1. FORECAST OF NEW DOMESTIC AUTO PRODUCTION MGT 267 Applied Business Forecasting Professor Mohsen Elhafsi Group 1: Hui Guo Minjia Xu Ao Gao

2. Introduction Objective Forecast the domestic new automobile production in the U.S. in 2015 and 2016. Methods Holt’s Method Multiple Regression Time Series Decomposition ARIMA Model Combined Model

3. Series – Dependent Variable Domestic New Auto Production (De-seasonalized)

4. Series – Independent Variables Urban Consumer Price Index (CPI-U)

5. Series – Independent Variables Unemployment Rate

6. Series – Independent Variables Gross Domestic Production (De-seasonalized)

7. Series – Independent Variables Disposable Personal Income (De-seasonalized)

8. Series – Independent Variables Inflation Rate

9. Series – Independent Variables Gas Price

10. Method – Holt’s Original series is de-seasonalized. A trend in original series.

11. Method – Holt’s Accuracy

12. Method – Multiple Regression Suitable for all type of data All 6 independent variables plus Time New Autos = 2,029.31+Time*14.13 + CPI * (-15.91) + Unemployment Rate * (-7.40 ) + GDP * 0.042602 + DPI * 0.000012 + Inflation * 12.32 + Gas Price*40.90 Remove Inflation and Gas Price New Autos = 1,079.17+Time * 7.25+ CPI * (-9.84 ) + Unemployment Rate * (-5.49 ) + GDP * 0.053008 + DPI * 0.000017

13. Method – Multiple Regression Only take Unemployment Rate, DPI and Gas Price New Autos = 216.42 + Unemployment Rate * (-20.21) + DPI * 0.00004 + Gas Price * (-45.91) Only take Unemployment Rate, GDP and Gas Price New Autos = -66.32 + Unemployment Rate * (-10.50) + Gross Domestic Product * 0.041994 + Gas Price *(- 35.63)

14. Method – Multiple Regression Accuracy

15. Method – Time Decomposition Fits data set with trend, seasonality, and cyclical factor. Holt’s Method for Cyclical Factor

16. Method – Time Decomposition Fits data set with trend, seasonality, and cyclical factor. Linear Regression for Cyclical Factor

17. Method – Time Decomposition Accuracy

18. Method – ARIMA Data set is required to be stationary. Try different method, compare the accuracy measures, choose the best two methods. ARIMA(p,2,q)

19. Method – ARIMA ARIMA (2,2,2)

20. Method – ARIMA Accuracy – ARIMA(2,2,2)

21. Combined Model Time Series Decomposition & ARIMA (2,2,2) Original = -0.13517 *ARIMA+1.14 * Decomposition

22. Accuracy Combined Model

23. Conclusion & Improvement After forecasting with all methods learned in class, we chose the most prospective four to further analyze. For our data, a combined model with Time Decomposition Holt and ARIMA provides the best results. We could try data that are not deseasonalized to test if we can get a more accurate forecast.