What Simulated ROI Means (Not a Profit Promise)
What does simulated ROI mean? Simulated (paper) ROI is the hypothetical return a model shows when its past analyses are replayed against real historical match results in a backtest: no money is involved, nothing is transacted, and the figure summarises only how well the model performed on the past. This page explains how 11Stat measures simulated ROI and hit-rate, why past simulated performance is not an indicator of future results, and why 11Stat reports calibration instead of tips. Everything here is football data analytics — not betting advice, and not a promise of any gain.
What exactly is simulated (paper) ROI?
Simulated ROI is football analytics' adaptation of "paper trading" from finance: every analysis the model produced in the past is recorded as if a fixed unit had been allocated to it, then scored against the real, settled results of those matches. The resulting percentage is purely hypothetical accounting — nobody staked anything, nobody gained or lost anything.
That makes simulated ROI a model quality metric, not an income figure. Its siblings are hit-rate (the share of analyses that landed) and calibration error. Together they answer one question only: "How close was this model to reality in the past?" They promise nothing about the future; the only thing the data permits is an honest report card on history.
How 11Stat runs a backtest: point-in-time and leakage-free
For a backtest to be honest, the model may only see data that was knowable before kick-off. 11Stat therefore uses point-in-time, walk-forward testing: for each historical day, the model is re-run with exactly the form, squad and league data available that morning, and its output is scored against the real final score published later.
The most insidious failure mode — look-ahead leakage, where the model accidentally sees information from the future — is blocked by a dedicated guard. Results are broken down per market (match outcome, goal counts, most likely score and so on), and markets that stay weak are automatically disabled. The full procedure is documented on the methodology page.
Why past simulated performance says nothing certain about the future
Three structural reasons. First, variance: football is low-scoring and luck-heavy; over a window of even a few hundred matches, a good model can look poor and a mediocre one brilliant. Second, regime drift: squads, managers, refereeing interpretations and season dynamics change, so a pattern that worked historically can quietly break.
Third, overfitting: the more finely a model is tuned to past data, the worse it may generalise forward. Because of this trio, no historical ROI curve can be projected ahead; 11Stat presents its simulated results as a diagnostic instrument, never as an indicator of future performance. Past performance is not a guarantee of future results — on this page that sentence is not legal decoration, it is the mathematics itself.
Hit-rate, ROI and calibration: what each one actually tells you
On its own, hit-rate misleads: a model that always picks the most likely outcome can boast a high strike rate while adding no information whatsoever. On its own, ROI is noisy: a handful of extreme results can swing the whole curve. The most durable metric is calibration: whether the events the model calls 60% actually happen about 60% of the time.
That is why 11Stat reports all three together, labelled with the time window and sample size. Wherever one metric shines alone, check the other two first; only when all three tell a consistent story have you genuinely learned something about the model. For the underlying concepts, see model calibration and the glossary.
Why 11Stat reports calibration instead of tips
The industry's default language is "today's picks"; 11Stat deliberately publishes something else: probabilities, confidence scores and their historical report card. Backtest windows that went negative are published too, because a chart that only ever shows good weeks is marketing, not measurement.
This transparency has practical teeth: weakening markets are switched off automatically, the 0-10 confidence score is derived from calibration data, and every analysis card shows which engines contributed. What the user receives is not a promise but an auditable measurement system. The how it works page walks through how the 12 engines reach consensus.
How to read any ROI figure responsibly
Wherever you encounter an ROI claim, interrogate it with these questions:
- How many matches in the sample? A window of a few dozen matches says almost nothing statistically.
- How was the period chosen? Showing only the best month is textbook selection bias.
- Was data leakage ruled out? If post-match information contaminated the test, the number is void.
- Simulated or real? Paper returns exclude real-world frictions such as timing and price movement.
- Are negative periods shown too? A system that never shows a bad window is not measuring anything.
Every simulated metric 11Stat publishes is designed to survive those five questions — and even then, it only ever describes the past.
Frequently Asked Questions
Does 11Stat guarantee profit or a positive ROI?
No. 11Stat does not and cannot guarantee any profit, gain or ROI. Every published ROI and hit-rate value is a simulated (paper) backtest metric measured against real past results, and it carries no forward-looking commitment of any kind. 11Stat takes no wagers and offers no betting advice.
Does a positive simulated ROI mean future gains?
No, it does not. Past simulated performance is not an indicator of future results: variance, regime drift and overfitting can break any curve that looked good historically. A positive window only says the model agreed with the data during that period — nothing more.
Is a high hit-rate the same as finding guaranteed winners?
No — nothing is guaranteed, and hit-rate cannot change that. A model can inflate its hit-rate simply by always choosing the lowest-risk outcome, adding no information. That is why 11Stat never reports hit-rate alone but alongside ROI and calibration, and never presents any outcome as certain.
Why does 11Stat publish negative backtest results as well?
Because showing only the good periods is marketing, not measurement. Negative windows reveal the model's limits, trigger the automatic disabling of weak markets, and set realistic expectations. A transparent report card always beats a curated showcase.
Why is simulated ROI not the same as a real-world return?
A paper test excludes real-world frictions: prices move minute by minute, timing slips, and conditions never stay as fixed as in a simulation. Simulated ROI therefore measures only the model's historical signal quality — and 11Stat does not handle money in any case; its product is data analysis.
How many matches does an ROI figure need to be meaningful?
There is no magic threshold, but windows of a few dozen matches are almost pure noise, and even samples of several hundred matches carry wide confidence intervals. 11Stat therefore publishes its metrics together with the sample size and period label — read those two numbers first.