Abstract:
This thesis is done mainly to explore the time series dynamics and lead lag relationships among the Istanbul Stock Exchange (ISE) Equity Market Index, called ISE30, session to session and daily returns, volume and volatility. In addition to the well known classical definition of the returns, a new definition of return is made, namely, the returns are also calculated by using the average values. Moreover, many variables, some requiring detailed information on individual stock basis were also calculated and included in the analysis. An expectation survey aimed at answering the question of how the market trade variables affect the expectations of brokers was conducted. This survey was found to provide very interesting hints about how the expectations of the market people form in case of different combinations of return, volume and other trade data variables. A very detailed analysis of the survey results are provided in this thesis. Additionally, distributional properties of return series are analysed for the whole period spanning 1997-2005. The period is divided into three sub-periods, namely the pre-crisis period, crisis period and post-crisis period and all the analyses are repeated to see whether the distribution and the sample moments of session to session and daily returns change between different data windows. Return series were mainly modeled by using Autoregressive (AR), Moving Average (MA) and Autoregressive Moving Average (ARMA) techniques. The return series were found to possess the so called "long memory" or "persistency" problem. The long term memory property was explored in detail and the series are transformed by using the fractional integration method (ARFIMA). After an univariate time series analysis of the returns, a multivariate analysis of the returns with the trade variables were conducted by using Vector Autoregressive Model (VAR). In summary AR,MA and ARMA models were found to have little explanatory power for close to close returns. On the other hand, the returns calculated by the average values were found to have significant serial correlations, a fact that makes the AR, MA and ARMA models more useful. ARFIMA method proved to be useful in some cases, while it did not help in some others. Although the inclusion of other variables in the VAR models contributed to the explanatory power, the improvement is generally regarded to be not so prominent. Thus it can be said that changes in volume and volatility were found to have limited explanatory power with regard to the mean return for the next period, a result that is contradictory to what was implied by the expectation survey.
