Open Conference Systems, ICQQMEAS2015

Font Size: 
Forecasting the CAC-40 Stock Index and its Returns: Empirical Evidence from Global and Local modeling
Christos Floros, Stelios Papadakis

Last modified: 2015-09-24


In this paper, we use global and local models to predict the price and returns of CAC-40 stock index (an international stock index). We experimentally show that predicting the stock index, itself, in terms of low absolute percentage error (APE) is feasible but impractical for forecasting the returns. We build a global model based on support vector machines, which achieves forecasting Mean absolute percentage error (MAPE) less than 1%. However, the forecasting of returns for the same period (using the same approach) is about 50%; this implies a “toss a coin to predict the result”. We formulate the claim that a global model cannot guarantee adequate forecasts of the returns. We propose an algorithm to identify those regions of the input space which are “predictable”. For each predictable region, we build a local model by using the data of the respective region. The forecasting of a new, unknown target, is feasible only if the new target belongs to a predictable region- and infeasible otherwise. The final output of our approach has three states a) sell, b) buy, d) hold (no suggestion). Experimental results on CAC-40 index shows that our model can be used as an effective decision support tool for financial analysis and trading

Full Text: PDF