discounting future healthcare costs and benefits (part 2)
TRANSCRIPT
Catastrophic risks and the value of health projects
Mark Freeman
Centre for Health Economics, York7th December 2017
Mark Freeman (University of York) Risk & Valuation December 2017 1 / 20
What is appropriate for valuing healthcare in LMICs?
Risk matters
Macroeconomic risk has a different valuation implication(precautionary saving) than project uncertainty (risk premium)
Risk should not be measured by standard deviation alone. Highermoments matter
Low probability, but potentially catastrophic, economic outcomes cansignificantly influence valuations
Risk premiums can be negative, increasing healthcare projectvaluation
National or international perspective?
Mark Freeman (University of York) Risk & Valuation December 2017 2 / 20
Risk-free or risk-adjusted discounting?
Risk-free discounting
Many governments value all social projects using the simple Ramsey rule(e.g. UK for horizons < 30 years). Based on the Arrow-Lind theorem
Risk-adjusted discounting
Others (e.g. French, Dutch and Norwegian), explicitly allow for risk invaluation, as project benefits generally relate to the overallmacro-economy (non-zero ‘beta’).
Central ground
Some argue that risk premiums are too small to concern us. The USdiscounts at 7% and 3% (2.5%, 3% & 5% climate change, 3% health).
Mark Freeman (University of York) Risk & Valuation December 2017 3 / 20
Stylistic example
Binomial process for consumption
Current aggregate consumption c0 = $1, 000. Consumption in 10 years ofcu (probability π) or cd (probability 1 − π).
Parameterizing
We set cu, cd and π so that future (10-year) consumption has anexpectation of $1,200 with a standard deviation of $300.
Vary π, let the mean and standard deviation determine cu and cd .
Mark Freeman (University of York) Risk & Valuation December 2017 4 / 20
The consumption process
0
500
1,000
1,500
2,000
2,500
0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100%
Futu
re c
onsu
mpt
ion
Probability of being in the up state
Consumption in the up state
Consumption in the down state
Low probability of severe consumption catastrophe
Mark Freeman (University of York) Risk & Valuation December 2017 5 / 20
An LMIC healthcare project
Another binomial process
Healthcare project with monetized benefit of bu or bd in ten years:
Pro-cyclical: bu occurs iff cu occurs; probability π
Counter-cyclical: bu occurs iff cd occurs; probability 1 − π
Also look at risk-free and acyclical projects.
Parameterizing
bu and bd set so the expected benefit is $2, with standard deviation to $1.They depend on π.
Mark Freeman (University of York) Risk & Valuation December 2017 6 / 20
Cyclical benefits
-1
0
1
2
3
4
5
0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100%
Bene
fit
Probability of being in the up state
benefit in up state (pro-cyclical)
benefit in down state (pro-cyclical)
Mark Freeman (University of York) Risk & Valuation December 2017 7 / 20
Counter-cyclical benefits - Mirror image
-1
0
1
2
3
4
5
0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100%
Bene
fit
Probability of being in the up state
benefit in up state (counter-cyclical)
benefit in down state (counter-cyclical)
Low probability of consumption catastrophe associated with high project benefits
Mark Freeman (University of York) Risk & Valuation December 2017 8 / 20
Valuing the healthcare project
Equilibrium
Social planner gets time separable utility e−ρtU(ct) from aggregateconsumption. In equilibrium, the present value, p, of the project:
U(c0 − p) + e−ρtE [U(ct + b)] = U(c0) + e−ρtE [U(ct)]
The utility function
Power utility with relative risk aversion = 2, rate of pure time preferenceρ = 1%.
Mark Freeman (University of York) Risk & Valuation December 2017 9 / 20
Present value of the risk-free project
0.00
0.50
1.00
1.50
2.00
2.50
3.00
3.50
0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100%
Pres
ent v
alue
of t
he p
roje
ct
Probability of being in the up state
PV (risk-free)
PV (risk-free, certain consumption)
Precautionary effect increases price of risk-free projects (extended Ramsey Rule)
Precautionary effect gets stronger with catastrophic consumption outcomes (Rietz 1988, Barro 2006 & 2009, Gabaix 2012)
Mark Freeman (University of York) Risk & Valuation December 2017 10 / 20
Present value of the pro-cyclical & acyclical projects
0.00
0.50
1.00
1.50
2.00
2.50
3.00
3.50
0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100%
Pres
ent v
alue
of t
he p
roje
ct
Probability of being in the up state
PV (pro-cyclical)
PV (independent)
PV (risk-free)
Positive (zero) risk premium for positive (zero) beta healthcare project (CCAPM)
Risk premium high in the presence of catastrophic risk (Rietz, 1988; Barro 2006 & 2009; Gabaix 2012)
High sensitivity of valuation to precise likelihood and outcome of catastrophe (Martin, 2013; Gollier, 2016)
Mark Freeman (University of York) Risk & Valuation December 2017 11 / 20
Present value of the counter-cyclical project
0.00
0.50
1.00
1.50
2.00
2.50
3.00
3.50
0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100%
Pres
ent v
alue
of t
he p
roje
ct
Probability of being in the up state
PV (counter-cyclical)
PV (risk-free)
Counter-cyclical projects more highly priced than risk-free assets (CCAPM)
Projects with strong payoffs in catastrophic states are very highly valued (e.g. Weitzman 2007 & 2009, Dietz 2011, Barro 2013, Pindyck 2013). Discount rate here negative.
