Bond Trading, Market Anomalies and Neural Networks: An Application with Kohonen Nets

Since the 70s, a wide stream of research has grown up on the presence of ``anomalies in the behavior of prices in financial markets. Both in the stock and the bond markets, many so called``puzzles were discovered, which were often explained resorting to factors such as agents heterogeneity, market microstructure and other institutional features or decision making behavior at the odds with the standard expected utility framework. In particular, in bond markets securities that are very similar in their features (maturity, duration, etc.) have been often found to display a widely different behavior in response to the arrival of news in the market, and such differences were often traced back to market ``anomalies such as the ``coupon effect, the ``benchmark effect or the seasoning process of each particular issue. Even in the most developed markets, market segmentation and illiquidity may turn out to be the determinant factor to discrimate between success and failure of a trading strategy. As a result, the decision making process of the agents in the market is more and more a question of dealing with complexity: many such ``puzzling effects have to be taken into account in the portfolio management problem and so many strange market responses are to be allowed for. Also, such decision making process is of a case-based kind, so that choices are made based on similarity relationships with events and situations that took place in the past.