Strategy study method

Design before results

Daily OHLC prices for QQQ, SPY, or TLT. Use the first 70% of observations for training and the remaining 30% for evaluation. Select the highest training Sharpe from 10/30, 20/60, and 50/150 moving-average pairs, with a common 150-observation warm-up.

Signals use prior-session information and execute at the next open. Returns are open-to-open. Entry and exit costs are controlled in basis points; default 10 bps each way. Inspect the engine for the exact implementation.

Comparison and limitations

The dashboard reports Sharpe, CAGR, drawdown, turnover, time in market, and a selected benchmark on the holdout calendar. Sharpe uses a zero risk-free rate. Prices exclude dividends; the strategy assumes no interest on cash, leverage, shorting, or market impact. Benchmark-specific assumptions appear with the chart.

The holdout is a demonstration, not a repeatedly reusable independent test. Changing costs reruns selection; changing the design after seeing outcomes creates a need for new untouched data. A favorable result is not evidence of live trading performance.

Reproduce a run

1. Open the workbench, choose the instrument, cost, and benchmark, and run the pipeline.
2. Inspect source dates, the training candidates, holdout results, and warnings.
3. Select “Export research record” to retain input bars, selected parameters, results, and benchmark inputs.
4. Download the engine and use the exported input bars and cost to call runResearch(record.inputBars, record.costBps). Compare its selected parameters and results with the record.

Open the workbench · Download engine · Download benchmark code

What comes next

Test independent periods, incorporate distributions and cash yield, examine parameter stability, and document paper-trading behavior. Those steps remain future work; the animated pipeline is a rules-based demonstration with no connected broker.