Mark Freeman (University of York) Risk & Valuation December 2017 12 / 20
A counter-cyclical healthcare project
Healthcare projects in LMICs to protect against economic collapse
Consider an LMIC that, in the event of an economic collapse (lowprobability event), expects a healthcare crisis (famine, malaria outbreak,etc.).
Project to protect against this outcome
Since this project’s benefits are greatest in catastrophic states, it has veryhigh value.
Mark Freeman (University of York) Risk & Valuation December 2017 13 / 20
What is the ‘beta’ of a healthcare project
National or international perspective
The probability of catastrophic economic outcomes, and the relationshipwith healthcare projects, within LMICs are greater/stronger on a nationalthan international perspective.
Proper international support would help mitigate national effects
Therefore, when undertaking these valuations, an assessment needs to bemade of the international policy response to a disaster
Mark Freeman (University of York) Risk & Valuation December 2017 14 / 20
Risk of future pandemic
Uncertain future risk
The number of deaths (as % population), N, in a future pandemic isunknown. Both the policy maker and health expert think N is lognormallydistributed.
Different parameter values
E [N] = m for the policy maker and E [N] = M > m for the expert.Var [N] = s2 for the policy maker and Var [N] = S2 for the expert.
Mark Freeman (University of York) Risk & Valuation December 2017 15 / 20
How does the policy maker respond to the expert advice
Bayesian learning
The policy maker hears the opinion of the expert and rationally Bayesianupdates her beliefs (e.g., French 1985; Lindley 1985; Genest and Zidek1986).
Bayesian posterior
The Bayesian posterior of the expert is also lognormally distributed in N.The mean and variance of this distribution, m′, s ′2 are known in closedform as functions of m,M, s,S (messy!)
Mark Freeman (University of York) Risk & Valuation December 2017 16 / 20
Example
m = 2%, s = 2%,M = 3.2%, S = 2%. Then m′ = 2.46% and s ′ = 1.23%
0
0.1
0.2
0.3
0.4
0.5
0 1 2 3 4 5 6 7 8 9 10
Prob
abili
ty D
ensi
ty F
unct
ion
N
Prior
Consensus
Posterior
Mark Freeman (University of York) Risk & Valuation December 2017 17 / 20
Economic damages
Consumption drops from ct to (1 − D(N))ct at some future time t in theevent of a pandemic. D(N) = 1 − exp(−θN2) for θ = 0.006585
0%
10%
20%
30%
40%
50%
60%
0 1 2 3 4 5 6 7 8 9 10
Econ
omic
dam
ages
, D(N
)
Percentage of population killed, N
Mark Freeman (University of York) Risk & Valuation December 2017 18 / 20
Main result
Expert opinion can result in decreasing WTP
After receiving expert opinion, a statistically and economically rationalpolicy maker with logarithmic utility will increase her expected value of N,but reduce her willingness to pay to avoid the potential pandemic if andonly if
M4
M2 + S2− s2 < m2 <
M4
M2 + S2
The example
In the example, m′ = 2.46% > m = 2%, yet the WTP drops from 5.13%to 4.86% of current income.
Mark Freeman (University of York) Risk & Valuation December 2017 19 / 20
Psychological evidence
Climate change communication reduces skepticism...
Expert communication is successful in reducing overall levels of climatechange skepticism through knowledge transfer (e.g. Lewandowsky, Gignac& Vaughan 2013, van der Linden et al. 2015; Ramney & Clark 2016)
... but not WTP
... but people become no more willing to pay for preventative action afterreceiving this communication (e.g. Deryugina & Shurchkov 2016; Kahan2016; Bolsen & Druckman 2016)
Mark Freeman (University of York) Risk & Valuation December 2017 20 / 20