Stock portfolio optimization using prohibited search algorithms and itinerant trader

In this thesis, modeling and forecasting of stock market fluctuations using the combination of neural network and conditional variance patterns (case, Tehran Stock Exchange) have been used from April 2008 to April 2012. According to the predicted results, this hypothesis is confirmed, but its accuracy is not as large as the composition of the neural network and the conditional variance pattern. In the return time series, both GRACH and ARCH conditional variances are rejected, but in the GRACH time series, ARCH is rejected. Given the artificial neural network simulation and conditional variance, the error value of the least squares is the return value of 18, that is, an error is required to estimate future returns. An important parameter of the opacifying factor is the dependence of our input and output at each stage, which indicates a number close to 1 and shows a complete dependence. According to the artificial neural network simulation and conditional variance, the least squares risk error value is 0.001, that is, to estimate the returns for the future, this error is error. Another important parameter of this regression table is R, which shows the dependence of our input and output in each stage, where 0 means a random relationship and 1 means dependence.