Modelos predictivos de lógica y lógica borrosa en índices bursátiles de América del Norte

2006. Trimestre Económico. Vol. 73, Nº 290, Pp 265 - 288

Franco Parisi F, Antonino Parisi F

Abstract:

This article continues with the research related to predict stock index, such as, genetic algorithms and neuronal networks. Parametrics or non parametrics, lineals and non lineals models, try to recognize patterns and relations that express themselves in a mathematical language, through the estimation of coefficients and their statistical significance. However, most of the agents in the stock market use a language that incorporates qualitative aspects to refer to, for example, the price of an asset, the yield of the investment, etc. In this context, the quantitative models have problems to absorb this information, which suggests the need to develop and analyze new techniques, in corporating this type of references. The methodology of fuzzy logic gives answer to this question because it’s based on the idea that the variables should be handled not as a number but as characteristics that they represent. We used historic series of daily prices of the North American stock index DJI and Nasdaq (USA), IPC (México) y TSE (Canada), corresponding to the period October 8, 1996 and January 7, 2005. We designed a model of logic and fuzzy logic, to forecast the stock indexes sign variations. The logic models and the fuzzy logic models reached a forecast capability statistically significant. In addition, both models achieved a significant and positive abnormal return when they were used as a trading strategy, even after transaction costs.

Palabras claves: lógica borrosa, funciones de pertenencia, conjuntos de pertenencia, reglas de comercio (trading), desfuzificación, porcentaje de predicción de signo, prueba de acierto direccional.

Acceder a la publicación completa