AriaS
Well-known member
- Messages
- 399
- Likes
- 124
IS (In-Sample): The historical data used to train, optimize, and build a trading strategy. The algorithm tweaks its parameters on this data to find what worked best in the past.
OOS (Out-of-Sample): A completely fresh period of historical data that the strategy has never seen. It is used as a blind test to see if the rules optimized in the IS period actually possess predictive power.
WFA (Walk-Forward Analysis): A systematic backtesting method where a strategy is repeatedly optimized on a chunk of historical data (IS) and then tested on the immediately following chunk of data (OOS, future-in-the-past). This entire window then “walks forward” through time.
In a standard WFA, data is broken down into overlapping windows to simulate continuous live trading and re-optimization:
Run 1: Optimize parameters on IS 1 → Test performance blindly on OOS 1
Run 2: Shift window forward. Optimize on IS 2 → Test performance blindly on OOS 2
Run 3: Shift window forward. Optimize on IS 3 → Test performance blindly on OOS 3
The final backtest result is the combined performance of all OOS periods stitched together.
WFE (Walk-Forward Efficiency): A metric calculated by dividing the annualized return of the OOS period by the annualized return of the IS period. A score close to or above 1.0 (or 100%) means the strategy performed just as well on unseen data as it did during optimization, proving high robustness.
CV (Cross-Validation): Partitioning data into multiple training and testing slices (often non-contiguous) to ensure parameter stability across different market regimes.
MDR (Max Drawdown Duration): The longest continuous amount of time a backtested strategy spends waiting to recover from a peak to a new high.
MC (Monte Carlo Simulation): Randomly shuffling the sequence of trades or altering backtest prices slightly to test the probability of extreme drawdowns or strategy ruin.
OOS (Out-of-Sample): A completely fresh period of historical data that the strategy has never seen. It is used as a blind test to see if the rules optimized in the IS period actually possess predictive power.
WFA (Walk-Forward Analysis): A systematic backtesting method where a strategy is repeatedly optimized on a chunk of historical data (IS) and then tested on the immediately following chunk of data (OOS, future-in-the-past). This entire window then “walks forward” through time.
In a standard WFA, data is broken down into overlapping windows to simulate continuous live trading and re-optimization:
Run 1: Optimize parameters on IS 1 → Test performance blindly on OOS 1
Run 2: Shift window forward. Optimize on IS 2 → Test performance blindly on OOS 2
Run 3: Shift window forward. Optimize on IS 3 → Test performance blindly on OOS 3
The final backtest result is the combined performance of all OOS periods stitched together.
WFE (Walk-Forward Efficiency): A metric calculated by dividing the annualized return of the OOS period by the annualized return of the IS period. A score close to or above 1.0 (or 100%) means the strategy performed just as well on unseen data as it did during optimization, proving high robustness.
CV (Cross-Validation): Partitioning data into multiple training and testing slices (often non-contiguous) to ensure parameter stability across different market regimes.
MDR (Max Drawdown Duration): The longest continuous amount of time a backtested strategy spends waiting to recover from a peak to a new high.
MC (Monte Carlo Simulation): Randomly shuffling the sequence of trades or altering backtest prices slightly to test the probability of extreme drawdowns or strategy ruin.
