Summary of Research

You can't step twice in the same river.
-- Heraclitus c.540-c.480 B.C.

AS A MICROECONOMIST at a business school, I have been drawn to the examination of markets:

My research has revolved around the application of economic theory to several social issues associated with these markets.

My environmental research is related to energy policy and to drug policy, through the issue of altering individuals' (or organisations') behaviour. In the case of the greenhouse effect, which links the use of carbon-based fuels to possibly severe environmental impacts, the issue is to encourage behaviour to conserve energy; in the case of drug policy, the issue is to encourage responsible drug use, or no illicit drug use at all, in some instances. In both cases the motivation is that more energy use (when less could achieve the same ends) is inefficient, and that excessive drug use (which for some people is the same as any drug use) is socially unacceptable.

Energy-policy and drug-policy research has also led to the study of instruments to reduce individuals' and organisations' energy use -- and more recently use of environmental resources -- and instruments to affect individuals' drug-taking behaviour. In place of the coercion of the criminal-justice system towards drug use and the exhortation of some environmentalists towards energy use, I have analysed the effectiveness and equity of market-based solutions

This work has led me to examine the changing nature of organisations and economic institutions, in response to the changing external environment, and to repeated strategic interactions, such as the rivalry among the small number of sellers in an oligopoly. My research into learning and adaptive strategic market behaviour, using game theory and machine learning, is a new approach. I have also derived new theoretical results in labour markets with easily identified groups among workers seeking employment, which may result in statistical discrimination.

Energy/Environmental Policy

My Ph.D. thesis included the study of disequilibrium economics in which non-renewable resources (such as fossil fuels) were input factors of production. The theory of non-Walrasian, disequilibrium economics and exhaustible resources resulted in the book, published as an Outstanding Economics Dissertation and an Outstanding Energy Dissertation, which was praised by Professor Robert Solow of M.I.T, Nobel laureate in Economics and recent president of the American Economics Association. I have a continuing interest in exhaustible resources, oil in particular, and this has led to the eleven papers on energy policy, and on the world oil market, five of which were written with co-authors (Professor Peter Swan, now of the University of Sydney, Professor Ray Ball, now of the University of Rochester, and Professor James Sweeney of Stanford University).

After my 1984 sabbatical, I wrote the definitive analysis of Australian energy policy, which led to an invitation from Dr Donald Abbott of New Brunswick, the editor of Energy Exploration & Exploitation, to write a study which was later published.

My interest in energy policy meant that energy-related environmental issues, such as the greenhouse effect, were an obvious application for me. Indeed, a joint consulting project for CRA Ltd. on the cost of Australia's attaining the Toronto targets for CO2 reductions has generated several papers, which have attracted much attention and been influential with policy-makers, culminating with a presentation before Federal Parliament. I was unable to accept Professor David Wood's invitation to present at an M.I.T. conference in 1990, but the work elicited a further invitation to publish on greenhouse insurance from Dr. M.A. Dorgham of The Open University, editor of the International Journal of Global Energy Issues. I have been invited to act as a reviewer in this area from the Journal of Environmental Economics and Management, the Economic Record, and The Energy Journal.

Drug Policy

I have been studying drug policy for twenty years, publishing one of the first papers to analyse the drug issue using a supply/demand framework. I have published several papers on Australian drug policy, which have been influential in the drug-policy debate in Australia. There are two significant achievements of this work: I obtained previously confidential and unique "inside" data on the structure of the illicit Australian heroin distribution system collected by the black-marketeers themselves, which together with other studies both here and abroad has permitted me to examine the costs and rewards of the drug-distributing and in some cases of the drug-using network. Professor Milton Friedman, Nobel laureate in Economics, has recently characterised these data as "particularly interesting and unique" My second achievement is the first serious attempt to estimate the social costs of illicit drug use in Australia (What Price Prohibition). This line of research culminated in my testimony before Federal Parliament, and provided the basis for my estimates of the cost of illegal drug use employed by the authoritative Federal Department of Community Services study on the costs of drug use, and praised by Professor Ethan Nadelmann of Princeton University. It also led Professor Nick Heather, then Director of the National Drug and Alcohol Research Centre, to invite me to act as co-convener and chair of the NDARC Annual Symposium at the Third International Conference on the Reduction of Drug-Related Harm in 1992.

In 1993, I was invited by the past president of the Law Society of N.S.W., Mr John Marsden, to give a presentation at a Society seminar, chaired by Mr Justice Michael Kirby, which resulted in the Society's revising its policy on cannabis use soon after. I have received invitations to act as a reviewer in this area from the Australian Journal of Political Science and, recently, from Dr Frank Stephen of the University of Strathclyde, editor of the Journal of Economic Studies. Public reaction to my work has included an editorial in the Sydney Morning Herald on 9 May 1989, and recently an editorial in the February 1994 issue of the Independent Monthly.

