You are responsible for an independent forecasting project. Your duties are to choose a monthly or quarterly time series, make a forecast, and evaluate your relative success.
You are required to turn in three reports.
The Project Description, due in class on Tuesday, March 28, will be a brief description of the variable you intend to forecast. You should include the following
· A precise description of the variable you intend to forecast
· The original source for the time-series (e.g. BLS)
· Where you obtained the historical series (e.g. BLS or FRED)
· Which precise date you intend to forecast (e.g. March 2017), where the observation will be released (e.g. BLS webpage), and the exact date of the release (e.g. April 29)
Your project must be turned in at least one day before from the date of the release.
The Project Report is your main forecast report. It is due at the latest by Tuesday May 2 (in class). You report will include the following
· A description of the data series, historical data, and time-series problems of the series
· A description of your forecasting method
· Presentation of your forecasting model and parameter estimates
· Forecasts of the series for the following year. If your series is monthly, this consists of 12 forecasts. If your series is quarterly, this consists of 4 forecasts.
· Both point forecasts and interval forecasts
· You should write your report as a paper. Do not include STATA do files or direct printout. You should include appropriate regression results, but write them in the format for a paper. Describe the models you considered, and why you selected them. Justify your choices. Describe how you selected your model. Present your best model estimates.
· There is not a specific length requirement.
The Forecast Evaluation, due by Thursday, May 4 (in class), will be a short forecast evaluation, where you compare your one-step-ahead forecast with the actual realization. Did the actual value fall within your forecast interval? Would a decision maker have been wise to follow your forecast?
Timing (VERY IMPORTANT):
The time-series you are forecasting must have a new realization sometime between April 27 and the morning of May 4. This is a very narrow window and greatly restricts the possible choices. This restriction is important as you must turn in your Project Report before the realization, and if this is too early in the semester you will not be able to take advantage of the techniques learned in the later parts of the course.
Examples of feasible forecast variables include those available in the following releases:
Metropolitan Area Employment and Unemployment (May 3 Bureau of Labor Statistics)
Personal Income and Outlays (May 1, Bureau of Economic Analysis)
GDP 1st quarter estimate (April 28, Bureau of Economic Analysis)
One component of a successful project is your choice of variable. Since the project involves a forecast evaluation, it is necessary to know that the realization will be unobserved when you submit the Project Report, but observed by the time you submit the Forecast Evaluation (May 4).
In order to make a successful forecast, there needs to be a significant historical record available to you. While there is no unique rule concerning the number of prior observations required, more observations are better than less.
You are free to pick any economic variable, as long as it meets the above criteria. As this is a course in economic forecasting, it needs to be an economic variable (and not, for example, a sports statistic).
One note of caution: Many students are eager to forecast financial series, such as stock prices or exchange rates. This is not recommended. Basic economic theory teaches that changes in asset prices are impossible to forecast, and this turns out to be true in the real world as well. Attempts to forecast the stock market are typically more frustrating than fruitful.
Many economic data sources can be found on the data pages of the resources for economists:
In particular, the Bureau of Labor Statistics and Bureau of Economic Analysis are good sources
Resources for Economists link
Federal Reserve Economic Data (FRED) link
They maintain a near-exhaustive inventory of economic data series. FRED data can be imported directly into STATA using the freduse tool. FRED, however, is not the original source, so you will also need to know where the original data is released (to know the release dates).
Bureau of Labor Statistics link
For historical data, pick an area (e.g. unemployment/national unemployment rate) then look for “CPS Databases”. Look for “One-screen data search”, select the series and data frequency, and “get data”. You will see a table of numbers. For some series, you can “change output options” to extend the sample. Under “more formatting options” you can change the format for downloading.
Bureau of Economic Analysis link
The BEA data is well organized.