Corporate Financing: An Arti¢cial Agent-based Analysis

We examine corporate security choice by simulating an economy populated by adaptive agents who learn about the structure of security returns and prices through experience. Through a process of evolutionary selection, each agent gravitates toward strategies that generate the highest payo¡s. Despite the fact that markets are perfect and agents maximize value, a ¢nancing hierarchy emerges in which straight debt dominates other ¢nancing choices. Equity and convertible debt display signi¢cant underpricing. In general, the smaller the probability of loss to outside investors, the more likely the ¢rm is to issue the security and the smaller the security’s underpricing. AMAJOR GOAL OF CORPORATE FINANCE RESEARCH is to explain corporate ¢nancing patterns. To explain security issuance preferences, researchers have extended the Modigliani andMiller (1958) analysis by allowing for the presence of market frictions such as transactions costs, taxes, and informational di¡erences among agents (DeAngelo and Masulis (1980), Castanias (1983), and Myers and Majluf (1984)). These extensions, however, maintain the assumption that the security payo¡ structure is common knowledge. The assumption of common knowledge places extraordinary demands on the abilities and knowledge of agents (Simon (1972), Russell and Thaler (1985), and Thaler (1992)). Agents must know not only the set of feasible states, but also the relationship among these states and the payo¡s on assets. In addition, they must be able to use this knowledge to compute the payo¡s on any potential portfolio of assets in any state of the world. Not surprisingly, actual economic agents do not appear to have these capabilities (see, e.g., Kahneman and Tversky (1979) and Kahneman, Slovic, andTversky (1982)). A realistic model of decision making should incorporate the fact that agents cannot precisely partition states of the world, that they can only construct fuzzy maps from states to outcomes, and that they lack the ability to fully exploit the knowledge they have by computing all of its consequences (see Mukerji (1997) for a detailed discussion). Despite these observed limitations in agent rationality, THE JOURNAL OF FINANCE VOL. LVIII, NO. 3 JUNE 2003 Noe is from the A. B. Freeman School of Business, Tulane University; Rebello is from the Department of Finance, Robinson College of Business, Georgia State University; andWang is from SAS Institute Inc. and Baruch College. The authors thank the seminar participants at Tulane University for their helpful comments on earlier drafts of this paper. Rebello would like to acknowledge research support from the Robinson College of Business. Special thanks are extended to Richard Green (the editor) and an anonymous referee for very insightful comments. Any errors are our own.