[Enoch Root]

Another massive challenge here is that there is far too much noise in the Corsi/Fenwick data.

One of the first things you do when trying to evaluate the predictive power of a statistic, is to isolate it and eliminate as many other variables as possible. With Corsi, that is pretty much impossible. Every aspect of a game is included in the data: all of the players, all of the bounces, all of the saves, all of the blocked shots, offsides, stumbles, sick or injured players, etc, etc. All of it is wrapped up into one nice big package.

Then predictions are made. The easiest prediction in the world is regression to the mean. If a team or player is scoring a lot of goals, or if a goalie is making a lot of saves, it is not hard to predict that they might slow down at some point.

The challenge is in determining when and why.

Currencies are a good example of what I mean. Everyone likes to predict where the dollar is going. And truth be told, it only has two places to go: up or down. And everyone likes to brag that they called it right. But unless they actually predicted when and why it moves, all they have done is land on the same side as the coin toss. There are an incalculable number of variables that affect currency prices. And all of the known facts/events/expectations are already priced in. So predicting what will happen is essentially impossible. But there are only two outcomes in the simplest sense. So people think they in fact did predict the outcome.

The same sort of thing happens here. People see a high PDO or low Corsi or whatever, on a successful team, and they predict that the success can't continue.

Well, guess what? Of course it can't continue! The most reasonable expectation - in EVERY circumstance - is a regression back to the middle of the pack.

So did Corsi accurately predict that Toronto would fall off a cliff? Maybe. Or maybe some of the myriad of other factors hidden in the data changed. Hard to say.

But their prediction was 'right'. So they are more confident in making their next prediction.

Will Calgary regress? Almost assuredly. But what does that mean? Do they start scoring less on the same number of shots? Do they start taking more shots? Do their goalies get better? Worse? Do they acquire different players and produce different results?

The answer is who knows? Corsi doesn't isolate the variables enough to give us any really useful information to predict when or why something might change or not be sustainable.

But it's the best people have. And people want to do analyses and make predictions and sound smart. And sooner or later, a given team's success will revert to the middle of the pack and everyone who predicted it can say they were right.