close
Last week
Aug 9, 2017

Soccer predictions: Past performance vs. Past odds

Does one season provide a big enough sample in a predictive model?

What's wrong with using past performance to predict outcomes?

How to use past odds to predict future performance

Why bettors should stick with expected points

Soccer predictions: Past performance vs. Past odds

Before the start of the season in European soccer leagues, bettors will begin assessing the relative merits of teams by referring back to the final tables from the 2016/17 seasons. How useful is last year’s league position when predicting match outcomes in a new season? Read on to find out.

The aim of any serious soccer bettor is to find data or use a model (usually a combination of both) that will provide accurate predictions of future performance.

Various methods can be used - be it analysing the importance of the first six games of the season or calculating expected goals - but the trick is finding reliable predictions that are accurate enough to return a profit in terms of betting. Is using last season's odds one such method?

Does one season provide a big enough sample?

A Premier League season spans ten months, 380 games in total, 36,000 minutes of clock time and roughly 10,000 attempts on goal.

It’s tempting to assume that the effects of randomness have largely disappeared from the actual outcomes and the number of points a side won in 2016/17 is an accurate assessment of their talent levels and therefore a credible indicator of their chances in the upcoming campaign.

However, there are numerous examples of a single, unexpectedly good or bad year being a poor predictor of a subsequent season - Leicester being the most extreme example, gaining 41 points in 2014/15, 81 in their title winning season and falling back to 44 in 2016/17.

Chelsea, Everton, Manchester United, Liverpool and West Ham have also performed similar gyrations in recent seasons and it would therefore appear that randomness does still influence a side’s final points total, even over a whole season.

A flurry of single goal victories, luck with injuries and a generous penalty differential can transform a season into one of unexpected success, but may not be sustainable in the longer run.

The probabilistic nature of goal attempts, whereby cast iron chances are sometimes squandered and long shots frequently find the back of the net are part of the attraction of soccer, but in the much longer term, actual outcomes tend to lean towards those expected from the quality and quantity of the chances created by a side.

A flurry of single goal victories, luck with injuries and a generous penalty differential, a less than normal error prone defence or a more disciplined distribution of red cards can transform a season into one of unexpected success, but may not be sustainable in the longer run. So is there anywhere else to look for a more reliable estimation for the new season?

Using past odds to predict future performance

Many astute gamblers have developed models that explore the underlying process of chance creation and defensive frailties that are thought to be better predictors of team quality and future performance.

But these often require extensive data collection that is rarely generally available and also needs to be updated after every match.

The Premier League sides who were perhaps flattered by their actual points total compared to Pinnacle’s estimation in 2016/17 included Chelsea, Burnley, Tottneham, WBA and Everton.

However, the influence of these model based outputs can be found, along with other novel, but predictive approaches, to team assessment in some of the pregame odds for the matches.

Pinnacle has described their methods of setting match odds here and historical archives of closing lines, available at such websites as Football-Data.co.uk is likely to contain some of the fruits of these advanced probabilistic models.

To make use of the insight that Pinnacle has incorporated into their odds and compare it to the actual records of Premier League teams, we need to convert the implied probabilities for each match outcome into league points.

Such a method is described in a previous article, but to briefly recap, Pinnacle’s small margin is removed from the implied probabilities for each possible outcome.

The adjusted probability of a win is multiplied by three and that for a draw by one. These are then added together to get the expected points haul for a side for a single game based on Pinnacle’s informed opinion.

As an example, the implied true probabilities for Manchester City vs West Ham in September 2015 was 0.814 for the home win, 0.126 the draw and 0.061 for a West Ham victory.

Manchester City’s expected points:

(0.814*3) + (0.126*1) = 2.568

West Ham’s expected points:

(0.061*3) + (0.126*1) = 0.309

West Ham won 2-1. 

Amid the excitement generated by Leicester’s title and poor seasons from many of the usual big teams, West Ham won 20 more points than they had averaged over the previous four seasons - Pinnacle’s closing odds were less impressed with the Hammers, even as they embarked on numerous unbeaten streaks.

There are numerous examples of a single, unexpectedly good or bad year being a poor predictor of a subsequent season - Leicester being the most extreme example

The sum of their expected points from all 38 games in 2015/16 based on Pinnacle’s closing line odds came to just 47 points. An improvement on the 42 actual Premier League points they had averaged between 2009 and 2015, but well below the actual leap forward they had achieved.

West Ham’s actual points total in 2015/16 of 62 promised a successful first season in their new home, but Pinnacle’s cumulative expected points total of 47 proved a more accurate predictor (West Ham gained just 45 points in 2016/17). 

Why bettors should stick with expected points

It’s easy to cherry pick an example to appear to prove a point, but we can look at the root mean square errors using a side’s actual points total or their expected points total from one season to predict their actual points total in the following season. 

Both understandably correlate to subsequent performance, but the RMSE is lower when comparing next season’s performance to a side’s expected points from the previous year over the past seven Premier League seasons.

In general, when judging how a team might fare in the upcoming year you would be better off looking at how Pinnacle expected a side to do - based on their closing lines from the previous year - than looking at that team’s actual league points and position.

The Premier League sides who were perhaps flattered by their actual points total compared to Pinnacle’s estimation in 2016/17 included Chelsea, Burnley, Tottenham, WBA and Everton, while Southampton and Leicester would not surprise many if they improved on their respective totals. 

BR Home
See the latest Premier League odds
Having graduated with a degree in Chemistry, Mark embarked on a career with a major UK brewery. However, his love for sports and numbers was always at the back of his mind. He has been writing about the statistical side of sports, mainly soccer and NFL, for over 20 years and has a particular interest in the randomness and uncertainty inherent in the numbers.
Betting Resources - Empowering your betting

Pinnacle’s Betting Resources is one of the most comprehensive collections of expert betting advice anywhere online. Catering to all experience levels our aim is simply to empower bettors to become more knowledgeable.