Dear Milad;

Thank you for your kind reply. I am using only one IV which is represented by stock market prices and I already converted them into return series for the period 1993-2013. So it is time series data with one variable (market index returns). No cross-correlation. Please see attached. Please note that I used Matlab and already divided the full sample randomly into 90%/10% training and testing period. Looking forward to hearing your reply soon.

I wonder if you can also help me to apply my Hybrid model . I received the codes from PhD student in our university and he used Matlab to build GA-ANN and I need to use discrete Poisson two-states Hidden Markov model. Sorry but I also need to know how to forecast the Extreme value theory in Matlab. I watched many webinars and searched in all mathworks files and in help files but it seems that EVT have been used only once in combination with copula model. I know how to apply this EVT in R but not in Matlab. I am reading highly related books to get the technical supports and to know more about all of those models. Moreover, the GARCH models have been applied directly for fitting of time series but not for forecasting. However, you cannot imagine how much time I am spending to search in all areas inside Matlab.

The below is just one time series for daily time horizon, but I am using daily, weekly, monthly, quarterly datasets from 4 stock markets.

You right Milad. I have series of stock prices for different time horizons in both domestic and $ currencies. each series is a sequence of stock prices for the period 1993-2013. Then I converted them into return series with log(return function). So the final dataset will be log return series and I already divided them into two training and testing subsets. So simply excel file with two columns: date (year from 1993-2013) and returns (log returns calculated from original prices). How many prices ? You mean how many observations? Ok it depends on the time horizon. So for daily its 5218, weekly 1044, monthly 240 and quarterly 80. For each time horizon the threshold of 0.90 is applied to split the observations (approximately). This means 5128*90%= training period and 5218*10%= testing period and the same for all other horizons.

Dear Milad;

Thank you for your help. Actually I cannot clarify the idea more than that and I just wanted to get the direct help to apply my models in Matlab and not to redefine the problem itself. I know that you're using your computational structure to build model but the in my opinion the model is simple and don't need that. What I need is just either the direct codes(or at least the general idea to start writing those codes). Also I just wanted you to direct me in which Matlab toolbox I must use for each group of model. Again simply the idea is to not to build the model but to use all of those mentioned models in Matlab. One series, one factor, (e.g. .0005 , -.006, .0045 , .............n) where n is number of observations (# of return observations in my case). After importing the data in Matlab excel will appear in matrix file. Then I should decide which tool box to use. How to train the model , how to predict the model, how to check the goodness of fit with error measures RMSE, MAPE, R square and AIC, BIC. Again thank you for your help and it seems that you consider the process too complex to deal with because as you said it is not your field of study. If it comes to me it is simple process but just needs special technical support to start applying the models.

Ok thank you. Do I need to write my own function over the space under the custom function. I know this toolbox but don't know how to use.