The failure of ARIMA models on stock-market behaviour

Abstract

The aim of this dissertation is that of discussing issues in time series model which adequately fits stock market behaviour. In the 1960s, Fama wrote a paper where he performed empirical research on real stock market data in order to argue in favour of the independence hypothesis and against any significant form of deterministic behaviour in stock markets. The aim of this dissertation will not be that of biasing itself towards either point of view but that of trying to see how both perform. In fact, we shall first attempt to fit an ARIMA model to our data by first constructing a sound theoretical background for such models and then testing our data using a specific time series software. Our next step will be that of performing hypothesis tests which will determine whether a unit root process is a better fit to our data that ARIMA models. If this will be so, then the hypothesis of independent price changes around which Fama's arguments are based will again be the most likely alternative. Finally further models which could also be applied to stock market behaviour and which might give us more information on its behaviour will be discussed.