Neljapäeval, 20. oktoobril 2022 toimus Eesti Panga avatud seminar, kus rahvusvaheliselt tunnustatud majandusteadlane Fabio Canova tutvustas koos Evi Pappaga valminud uurimustööd „Kulukad looduskatastroofid, energiatarbimine ning eelarvepoliitika“ (Costly disasters, energy consumption, and the role of fiscal policy).
5. Questions
Are disasters associated with
output recessions and
unemployment increases?
How did (fiscal) policy respond?
Were recessions shorter
(smaller) when fiscal policy was
more pro-active? Was local
debt affected?
Should budget rules be
suspended after catastrophic
events? Should there be
budget stabilization funds for
disasters?
Similarities with COVID19
(Ukraine war): potentially large
and persistent strain on local
fiscal balance.
Financial stability issues added
if insurance companies fail.
10. Event study
exercise I
Trace out the spatial distribution of state macro
variables (scaled by the US average, normalized by pre-
desaster level) around disaster dates. Consider:
All events
Events destroying capital stock, affecting agriculture,
hurricanes only.
Costly events: more than 2200 dollars per capita or
more than 5% of GSP (32/33 episodes).
Events occurring in years with more than 10
calamities (1998, 2008, 2011-2013, 2015-2017).
17. IV-Local
projection
exercise
Measure state macro variables responses (scaled by the US
average) to a disaster cost shock:
All events, state by state.
Conditional regression of y(t+h), h=-1,0,1,2,3 on the costs
(instrumented by a cost shock), rather than a dummy.
State conttols included.
Disaster cost shock proxy for an unexpected disruption
of the capital stock. Computed using a lag of costs and a
regional and a US factor
18. Responses to disaster cost shocks, per-capita costs
0 1 2 3 4
GSP
-6
-4
-2
0
2
4
6
10
-3
0 1 2 3 4
Urate
-0.015
-0.01
-0.005
0
0.005
0.01
0.015
0 1 2 3 4
State G
-5
0
5
10
-3
0 1 2 3 4
State Welfare
-0.015
-0.01
-0.005
0
0.005
0.01
0 1 2 3 4
Federal Tansfer
-0.01
-0.005
0
0.005
0.01
0.015
0 1 2 3 4
State Debt
-0.01
-0.005
0
0.005
0.01
0 1 2 3 4
State Revenue
-5
0
5
10
-3
0 1 2 3 4
Personal Income
-6
-4
-2
0
2
4
10
-3
zero mean lower range upper range
20. Responses to disaster costs shocks, per-capita costs, conditional on energy
0 2 4
GSP
-6
-4
-2
0
10
-3
0 2 4
Urate
0
0.01
0.02
0.03
0 2 4
State G
0
5
10
15
10
-3
0 2 4
State Welfare
-4
-2
0
2
10
-3
0 2 4
Federal Tansfer
0
0.02
0.04
0 2 4
State Debt
0
5
10
15
10
-3
0 2 4
State Revenue
0
5
10
10
-3
0 2 4
Personal Income
-4
-2
0
10
-3
0 2 4
e-consumption
-0.02
-0.015
-0.01
-0.005
0
0 2 4
e-residential
-10
-5
0
10
-3
0 2 4
e-commercial
-0.02
-0.01
0
0.01
0.02
0 2 4
e-industrial
-0.04
-0.03
-0.02
-0.01
0
0 2 4
e-transport
-8
-6
-4
-2
0
2
4
10
-3
zero
mean
Lower range
Upper range
21. Which are the states with negative energy responses?
22. Why is energy
consumption
falling in some
states but not in
others?
Investigate three
hypotheses:
• Vulnerability and poor
maintenance of power
facilities.
• Home ownership vs. rental.
• National Flood Insurance
Policy (NFIP) and special state
insurance programs.
30. Does the the
fiscal framework
matter for the
way the state
economy respond
to disasters?
• Tight vs. loose budget (debt)
restrictions states.
• Existence of budget stabilization
(rainy days funds) or not.
