Scorecasting Economists

Lower League Forecasts, October 24

Lower league football is in the midst of a busy spell, as teams play their third match in a week this weekend, with another to follow on Tuesday – Covid-19 excepting.

What might happen? While the uncertainty surrounding football this season might appear to be Covid-19 related, there is no actual evidence for it. So we’re not adjusting our forecasts.

Here they are for Leagues One and Two, following the usual format (abbreviated team names, expected goals (via a Poisson regression), match outcome probabilities, most likely scoreline picks, then the probability of each scoreline occurring, and finally bookmaker odds from Oddsportal):

Championship forecasts 23/10-24/10

The Championship returns tonight with forecasts in the table below. Nottingham Forest play Derby County (as Carl Singleton has reminded me, ‘the Brian Clough trophy’) in the first match with Forest predicted to win, with bookmakers agreeing with us.

This prediction is calculated using past results, including goals, whether a team wins or not in their most recent match and also in the last 6 matches. The teams Elo rating is also used.

Expected goals are those drawn from a Poisson model based on past results given the information listed above from recent games, and Elo ratings. For instance, Reading are predicted to score 1.36 goals and Rotherham 0.87.

We then sum up all these probabilities to give an overall probability of each outcome, under “Outcome Probs”.

The most likely scoreline is then in the “Score Picks” columns under “Most”. This is often 1-1, since 1-1 historically is the most likely scoreline, though not exclusively, with Reading predicted to win against Rotherham.

The column “Cond” is the most likely scoreline conditional on the most likely outcome occurring – that’s usually, but not always, a home win since our forecast model builds in the home advantage.

The final two columns are the average bookmaker prices for match outcomes according to Oddsportal

Bookmakers seem to think a Norwich victory against Wycombe with a greater degree of certainty (74.6% chance)  than than us (a mere 57.4% chance!) but we both agree that the outcome will be similar.

Premier League forecasts for Weekend of 23/10-26/10

Premier League action returns tonight with the chance for Aston Villa to temporarily go top of the league if can avoid defeat against Leeds. We think a Villa win is likely. Why this is, is explained below.

This prediction is calculated using past results, including goals, whether a team wins or not in their most recent match and also in the last 6 matches. The teams Elo rating is also used.

Expected goals are those drawn from a Poisson model based on past results given the information listed above from recent games, and Elo ratings. For instance, both Liverpool and Man City are predicted to score at least 2 goals (2.24 goals for Liverpool and 2.14 goals for Man City).

We then sum up all these probabilities to give an overall probability of each outcome, under “Outcome Probs”.

The most likely scoreline is then in the “Score Picks” columns under “Most”. This is often 1-1, since 1-1 historically is the most likely scoreline, though not exclusively.

The column “Cond” is the most likely scoreline conditional on the most likely outcome occurring – that’s usually, but not always, a home win since our forecast model builds in the home advantage.

The final two columns are the average bookmaker prices for match outcomes according to Oddsportal

We can see below that our model closely agrees with bookmakers that Liverpool are going to win at home to Sheffield Utd for instance. A clear 2-0 victory is predicted as the most likely result (despite Liverpool being without Van Dijk and Alisson – because the model doesn’t take into account individual player absence).

Generally speaking though there isn’t a lot of difference between our model predictions and the bookmakers odds this week.

Champions League Forecasts

The Champions League gets underway tonight. Our forecasts are posted here in the usual format. Please ignore PSG’s forecast though, as a bug that entered last season appears to still be present – the model’s a little rusty in terms of getting up to speed for European club competition this season.

Related, for some reason the bookmaker prices for these matches haven’t made it, either.

As each time, we have forecasts from Jean-Louis Foulley, whose forecast for PSG against Man United will be much more reliable than ours, for sure:

Lower League Forecasts, October 17

The Covid-19 season rumbles on. After a week of intense off the field trials and tribulations, we’re on the field. What can we expect? More of the same?

If so, our forecasts will be half decent, since they’re based on past results. It’s not a crystal ball, it’s just projecting patterns in the past into the future – extrapolating, the essence of forecasting. They churn out each week, they’re based on Elo ratings, and information from recent games (goals scored and outcomes from the most recent match, and also the last six).

Here, we look at Leagues One and Two, the third and fourth divisions of English football.

Below are the League One forecasts. The table has further left the abbreviations for team names taken from Reuters (though with a bug for AFC Wimbledon, who are AW’, and Bristol Rovers, who are BR). Expected goals are those drawn from a Poisson model based on past results given the information listed above from recent games, and Elo ratings. Hence Fleetwood are expected to score 2.16 goals, Doncaster just 0.88. This then feeds into scoreline probability forecasts, nine of which are listed for the nine most common scorelines.

