1200
1000
800
600
400

R0 lowered to 1.1
200 R0 lowered to 0.8
R0 lowered to 0.5
R0 lowered to 0.3
0
2020
62 / 68
California, ever infected reducing R 0 at the end of observations

R0=2.4,1.0 = 0.003, =0.07, =0.2, %Infected(today)= 1
140
R0 lowered to 1.0
R0 lowered to 0.8 10-Apr-2020
120
R0 lowered to 0.5
R0 lowered to 0.3
100
80
60
40
20
0
2020
63 / 68
Reopening and Herd Immunity
42 / 46
Percent Ever Infected would be very informative
— Percent Ever Infected (today) —

δ = .001 δ = .003 δ = .005
New York 58 21 13
California 2 1 0
Italy 35 12 7
Spain 40 14 8
U.K. 22 8 5
France 34 13 8
Hubei, China 5 2 1
S. Korea 0 0 0
Sweden 14 5 3
Louisiana 22 7 4
Florida 3 1 1
Washington 8 3 2
Michigan 19 6 4
Massachusetts 15 5 3
43 / 46
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)
44 / 46
Herd Immunity and Opening the Economy?
Percent R0 (t+30) Percent
Susceptible with no way back
R0 R∗0 t+30 outbreak to normal
New York 4.2 0.9 57.7 1.7 26.0

California 4.0 1.0 96.9 1.0 -0.3
Hubei, China 2.5 0.5 98.2 1.0 25.0
Italy 5.3 0.7 85.3 1.2 10.0
Spain 5.9 0.3 84.1 1.2 15.3
U.K. 3.6 1.5 49.1 2.0 25.8
France 4.4 1.7 34.3 2.9 45.0
S. Korea 1.8 0.9 99.8 1.0 11.0
Sweden 3.2 1.3 68.5 1.5 8.8
Louisiana 3.5 0.6 87.0 1.1 18.7
Florida 4.0 1.3 89.1 1.1 -5.3
Washington 1.4 1.4 81.3 1.2 -456.2
Michigan 3.7 0.5 86.7 1.2 19.8
Massachusetts 2.4 1.6 44.7 2.2 76.7
45 / 46
Conclusion
Thanks!
46 / 46

Slides Covid PDF

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Slides Covid PDF

Uploaded by

Copyright:

Available Formats

Estimating and Simulating

a SIRD Model of COVID-19

Jesús Fernández-Villaverde and Chad Jones

April 13, 2020

◦ Social distancing via a time-varying β

• Estimation and simulation

◦ Different countries and states

• How much can we relax social distancing?

• Number of people who are (stocks):

• Constant population size is N

• Susceptible get infected at rate βIt /N

New infections = βIt /N · St

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

• Resolution happens in one of two ways:

∆It+1 = βSt It /N − γIt

• Let s0 ≡ S0 /N = fraction susceptible

• Terrible recycled notation: “Reproduction number” R0 ≡ β/γ

• If R0 s0 > 1, the disease spreads, otherwise declines immediately

• Initial exponential growth rate is β − γ

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

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

• What about a time-varying infection rate βt ?

◦ Disease characteristics – fixed, homogeneous

• Model: assume two key parameters β0 and β ∗

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

⇒ can think about initial R0 = β0 /γ

• Parameters that are fixed and homogeneous

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

• Parameters that vary across countries/states

◦ Equally weighted SSR for Cumulative deaths (logs) and

• 7 days of forecasts: ROY-G-BIV (old to new, low to high)

• For robustness graphs, same idea

◦ Black = baseline (e.g. δ = .003)

Different futures for daily

But very different futures!

12 λ = .07/.05 better fit for daily

Slightly different futures!

γ = .2 better fit for daily deaths

– After peak, forecasts settle down.

400 (noisy and unreliable)

(noisy and unreliable)

(noisy and unreliable)

35 (noisy and unreliable)

90 (noisy and unreliable)

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

35 (noisy and unreliable)

70 (noisy and unreliable)

(noisy and unreliable)

60 (noisy and unreliable)

• Basic SIRD model is deterministic

◦ There is an implicit appeal to a law of large numbers holding

• Thus, assessing uncertainty is not straightforward.

• A simple empirical Bayesian method

• We tried to compute standard errors.

U.S., uncertainty regarding R 0

1200 R0 today = 1.9

New York, uncertainty regarding R 0

California, uncertainty regarding R 0

Spain, uncertainty regarding R 0

U.S., uncertainty regarding keeping R 0 constant

New York, uncertainty regarding keeping R 0 constant

California, uncertainty regarding keeping R 0 constant

Spain, uncertainty regarding keeping R 0 constant