ARMA forecasting

Hello all. I have a time series, and I have to apply the four steps of ARMA estimation ( chose the orders, compute the coefficients of the suitable ARMA model), and after that I have to predict the next 10 values. I am a beginer in this domain, and I have insatlled the Matlab 2010b. Thank you for your Help

[Walter Roberson]

http://www.mathworks.com/matlabcentral/answers/29922-why-your-question-is-not-urgent-or-an-emergency

[Wayne King]

Hi Marina, do you have the System Identification Toolbox?

[Marina]

yes I have it

[hamed Aghapanah]

Dear,

Can it forcasting a stochastic signal?

[Abolfazl Nejatian]

this is my forcasting code which is allow you predict your time series data with LSTM, CNN, and MLP Networks.

https://www.mathworks.com/matlabcentral/fileexchange/69506-time-series-prediction

Answer 1

[Rajiv Singh]

Use ARMAX function in System Identification Toolbox to estimate an ARMA model, as in model = arma(data, [na nc])

In R2012a: use FORECAST function to forecast the response. See http://www.mathworks.com/help/toolbox/ident/ref/forecast.html

Previous releases: See http://www.mathworks.com/matlabcentral/answers/12186-forecasting-function

[Marina]

thank you. but my question is how to chose the ordes na and nc? (I think I have to calculate ACF and PACF) and after that how to calculate my coeficients

[Rajiv Singh]

System identification toolbox has no automated way for you to pick na/nc. The ARXSTRUCT/SELSTRUC commands let you for this for a ARX model structure. You could subject your data to this command and determine "na" that way. Then you can pick nc<=na as a start. The "optimal" values of na and nc orders are ultimately going to be those for which the prediction errors are minimized against a validation data set.

See COMPARE and RESID commands. The most basic test of optimality would be that the 1-step ahead prediction fit to validation data is maximized, as revealed in the COMPARE plot.