Business Forecasting Essay

3645 Words Aug 22nd, 2012 15 Pages
Content Introduction 1 Part 1. Examine the data, looking for seasonal effects, trends and cycles 2 Part2. Dummy Variables Model 3 Linear trend model 3 Quadratic trend model 5 Cubic trend model 7 Part 3. Decomposition and Box-Jenkins ARIMA approaches 8 First difference: 10 a. Create an ARIMA (4, 1, 0) model 10 b. Create an ARIMA (0, 1, 4) model 11 c. Create an ARIMA (4, 1, 4) 11 d. Model overfitting 12 Second difference 13 Forecast based on ARIMA (0, 1, 4) model 13 Return the seasonal factors for forecasting 14 Part 4. Discussion of different methods and the results 15 Comparison of different methods in terms of time series plot 15 Comparison of different models in terms of error 17 Assumptions and the …show more content…
| | B | Std. Error | Beta | | | 1 | (Constant) | 16582.815 | 866.879 | | 19.129 | .000 | | TIME | 765.443 | 26.000 | .970 | 29.440 | .000 | a. Dependent Variable: creditlending |
According to the output, it is obvious that Q1, Q2, Q3 has been removed from the regression model. This is because that instead of removing the non-significant variables step by step, Stepwise method in SPSS is used to get the final model in which shows that the p-value of TIME is significant at 5% level, while non-significant model such as Q1, Q2 and Q3 are directly taken out. In addition, the adjusted R-square is 93.9% which is a very good fit. Hence, dummy variable model is built as:
Trend-cycle = 16582.815 + 765.443*TIME + error

By putting TIME dummy variable into this formula, we can draw a new time-series chart with an assumption that the error is white noise. The time-series plot (see next page) displayed both raw data trend-cycle and forecast trend of this model. It is clear that the modeling data fluctuate around this model and shows a relatively acceptable fit, while in terms of holdback data, it displays a bad fit which the model maintains an up-ward trend whilst the holdback data falls down. As for residual component, according to time-series plot and ACF of unstandardized residual, there still has a significant structure left. This is to say that there are significant autocorrelations on the ACF plot

Related Documents