Comparing forecasts of ARMAs and ANNs on OMXS30. Examining from a economic and statistical point of view.

Abstract

Forecasting is of great importance within economics and a vast number of papers have been published on financial forecasting. One of the most used forecasting models in economics is the autoregressive moving average (ARMA). This study compares the ARMA models to artificial neural networks (ANNs). ANNs have proven to be successful in other fields and have increased in popularity with the increase of computing power in recent times. The study includes six different versions of the ARMA model and three different ANNs. These models are examined using statistical and economical measures in order to determine their forecasting performance. The study shows a discrepancy between these two types of performance measures and also shows the difficulty of evaluating a forecast from a financial perspective. The results are inconclusive and dependent on the purpose of the evaluation.