Exploring Value Betting in Football Match Odds – Exit Strategies

[Fugazi]

My theory is the gambler is more likely to back than lay, so easier to get matched for value on lay bets

[ShaunWhite]

My theory is the gambler is more likely to back than lay, so easier to get matched for value on lay bets

I'd agree that layers are regarded as being the more sophisticated gamblers, but as such they also have the biggest wallets and competition between them is so fierce they'll undercut each other's margins to the point that efficiency is maintained. And as models give you the info to be equally effective either side of the book, there's no one side that's easier to find value on than the other.

[Kai]

If these markets were packed with gamblers trading would be a joy, not sure you should always count on them being there

Draw up a simple list of market participants and see who should beat whom, you may get a rough sense of where you stand in the market's food chain

But it's not necessarily a linear hierarchy, like this popular one isn't :

rock beats scissors
scissors beat paper
paper beats rock

[TupleVision]

I've improved on my inplay model in that it now has memory, so any user inputs, SoT, xG, Possession etc is called and if one side is showing strong improvement and it is a team I have layed for example the model will now recommend a point to exit if the bet drops to -ve

Question: I'm having to manually input these stats myself every few mins and it's a bit of a faff and obviously prone to human error. Anyone had any success using Sofascore API for anything? I don't want to open a whole can of worms as I know what I'm like, once I start I can't stop.

[wearthefoxhat]

I've improved on my inplay model in that it now has memory, so any user inputs, SoT, xG, Possession etc is called and if one side is showing strong improvement and it is a team I have layed for example the model will now recommend a point to exit if the bet drops to -ve

Question: I'm having to manually input these stats myself every few mins and it's a bit of a faff and obviously prone to human error. Anyone had any success using Sofascore API for anything? I don't want to open a whole can of worms as I know what I'm like, once I start I can't stop.

No experience with the sofa-score app, sounds like the way forward though as you've made some good progress.

If not, you need an army of university of volunteers or people that need some work experience to do the input for you, or, specialise on certain leagues/games and focus on their inputs/decisions.

[TupleVision]

If anyone is interested this is the live in-play betting model using a zero inflated poisson model with a bit of bayesian thrown in to track momentum with a dynamic approach:

How It Works:
User Inputs:

• Pre-match stats: Expected goals (xG), goal averages, possession, and shots on target.
  Live match data: Current score, elapsed time, in-game xG, possession, shots, and live odds.
  Fair Odds Calculation:

• Uses a zero-inflated Poisson model to estimate expected goals.
  Adjusts for time decay, favoring recent in-game data over pre-match stats.
  Computes true probabilities for Home, Away, and Draw outcomes.
  Momentum-Based Adjustments:

• Tracks recent trends in xG, shots on target, and possession.
  Detects momentum peaks, where a team is dominating but odds haven’t adjusted.
  Identifies reversals, signaling when a team that was dominant is losing control.
  Value Lay Bet Detection:

• Compares fair odds vs. live odds to spot mispriced markets.
  Uses Kelly Criterion to calculate optimal lay stakes based on detected edge.
  Alerts when market overreactions create profitable betting opportunities.
  Optimal Betting Window:

• Highlights key moments in the match where lay bets are statistically strongest.
  Adjusts for game state, ensuring bets are placed at the right time.

I've since had a think about my approach to value betting and agree with GPT's summation of pre-match, therefore I think I'm going to stick with in-play value finding for now...

Model 2: Pre-Match Football Betting Model
(Your second model, which is purely pre-match and calculates fair odds before kickoff.)

Advantages:

No need for in-play monitoring: You can place bets before the game and avoid stressful live trading.
Lower variance: Since bets are placed before kickoff, sudden events (like red cards) won’t impact decisions.
Works with available public data: Unlike in-play models, pre-match data (xG, form, past performance) is more widely available and consistent.
Disadvantages:

Markets are highly efficient pre-match: Bookmakers and exchanges use massive datasets, making it harder to find mispriced odds.
Edges are small: Without market overreactions, it's harder to consistently find high-value bets.
No adaptability: If a key player is injured in warm-ups or tactics change, your bet is already placed.
🛠 Profitability Potential:

This model is lower risk but also lower reward.
Edges are much smaller because pre-match markets are harder to beat than in-play markets.
It is scalable (you can place many bets across multiple games without constant monitoring).