Relatively recently, a Current Employment Report has been released, 1st Quarter of 2015. In February, I made a forecast of 1st Quarter 2015 headline employment. The forecast amount was 63,010. Headline employment for the period was 62,760, which means the forecast was within my margin of error of 1,050 (95% confidence). To reflect, my forecasts of the average hourly private sector wage, employment, and CPI were all overstated. The employment forecast was not so bad, but I'd like to compare it to a naive forecasting methodology (rolling average annual rates of increase for windows of different periods of years):
VAR forecast: 63,010 (+250)
1-year window: 62,210 (-550)
2-year window: 62,320 (-440)
3-year window: 62,450 (-310)
4-year window: 62,300 (-460)
5-year window: 62,390 (-370)
Comparatively, this one period shows a better performance for the VAR. Nevertheless, for now, consider the following forecasts for 2nd Quarter 2015 (which has already past, but is not published yet, so still unknown):
1-year window: 62,920 (+160 jobs)
2-year window: 62,730 (-30 jobs)
3-year window: 62,450 (+80 jobs)
4-year window: 62,300 (-10 jobs)
5-year window: 62,720 (-40 jobs)
If these forecasts are to be believed, there will probably be little change in headline employment between March and June 2015. A major drawback to this form of forecasting is that, since it is based on lagged rates of change, once a growth trend ends, the forecast will be thrown off. If all one uses is a simple extrapolative technique, then even a rate of change which should be able to be anticipated because of known or suspected (perhaps by some type of model with structural variables) relationships between available data would not be accounted for.
In the future, I will probably add back a VAR model to forecast a number of variables together, but some simplified technique would probably be worthwhile, just to consider an alternative benchmark.