Optimised Regret Avoidance: Does It Matter What Policy-Makers Think About the Environment? Peter Read, Massey University Department of Economics, New Zealand and Visiting Fellow, Centre for Resource and Environmental Studies, Australian National University, Canberra
ADDRESSING UNCERTAINTY
There is no agreed methodology for taking policy decisions under uncertainty (Beckerman, 1995) and the approach adopted here, Optimised Regret Avoidance (ORA) builds upon the method of minimax regret (MMR) sometimes employed in business. MMR is well known to be a highly risk-averse procedure. The stock market would wish management to be risk-neutral, with portfolio holding providing insurance to investors and expected profits maximised. However, management has other motivations and, outside the peculiar world of perfect competition, policy might favour risk aversion even in commercial decision taking, given that the cost of failure falls on many besides stockholders. In relation to environmental policy, the potential for irreversibility, and for poorly understood complex behaviour in the ecosystem upon which policy impacts, imposes an additional burden of care, particularly where very long term problems are involved in which the appropriate constituency includes voters yet unborn.
The Climate Change Convention signed in Rio in 1992 calls for
cost-effective measures to be taken in response to threats of
potentially serious impacts, including irreversibilities, even
without full scientific certainty of the threat in question. ORA
was conceived in response to the greenhouse gas problem (Read,
1994, Chapter 4) although it acquired a name and an acronym
subsequently (Read, 1994a). It adds to the recognised
precautionary character of minimax regret through a non-linear
measure of acceptability: diminishing marginal welfare of
material consumption is inverted to yield an infinite
"objection" to extinction (assumed equivalent to zero
consumption) resulting from climatic catastrophe. This means that
policies involving non-zero risk of catastrophe are rejected -- a
belt and braces approach which, driven to an extreme, may yield
no feasible policy. But it may also provide a basis to motivate
risk assessment: we require 1:109 probability of
nuclear reactor catastrophe. On the other hand ocean-atmosphere
models suggest that only around three times pre-industrial levels
of CO2 are required to arrest the Gulf stream (which
carries approximately one third of the N. Atlantic region's heat
supply) with prospects of mass migration and land conflicts.
PROCEDURE
In the present paper we do not, as with the analyses of
climate change adaptation and mitigation costs based on
mainstream climate modeling, arrive at costs (losses of welfare)
that are more than marginal to the quantum of material
consumption and accordingly we do not conduct the analysis in
terms of the objection concept. The focus of attention is upon
the regret in terms of a social welfare function (SWF) -- i.e.
losses of social welfare relative to an optimum policy -- that
may arise from uncertainty as regards the SWF, maximisation of
which is taken to be the objective function of policy makers. In
particular we suppose ignorance as to whether the true objective
function -- that is to say the criterion for policy that would be
adopted by a society that was fully informed as regards the
present and future productive system, as regards the
environmental consequences of social action, and as regards its
own preferences --should be taken to be environmentalist or
consumerist. As is customary with economists -- a strength in
relation to questions which are beyond current knowledge --
empirical data is replaced with assumptions, and the results
obtained are suggestive rather than indicative. In particular
they may suggest areas in which further research would be useful.
A simple, but significantly non-linear, model of the (global)
economic system and its interaction with the environment is set
up and its behaviour is optimised by non-linear dynamic
programming, using the CONOPT solver within the GAMS system,
under alternative objective functions representing consumerism
and environmentalism. Regret matrices are developed under myopic
assumptions -- i.e. the sudden discovery in time period 2 (after
2030 by when the suggested further research may have born fruit)
that the objective function is different from that used in the
current four decades -- under foreknowledge of changed social
preferences, and under a hedging strategy in which the underlying
true objective function is completely unknown in the earlier
period (so that the two possibilities are given equal weighting
in the eventual revelation). Obviously there is a continuum of
possible objective functions and further work may involve
sensitivity analysis of the large number of parameters in the
model, which may be a guide to prioritising the suggested further
research.
THE MODEL (an acronym??)
Economy and Environment
The following equations are provided in a simplified version of the GAMS coding employed, together with explanatory comment. Time periods are decades 1990 to a 2300 horizon TT.
Additionally, each of the variables is constrained by upper
and lower bounds as is required to initiate the CONOPT solver.
Alternative Objective Functions (Social Welfare Functions)
A. Wenv = SUM(T,
DISC(T)*L(T)*LOG[((E(T)**alpha)*((C(T)/L(T))**(l-alpha)))]
A Cobb-Douglas style welfare with some (alpha= .25) public good environmental weighting and some weighting for per capita consumption, summed to TT, zero discounted (DISC = 1) and weighted by population. An unchanging SWF, equivalent to certainty of the environment perspective.
B. Wcon =
SUM(T, DISC(T)*L(T)*LOG[C(T)/L(T)]
As A. but no weighting for environmental satisfactions - an unchanging consumerist perspective.
C. WC-E = SUM(T,
DISC(T)*L(T)*LOG[Wcon(T<5), Wcnv(T>4)]
Foreknowledge that in 2040 that the true SWF becomes environmentalist, contrary to previous experience.