Game Theory and Economic Evolution

Oligopolistic markets, with price-setting firms involved in a strategic "dance" with their rivals, selling goods or services that are close (but not perfect) substitutes, include markets for brand-differentiated items, such as ground coffee at the retail level, air travel, gasoline, and cola-based soft drinks. Over the last fifteen years there has been a very fruitful interaction between the theories of non-cooperative games and oligopolistic behaviour, resulting in use of the concepts of dynamic equilibria, which has meant that phenomena such as price wars can be characterised as the outcomes of non-cooperative games. I have been less concerned with characterisation of equilibria, and more with estimating the relationship between market structure and strategic behaviour, a disequilibrium approach of long-standing interest to me.

Advances in game theory have provided an impetus for renewed investigation of the strategic behaviour of oligopolists as players in repeated games. I have used marketing data to examine how players in iterated oligopolies respond to their rivals' behaviour, and used machine learning to derive improved contingent strategies for such markets, which provide insights into the evolution of such markets and the patterns of behaviour observed. In such markets, external stimuli may lead to prices changing from one period to the next, but for most periods, the player's new price level and the levels of its other marketing instruments are determined by rivals' past actions. One paper addresses the issues of (a) how past actions determine present actions, (b) which past marketing variables are relevant, and (c) how far back economic actors remember when determining next period's levels.

My win in the Second Competitive Computer Strategy Tournament at M.I.T. with a new and unique solution and my reading of Professor Robert Axelrod's then unpublished work on the use of machine learning to simulate solutions to the iterated Prisoner's Dilemma, a two-person strategic situation, as a development of his computer tournaments of the early '80s, pointed me to a fruitful area of research.

Using the Genetic Algorithm (GA), a new optimisation technique from machine learning, I became the first economist in the world to publicly present the results of using the GA to examine an economic issue in 1988: A paper read at the Econometric Society's Australian Congress in Canberra (August), and a paper read at the Allied Social Sciences Associations meetings in New York (December), again under the ægis of the Econometric Society. One paper has been provisionally accepted for publication by Computational Economics, which will please Professor Robert Axelrod of the University of Michigan, who wrote to me in 1989 that this paper deserved a wide audience. Representing these strategies led me to the concepts of finite automata and their graphical representations from computer science. This in turn led me to means of conceptualising and measuring the complexity of such strategic behaviour

The GA simulates "evolution" of solutions to maximization problems, using a kind of "natural selection" in deriving the next "generation" of possible solutions. As such, it provides an operational link between the study of economic institutions and evolutionary processes. I define the firm as a bundle of decision rules, and use the GA to generate distinct firms or organisations and letting the behaviour of the decision-rule interactions emerge as the computer experiment progresses. My research implements a simulation of a Schumpeterian process, where the GA can generate "good" bundles, given each environment of competitors, through natural selection.

Strategies (sets of rules mapping from state to action) for playing repeated games can be represented as strings in such a way as to allow use of the genetic algorithm (GA) to search for higher-scoring strategies. This is done by pitting the strategies against each other and using the highest-scoring strategies to derive a new generation of strategies in a simulation of natural selection and sexual reproduction.

An increasing number of economists are now using GAs and related tools, including Professors Jasmina Arifovic (McGill), Brian Arthur (Stanford), John Miller (CMU), Thomas Sargent (Chicago), Margaret Slade (UBC), and a number of economists associated with the Santa Fe Institute, which I visited in February and March, 1993.

Within the framework of repeated games, rationality demands a high level of pattern recognition, memory, and computing. The concept of "bounded rationality" in which the memory, or the computing ability, or competence at pattern recognition of the actors are limited, may, however, be sufficient to generate the observed behaviour, particularly dynamic oligopolistic behaviour. I model economic actors as boundedly rational stimulus­response automata, which can be estimated from historical data. Several papers have recently been positively reviewed by Professor Robert Wilson of Stanford University.

Advances in game theory and oligopoly theory and the development of machine learning have only recently allowed such a research program to be undertaken. No-one else has examined the estimation of automata from patterns of stimulus­response behaviour in repeated strategic interactions. No-one else has used such estimated machines with machine learning to search for better contingent strategies. Such research advances understanding of oligopolistic behaviour and the theory of repeated strategic interactions, to improve the efficiency of resource allocation.

Professor David Fogel of the University of California, San Diego, has invited me to join the program committee of the 1995 IEEE Conference on Evolutionary Computation. Professor Daniel Leonard of The Flinders University, editor of the Economic Record, has invited me to act as a reviewer in this area.

Project with Professor David Midgley

I was invited by David Midgley, Professor of Marketing at the AGSM, with his interest in the retail market behaviour of sellers, and Lee Cooper, Professor of Marketing at U.C.L.A., with his marketing models and his access to marketing data sets, to use the machine-learning genetic algorithm to search for better contingent strategies on the part of sellers in a mature iterated oligopoly. We used weekly data on prices, marketing instruments, market shares and sales, by sellers in the U.S. retail market for ground caffeinated coffee, with Lee Cooper's model to predict weekly profits to test our models in ways that would not have been possible with more aggregated data. This research was supported by a small ARC grant, and we have now applied for a large grant to continue the work, and to extend it to other markets.

