Arbusers

What are Expected Goals (xG)?
Expected Goals or (xG) is a statistical measure used to evaluate the quality of chances created and conceded by a team or player. It estimates the probability of a specific shot being scored based on the location and type of shot and assigns a value to each shot between 0 and 1.

How is xG calculated?
XG is calculated using data from a large number of shots and is based on the historical probability of a shot being scored from a specific location and situation. The higher the xG value of a shot, the more likely it is to be scored.

The variables that affect the likelihood of a goal being scored are:
Distance to the goal
Angle to the goal
One-on-one situation
Big chance
Body part (header or foot)
Type of assist (cross, pull-back etc)
A pattern of play (open play, direct free kick, throw-in etc.)
The amount of pressure they are under from opposition players
The goalkeeper’s position relative to the direct path of the shot to the centre of the goal

For example, a chance with an xG rating of 0.35 means that a player would be expected to score from it 35% of the time, or one-in-three chances.
Similarly, a chance with an xG rating of 0.5 should be scored 50% of the time, and so on.
Some situations are unique and require independent modelling.

Penalties are assigned a fixed xG value based on their overall conversion rate (0.79 xG) , while direct kicks and headed chances have their own separate models based on specific situations. Set pieces and open-play situations are valued differently for headed chances.


(source:Pinnacle)

How can we use xG?
Analyzing team and player performance: By analyzing the xG values of shots taken and faced by a team or player, it is possible to evaluate their performance and identify trends and patterns in their play style. xG can be used to identify teams that are consistently creating high-quality chances. By analyzing a team's xG per game, it is possible to spot teams that are creating a lot of chances that have a high probability of being converted into goals. Conversely, if a team has scored fewer goals than their xG suggests they should have, this could indicate that they are underperforming and are likely to improve in future matches.

Evaluating odds: By comparing the xG values of shots taken by a team with the odds offered by the bookmaker, it is possible to identify value bets. For example, if a team has a high xG but is being offered long odds to win a match, it may be a fair value bet.

Analyzing match statistics: In addition to analyzing xG values, it is also important to consider other match statistics when betting, such as a shot on target and possession. By analyzing these statistics in conjunction with xG, it is possible to get a more comprehensive understanding of a team or player’s performance and make more informed betting decisions. By analyzing a team's xG and xGA (expected goals against), it is possible to recognize which teams are strong at creating chances and which teams are strong at preventing chances. This information can be useful when analyzing a team's future performance and identifying potential betting opportunities.

Other x* measures
Expected goals for ( xGf)
Expected goals against (xGa)
Expected goals assisted (xA)
Expected points (xPts)

Strange marketing material guys.

My take on this: it’s a good metric for building a soccer prediction model. If it’s the only one you use, you will probably be very competitive with state of the art models from ten or fifteen years ago…

There have been some interesting works on “non shot xG” in the recent years - valuing actions that did not lead to a shot as well (for example, getting close to the goal, but intercepted just before the shot). A good and simple variant to look into is xT - “expected threat”.

=======

Now for questions - those of you who bet sports other than soccer - do you use metrics similar to this? For example “expected score” in basketball; etc? How is it working for you?

ex-hft wrote: ↑

Tue Jan 10, 2023 10:28 pm

Now for questions - those of you who bet sports other than soccer - do you use metrics similar to this? For example “expected score” in basketball; etc? How is it working for you?

@ Ex-hft,

Definitely YES, but I can't accept the published metrics at all, all my projects are built from the scratch. Definitely, markets in sports such as American Football, Soccer, Basketball, Cricket, Aussie Rules and Volleyball have something similar, but if you rely on published metrics, you're definitely off the board.
Later I may post some impressions of my business experience in 2022 in a suitable thread here on the forum and if I have enough time I may post some thoughts.

