Project Part5: Model Identification
FOR THIS PROJECT PART, PLEASE USE YOUR Y VARIABLE ONLY.
- Model Identification (page 407): Run your ACF and answer the following questions based on the ACF. (i) Is your time series data (Y), integrated or not? (ii) Please write the integration of order? I(0), I(1), I(2)? (iii) Is your data stationary or non-stationary?
- Take difference(s) if necessary to get your data stationary. Plot the differenced data. You don’t need to plot it if your data is I(0).
- Plot ACF and PACF of your STATIONARY data.
- By comparing the ACF and PACF plots on pages 401-403 to your plots, determine your the preliminary ARIMA model (starting tentative values for nonseasonal p and q).
- If you have seasonal component, you need to fit the seasonal model as well by clicking the relevant bottom on ARIMA window and filling in the relevant p and q for seasonality.
- Model Estimation (page 409): Run the first ARIMA model using MINITAB -Stats>Time Series > ARIMA (copy and paste the output).
- Model Checking (page 410): Check your model for accuracy.
(a) Significant AR and/or MA components (indicated by the t-values in the 1st part of the output).
(b) Random errors (large p-values will indicate random errors, small p-values will indicate pattern)
- If your model is not adequate yet, keep trying different AR, MA or combination models until adequacy is reached.
- Once the model is adequate, plot the residuals for the final model.
- Forecasting with the Model (page 411): Once the model is adequate, forecast 10 out of sample forecasts for the final model.
- Compare your 10 forecasted values from the ones you got with regression analysis. How different/similar are they? How do the model errors compare?
- Explain which method you would choose to estimate and forecast your revenue with and WHY?