When should you trade corners?

The corner kick is a staple event of a football match, with regular occurrences of it seemingly regardless of the action or momentum.

Many teams still utilise corners as a big part of their attacking plan. Some, like FC Midtjylland of Denmark, have mastered them in recent seasons and reaped the rewards.

However, for analysts and traders it can be quite difficult to judge games that are likely to have a high number of corners, especially pre-match. In this piece we will investigate a few standard ways of predicting corners.

Later, we will use an Asian Handicap model to see if games where one team is expected to dominate produce more corners. For now, let’s just look at the average corners per game across 7 leagues in 2014/15:

AverageCorners

Nothing too startling here, although it is perhaps surprising that in the Bundesliga, a league renowned for attacking and with a high goal average (as shown below), there are comparatively few corners:

AverageGoals

Getting into the Asian Handicap model, it is important to note that while some leagues had teams with much more dominance (such as Spain, where Barcelona and Real Madrid occasionally had a -3.5 handicap), most games fell between -0.5 and +0.5 for the home team:

GamesPerAsian

Logic dictates that if one team is expected to dominate then they will attack more, and also that a team who attacks more will win more corners. I looked at the numbers to see if this was true, combining the far-reaching dominances, so that there was more data to look at (anything beyond +/- 1 either way are combined):

AveragesCornersPerHandicap

While the average corners for a handicap of -3.5 through to -1 and -0.75 are slightly higher than the overall of 10.33 per game, it is not statistically significant and well within 1 standard deviation (3.49).

However, it is very interesting that in games where the home team were handicapped greater than +0.75, the number of corners drops dramatically. There is a roughly linear curve going upwards away from 0 supremacy for all other levels.

So, while there may be a slightly higher chance of getting more corners, particularly when the home team is thought to be dominant, there is nothing that really stands out.

There is one more area to look at, though, what if the team that was expected to dominate do not? I looked at games where the team with a handicap greater than -0.75 did not win:

Asian075DidNotWin1

Fewer games fall into this category and this makes it more volatile, but it does show that teams that are at home with a greater than -0.75 handicap tend to create more corners when they don’t win. Though this is not significant it is still a hike from the average by 1.11 corners per game (11.44 to the average of 10.33 across all games).

Observing things at league level increases the volatility of the data due to the reduction in the number of games, but still offers some more interesting findings.

The English Championship in particular shows a further increase in the number of corners when the dominant team doesn’t win. The average is close to 2 corners per game greater in this case, although it is worth noting that the number of corners in this league is above the average for all leagues to begin with:

Asian075DidNotWin

Game state will play a large part in this and it may be that the earlier the non-dominant team goes ahead that more corners are created as the home team tries to get back into the game.

If this occurs it could well be worth backing more corners if the away team takes an early lead and then laying this off if the home team equalises and market value persists.

StrataBet now contains detailed data on how corners are won and lost, so in a future blog I will look to explore how and where they are created. I will also see if any players stand out who could be useful for trading purposes.

Dave Willoughby

Are Tottenham justifiable favourites for 4th?

Since their impressive performance in the 1-1 draw at rivals Arsenal two weeks ago, Tottenham have moved ahead of Liverpool in the market to become favourites for 4th place. Mauricio Pochettino’s team had largely flown under the radar up to that point, with most media coverage revolving around Harry Kane’s supposed drought and how they consistently field the youngest starting XI in the Premier League.

Casting a glance back over their 12 league games to date, our Fair Table actually believes Tottenham to be 3 points worse off in the Actual Table. It even goes so far as to suggest they should be one of only two undefeated teams in the division, along with Manchester City. In reality they have only lost once, of course, falling to a narrow 1-0 defeat at Manchester United back on the opening day:

Tottenham_Actual-Results-vs.-Fair-Results
A comparison of Tottenham’s Actual Results/Points and Fair Results/Points in 2015/16

An argument could be made that Arsenal deserved to win the aforementioned North London Derby, based on the momentum swing in the last quarter of the contest, but Tottenham’s performance to that point of the game would have made a defeat incredibly harsh. Otherwise, their matches have largely been tightly contested affairs, with the incisive performance at Bournemouth being the only time we believe them to have been truly worthy of a winning margin of more than 1 goal.

Given this belief, are Tottenham’s performance metrics unsustainable for a team with ambitions of a top 4 finish?

