Ranko's Champions League Prediction Study

A two-season test of a results-only model built for inter-league competition

Why this is the model's primary test

Ranko was invented to compare clubs from different domestic leagues. It connects domestic competitions through UEFA results and converts that network into a common rating scale.

64
recorded Champions League matches
62.5%
Ranko accuracy: 40 correct results
+4
correct picks versus day-before BWIN
38 / 60
matched closing average in the closing-odds subset

Scientific paper (PDF) Raw data and formulas (Excel) 2026/27 forecast centre

Champions League Results – 2024/25 and 2025/26

All forecasts refer to the 90-minute home/draw/away result. Ranko and bookmaker probability vectors were normalized before scoring. The primary benchmark is timing-matched day-before BWIN odds. BetExplorer closing averages are reported separately where a closing record could be matched. This is a recorded league-phase subset, not a complete tournament census.

Primary Champions League cohort (64 tracked matches)

ForecastCorrectAccuracyBrierLog loss
Ranko day-before40 / 6462.5%0.55701.043
BWIN day-before36 / 6456.3%0.55390.943

What the paired statistics say: Ranko's accuracy difference versus BWIN is +6.25 percentage points (paired bootstrap 95% CI -7.81 to +20.31; exact McNemar p=0.503). Ranko was uniquely correct 12 times and BWIN eight times.

Interpretation: the observed Champions League top-pick accuracy is competitive in this sample. Because the confidence intervals include zero, the data do not establish superiority or statistical equivalence. The full paper reports probability scores, limitations and all secondary analyses.

Matchday-cutoff sensitivity

The primary 64-match Ranko-versus-BWIN cohort remains intact; every reasonable minimum matchday represented in the ledger is shown.

Minimum matchdayMatchesRankoBWIN
2+6462.5%56.3%
3+6163.9%59.0%
4+5563.6%56.4%
5+4266.7%61.9%

Closing-market sensitivity subset

BetExplorer closing-average odds could be matched for 60 of the 64 CL rows. This subset is later-timed and is not used to decide which rows enter the primary comparison.

ForecastCorrectAccuracyBrierLog loss
Ranko day-before38 / 6063.3%0.55361.039
BWIN day-before34 / 6056.7%0.55520.945
Closing bookmaker average38 / 6063.3%0.54670.931

Method and Reproducibility

Model. Ranko is a proprietary results-only cross-league rating and forecast system. Its declared inputs are completed top-five domestic-league and UEFA match results; it does not use bookmaker prices, lineups, injuries, transfers, weather, expected goals or event-level match data. The forecast-generation algorithm and parameters are confidential.

Reproducibility boundary. The published ledger contains the fixed Ranko probabilities, benchmark probabilities and realized outcomes required to reproduce every evaluation statistic. Forecast generation remains proprietary; evaluation inputs and calculations are public.

Study records. The BetsAnalysis ledger was operationally maintained before matches. Its CSV export does not contain cell-level edit timestamps, so the exported file alone cannot independently prove when each cell was entered. A stronger next study will use an immutable timestamped archive and a protocol frozen before results are known.

Benchmarks. BWIN prices were recorded for the day before the match. BetExplorer supplied the closing bookmaker average. All implied probability vectors were normalized to sum to one before scoring.

Metrics. Log loss punishes confident wrong predictions hardest and is the standard forecasting score. Brier score is the mean squared error of the probabilities. Accuracy is how often the highest-probability outcome actually happened.

Uncertainty. The report uses paired bootstrap confidence intervals and exact McNemar tests because every method predicts the same matches. No claim of superiority or equivalence is made from a non-significant result.

Probabilities are model estimates for information and entertainment, not betting advice.