Don't be confused, when I mention high-scoring sports like Tennis, Basketball and so on, these xG-like metrics are still valid and very helpful from a different point of view, aiming for shaping better self-computation of the odds during in-play mode! Naturally, low-scoring events like Soccer games are much more favored here.

The Basketball game itself can have over 200 or 300 metrics, and only building the quality time-proven filters plus constantly improving the rest of the filters plus inventing additional filters can give you some hope of beating the market. This game is a Masterpiece like the music of Beethoven or Vivaldi - each time you can perceive it in a different way than the Author created it.

Beware of Lopetegui's modus operandi, if you rely on the so-called raw xG Stats, remember Liverpool vs Wolverhampton in the FA Cup (2:2). There is an opportunity to build more adequate xG-like Stats.

arb12 wrote: ↑

Wed Jan 11, 2023 5:19 pm

Beware of Lopetegui's modus operandi, if you rely on the so-called raw xG Stats, remember Liverpool vs Wolverhampton in the FA Cup (2:2). There is an opportunity to build more adequate xG-like Stats.

That’s very interesting (as well as the other posts above). As I’ve mentioned in other posts, my approach is now much more data-driven, and less reliant on feature engineering. But n earlier stages, I was using xG, and saw some improvement from “non shot xG” - dangerous positions which did not lead to a shot.

Is that what you mean, or something else entirely?

ex-hft wrote: ↑

Wed Jan 11, 2023 6:56 pm

arb12 wrote: ↑

Wed Jan 11, 2023 5:19 pm

Beware of Lopetegui's modus operandi, if you rely on the so-called raw xG Stats, remember Liverpool vs Wolverhampton in the FA Cup (2:2). There is an opportunity to build more adequate xG-like Stats.

That’s very interesting (as well as the other posts above). As I’ve mentioned in other posts, my approach is now much more data-driven, and less reliant on feature engineering. But n earlier stages, I was using xG, and saw some improvement from “non shot xG” - dangerous positions which did not lead to a shot.

Is that what you mean, or something else entirely?

Hi again Ex-hft,

Especially for Julen Lopetegui, I have been closely observing his working style over time so far at Real Madrid, Sevilla, España, Wolves, etc.

One of my working basics in a nutshell - everything is caused by something else. Coaching decisions are much more interesting to look at as a cause than simply building xG-like Stats that are a consequence of combinations of very complex factors (coaches, players, tactical approaches, etc.).

Introducing the Coaching matrix into the model will generate some better probabilistic results than simply following stats like xG or similar, which are themselves caused by other complex factors. I'm not speaking against xG here at all, just looking to create some better Stats from scratch due to very long observations. That applies not only to Soccer! That applies everywhere, counting the specifics in every kind of sport.

That's why all the other variables are important, including the so-called "non-shot xG" you mentioned and similar Stats. I admire your decision to use them! A very important point here is to break them down over time and use a wise rationale behind them before using them in your market strategy in general and further market entries/exits.

For example, I advise you to closely observe and further backtest various battles between teams coached respectively by a "direct soccer"-style lover vs a "Tiki-taka"-style master coach. Or a battle between a team, performing a "Catenaccio old-school"-style, led by an extremely experienced coach vs. a team, performing a "New hidden semi-Catenaccio"-style led by a newcomer and very ambitious but inexperienced coach. Or thousands and thousands of other existing combinations! You'll discover a completely new world then.
Furthermore, these initial tactical approaches and their further different implementations by the players are very different in Argentine, Greece, Belgium, Taiwan, and so on. So, that's another pitfall for you. That one persists even in different tiers within a country. What about Aston Villa vs Stevenage Borough (1:2) for the FA Cup days ago?


Assure you that it takes years to fully recognize and classify these types of data. This is why quality analysis should be used alongside data-driven technical analysis. But it's much more important to build a strong rationale behind decisions about when and why to include "non-shot xG" or other Stats you've created.
Once you've mastered that, you need to keep the things you've invented evolving over time.

Much later it is very easy and fun.

Definitely much to learn here