Effectiveness of turkish derivatives market and forecasting comparative prices for the contracts

In derivative markets developed for eliminating uncertainty and risk arising from financial markets, confidence in the market is the most important determinant of market functioning as in others. Market credibility is closely related to the right decisions that are made, after the right information is received by all components of the market. This can only be achieved by effectively processing markets. If the market is not effective, it is possible to make forecasts for the future period using past period price movements since all the information available on the market has not been fully and accurately reflected in the prices of financial assets. Forecasting the future values of financial assets traded in ineffective markets is at the top of the issues that decision-making units are interested in. Because forecasting the future prices of contracts that are traded on derivative markets that have two basic functions such as risk management and future price formation is very important for both the country's economy and market investors. In this study, firstly, the effectiveness of the Turkish Derivatives Market was tested by applying the Augmented Dickey-Fuller (ADF), Phillips-Perron (PP) and Kwiatkowski-Phillips-Schmidt-Shin (KPSS) linear unit root tests and Kapetanios, Shin ve Snell (KSS) nonlinear unit root test. As a result of all unit root tests, it was concluded that the series did not have unit roots, that is, they did not show random walk, so that the market was not effective. Then, the method that shows the highest performance is tried to be determined when forecasting the end of day settlement price of the TL/Dollar and Bist-30 contracts which is traded in the Derivatives Market . For this purpose, Box-Jenkins (ARMA - Autoregressive Moving Average), autoregressive conditionally heteroscedasticity models (ARCH, GARCH, EGARCH, ...), and finally, artificial neural networks method has been applied to the data which is provided from Borsa Istanbul Inc. and covering the dates between 04.02.2005 and 31.12.2015. The forecasting results produced by the time series analysis methods are compared with the results of the artificial neural network model which has the best performance by employing different architectures, layer numbers, cell numbers in layers, activation functions and learning methods. According to the results of analysis, RBF-1-BL artificial neural network model performed better than ARMA (4,4) and ARCH (1) model for TL/Dollar contract series. For the Bist-30 contract series, TDNN-1-B-L artificial neural network model has higher predictive performance than ARMA (4.5) and ARCH (1) models. Brock, Dechert ve Scheinkman (BDS) linearity tests conducted in this study have shown that the contract series are non-linear. Thus, artificial neural network models, which are capable of working with both linear and nonlinear series, are powerful alternative methods in the forecasting of the future period.