xFIP finds a pitcher's FIP, but it uses projected home-run rate instead of actual home runs allowed. The home run rate is determined by that season's league average HR/FB rate.
For example: In 2002, Randy Johnson had a 2.66 FIP and a 2.44 xFIP -- the difference being that he allowed a 12.9 percent HR/FB rate, when the league average stood at 10.7 percent.
Where "FIP constant" puts FIP on the same plane as league-average ERA: ((Fly balls / league average rate of HR per fly ball x 13) + (3 x (BB + HBP)) - (2 x K)) / IP + FIP constant.
Why it's useful
Like its cousin, FIP, xFIP can be used to portend future performance (as opposed to simply evaluating past results). However, xFIP and FIP differ in how they penalize pitchers for home run allowance. xFIP is predicated on the notion that pitchers have more control over how many fly balls they allow than how many of those fly balls leave the park. As a result, xFIP substitutes a pitcher's homer tally with an estimation of how many long balls that pitcher should have permitted given the number of fly balls he induced.
To determine the latter part of the equation, xFIP assumes a pitcher should have allowed a league average HR/FB rate, which was 15.3 percent in 2019. This assumption is drawn because HR/FB rate can fluctuate a lot from year to year, with pitchers often regressing back toward the league average rate.
The xFIP leaderboard is a nice tool to find buy-low pitching candidates for the rest of the season or heading into a season. That's because xFIP can be a fantastic predictor of future pitching success. xFIP is predictive because its components are related closely to a pitcher's skill level and, thus, tend to be more sustainable than batted ball luck or strand rate, which rely on external factors.