a more complex window, to select the variables (simple) and some details of the model (complex). First: define the type of dates ... select the desired transformation of dependent variable, before it is modeled: ... 7c) the confident interval width.
Click on "Analyze", then on "Forecasting", and finally on "Create Models" They appear two windows: - a simple one, to define the dates - a more complex window, to select the variables (simple) and some details of the model (complex) First: define the type of dates (the type of periodicity of the data), and close the window Second: select the variables, that is 1) the variable target of your analysis must to be selected as "dependent" (the core of the ARIMA) 2) the exogenous variables must to be selected as "independent" Third: to choose the details of the model you must operate again in the variables page, and in other 5 pages of the same window (statistics, graphs, output filter, saving, options) 1) method (in the variables page): select ARIMA 2) criteria (in the variables page): three sub-pages 2a) model (first criteria sub-page): - specify the desired orders [autoregressive (p), difference (d) and moving average (q)] - select the desired transformation of dependent variable, before it is modeled: i) no transformation ii) square root iii) natural log - indicates if you want a constant in the model 2b) transfer function (second criteria sub-page): it is useful to specify the manner in which past values of independent (exogenous) variables are used to forecast future values of the dependent variable - numerator order, for lags: for example, if you specify 1 means that the model consider not only the current time but also one lag time - denominator order, for deviations: for example, if you specify 1 model predicts the current value of the dependent considering also the deviations from the mean value of the independent series one time period in the past - difference order: it is necessary when there are trends, and it is used to remove their effects -delay: setting a delay causes the independent variable's influence to be delayed by the number of intervals specified; for example, if the delay is set to 3, the value of the independent variable at time t doesn't affect forecasts until three periods have elapsed (t+3) -transformation: you can specify a transfer function for the independent variables i) no transformation ii) square root iii) natural log 2c) outliers (third criteria sub-page): - choose if you want on not detect the outliers - if yes, you must select the option to perform automatic detection of outliers: you can choose one or more of the following i) additive ii) level shift iii) innovational
iv) transient v) seasonal additive vi) local trend vii) additive block - choose if you want to model some specific time points as outliers (that is: select this option to specify particular time points as outliers); use a separate row of the outlier definition grid for each outlier, and enter values for all of the cells in a given row. 3) statistics page: you can choose a lot of statistics 3a) choose the display of fit measures, of the Ljung-Box statistic, and of the number of outliers 3b) choose the fit measures 3c) choose the statistics for comparing models (you can run more than one model) 3d) choose statistics for a single model 3e) choose the display of forecast 4) graphs page: you can choose several methods for diplay of plots 4a) for the comparison of models 4b) for a single model 4c) for series 4d) for residual autocorrelation function 4e) for residual partial autocorrelation function 5) output filter page: you can choose if include in the output 5a) all the models (remember that you can run more than one model) 5b) the criterions to include a selection of the best fitting model 5b) the criterions to include a selection of the poorest fitting model 6) saving page: you can choose if you want save model predictions, confidence intervals, and residuals as new variables in the active dataset; VERY IMPORTANT: in the saving page you must also indicate the name of export output file (it is a XML file); only if you generate this file it will be very easy to make simulations, simply recalling this file; if you do not create this file, it will be necessary to rewrite the model for each simulation. 7) options (forecast) page: you can choose 7a) the forecast period 7b) how to treat the missing values (if they exist) 7c) the confident interval width 7d) the prefix for model identifiers in output (models are distinguished with unique names consisting of a customizable prefix along with an integer suffix) 7e) the maximum number of lags shown in residual autocorrelation function and residual partial autocorrelation function output.