Evaluation methods
Forward Performance vs Backtest Performance
Backtesting asks how configured rules behave in a historical simulation. Forward observation asks how an implementation behaves after configuration. They answer related, but different, questions.
Backtest: examine the model and its assumptions
- Historical data quality
- Missing ticks, inconsistent prices or incomplete sessions can change signal timing and modeled fills. A backtest needs a disclosed data source, resolution and evaluation period.
- Spread, slippage and execution modeling
- Modeled transaction costs, order timing and fill rules affect algorithmic results. A simulation that assumes ideal fills does not establish achievable execution.
- Parameter sensitivity and overfitting
- Rules selected to fit one historical window may not generalize. Sensitivity checks and separate evaluation periods help expose dependence on a narrow parameter choice.
- Regime dependence
- A trend, range or volatility regime can favor particular rules. Historical profitability in one regime does not establish robustness in another.
Forward / Demo Reference: observe the configured implementation
- Observation after configuration
- The Forward Performance Record documents system behavior during an observed period after the strategy has been configured, rather than reconstructing that period as a simulation.
- Broker and demo execution conditions
- The Public Demo Reference Account is not a real-money account. Demo fills, spreads, order handling and liquidity representation can differ from real-money execution.
- Implementation behavior
- Forward observation can expose signal processing, order handling and position-management behavior. It remains specific to the stated system, instrument and platform configuration.
- A record, not a forecast
- A forward period does not eliminate model risk, operational failures or changing market conditions. Observed outcomes do not guarantee that later periods will resemble them.
Use both without treating them as equivalent
Backtesting can investigate rules across historical conditions; forward observation can examine the configured implementation over a subsequent period. Neither alone guarantees future results. The current public dataset contains a static Quant forward/demo record, not a published matched backtest series. A forward record can be stored as historical documentation without becoming a backtest.