quantheus/eurusd_45hypotheses_research

This repository contains a systematic backtesting study of 45 trading strategies applied to the Euro/US Dollar currency pair on 5-minute price bars, spanning January 2018 through April 2026. The author ran this research as part of proprietary trading work, with the specific goal of finding strategies that meet the constraints of prop trading firms: a maximum daily loss of 5%, a maximum overall drawdown of 10%, and consistency across multiple years rather than one big winning period. Roughly 90% of the strategies failed this standard, and the notebook documents all of them including why each one did not work. The methodology uses a walk-forward structure, which is considered a more honest test than simple backtesting. Parameters are chosen using data from 2018 through 2021. The strategy is then applied once to the 2022 through April 2026 period and evaluated without any further adjustments. A strategy only passes if it achieves a profit factor above 1.20, survives a Monte Carlo stress test that randomly reorders its trades 10,000 times and checks whether drawdown stays within limits, and shows positive results in at least 3 of the 4 out-of-sample calendar years. None of the 45 strategies met all three criteria. The 45 hypotheses cover a wide range of approaches. Session-based strategies try to trade around when major markets like London and New York open or close. ICT and Smart Money Concepts (SMC) are a popular trading framework that describes price action in terms of liquidity grabs, fair value gaps, and order blocks. Statistical strategies look for mean reversion based on indicators like RSI and VWAP. Calendar anomalies test whether certain days or times of week have predictable behavior. The study found that ICT and SMC concepts, while popular among retail traders, did not produce systematic edges when expressed as fixed algorithmic rules. Calendar effects mostly did not generate enough trades to draw reliable conclusions. The key methodological observation is that a high-performing in-sample result with a small number of trades predicts failure, not success. Every strategy with a profit factor above 1.5 in-sample but fewer than 30 trades collapsed when applied to out-of-sample data. The repository contains a single Jupyter Notebook with all 45 hypotheses, their signal logic, parameter grids, result tables, and equity curves. Running it requires historical EURUSD 5-minute data, which can be exported from MetaTrader 5 or downloaded free from Dukascopy.