44. Research, testing, your own strategy

Research, glossary, history

The Research page is the platform’s scientific footnote: a curated list of nine studies from Dixon-Coles to the original of the Kelly criterion, with summaries, including our own publication. The description of the latter even states it: the combination found best in the paper was not confirmed as a lasting edge by our later, 50,000+ match walk-forward research — expect this kind of self-correction from anyone who cites data.

The Glossary is the course’s dictionary with fourteen basic concepts, and on the History page you can look back at which model predicted what on which match and what came of it, in raw, filterable form.

Market efficiency by league

The efficiency page is the map of chapter 14 broken down by league: average margin, the market’s Brier score, favourite statistics, and the favourite-longshot curve, on which it is visible how well the probability priced by the market covers the actual frequency. The per-band flat ROI list is typically negative throughout — this is the price of the margin itself — and if the high-odds bands are even worse, that is explained by the favourite-longshot bias (chapter 35).

Strategy tester: check it with your own hands

The strategy tester is the public version of our research backtest engine: you choose the league, season, form window and number of simulations, and the engine, in walk-forward mode, predicting strictly only from data known before the match, measures through the models. In the result table, next to every model’s Brier score there is also the reference of random guessing: if a model does not beat this meaningfully, its confident percentages are worth little.

This is the theory of chapter 34 made tangible: there is no look-ahead data, no hindsight-selected sample, and "accuracy" by itself is misleading — the always-home-tipping strategy also hits around 45%.

My strategy: from a filter to a system

On the My strategy page you can save your filter settings by name (model, EV and odds band, minimum model probability, league, bookmaker), and the screener lists the upcoming matches that meet the conditions. You can also activate a strategy: the machine automatically tips for you in the tipping game according to your rules, immediately or right before kick-off.

This is the main tool of the capstone project of chapter 47: you assemble a rule system, run it in the tipping game without money, and measure with the journal and your calibration panel what it actually knows. If it does not work there either, it will not with money.

A positive EV is the gap between your estimate and the market price, not a promise of return: it is only as good as your estimate, and our own measurement shows the market's closing price is more accurate than our models. Every tool here was built for research, not as betting advice. The worth of your strategy comes from the measured numbers, not from your conviction.