Ive prepared the comparison between playoff predictions based on DVOA power ratings and NFL-Forecast playoff predictions in graphical time-series form , from week 9 through week 13. The filled circles represent DVOA predictions while the open circles represent my predictions. Hopefully the team-based color coding that I used is self explanatory. I’ll include the Week 13 DVOA ratings as soon as Mike Harris posts them.
The first set of graphs shows the division races (Ive shown only teams that have had a significant chance since week 9). All 8 division races appear to be converging to selecting a champion, using both the DVOA and NFL-Forecast projections. DVOA ratings did significantly better in predicting the AFC West race, recognizing the superiority of the Chargers compared to the Broncos. In the NFC East race, my software had greater misplaced confidence in the Giants than DVOA, but transferred this misplaced confidence to the Eagles. Neither system really picked out the Cowboys as the division winner significantly better that the other. In the rest of the races, the predictions were comparable, with DVOA converging to apparent division winners slightly faster than NFL-forecast, with the exception of the NFC West race. My system still holds a bit of hope for the Jets, to eek out the AFC East title, while the DVOA has declared the race all but over. In contrast, the Wild Card races are not converging, with only 4 weeks remaining in the season.
The Wild Card race this year appears to be a white noise process, with teams fortunes rising and falling on a weekly basis. The early front runners as predicted by both systems, the Broncos and Giants, have lost much of their lead. It will be interesting to see if either system outperforms the other in the WC race, but the lack of convergence would seem to indicate that any better performance of one system over the other would be attributable to luck rather than reproducible predictive ability. Nevertheless, forecasts like these during dynamic non-converging playoff races do help quantify the race if it is understood to be a snapshot in time.
Only preliminary conclusions can be made, since the playoff races are not over yet, and even when they are, a single season is still a pretty small data set. But this playoff forecasting exercise appears to be pointing to a small but significant advantage of DVOA compared to win-only systems in predicting future results.