We focused on the process of response and learning of the players, and used repeated games and machine learning to model adaptive behaviour over time in a synchronised, iterated oligopoly. Using empirical data on the weekly prices and promotional instruments of the three largest and several smaller coffee sellers in a regional U.S. retail market, and using a marketing model to predict brands' market shares and profits in response to others' actions in any week, we examined the adaptive strategic behaviour of the three largest brands. We modelled the brands' strategic behaviours as finite automata with memory of previous weeks' actions, and used the genetic algorithm to derive automata which were "fit" given their environment, as described by their rivals' actions in the past.

We modelled the three largest sellers with three "populations" of strategic mappings. We used the actual profits of each seller, given the others' activities, as the payoff matrices in the repeated game. We introduced institutional constraints on head-to-head discounting and "saturation" effects in the demand for coffee over time. We derived artificially intelligent brand managers and compared their actions (and profits) with those observed historically, in computer experiments. No-one else had done this for a single population, let alone three.

We have presented reports of this work at Stanford University, the Santa Fe Institute, the TIMS Marketing Science conference in St. Louis, in March 1993, and at the Conference on Computer Science Applications in Economics, Econometrics, Statistics and Finance, jointly sponsored by the Society for Computational Economics and the Society of Economic Dynamics and Control, in June 1993, in Nafplion, Greece. The AGSM Working Paper, "Adaptive Behaviour in an Oligopoly" was presented at the 1994 Australasian Meeting of the Econometric Society, at the University of New England in July 1994. Most recently, we have had word that a paper of ours will appear in Management Science, later in 1996.

The Economics of Labour Markets

A further area in my research is the economics of labour markets. In particular I have been interested in statistical discrimination, wherein employers might use a cheaply observed characteristic, such as gender, as a proxy for a costly-to-observe characteristic, such as productivity. In the course of this theoretical work, I have recently solved some problems of modelling firms' production functions when they employ a mixed workforce (Paper 2), and found a new source of potential discrimination. Professor Wilfred Ethier of the University of Pennsylvania, editor of the International Economic Review, has invited me to act as a reviewer in this area.

Work in Progress

Work in progress falls mainly into the game theory/economic evolution/oligopoly area of investigation; I have been invited to write one paper on optimal hedging against the greenhouse effect, and I want to explore the demand functions underlying observed market behaviour in drug black markets, and use the unique data to explore the structure and transactions within what is a completely unregulated, if prohibited, market. I have no other plans in these areas, or the labour market area.

In providing the machine-learning genetic algorithm with appropriate environments, given limited data, I have become interested in the econometrics of estimating finite automata as descriptions of firms' stimulus­response strategies in repeated interactions. Specification of the model is crucial: the number of rounds remembered, the thresholds of perception or action, the issue of the number of possible states. Synthetic data and historical data can be used to explore this area.

The economist's intuition is that since more complex automata are able to play more complicated strategies (by remembering longer or counting longer, for instance), they should be able to perform better. But if there is a cost to complexity (the cost of design, the cost of choice, the cost of maintenance) then there may be a point of diminishing returns. Experiments in which more complex strategies can evolve and compete, but with some penalty for their complexity, should reveal whether the intuition holds.

Computer experiments (or simulations) allow us to examine so-called emergent phenomena, which owing to intractable complexity may be beyond our abilities to analyse with closed-form mathematical descriptions. The increasing use of these techniques would be standardised by an agreed protocol, as I have recently discussed with Professors Blake LeBaron (of the University of Wisconsin) and Brian Arthur (of Stanford and the Santa Fe Institute). Since discussion and redrafting are desirable in devising such a protocol, it will not quickly emerge.


I have been a co-researcher with Professor Peter Swan (now head of the Finance Department, University of Sydney) on energy/environment policy. We continue to work on the use of reserves/production ratios in the world oil market, as an application of the theory of exhaustible resources which yields an elegant, useful relationship, recently published. In the light of this work, one paper is being redrafted and will be submitted for publication to Resources and Energy Economics.

Major contributor

Dr Jean Cross, Professor of Safety Science, and I have recently been awarded a grant of over $100,000 by Worksafe Australia to examine the costs and benefits of safety programs. Professor Cross is the principal researcher of this project, which will examine four Australian firms as case studies in order to put estimates on the value of safety studies, and to develop procedures for routinely measuring these values. I am the major contributor, as an economist with an interest in firms' costs, especially where unplanned events, such as drug abuse, are concerned.

Other contributions

I am a member of stage two of the Feasibility Research into the Controlled Availability of Opioids, being conducted by the National Centre for Epidemiology and Population Health at the A.N.U., coordinated by Dr Gabriele Bammer, under the Directorship of Professor Robert Douglas. This is a unique study to examine exhaustively all possible consequences of controlled availability of the previously prohibited opiates, such as heroin, before such a program is actually undertaken in the A.C.T. The main product of the program is information, to answer some of the otherwise unanswerable questions about relaxing the prohibition. Although few of the team are economists, the organisers have recognised the desirability of having market experts involved since the first meeting in Canberra in 1990, which I attended, at the suggestion of Professor Duncan Chappell, director of the Australian Insitute of Criminology.

Back to Robert's page.
Last Updated 25 March 1996
Robert Marks,