Alternative distributed models for the comparative study of stock market phenomena

Abstract

In this paper we present a methodology of study of complex phenomena emerging in stock markets. This methodology is based on the use of distributed, multi-agent models with minimal knowledge representation and reasoning capabilities that has proven a powerful modeling tool for complex biological systems. Unlike neural and "neo-connectionist" models, ours' allow a comparative and incremental evaluation of their validity and relevance to the observed phenomena. The possibility of their application to the modeling and study of stock market phenomena is demonstrated on a simple example of a central agency that regulates the behavior of the investors : we show how a "blind" or myopic behavioral model reproduces results found in the literature and how the mutation of the model according to the parameters' values or the adaptation structures gives rise to a series of complex phenomena comparable to those observed in reality. IntroductionWhy do prices fluctuate and under what circumstances is it possible to control them ? What happens if some "irrational" investors enter the market and do they survive ? Is it possible to predict economic crises or other phenomena from observation of a market for a limited period ? The first step to answering any one of these questions is to find suitable economic agent models supporting experimental evidence (excess volatility, survival of technical analysts, etc.) while offering the possibility to control for behavioural investment factors. This paper attempts to build a structure which will allow the researcher to closely simulate real financial markets and highlight their drawbacks and their resilience to shocks.Our methodology is based on the use of distributed, multi-agent models with minimal knowledge representation and reasoning capabilities that has proven a powerful modeling tool for complex biological systems. Unlike neural and "neo-connectionist" approaches, ours' allows a comparative and incremental evaluation of their validity and relevance to the observed phenomena. Algorithmic models rely on the modeler's knowledge about the modeled system, while evolutionary ones rely on the potential to discover structures fit to the systems at hand. While the latter appear more robust to system tuning, the former demonstrate both minimality in respect to the relevant aspects of the problem and incrementality in respect to modeling.Standard financial economics literature is centered around the paradigm of homogeneity. Investors, consumers or traders are supposed to be identical in beliefs. They can vary in wealth endowments (rich and poor), utility functions (risk averse or risk neutral) and information sets (insiders versus the uninformed public) but they all pertain to the same knowledge about the economic environment. Common knowledge of the structure of the market is implicitly assumed and no diverse priors due to biological differences are allowed. In consequence, transactions occur only because of different approaches to risk (hedging) ; common knowledge of differential in...

Overconfidence in investment decisions: An experimental approach

“…Regarding financial decisions overconfidence was analytically (e.g., Barber and Odean 1999, Benos 1998, Caballé and Sákovics 2000 and experimentally (e.g., Adams et al 1995, Benos and Tzafestas 1997, Camerer and Lovallo 1999 studied and also confirmed for field data (e.g., Odean 2000, Statman andThorley 1999). Overconfident investors trade too much (Biais et al 2000, Odean 1998, deviate from Bayes' rule when aggregating information (Nöth and Weber 2000), and overreact to private and underreact to public signals (Daniel et al 1998).…”

Overconfidence in Investment Decisions: An Experimental Approach

“…Regarding financial decisions overconfidence was analytically (e.g., Barber and Odean 1999, Benos 1998, Caballé and Sákovics 2000 and experimentally (e.g., Adams et al 1995, Benos and Tzafestas 1997, Camerer and Lovallo 1999 studied and also confirmed for field data (e.g., Odean 2000, Statman andThorley 1999). Overconfident investors trade too much (Biais et al 2000, Odean 1998, deviate from Bayes' rule when aggregating information (Nöth and Weber 2000), and overreact to private and underreact to public signals (Daniel et al 1998).…”

An empirical evaluation of the overconfidence hypothesis

“…Experimental studies also have demonstrated that people tend to underestimate their error variance in making predictions, and overweight their own forecasts relative to those of other forecasters (e.g., Lichtenstein et al, 1982;Yates, 1990;Batchelor and Dua, 1992;Benos and Tzafestas, 1997). To empirically evaluate the above hypothesis, we need to identify private and public information.…”