Σ.Μ.Ε., όγκος συναλλαγών και ανοικτά συμβόλαια : τυχαίος περίπατος και πρόβλεψη στο Χ.Π.Α.

Abstract

The major purpose of this paper is dual: testing the Random Walk Hypothesis and the forecasting
power of the linear regression and GARCH models. The data refer to the stock index FTSE/
ATHEX20, for a five year period, from 2002-2006, contributing robustness to the findings. In
particular, the paper finds strong evidence of random walk patterns, i.e. stationarity on first differences
and cointegration according to Johansen. Furthermore, there is evidence of one way Granger
causality, with various lags, proving that the future markets lead the spot, by reacting faster to the
news. The paper further examines the relationships among a set of four variables, looking into possible
patterns in their behavior and deviations from the long term cointegrated equilibrium. Through
gradual improvements, the constructed model evolves from a simple linear regression to a GARCH
(1,1) one, corrected for structural break and autocorrelation. Surprisingly, the forecasting power of
the advanced GARCH model is not superior to that of the simple linear one, showing that such
costly and time consuming methods do not yield respective results and thus can be avoided if price
movements are the major goal of forecasting.