To investigate we can look at their current numbers against the ones from their 2013/14 and 2014/15 campaigns, in addition to those of the teams who finished 3rd and 4th in those seasons (Arsenal/Manchester United and Chelsea/Arsenal respectively):

Tottenham_Hotspur-15-16_vs
A comparison of Tottenham’s 15/16 stats and those from 13/14 and 14/15

The thing that instantly jumps out here is just how strong Tottenham have been defensively, with their concession of Goals verging on the “best” numbers posted by the 3rd/4th place teams from the past two seasons. Further still, their concession of Great and Good Chances is actually significantly better than those teams, almost verging into “too good to be true” territory. They score less impressively on the less important metrics (Attempts, Corners and Key Entries), but defending seems to be their key strength overall.

Not that their offensive numbers are at all bad, though, with only Great Chances falling outside of the bounds of what would usually be necessary to achieve a 3rd or 4th place finish in the division. Creativity still does appear to be an issue they need to address and one big question for Pochettino is if he is able to remedy this without sacrificing too much of their solidity, should he even need to at all.

Another question that might be asked of the Argentinian is if his trademark high-energy, high-pressing style can be maintained over the course of a whole season. A look into his time in the English Premier League so far can attempt to answer that one:

Mauricio-Pochettino
A comparison of Mauricio Pochettino’s half-season performance metrics since coming to England

The main takeaway from these tables is that during his time in England, Pochettino’s teams have scored and conceded more during the 2nd half of a season. The big outliers in the second graph were actually the number of Great Chances created and conceded, but this data does suggest that some regression to the mean will not be catastrophic for their chances of a top 4 finish.

In conclusion we believe that it is difficult to argue with the market positioning Tottenham as favourites for 4th and that a current price of 2.5 still retains some value. We will focus on their main rivals in a follow-up piece soon, with particular attention paid to Chelsea, Liverpool and Leicester.

Rich Huggan

Which Scottish Premiership teams are good and which are lucky?

Data is becoming more and more available in football and is now largely accessible to the average fan. However, as is clear to most people, not all data is useful.

Shot data is a prime example of this.

The standard way of recording shots on target or off target can often give a misleading view of what happened during a game, particularly if you haven’t seen the shots in question.

In StrataBet, we don’t just record shots on/off target. We record the quality of chances.

Rich Huggan did an excellent job of explaining that here, but to summarise:

  • A “Great Chance” is a situation that a player would be expected to score from.
  • A “Good Chance” is a situation that a player could score from but would not necessarily be expected to.
  • An “Attempt” is a situation that a player would not be expected to score from.

This is simple, but extremely effective.

Rich’s blog gave detail on the conversion rates of these chances. It showed a linear trend that provides an excellent basis to investigate which teams are over-performing or under-performing.

This season I have been working on the Scottish Premiership, so have taken the chance to look at which teams have been “good” and which have been “lucky”.

The following matrix compares the “Difference in Expected Goal Difference”, using StrataBet chance data to determine whether a team has more or less goals than expected, and “PDO”, a metric in Football Analytics that is essentially 10 x (Shooting Percentage + Save percentage) – further details available on James Grayson’s blog here.

Good-vs.-Lucky

The matrix is split into 4 quadrants – to the right of the vertical axis are teams who have outperformed their Expected Goal Difference. These teams have made more of their chances than would be expected, while teams to the left of the axis have not scored at a rate consistent with the chances they have produced.

Being above the horizontal axis means a team is performing well – scoring lots of goals and/or saving a high number of shots on target. This is likely to regress as the season continues, as it is believed to be unsustainable.

Teams in the top right quadrant have a better goal difference than the expected chances they have created/conceded, while they also have a higher than expected Shots on Target and Saves ratio. Teams in the lower left are the opposite.

Those in the upper left quadrant have a higher than expected shots/save ratio but have not scored as many/have conceded more than would be expected from the chances against. The opposite is true of the bottom right.

To take a team as an example, Dundee United are massive outliers in the bottom left corner and currently sit in 12th place. They have a Difference in Expected Goal Difference of -14.

StrataBet-Data_Over-Under-Performance

When comparing them to St. Johnstone (4th place, Difference in Expected GD of +9) we see that Dundee United have created just 4 less Great Chances than St. Johnstone and have had more Good Chances (+10)/Attempts (+12), giving them an identical Expected Goals For (18). However, St. Johnstone have actually scored 29 while Dundee United have only scored 10.

They have faced exactly the same number of Great Chances while Dundee United have conceded more Good Chances (+14) but less Attempts (-10). Dundee United would have been expected to concede 22 goals while St. Johnstone would have expected to concede 20. In actuality Dundee United have conceded 28 and St. Johnstone 22.

Of course this is just one example, but in general this kind of information is key in providing insights into which teams are underperforming and likely to improve, while also highlighting which teams are likely to decline as the season progresses.

Dave Willoughby

Premier League Relegation: has the market got it right?

