Forecasting stock prices time series based on independent factors using ARIMA model I am trying to forecast time series of stock for a particular case in which closing value of the stock depends on independent factors which is in which infact another time series.
Situation is like I have to predict value for tomorrow's stock based on some independent variables which are also defined for each day. One time series dependent on another independent time series. But it might not be mapped series. Value of dependent series can be more correlated to independent variable's value of yesterday or sometime in past but also dependent on past values of series itself.
I am going for ARIMA modeling through SPSS's expert modeler for working out on this problem.
Please someone throw light on this and how I should go about working on this or primarily how to approach this problem?
Spare me for this question is very basic in stats and I am just a beginner.
SOS= Scared of stats.
Sincere Thanks.
 A: You should investigate Transfer Function Modelling ( A.K.A. Multiple time series modelling with 1 endogenous series ) and take care to identify via INTERVENTION DETECTION any Pulses, Level Shifts , Seasonal Pulses and or Local Time Trends. Furthermore finding break-points in parameters would be a useful diagnostic while also checking for deterministic variance changes in the errors.
A: It probably goes without saying that forecasting equity returns is extremely difficult. The most relevant factor in determining the success or failure of your analysis will certainly be the quality of your economic analysis -- that is, the quality of your independent variables/model selection. The quality of your statistical model is secondary.
I would suggest first considering altering the problem to the prediction of excess returns, or returns over some benchmark, like the S&P 500. Otherwise, the majority of the problem is predicting global supply and demand for equities as an asset class, not the specific equity under consideration.
If the economics and fundamentals behind a model are solid, most people just use linear regression. That is probably your best bet. Simply regress the future stock return on the present values of the independent variables.