46. Event study econometric model
𝑦𝑦𝑖𝑖,𝑡𝑡+ℎ = 𝑎𝑎𝑖𝑖,ℎ𝑥𝑥𝑖𝑖,𝑡𝑡 + 𝑒𝑒𝑖𝑖,𝑡𝑡+ℎ
𝑖𝑖 = state, 𝑡𝑡 =time, ℎ =horizon; 𝑦𝑦𝑖𝑖,𝑡𝑡+ℎ, variables of
interest (scaled relative US average, normalized at t-1
level); 𝑥𝑥𝑖𝑖,𝑡𝑡 is the disaster per-capita (per-gsp) dummy
at time t for state i.
47. Event study
exercise II
Equivalent to running a regression of y(t+h) h=-
1,0,1,2,3 on a dummy for the year that there
is an event, state by state, without constant or
controls.
Why not use diff-in-diff? Units may be treated
or controlled at different time. There is a spatial
and temporal pattern in the treatment (it is
not random).
Post-treatment average dynamics messed
around. (Sun and Abraham, 2021).
53. LP exercise: the econometric model
-
𝑦𝑦𝑖𝑖,𝑡𝑡+ℎ = 𝑎𝑎𝑖𝑖,ℎ + 𝑐𝑐𝑖𝑖,ℎ𝑤𝑤𝑖𝑖,𝑡𝑡−1 +𝑏𝑏𝑖𝑖,ℎ𝑦𝑦𝑖𝑖,𝑡𝑡−1 + 𝑑𝑑𝑖𝑖,ℎ𝑥𝑥𝑖𝑖,𝑡𝑡 + 𝑒𝑒𝑖𝑖,𝑡𝑡+ℎ
𝑖𝑖 = state, 𝑡𝑡 =time, ℎ =horizon; 𝑦𝑦𝑖𝑖,𝑡𝑡+ℎ, variables of interest (scaled relative US average); 𝑥𝑥𝑖𝑖,𝑡𝑡 is
the disaster per-capita cost at time t for state i; 𝑤𝑤𝑖𝑖,𝑡𝑡−1 are controls (at least, scaled state
output, scaled state unemployment).
- Instruments: 𝑧𝑧𝑖𝑖,𝑡𝑡=(1, 𝑤𝑤𝑖𝑖,𝑡𝑡−1,ℎ, 𝑦𝑦𝑖𝑖,𝑡𝑡−1, 𝑢𝑢𝑖𝑖,𝑡𝑡).
- 𝑢𝑢𝑖𝑖,𝑡𝑡 obtained from (𝑓𝑓𝑡𝑡 = spatial factor):
𝑥𝑥𝑖𝑖,𝑡𝑡 = 𝛼𝛼𝑖𝑖 + 𝛽𝛽1𝑖𝑖𝑥𝑥𝑖𝑖𝑖𝑖−1 + 𝛽𝛽2𝑖𝑖𝑓𝑓𝑡𝑡 + 𝑢𝑢𝑖𝑖,𝑡𝑡
- Normal prior on (ai,h, bi,h, , ci,h, di,h) with zero mean and fixed variance (IV ridge estimator).
- Use a diffuse prior on (𝛼𝛼𝑖𝑖, 𝛽𝛽𝑖𝑖).
- Unit by unit estimation (no pooling because of heterogeneity), see Canova (2021).
61. Home insurance industry
41.11
35.49
39.92
41.24
44.39
56.48 56.25
54.08
63.85
0
10
20
30
40
50
60
70
2012 2013 2014 2015 2016 2017 2018 2019 2020
Loss
in
billion
U.S.
dollars
Note(s): United States; 2012 to 2020
Further information regarding this statistic can be found on page 39.
Source(s): Insurance Information Institute; SNL Financial; ID 428998
11
Incurred losses for homeowner insurance in the United States from 2012 to 2020 (in billion U.S.
dollars)
Incurred losses for homeowner insurance in the U.S. 2012-2020