Scorecasting Economists

The Biggest Ever FA Cup Shocks

We’ve all done it – tried to argue that that Cup shock our team produced was bigger than some other team’s. Or tried to argue that our hated rival’s humiliating defeat was well worse than any we suffered.

There are more objective ways to think about this. Simon Gleave uses bookmaker odds, which are known to be good predictors of football match outcomes. But that only works as far back as we can get bookmaker odds for – usually about twenty years or so. But the FA Cup has been going on since November 1871 – how can we compare Crystal Palace’s 1907 win at Newcastle United to Wigan’s Cup 2013 Final win over Man City?

One way is to use league positions. But these are only as good so long as teams are in the Football League system. Crystal Palace were in the Southern League in 1907. Another is to use Elo ratings. These are an ingredient in our Scorecasting model, and they can be applied to all sorts of contexts, sporting or otherwise, to get a relative measure of the quality of participants in a tournament.

Elo ratings generate a prediction for each match, and hence a prediction for each FA Cup match, retrospectively. For example, when Newcastle hosted Crystal Palace on January 12 1907, the prediction was 0.79. Roughly speaking, Newcastle were 79% likely to win (we’re ignoring the draw here). Alternatively, when Wigan hosted Man City on February 19 2018, the prediction was 0.12. Wigan were roughly speaking 12% likely to win, City 88%.

We can then denote 1 for a home win, 0 for an away win, and calculate the “size” of the shock by taking the difference. Hence when Palace won at Newcastle, the shock was 0.79, while when Wigan beat Man City, it was 0.88.

Doing this, we can rank shocks. Here is the all-time top ten:

Wrexham’s famous win against Arsenal in 1992 is the biggest ever FA Cup shock. Bradford’s stunning win at Chelsea in 2015 is the second biggest. Quite a few of these happened in the time of bookmaker odds, but a number didn’t – not least Colchester’s win over Leeds United in 1971, Halifax’s win over Man City in 1980, and Bournemouth’s win over Man United in 1984.

But there aren’t many away wins in the list, and a potential explanation for this is that we do not control for the home advantage. Bradford’s win at Chelsea was all the more stunning because it was at Stamford Bridge rather than in Bradford’s own surrounds at Valley Parade. Indeed, Crystal Palace’s win at Newcastle is the 70th largest ever shock, but the 20th largest away win shock.

So how should we control for the home advantage – teams win more often at home than they ought to given their relative quality?

One answer is to use a common forecast evaluation technique – the Mincer-Zarnowitz linear regression method. This regresses outcomes on forecasts, plus other information that may have mattered. It hence corrects for biases in forecasts, if there are any. In our case, home advantage creates a bias – one that has varied through time, since home advantage has fallen over the years. Once we correct, here are the biggest shocks:

Significant amounts change. Not least, the top nine shocks are all away wins, and the ordering has changed quite dramatically. The first column is the original ordering, and shows that Wrexham’s win has dropped to 14th biggest.

The ordering between home and away shocks has changed. Oldham’s 2013 3-2 win over the Liverpool featuring Suarez and Gerrard is now the largest home shock, perhaps reflecting that home advantage wasn’t as big in 2013 as it was in 1992. And Bradford’s win drops from 2nd to 4th. Wimbledon’s 1975 win at Burnley is the second biggest, but the largest, perhaps surprisingly, is a win for Second Division Norwich City at Man United in 1967. United won the league that season, and Dennis Law scored their goal in a 2-1 defeat. But the Giant Killer’s website puts this only at 64th biggest shock

Finally, Crystal Palace’s 1907 win at Newcastle drops down, but another Southern League shock enters at number three here – QPR winning at Wolves in 1900.

Champions League 2019-2020 season conclusion

This evening, the Champions League returns after its Covid-19 enforced suspension.

Nothing is normal, of course. Not least the look of our forecasts:

KDR, ICL, TFI an LIN are all minnows of the global game. Yet they get to play on the same heady Champions League days/nights as Manchester City, Real Madrid, Juventus and Lyon. That’s right – the 2020-2021 Champions League has begun, even while the 2019-2020 one hasn’t even finished yet.

Juventus have a 1-0 deficit to overturn, and they’ll likely do it, but may need extra time to force overall victory. Man City are 58% to beat Real Madrid, too.

Jean-Louis Foulley has as always produced his forecasts – see his recent presentation on them, too, here.

Championship, R46 (22 July, 2020) — Bees or Baggies? Model suggests West Brom remain favourites to go up, although Brentford are marginally more likely to win tomorrow.

The table below gives the Model’s forecasts for Round 46 of the English Championship, along with forecast probabilities estimated from current online bookmaker odds.

EFL Championship R46 2020 forecasts, RED
  • Expected Goals: the forecast average number of goals the Model expects for Home and Away teams
  • Outcome Probs: the model predicted % chance of either a Home or Away win, with 100 minus these two numbers being the % chance of a draw.
  • Score Picks: the Most likely forecast scoreline outcome, as well as the most likely conditional on the most likely result outcome happening.
  • Home wins / Draws / Away wins: the predicted % chance of various potential scoreline outcomes of the match.
  • Mean odds: Estimates of online bookmakers’ average probability forecasts for the Home and Away teams to win, which could be compared with Outcome Probs.

Premier League, R37 (21-22 July) — Villa need a win against Arsenal, but only have a 27% chance

The table below gives the Model’s forecasts for some of Round 37 of the Premier League, along with forecast probabilities estimated from current online bookmaker odds.

