Alisson’s status as the world’s most expensive goalkeeper lasted roughly three weeks until Chelsea snapped up Kepa Arrizibalaga just before the start of the season.
Each decision was based on a clear need – Chelsea to replace the Real Madrid-bound Thibaut Courtois, and Liverpool to ensure what happened in the 2018 Champions League Final wasn’t ever repeated on Jurgen Klopp’s watch.
But is it all that straightforward?
Kepa’s transfer fee had everything to do with his buyout clause, but Chelsea must have felt he was at least approaching Alisson’s level to justify the cost. In hindsight, most would say that Alisson has been the better keeper, but how could we have objectively and meaningfully assessed this at the time? And how might we have considered less expensive options that could produce similar results?
This is difficult, because traditional goalkeeper measurements such as clean sheets and save percentages aren’t truly indicative of performance. A goalkeeper’s value is difficult to quantify, even with advanced metrics such as expected saves (xS), because they only measure how a player performed against the shots they faced. Keepers may have completely different types of saves to make depending on the defensive style of their team and the opponents they face.
So rather than using metrics that might not capture these variables, we can simulate each goalkeeper for every shot, then compare who would concede the fewest goals. To do this, we must be able to accurately simulate how one keeper would manage against another keeper’s shots – or better yet, how each keeper would perform against every shot in a uniform sample. That’s exactly what we’ll do here.
The following table shows an unsurprising duo at the top. When measured against the same shot sample of 2,754 shots from the 2018/19 season, David de Gea and Alisson are capable of saving a team more than a goal every four matches. They truly are season-altering shot-stoppers for any club:
Before diving too deep, we’ll briefly explain the method, then get back to a larger Premier League goalkeeper analysis with a first-of-its-kind shot-stopping simulation.
Here’s how it’s done.
Using tracking data and machine learning, we can create a set of features that measure keepers’ strengths and weaknesses when facing shots in different situations. For example, are they stronger to their left or right sides? How do they handle 1-vs-1 situations? Are they easily beaten at their feet?
Once we generate these features, we combine them into what we call a Player Embedding, which creates a unique data-driven goalkeeper identity. When we plot these embeddings, we’re able to pick up on the unique nuances of each one (see the published paper linked below for full details on how this is calculated).
If we combine these embeddings with a normal xS model, we significantly improve our accuracy in predicting whether a goalkeeper will save a shot on target. This allows us to move into the area of personalised player prediction. As a result, if we want to know how Goalkeeper A would cope facing Goalkeeper B’s shots – or in our case, how Alisson would handle Kepa’s shots – we can now take each player’s unique identity and swap them to simulate what we’d expect to happen.
Now back to the fun part.
NEXT: Why Kepa’s not quite stacking up…