Remarks over the Error-Based Learning Process in Financial Markets with Application on FTSE 100 Market DIMA BOGDAN a , PIRTEA MARILEN GABRIEL a , CRISTEA ŞTEFANA b a. Finance Department, Faculty of Economy and Business Administration a., b. West University of Timişoara J. H. Pestalozzi nr. 16, 300115, Timişoara ROMANIA bogdandima2001@gmail.com http://www.feaa.uvt.ro Abstract: - A critical issue in financial markets’ research field is the debate between the academic orthodox approach of the Efficient Markets’ Hypothesis and the critics rising from the behavioral finance paradigm and practice. The importance of this debate consists in the implications of the adopted point of view on the assessment of the financial markets’ predictability degree. [4], [5] proposed a unified approach labeled as Adaptive Markets Hypothesis. If in such framework the markets are considered to display, at least in a certain sense, a significant degree of predictability, then, one of the major difficulties in supplying empirical evidences is the requirement of knowing ex ante the “exact” forecast model used by the economic subjects. The purpose of this study is providing a solution to this problem inspired by the transduction’s (supervised learning) algorithms. Our main output consists in the thesis that the forecasting errors matter for price formation in financial markets. Key-Words: - Financial markets, FTSE 100, Adaptive Markets Hypothesis, forecasting algorithms, forecasting errors, adaptive mechanisms 1 Introduction Nowadays, one of the most important issues in the field of financial markets’ analysis is the divergence between two major conceptual frameworks: the Efficient Markets Hypothesis - the notion that markets fully, accurately, and instantaneously incorporate all available information into market prices and the behavioral finance which tries to account for the behavioral idiosyncrasies of the markets participants. Lo [2004, 2005] proposed a new conceptual framework which reconciles theories that imply that the markets are efficient with behavioral alternatives, by applying the principles of evolution - competition, adaptation, and natural selection - to financial interactions. This framework was labeled as Adaptive Markets Hypothesis. As Lo [2005,1] states “Based on evolutionary principles, the Adaptive Markets Hypothesis implies that the degree of market efficiency is related to environmental factors characterizing market ecology such as the number of competitors in the market, the magnitude of profit opportunities available, and the adaptability of the market participants”. If such a framework is viable that it implies the postulate of the existence of several adapting learning mechanisms on financial markets. Or, in other words, it presumes that the economic subjects are able to learn from their forecasting mistakes according to the accuracy, relevance, volume and structure of the market information. However, there are important difficulties in order to empirically test this hypothesis, since such a test requires knowing ex ante the “exact” forecasting model used by the market participants. The purpose of our study is to advance a possible solution to this problem by employing a two-stage approach: (1) a competitive framework of identifying some plausible forecasting algorithms and (2) an ex post test of their forecasting error relevance in the formation of observed level of prices. 2 The analytical framework In order to identify a feasible solution for the error- based learning process a simply competitive framework could be involved. Such a framework could imply the next steps: 1. The ex ante selection of a forecasting algorithms’ set {F i }; 2. The estimation of the “optimal-fitting” parameters on a learning window with length l of observed data so the estimation errors for the current level of the forecasted variable y t , ε t , are minimal regardless their sign: ( ) 1 t t t y F Min ε = − → 3. The computation of each individual algorithm’ forecasting performance with these parameters for q steps ahead according with a pre-defined performance criterion and Proceedings of the 13th WSEAS International Conference on COMPUTERS ISSN: 1790-5109 283 ISBN: 978-960-474-099-4