Walk-Forward Efficiency Ratio
A single number comparing how a strategy performed on out-of-sample walk-forward periods versus its in-sample optimization periods, used as a quick check for how much of a backtest's apparent edge survives outside the data it was tuned on.
Walk-forward testing repeatedly optimizes a strategy's parameters on one window of data, then tests those fixed parameters on the following out-of-sample window, rolling forward through history. The walk-forward efficiency ratio boils all those out-of-sample runs down to one summary number: how much of the in-sample performance actually showed up out-of-sample.
The calculation
It's typically computed as the average out-of-sample return (or Sharpe ratio) across all walk-forward windows, divided by the average in-sample return across the same windows, often expressed as a percentage. A ratio near 100% means the strategy performs about as well on unseen data as it did during optimization — a good sign. A ratio well below 100%, or negative, means the optimization is finding parameters that fit the in-sample noise and don't generalize.
Worked example
A strategy is walk-forward tested across ten rolling windows. The average in-sample annualized return across all ten optimization windows is 22%, while the average out-of-sample return across the following ten test windows is 9%. The walk-forward efficiency ratio is — a warning sign that more than half of the apparent in-sample edge is disappearing once the parameters face data they weren't tuned on.
The walk-forward efficiency ratio divides average out-of-sample performance by average in-sample performance across rolling walk-forward windows; a low ratio flags a strategy whose optimized parameters are fitting in-sample noise rather than a pattern that persists out of sample.
Related concepts
Further reading
- Pardo, The Evaluation and Optimization of Trading Strategies, ch. 9