Research Paper Instructions
- Define the chosen variable
- New one-family houses sold (NHS)
- Collect the data for the variable
- Data source
- www.economagic.com or other sources
- Monthly data and not seasonally adjusted (NSA)
- Time period –January 1975 to the present
- Enter the data in Excel according to ForecastX format
- 1/1/1975 for Jan. 1975 and 2/1/1975 for Feb. 1975
- Plot the graphfor whole series
- ForecastX — Preview
- ACF for the whole series
- ForecastX –Analyze
- Identify the trend
- ACF for the first differenced series
- ForecastX – Analyze – Differencing – Non-Seasonal >1
- Identify seasonality
- Decide the sample period for model selection
- Historical period: 1/1/1975-12/1/2010
- Holdout period: 1/1/2011-12/1/2011
- Select models according to the data pattern
- Table 2.1, p. 58
- Time-series models: Modified naïve model, Winters’exponential smoothing, Time-series Decomposition, and ARIMA
- Regression mode:
- NHS = f (IR, DPI, dummy variables)—need data for IR (30-year mortgage rate), DPI(disposable personal income) and dummy variables (for seasonality)
- Perform estimation and forecasting for the data in the historical period, 1/2001 – 12/2010
- ForecastX – Forecast Method – choose models
- ForecastX – Statistics — RMSE
- For regression model, standard report shows the forecasts for independent variables
- Compare MAPEs and RMSEs of different models for the historical and holdout periods
- Perform ex-ante forecast – 11/2011-12/2012
- Use the whole series, 1/2001 – 10/2011
- Either choose the best model or combine two models