Empirical calibration: how each signal type performs against actual price outcomes. Not a backtest — forward-tested on live data.
Most trading systems publish backtests. Backtests are fiction — curve-fitted to historical data and worthless for predicting future performance. This page shows forward-tested empirical results: signals detected in real-time, then measured against actual price changes at 1, 3, and 7 days later.
Every signal that fires gets a price outcome recorded. There's no cherry-picking — winners and losers both appear. This is the honest calibration data that tells us which signal types have actual predictive power.
Signal accuracy data will populate as signals fire and their price outcomes are measured.
Each signal gets a 1-day, 3-day, and 7-day price outcome recorded automatically.
A signal type with a 65% win rate and +3% average 7-day return has real predictive power. A signal type with a 48% win rate and -0.5% average return is noise — no matter how sophisticated the underlying model sounds.
By publishing this data publicly, Fin45 creates accountability. If a signal type stops working (alpha decay), the numbers show it immediately. This is the opposite of black-box trading systems that never reveal their actual hit rates.
No. Backtests use historical data and are prone to overfitting. This page shows forward-tested results: signals detected in real-time, with price outcomes measured after the fact. There's no look-ahead bias because the signal fired before the outcome was known.
A signal is classified as a winner if the stock price moved at least 2% in the predicted direction within 7 trading days. This threshold avoids counting noise as wins while being achievable for most meaningful moves.
Individual signal types are designed to fire frequently with lower precision. The confluence system combines multiple low-precision signals into high-conviction trades (requiring 3+ signal types to agree). Think of each signal type as one voter in an election, not a standalone trading system.