Slow Learning in the Market fopr Lemons

Human-participants experiments using markets with asymmetric information typically exhibit a “winner’s curse,” wherein bidders systematically bid more than their optimal amount. The winner’s curse is very persistent; even when participants are able to make decisions repeatedly in the same situation, they repeatedly overbid. Why do people keep making the same mistakes over and over again? In this chapter, we consider a class of one-player decision problems, which generalize Akerlof’s (1970) market-for-lemons model. We show that if decision makers learn via reinforcement, specifically by the reference point model of Erev and Roth (1996), their behavior typically changes very slowly, and persistent mistakes are likely. We also develop testable predictions regarding when individuals ought to be able to learn more quickly.

Minimal Intelligence Agents in Double Auction Markets with Speculators

This chapter explores the minimal intelligence conditions for traders in a general double auction market with speculation activities. Using an agent-based model, it is shown that when traders and speculators play together under general market curve settings, zero-intelligent plus (ZIP) is still a sufficient condition for market prices to converge to the equilibrium. At the same time, market efficiency is lowered as the number of speculators increase. The experiments demonstrate that the equilibrium of a double auction market is an interactive result of the intelligence of the traders and other factors such as the type of the players and market conditions. This research fills in an important gap in the literature, and strengthens Cliff and Bruten’s (1997) declaration that zero is not enough for a double auction market.

Influence Diagram for Investment Portfolio Selection

The goal of an artificial intelligence decision support system is to provide the human user with an optimized decision recommendation when operating under uncertainty in complex environments. The particular focus of our discussion is the investment domain—the goal of investment decision making is to select an optimal portfolio that satisfies the investor’s objective or, in other words, to maximize the investment returns under the constraints given by investors. The investment domain contains numerous and diverse information sources, such as expert opinions, news releases, economic figures and so on. This presents the potential for better decision support but also poses the challenge of building a decision support agent for selecting, accessing, filtering, evaluating and incorporating information from different sources, and for making final investment recommendations. In this study we use an artificial intelligence system called influence diagram for portfolio selection. We found that the system outperform human portfolio managers and the market in the year of 1998 to 2002.

Fundamental Issues in Automated Market Making

The goal of this chapter is to establish an analytical foundation for electronic market making. We use two classes of models to reason about this domain: structured and relaxed. In our structured model, we will formalize the decision process of a dealer and then use a simple class of trading strategies to highlight several fundamental issues in market making. In our relaxed model, we treat the dealer’s quotes and transaction prices as a simple time series. We apply statistical techniques to discern possible structure in the data and then make conclusions about the dealer’s optimal behavior. Our main interest is a normative automation of the securities dealer’s activities, as opposed to explanatory modeling of human traders, which is the primary concern of earlier studies in this area.

Financial Modeling and Forecasting with an Evolutionary Artificial Neural Network

In this chapter, I consider a design framework of a computational experiment in finance. The examination of statistics used for economic forecasts evaluation and profitability of investment decisions, based on those forecasts, reveals only weak relationships between them. The “degree of improvement over efficient prediction” combined with directional accuracy are proposed in an estimation technique, as an alternative to the conventional least squares. Rejecting a claim that the accuracy of the forecast does not depend upon which error-criteria are used, profitability of networks trained with L6 loss function appeared to be statistically significant and stable. The best economic performances are realized for a 1-year investment horizon with longer training not leading to enhanced accuracy. An improvement in profitability is achieved for models optimized with genetic algorithm. Computational intelligence is advocated for searching optimal relationships among economic agents’ risk attitude, loss function minimization in the learning process, and the profitability of trading decisions.

Pricing Basket Options with Optimum Wavelet Correlation Measures

This chapter describes a new procedure for designing optimum correlation measures for financial time series. The technique attempts to overcome some of the limitations in existing methods by looking at correlations among wavelet features extracted at different time scales from the underlying time series. New correlation coefficients are further optimised with help of artificial neural networks and genetic algorithms using a nonparametric adaptive wavelet thresholding scheme. The approach is applied to the problem of pricing basket options for which the pricing formula depends on accurate measurements of correlations between portfolio constituents. When compared with standard linear approaches (i.e., RiskMetrics™), an optimised predictive wavelet correlation measure offers potentially large reductions (over 50% in some cases) in static delta-hedging errors.

Optimization of Individual and Regulatory Market Strategies with Genetic Algorithms

An optimal policy problem is formulated for evolutionary market settings and analyzed in two applications at the micro- and macrolevels. First, individual portfolio policy is studied in case of a fully computerized, multiagent market system. We clarify the conditions under which static approaches—such as constraint optimization with stochastic rates or stochastic programming—apply in coevolutionary markets with strictly maximal players under scaled genetic algorithms. Convergence to global optimum is discussed for (a) coevolution of buying and selling strategies and for (b) coevolution of portfolio strategies and asset distributions over market players. Because only a finite population size in our setting suffices for the asymptotic convergence, the design criteria for genetic algorithm given (explicit cooling scheme for mutation and crossover, exponentiation schedule for fitness-selection) are of practical importance. Second, system optimization policy is studied for a model economy of Kareken and Wallace (1982) type. The income redistribution, monetary and market regulation policies are subjected to a supergenetic algorithm with various objective functions. In particular, the fitness function of a policy (i.e., a supercreature) is computed by means of a conventional genetic algorithm which is applied to the market players (creatures) in a fixed evaluation period. Here, the underlying genetic algorithm drives the infinite market dynamics and the supergenetic algorithm solves the optimal policy problem. Coevolution of consumption and foreign currency saving policies is discussed. Finally, a Java model of a stationary market was developed and made available for use and download.

A Physics of Organizational Uncertainty

This chapter constructs a dynamic model of a multinational enterprise (MNE) to quantify the effects of various capital control policies on a firm’s debt and equity positions, innovations, and outputs at the headquarters and subsidiary. The model is calibrated to the US Foreign Direct Investment (FDI) Benchmark Survey and the IMF’s Exchange Arrangements and Exchange Restrictions so that it reproduces the average US FDI and technology flows to foreign subsidiaries. Both steady-state and transition analyses suggest a significant impact of capital controls on an MNE’s operations. Lifting capital restrictions produces an inflow of capital and technology into the less developed countries, leading to an increase in the steady-state FDI position and production. Simulation experiments reveal that even short-term capital controls have long-lasting negative effects.

Capital Controls and Firm's Dynamics

This chapter constructs a dynamic model of a multinational enterprise (MNE) to quantify the effects of various capital control policies on a firm’s debt and equity positions, innovations, and outputs at the headquarters and subsidiary. The model is calibrated to the US Foreign Direct Investment (FDI) Benchmark Survey and the IMF’s Exchange Arrangements and Exchange Restrictions so that it reproduces the average US FDI and technology flows to foreign subsidiaries. Both steady-state and transition analyses suggest a significant impact of capital controls on an MNE’s operations. Lifting capital restrictions produces an inflow of capital and technology into the less developed countries, leading to an increase in the steady-state FDI position and production. Simulation experiments reveal that even short-term capital controls have long-lasting negative effects.

An Application of Multi-Agent Simulation to Policy Appraisal in the Criminal Justice System

This chapter reports on a multi-agent approach to the construction of a model of the English criminal justice system. The approach is an integration of model-building with ways of enabling people to engage in strategic policy making and take into account the complex interactions of the criminal justice system. From the workings of the police to court procedures to prisons, decisions in one area of the criminal justice system can be crucial in determining what happens in another area. The purpose was to allow assessment of the impact across the whole justice system of a variety of policies.