Recently I published a list of my top young prospects from the English Premier League’s 2014-15 season on the OptaPro blog. As I wrote there, the idea of the list was to cast a broad net, so that players of great potential would be identified as early as possible. This approach works well for clubs that can afford one or two false positives – supposed stars who turn out to be duds. But what about clubs that can’t afford to make a mistake?
No model is absolutely foolproof, but I thought it would be interesting to look at a slightly different cutoff for inclusion in my list. I wanted to find players who looked like they could do it all, contributing to shots and advancing the ball into dangerous areas at rates well above the league average, while defending equally well.
Not surprisingly, it’s a shorter list. All the players listed here were 20 years old or younger on 1 August of the season that led to their selection, and all played at least 500 (not 360) minutes. Some played that number of minutes or more in earlier seasons but didn’t make the cut. The other caveats I described on the OptaPro blog still apply. Here are the prospects:
2010: Rafael, Smalling, Sturridge, Wilshere
2011: Oxlade-Chamberlain, Ramsey
2012: Caulker, Coutinho, Jenkinson, Nastasic, Shelvey, Sterling, Ward-Prowse
2013: Chambers, Davies, Deulofeu, Shaw
2014: Bellerín, Can, Januzaj, Manquillo
Again, as before, if the names here aren’t surprising, that’s good. The models are useful if they can identify known stars in a league that’s well understood; then the models can be applied to other leagues with trust in the results.
One player who didn’t appear on the OptaPro list did make this one: Javier Manquillo. Many others didn’t, though, and some appeared in different seasons, so there may be quite a few false negatives implied here. But remember that this list is designed for clubs with limited budgets. They can’t afford to make mistakes, and they needn’t buy every star on the market, either. In other words, the key is to reduce the false positives as close as possible to zero; the false negatives don’t matter.
By those criteria, I think this list works pretty well. I’d be glad to have any of the above players in my squad, and most of them were identified at reasonable times. A word of caution, though: As I explained on the Analytics FC podcast, players in some positions – especially central defenders – may not develop every part of their game so early in their careers. So while this different cut of the models might excel in picking fullbacks and midfielders, the original version might work better elsewhere on the field.