After eleven rounds of the 2015/16 English Premier League there are some clear ideas emerging about who will be relegated, but has the market got it right?

We will look at the Market Implied Probability Of Relegation, the Actual Data and the Fair Data to find out, before offering one possible trade idea:

Market Implied Probability Of Relegation

RelegationProbability

Despite both clubs making managerial changes in recent weeks Sunderland (73.53%) and Aston Villa (69.44%) are considered favourites for the drop, while one of Bournemouth (45.45%), Newcastle (38.02%) and Norwich (37.04%) are considered the most likely to go with them.

West Brom (14.29%), Stoke (8.33%) and Watford (8.33%) are all thought to be at risk too and the implied probability of any other club in the division to be relegated is just 5.57%.

A brief look at the data can provide some explanation of this, as well as opening up a number of interesting back and lay opportunities:

Actual Data

ActualData

To take one example, Watford’s implied relegation probability of just 8.33% from the first graph is a surprise on first sight, but the traditionally available performance metrics actually support the position.

Quique Sanchez Flores’ side have already amassed 16 points, have the best goal difference/defence of all the “at risk” clubs and are in the best immediate form.

Still, to see a newly promoted team at the same price as Premier League regulars Stoke and considered almost half as likely to drop as West Brom is startling at this relatively early stage of the season.

Further questions over their price are raised by a deeper look into the performance metrics:

Fair Data

FairData

The “fair” data believes Watford are currently two positions higher and two points better off than they should be.

While it agrees that they have deserved to take seven points from their last five fixtures, this data also suggests that they have been operating at an unsustainable level in both defence and attack.

They rank 1st for Goals Against, but =3rd for Chances Against and are =5th for Goals For, despite being 8th in Chances For. This translates to them creating fewer Chances per game than every other team in the league.

Trade Idea

Back Watford for relegation at 12.00 or better.

Schedule

-Leicester (a), Man Utd (h), Aston Villa (a), Norwich (h), Sunderland (a), Liverpool (h), Chelsea (a), Tottenham (h), Manchester City (h), Southampton (a) are their next ten games.

-Their price is expected to come in after the next two fixtures, with potential to move back out again before Liverpool (h). This offers a quick exit point and a secondary entry point for the more cautious.

Other Considerations

-Odion Ighalo’s record of 0.72 goals per 90 (second only to Jamie Vardy at 0.82) and 0.93 goals/assists per 90 (third behind Riyad Mahrez at 1.16 and Mesut Özil at 1.06).

-Ighalo and Troy Deeney are responsible for 80% of their total goals.

-Watford have barely been impacted by injuries to date, ranking 6th behind Swansea, Leicester, Chelsea, West Brom and Norwich in this category.

Analyst Data Part III: How do I use it?

By this stage we know what Analyst Data is and why it is valuable, but you might now be wondering, “how do I use it?”

To answer this question we asked two of our company directors to explain how Analyst Data helps them in their day-to-day work at Stratagem Technologies:

Andreas Koukorinis (Head of Trading)
StrataBet Invest data helps us develop insights that are directly incorporated into our trading strategies. The collection of this data is specifically focused for trading, not just ad-hoc mathematical analysis.

That means that our analysts take into consideration those variables that are directly related to the outcome of specific situations. We believe that without granular understanding of the quality of certain variables (such as shots), cause and effect will be confused (such as the comparison between shots that generate goals and chances that generate goals).

Some of our ideas are related to the medium or long-term performance of teams and how to exploit them in the markets. We believe that by breaking down the drivers of performance (such as chance creation or attacking efficiency), we can help identify situations where teams will revert to their long-term performance level and exploit mispricing in the market.

The easiest way to explain this is by constructing a factor driven “fair score” that will neutralise the effect of chance in the long run. There will be periods where chance will allow teams to experience exceptional positive or negative results, but by incorporating this type of data into our trading framework we can quickly identify them.

We believe that there will be a market-shift in using this type of data, which is part of the reason we are investing heavily to develop the framework of data collection fused with human expertise.

Dan Edwards (Head of Modelling)
The final score of a football match tells us what happened and is often, directly or indirectly, what we are trying to predict.

Unfortunately football’s low-scoring nature means that the outcome informs very little about how the match developed. Was the result fair? Was the winning team lucky?

We can begin to answer these types of questions by introducing more data, such as shots or cards, but ideally we require something more insightful.

The Analyst Data offered by StrataBet Invest measures how successfully the two teams created/denied goal scoring opportunities and allows us to mitigate the role that luck played in our analysis.

This type of information is invaluable for the modelling of football matches as it highlights teams that are being under or over-rewarded in match results relative to the level of their play:

Actual-vs-Fair-Points2