We sum up all these probabilities to give an overall probability of each outcome, under “Outcome Probs”. The most likely scoreline is then in the “Score Picks” columns under “Most”. This is often 1-1, since 1-1 historically is the most likely scoreline, though not exclusively. The column “Cond” is the most likely scoreline conditional on the most likely outcome occurring – that’s usually, but not always, a home win since our forecast model builds in the home advantage.

The final two columns are the average bookmaker prices for match outcomes according to Oddsportal.

This is a weekend of strong home teams in League One. Only Sunderland at Swindon, and Hull at Rochdale are expected to win, according to the bookmakers. Our model agrees on Sunderland, but dissents on Hull, giving plucky Rochdale a 49% chance of upsetting newly relegated Hull.

In League Two, below, there’s a Lancashire Derby taking place at Bolton, as they host Oldham. The bookmakers rightly fancy Bolton, who have picked up from a slow start, though seem to find goals hard to come by. Our model expects a tighter encounter, with each team expected to hit 1.4 goals. Oldham have scored many goals, but have conceded many more. What will happen this afternoon?

There’s also a battle of the newly promoted clubs, as Harrogate host Barrow. League Two has been fairly unpredictable so far, though with one exception – Oldham always lose. Will Grimsby follow up their impressive win against Cheltenham? Can Morecambe, who struggled last season but sit atop the league, carry on at Crawley?

Premier League Week 5

Teams return from the international break this weekend with the Merseyside derby kicking things off shortly.

Bookmakers generally seem to agree with us that it will be a likely Liverpool away win, despite them losing 7-2 to Aston Villa and Everton being in excellent form.

Home wins are predicted for Chelsea, Man City, Tottenham and Sheffield Utd in our mode. And are also most likely according to bookmakers (although this last one the bookmakers have only a 45% chance of a win for Sheffield (see Mean Odds, Home).

A score of 2.5 on expected goals for Tottenham against West Ham suggests a decent chance of a high-scoring match (West Ham also have been playing better than expected but their chances of a win range from less than 15% (our model, see Outcome Probs, Away) to under 20% (bookmakers average odds (Mean Odds, away).

 

Championship predictions

It looks like there should be quite a few draws as the likely result this week in the Championship and not many away wins (see the ‘Score Picks, Most’ section).

On expected goals though, a bunch of teams stand out – Bournemouth, Blackburn, Brentford and Swansea all expected to out-score their opponents from whatever home advantage there is with no crowds. Brentford v Coventry though is predicted to be a draw (see Score Picks, ‘Most’), although if there is a win (see ‘Cond’ in the Score Picks section), then it will be a win for Brentford).

Bookmakers broadly agree with our model, with Derby v Watford, Luton v Stoke City, Rotherham v Norwich, the exceptions where our model predicts wins for the home team and they predict away wins.

Lower League Forecasts, October 9-10

In this strange season, the first international break has arrived, which has caused a small amount of schedule change in League One that our “Supercomputer” hasn’t been able to make sense of – Blackpool’s game with Ipswich being brought earlier due to Charlton and Sunderland postponing their matches with Ipswich and Blackpool respectively due to the international break.

One match has already taken place, Fleetwood thrashing Hull 4-1 last night. The bookies were taken aback by this, though we had Fleetwood 57% for the win. For what it’s worth, there are similar disagreements for Gillingham’s game with Oxford (we think Gills, bookies think Oxford), and Northampton’s game with Peterborough (we think it’ll be closer than the bookies who strongly fancy Peterborough).

In League Two, Morecambe have had a fine start. But neither the model (17.5%) nor the bookies (28.6%) think they’ll continue with that at winless Oldham, who only have their single point thanks to a last-minute equaliser at Colchester. Maybe I need to place a bet on this, given the inevitable disappointment come 5pm…

Lower League Forecasts, October 3, Game Week 4

We continue at Scorecasting Economists to provide forecasts before matches have taken place. It’s something we enjoy doing – it’s interesting to try and predict something, to record it, and then to see how you did. You can compare against bookmakers, for example, which are another source of forecasts. We do that in our tables each week. In the coming weeks we will (re)develop our pages here on evaluating forecasts – are we even any good?

For now, though, here are our lower league forecasts. We provide a most likely scoreline, which very commonly is 1-1, since this is the most common scoreline. Given this, we also provide a conditional most likely scoreline – the most likely scoreline with the team most likely to win, winning. Henc fo League One below, 1-1 is most likely between Wigan and Doncaster, but because Wigan are most likely to win, a 2-1 Wigan win is the conditional most likely score:

In League Two, we have:

Covid-19 denies us what would have been a fascinatingly tight match between Grimsby and Bradford.