A Profitable Model For Predicting the Over/Under Market in Football
-
firlandsfarm
- Posts: 3564
- Joined: Sat May 03, 2014 8:20 am
I uploaded the paper to GPT and Claude and this is what they said
ChatGPT
My conclusion is:
This is a credible and worthwhile paper, but I would treat it as a model-design paper, not a ready-to-use betting system. Its most valuable takeaway is that chance-creation proxies beat goals as predictive inputs for this market. Its biggest weakness is that the model partly leans on market odds as an input and relies on maximum quoted prices being achievable in practice. The paper itself also notes that the edge seems to level off in later years, which is another warning against assuming it still works unchanged today.
In plain English: interesting, credible, and useful — but not proof that you could simply copy it and print money today.
Claude
10. Overall Verdict
10.1 This is a rigorous, well-constructed academic study with a large sample (68,672 bets over 13 years, 10 leagues), careful statistical methodology, and honest acknowledgement of limitations. Its publication in a peer-reviewed forecasting journal adds credibility.
10.2 The central finding — that match statistics (particularly shots + corners) produce significantly better goal predictions than goals themselves, and that this generates a demonstrable betting edge — is both novel and statistically robust.
10.3 However, the edge is conditional on maximum available odds, meaning it requires disciplined odds-shopping or exchange use, and it appears to be declining over time as the information has become more widely used.
10.4 For a technically capable bettor with access to data feeds, odds comparison tools, and the ability to implement the GAP rating and logistic regression model, this represents a credible but diminishing systematic edge. For a casual bettor, the overhead of implementation relative to the ~0.8% margin per bet is unlikely to be worthwhile.
10.5 The paper's broader contribution is methodological: the GAP framework is extensible to other markets (Over/Under 1.5, 3.5, corner totals, etc.) and other sports, and the principle that proxy statistics outperform the final score as a predictor is likely to hold across many sporting contexts.
I then asked Claude how it would apply the principles to "other sports" and what "sporting contexts" it had in mind with "the principle that proxy statistics outperform ... as a predictor is likely to hold across many sporting contexts." Here's what it replied ...
Sport-by-Sport Application
Baseball
Ice Hockey
American Football
Basketball
Rugby Union and League
Tennis
Cricket
Golf
Horse Racing
It proposed certain alternative metrics for each sport but too many/long to list here.
ChatGPT
My conclusion is:
This is a credible and worthwhile paper, but I would treat it as a model-design paper, not a ready-to-use betting system. Its most valuable takeaway is that chance-creation proxies beat goals as predictive inputs for this market. Its biggest weakness is that the model partly leans on market odds as an input and relies on maximum quoted prices being achievable in practice. The paper itself also notes that the edge seems to level off in later years, which is another warning against assuming it still works unchanged today.
In plain English: interesting, credible, and useful — but not proof that you could simply copy it and print money today.
Claude
10. Overall Verdict
10.1 This is a rigorous, well-constructed academic study with a large sample (68,672 bets over 13 years, 10 leagues), careful statistical methodology, and honest acknowledgement of limitations. Its publication in a peer-reviewed forecasting journal adds credibility.
10.2 The central finding — that match statistics (particularly shots + corners) produce significantly better goal predictions than goals themselves, and that this generates a demonstrable betting edge — is both novel and statistically robust.
10.3 However, the edge is conditional on maximum available odds, meaning it requires disciplined odds-shopping or exchange use, and it appears to be declining over time as the information has become more widely used.
10.4 For a technically capable bettor with access to data feeds, odds comparison tools, and the ability to implement the GAP rating and logistic regression model, this represents a credible but diminishing systematic edge. For a casual bettor, the overhead of implementation relative to the ~0.8% margin per bet is unlikely to be worthwhile.
10.5 The paper's broader contribution is methodological: the GAP framework is extensible to other markets (Over/Under 1.5, 3.5, corner totals, etc.) and other sports, and the principle that proxy statistics outperform the final score as a predictor is likely to hold across many sporting contexts.
I then asked Claude how it would apply the principles to "other sports" and what "sporting contexts" it had in mind with "the principle that proxy statistics outperform ... as a predictor is likely to hold across many sporting contexts." Here's what it replied ...
Sport-by-Sport Application
Baseball
Ice Hockey
American Football
Basketball
Rugby Union and League
Tennis
Cricket
Golf
Horse Racing
It proposed certain alternative metrics for each sport but too many/long to list here.