Tracking the results when the
Turnover Difference and CSM agree

Two of the most successful measures to use in predicting the outcome against the spread that we have ever devised are the two "contrarian" indicators: Turnover Difference, and CSM (namely Cumulative Spread Margin, formerly known as Contrarian Bridgejumping).

These tools have both been solid for a span of twenty years now, and while they have had their ups and down in given seasons, there can be little doubt that they have worked well in the big picture. Lately indeed the results have been better than their lifetime average, with 2001 and 2002 being strong contrarian years.

For a refresher on these tools, please see the Turnover Difference Revisited and Contrarian Bridgejumping Revisited articles in the subscriber research section.

To review, the basic turnover difference method requires a 1/2 game (per game average) difference in net turnovers between the two opponents in a matchup to qualify as a play, with the worse turnover team being the call. For the Contrarian Bridgejumping/CSM approach, there needs to be a team with a six points worse than its opponent "average net against the line per game". In both cases you only want to trust the recommendation in the week 5 to week 12 time period (the heart of the season). For reasons that are not entirely clear, the methods work less well from week 13 on.

The turnover rules are good to the tune of 57% over the last twenty years and the CSM picks have been 55.5% (with back to back 60% campaigns). That's some strong picking considering you are dealing with quite a few picks in a given week, and looking at one meager stat per team! Nevertheless the temptation is always strong to see if things could be even better, and so we have run some to see what happens in various situations when you consider both indicators at the same time.

First up then is testing the results when both the turnover differnce and the CSM have an active play on the same team in a game:

Comments: Pretty good then, with the overall performance improving to 59% when you have both factors working at the same time. Retesting for the 2001 and 2002 seasons shows a 41-24 mark, good then for a 63% run (not surprising in that we know the contrarian methods were individually solid in those years). What about in the week 13 on timeframe? The numbers go way down to 52% and should be considered non-predictive.

One interesting zone to note is that games with a 5 to 9.5 spread were impressive: 63% in the initial ten year sample, and over 60% in the two-year retest. Likewise "double contrarian" favorites were hot in the '91 to '00 period, 63%, and also retested at better than 60% in the two most recent seasons.

We ran a slew of other tests, looking for combined values and incorporating the 'lesser' turnover tools of interceptions and fumbles alone, and came up with the following results:

-- INT picks + CSM agree: 56%
-- FUM picks + CSM agree: 61%
-- All three T/O models agree: 56%
-- All three T/O + CSM agree: 57%

So, what's the conclusion? Using Turnover Difference and the Contrarian Bridgejumping factors together can add some extra punch to an already solid approach. As far as individual turnover elements combined, fumbles are the key! In particular, when the T/O fumble pick and overall T/O pick agree without an INT pick, the numbers are outstanding!