So how easy is it to predict what a Corsi% will be from one season to the next?
There are 4 scatter plots below showing year over year Corsi numbers for defensemen and forwards (who played at least 20 games last season and the next season). It's then broken down into players returning to the same team, or players who have changed teams (minus those who played for more than one team during the season). The total sample size from all four plots is 4885. Some players do appear more than once, but different ages.
The R^2 for forwards who have changed teams is half of what it is for those returning to the same team. Going to a new team means new linemates and a new coach who uses the player in new situations. It's clear from the pictures above that this transition creates greater variance in shot differential, making it very difficult to accurately forecast this statistic for those moving to a new team. The data above only looks at "this year" and "next year" and doesn't chart whether the players revert to career averages the longer they stay with the new franchise.
There are far more outliers when players change teams, where we see more large swings from good to bad, and visa versa. One of the best examples is David Clarkson, who put up a 62% CF in his last season in New Jersey, then 43% his first season in Toronto. If the Leafs gave Clarkson that giant contract hoping for a good "possession" player, they were in for a rude awakening.
The first thing that becomes immediately clear looking at the images above, there is almost no correlation between Corsi For year over year for defensemen who change teams. Those returning to the same team have a much stronger correlation, where they are more likely to be used in similar situations with the same partner. This does strongly suggest that a defenseman's CF% is heavily dependent on the team he plays for.
TSN writer Travis Yost has claimed in the past that "sweaters don't make the D-Man". I disagreed with him at the time, but had not yet constructed the data set to fully disprove his assertion. There is unquestionably greater variation in year over year Corsi for those who change teams versus the ones who return to the same situations. This does beg the question, how much of the shot differential battle does a single player really have if there is weak correlation after changing teams? Their combination of linemates and the average skill level of opponents they are matched up against is what drives an individual player's Corsi Number.
It seems plausible that many hockey statistical analysts are over ascribing individual player affect on the shot differential, as it tends to depend on the other 9 guys on the ice.