Football Musings

[TupleVision]

Referring to last nights' game Nice v Auxerre at HT, the sheet accurately reflected the HT odds market based on Pre-off Poisson data.

I bolted on my interpretation of the Ben Michaels In-play xG spreadsheet, using my own xG calculations (not flashscore). He talks about how the xG can revert to the mean, and a trade can be made based on this assumption. Trading is opinion/maths based that needs to be backed up by data where possible.

Interesting but I'm not sure how it can be mean reverting as xG only ever increases during a game? If a shot is missed it doesn't subtract from the previous xG?

Agree with your earlier assertion of xG though I incorporate it myself as a pre-match metric that I then compare to in-play, this plays a role in my model but certainly isn't weighted as the highest metric it's just a factor alongside many other inputs.

Essentially, xG reversion means that teams performing significantly below their xG at halftime often improve, while those overperforming tend to slow down, provided, normal game conditions persist.

@Pre-off - Home xG 2.50 Away xG 1.50.
@HT Home xG 0.50 Away xG 1.00
The current score is 0-1.

Backing the home team to score in the second half could have value if odds reflect their poor first half rather than their overall ability.

In-play markets might overreact to the halftime xG, so if the home team’s price is inflated, it could present an opportunity.

If the match dynamics suggest the home team is improving (momentum, attacking pressure), a late goal for the home side is more likely than the odds might suggest.

Oh I see, yeah I must admit I've been a bit back and forth for a few days with how to use xG in my model, taking the total xG over an entire game sometimes seems to over inflate a team's chances if they've had a couple of golden opportunities in the first half and are dog shite in the second half. I've just reduced the weighting it has for now until I can come up with something else but this has certainly given me a couple of ideas, thanks.

[wearthefoxhat]

Here's the latest progress on the In-play excel sheet.

With this one, I attempt to reverse engineer the Pre-off Pinnacle odds to produce an xG that then takes into account the league's attack/defence data...etc

Reckon, the early exchange market makers and maybe betfair themselves (seeding?), refer to Pinnacle initially, then refine their own models accordingly. For this exercise, It then serves as an anchor for in-play decisions at half-time or anytime during a game.

Game selection is important, I chose Port Vale v Morecambe yesterday. Port Vale, need to win their home games to improve their chance of play-offs with a strong chance of automatic promotion if other teams falter. Morecambe are looking shaky for relegation and will likely try and park the bus to gain a point. Liquidity was okay for match odds and correct scores.

Pre-off the top 4 correct scores grouping was 1-0, 2-0. 1-1. 0-0

PortV Pre.png

----

At half-time, the picture changes a little. (Editing the remaining minutes to 45). It was 0-0, and by creating my own xG (using sofa-score metrics), I can compare the xG with the pre-off xG and see what the in-play review section indicates.

2 goals expected, with it leaning towards an overdue goal for Port Vale. A draw is predictable too.

The top 4 correct score grouping is now: 0-0, 1-0, 2-0, 0-1.

Once the betfair half-time odds are input, the draw match odds 2.94 compare favourably for value against the models' 2.46. With the home/away odds there's not much in it either way.

Port V HT.png

------

The FT score was 1-0 to Port Vale with the winner scored in the 89th minute.

The attack/momentum graph indicates Port Vale's dominance, but were nearly caught out during 1 big chance and pressure from Morecambe.

My own xG for the whole game was Port Vale 1.34 v Morecambe 0.51

Port V FT Atteff.png

-----

One other element of the sheet is that I can take a view at anytime during the game, not just half-time, by adjusting the remaining minutes, sofa-score metrics xG, and then comparing for value, what the exchanges offer in various markets.

[wearthefoxhat]

The reverse engineering is starting to look alright. The over/unders match up well, so it's fair to say, the xG and correct scores will be "accurate", and should be reflected in-play/HT too. Spotting the value prices, if the market is skewed, should be fairly straight forward.

Red cards or impactful yellow cards, have to be considered too. (Something I'll have to add in going forward)

Using tonights' Port Vale v Barrow for reference. Has a similar look to the game previewed last time.

PV v Barr Pre.png

[wearthefoxhat]

@ Half-Time, odds again match up well. Indicates an overdue goal for Port Vale and a Draw, but odds, generally, not offering any value.

Be interesting to see how it develops late into the game.

PV v Barr HT.png

[wearthefoxhat]

In the last 10 mins, again accurate, but no real value.

PV v Barr 80 mins.png

----


The 2nd half was better for Port Vale. My own xG 0.88 v 1.03, so, on balance, the result was probably about right, with no surprise if Port Vale had got the draw.

PV Res.png

-----


Reckon I've got the calculations working well now. They stand up against the market both pre-off and in-play. Game selection will be key and the test will be with free scoring leagues like the English Premiershio and German Bundesliga.

[wearthefoxhat]

Here's the HT: Notts Forest v Man Utd game tonight.

Indication is, a goal's overdue, maybe, for Man Utd, if, they repeat their first half performance.

NF v MU HT.png