EPL Forecasts R37 2020 update, RED
  • Expected Goals: the forecast average number of goals the Model expects for Home and Away teams
  • Outcome Probs: the model predicted % chance of either a Home or Away win, with 100 minus these two numbers being the % chance of a draw.
  • Score Picks: the Most likely forecast scoreline outcome, as well as the most likely conditional on the most likely result outcome happening.
  • Home wins / Draws / Away wins: the predicted % chance of various potential scoreline outcomes of the match.
  • Mean odds: Estimates of online bookmakers’ average probabbility forecasts for the Home and Away teams to win, which could be compared with Outcome Probs.

Premier League, R37 (18-22 July) — Norwich looking for some consolation today? A puncher’s chance (26%)

The table below gives the Model’s forecasts for Round 37 of the Premier League, along with forecast probabilities estimated from current online bookmaker odds.

EPL Forecasts R37 2020 RED
  • Expected Goals: the forecast average number of goals the Model expects for Home and Away teams
  • Outcome Probs: the model predicted % chance of either a Home or Away win, with 100 minus these two numbers being the % chance of a draw.
  • Score Picks: the Most likely forecast scoreline outcome, as well as the most likely conditional on the most likely result outcome happening.
  • Home wins / Draws / Away wins: the predicted % chance of various potential scoreline outcomes of the match.
  • Mean odds: Estimates of online bookmakers’ average probabbility forecasts for the Home and Away teams to win, which could be compared with Outcome Probs.

Championship, R45 (17-19 July, 2020) — Baggies ready to Boing? 45% chance of 3 points tonight at Huddersfield

The table below gives the Model’s forecasts for Round 45 of the English Championship, along with forecast probabilities estimated from current online bookmaker odds.

EFL Championship R45 2020, RED
  • Expected Goals: the forecast average number of goals the Model expects for Home and Away teams
  • Outcome Probs: the model predicted % chance of either a Home or Away win, with 100 minus these two numbers being the % chance of a draw.
  • Score Picks: the Most likely forecast scoreline outcome, as well as the most likely conditional on the most likely result outcome happening.
  • Home wins / Draws / Away wins: the predicted % chance of various potential scoreline outcomes of the match.
  • Mean odds: Estimates of online bookmakers’ average probability forecasts for the Home and Away teams to win, which could be compared with Outcome Probs.

Championship, R44 (14-16 July, 2020) — Rams need a win, but only 30% chance of that tonight at Cardiff

The table below gives the Model’s forecasts for Round 44 of the English Championship, along with forecast probabilities estimated from current online bookmaker odds.

EFL Champ Forecasts R44 2020, RED
  • Expected Goals: the forecast average number of goals the Model expects for Home and Away teams
  • Outcome Probs: the model predicted % chance of either a Home or Away win, with 100 minus these two numbers being the % chance of a draw.
  • Score Picks: the Most likely forecast scoreline outcome, as well as the most likely conditional on the most likely result outcome happening.
  • Home wins / Draws / Away wins: the predicted % chance of various potential scoreline outcomes of the match.
  • Mean odds: Estimates of online bookmakers’ average probability forecasts for the Home and Away teams to win, which could be compared with Outcome Probs.

Premier League, R36 (14-17 July) — Chelsea given 73% chance of taking advantage of relegated Canaries

The table below gives the Model’s forecasts for Round 36 of the Premier League, along with forecast probabilities estimated from current online bookmaker odds.

EPL Forecasts R36 2020, RED
  • Expected Goals: the forecast average number of goals the Model expects for Home and Away teams
  • Outcome Probs: the model predicted % chance of either a Home or Away win, with 100 minus these two numbers being the % chance of a draw.
  • Score Picks: the Most likely forecast scoreline outcome, as well as the most likely conditional on the most likely result outcome happening.
  • Home wins / Draws / Away wins: the predicted % chance of various potential scoreline outcomes of the match.
  • Mean odds: Estimates of online bookmakers’ average probabbility forecasts for the Home and Away teams to win, which could be compared with Outcome Probs.

Championship, R43 (10-12 July, 2020) — Last chance for Luton? They have a 29% chance of beating Huddersfield tonight.

The table below gives the Model’s forecasts for Round 43 of the English Championship, along with forecast probabilities estimated from current online bookmaker odds.

Champ forecasts EFL, R43 2020, RED
  • Expected Goals: the forecast average number of goals the Model expects for Home and Away teams
  • Outcome Probs: the model predicted % chance of either a Home or Away win, with 100 minus these two numbers being the % chance of a draw.
  • Score Picks: the Most likely forecast scoreline outcome, as well as the most likely conditional on the most likely result outcome happening.
  • Home wins / Draws / Away wins: the predicted % chance of various potential scoreline outcomes of the match.
  • Mean odds: Estimates of online bookmakers’ average probability forecasts for the Home and Away teams to win, which could be compared with Outcome Probs.

Premier League, R35 (11-13 July) — Watford on the brink: only 30% chance of 3 points at home to Newcastle

The table below gives the Model’s forecasts for Round 35 of the Premier League, along with forecast probabilities estimated from current online bookmaker odds.

EPL Forecasts R35 2020, RED, 2020
  • Expected Goals: the forecast average number of goals the Model expects for Home and Away teams
  • Outcome Probs: the model predicted % chance of either a Home or Away win, with 100 minus these two numbers being the % chance of a draw.
  • Score Picks: the Most likely forecast scoreline outcome, as well as the most likely conditional on the most likely result outcome happening.
  • Home wins / Draws / Away wins: the predicted % chance of various potential scoreline outcomes of the match.
  • Mean odds: Estimates of online bookmakers’ average probabbility forecasts for the Home and Away teams to win, which could be compared with Outcome Probs.