D. WE-C =
SUM(T, DISC(T)*L(T)*LOG[Wenv + (T<5), Wcon
(T>4)]
Foreknowledge that in 2040 that the true SWF becomes consumerist, contrary to previous experience.
E. WCav =
SUM(T, DISC(T)*L(T)*LOG[(Wcnv + Wcon)/2 |
(T<5), Wcon | (T>4)]
A hedging strategy to 2040 when a consumerist SWF is discovered.
F. WEav =
SUM(T, DISC(T)*L(T)*LOG[(Wenv + Wcon)/2 |
(T<5) Wenv | (T>4)]
A hedging strategy to 2040 when an environmentalist SWF is discovered.
For each of C. to F. the stream of welfare inputs is
calculated in relation to alternative true SWF's A. or B. as well
as in terms of the specified objective function, which of course
drives the optimisation process and determines the time path
taken by the system. For the myopic case A. and B. were run for
four decades and then restarted at 2030 using the 2020 output of
B. and A. respectively. Some dynamic inconsistency arose and the
results reported below for the myopic case are approximate.
Each of these functions is associated with an
inter-generational equity condition that per capita welfare is
non-decreasing. In addition to 6. and zero discounting, this
represents an effective concern for unborn voters: there is some
psychological evidence for well-being derived from comparisons
with others, and the awareness of previous higher standards of
living may have more significance for posterity than the absolute
living standards which earlier generations' investments hand
down.
Some comment is needed in relation to zero discounting and
the use of an arbitrary time horizon which appears to
disenfranchise the voters of the 24th and later centuries.
Firstly it may be noted that the 2300 horizon is sufficiently
distant for the quasi-steady state "turnpike" growth
path to establish itself, so that more distant time horizons have
little impact on policy over the next several decades during
which achieving the transient path onto the turnpike is the
proximate task. Secondly, and relatedly, today's decisions do not
bind tomorrow's and decisions as regards the best course of
action now fall to be revised in the light of changing
circumstances in a process of sequential decision taking. And
thirdly, infinite horizon problems cannot be addressed without
sufficiently rapid discounting to achieve convergence.
Demonstrating such convergence requires analytic tractability,
which is defeated even by so simple a formulation as 1. to A.
above (an aspect which has left two- or more-argument SWF's out
of the running in the bulk of the economic literature, despite
the obvious significance of environmental factors in most
people's enjoyment of life). It is felt that truncation of the
problem at a point after the 'turnpike' has been reached is a
small penalty in relation to overcoming the limitations of
analytic tractability. However, it should be noted that the zero
discounting of welfare used here does not imply a zero discount
rate in investment appraisal, or that the rate used for public
and private sector appraisals should be the same.
RESULTS
The results of these computations are given in the following table where the various policy options need explanation. A whole hog policy consists of pursuing the 'wrong' SWF after the correct SWF becomes known in 2030. This may seem an illogical option but it should be recollected that economists have been saying for decades that the National Income accounts are a defective measure of social welfare. A myopic policy consists of pursuing one SWF until the true SWF is revealed, after which the true SWF is followed. Its computation involves restarting the optimisation process at 2030 and the dynamic inconsistency is revealed by the increase in total welfare which results when restarting with the same policy as for the initial period. This effect has been compensated for but one case nevertheless yields negative regret. However, as with whole hog policies, the average regret from myopic policies is an order of magnitude greater than for a hedging approach, where the policy maker gives equal weight to both proposed SWF's until the true one is revealed (Equations E. and F.). Not included in the table is the computation result in which the policy maker is assumed to optimise on the assumption that the true SWF will change in 2030, since this seems an implausible state of mind for the policy maker.
It may also seem that the penalty (in proportionate terms -
the two measures of social welfare are not commensurate) from
assuming a consumerist SWF when the environmentalist SWF is true
are greater than the converse. However, that is not a safe
conclusion to draw without an exhaustive sensitivity analysis.
But it may be noted that the smooth substitutability implicit in
the Cobb-Douglas formalism adopted so far, for both production
and the SWF, does not capture substantial environmentalist
concerns (Common, 1995) that some substitutions may not be
feasible. If more constrained substitutability is the reality,
the penalties from continuing to assume a consumerist SWF (if
untrue) could become much greater. Also, the regrets may seem
small but the unit of account measures aggregate social welfare
for the next 300 years and the regrets are of the same order of
magnitude as the estimated cost of climate change. Furthermore
very different patterns of public and private investment emerge
in the near term.
References
Beckerman, W. (1995) Review of Read, 1994, in Economic
Journal, May 1995 (see also corrigendum November, 1995).
Common, M.S. (1995) Sustainablity and Policy, C.U.P.,
England.
Read, P. 1994 "Responding to Global Warming: the
Technology, Economics and Politics of Sustainable Energy",
ZED Books, London and New Jersey, February.
1994a "Optimised Regret Avoidance (ORA)" Paper,
International Energy Workshop, Honolulu, June.