Turnovers are one of the most important statistics in the NFL. They're also one of the most unpredictable. Every season, we see teams that are registering a lot of turnovers suddenly stop getting takeaways altogether in the middle of the campaign. Or teams without many turnovers will abruptly start turning the ball over. Future turnover totals are a lot harder to predict than other stats such as yardage or completion percentage.
One big reason that turnover numbers are inconsistent has to do with fumbles. Essentially, recovery of a fumble has no predictive value. Teams that recover a high percentage of fumbles in one season will sometimes recover very few fumbles the next, even if there were more fumbles in their games. The reverse is also true.
Interceptions are almost as unpredictable as fumble recoveries. Over the past five years, the correlation coefficient (explained here) for offensive interceptions from the first half of the season to the second half is just 0.17. The correlation on defense is even worse, just 0.08. The defenses with the most first-half interceptions over the past five years averaged 12.9 interceptions in the first nine weeks of the season, then just 6.6 interceptions in the last eight. The defenses with the fewest first-half interceptions went from 2.5 picks in the first nine weeks to an average of 5.9 in the last eight.
We know that we can "predict" fumble recoveries by looking at raw fumble totals. Roughly half of all fumbles end up recovered by each team, although the rates are different for different types of fumbles (including sacks, aborted snaps, rushes, pass completions). Is there a stat that helps predict interceptions in the same way that total fumbles predict fumble recoveries?