A backtest is not a verdict on a strategy. It is a test of one precise specification on one precise dataset. Change the timeframe, lookback, sizing, or exit rules and you are testing something else.
The idea we described
The test began with a plain-English request: trade SPY long when the 20-day simple moving average crosses above the 50-day average. Exit when it crosses back below, or when a 1% stop loss or 2% take profit is reached. Allow one position at a time and use no leverage.
Test one: the intended daily timeframe
Once the strategy's base timeframe was set to one day, the available seven-day lookback produced zero trades. That is not surprising: a 50-day moving average needs far more than seven daily observations before it can produce a meaningful crossover signal.
The right interpretation is not “the strategy does nothing.” It is “this sample cannot test the strategy.” A result without enough warm-up data is inconclusive.
Test two: the default one-minute timeframe
Before correcting the base timeframe, the same numeric parameters ran on one-minute bars over seven days. That changed the hypothesis from a slower daily trend signal into a rapid intraday system. It generated 51 trades and lost money in the sample.
This does not disprove the daily idea either. It shows how silently substituting one-minute bars can create a different, noisier system. The same numbers are not the same rules when the unit changes.
A useful backtest starts by proving that the machine tested what you thought you asked for.
What we would do next
- Match the lookback to the longest indicator.A 50-day average needs enough daily history for warm-up plus an evaluation period.
- Freeze the specification before comparing results.Symbol, timeframe, sizing, exits, and data window should be recorded with every run.
- Keep failures in the research record.Zero-trade and losing tests prevent false confidence and expose configuration mistakes early.
- Paper trade before considering live capital.Historical behavior does not guarantee future execution or performance.
The real advantage of a connected workflow
TradeLab keeps the strategy description, adjustable parameters, backtest, trade history, and deployment controls attached to the same strategy. That makes it easier to see when a test has drifted away from the original idea—and easier to correct it before money is involved.
Tests were run in TradeLab on July 25, 2026 using SPY, $50,000 starting capital, and 10-share position sizing. The daily run used a seven-day lookback and returned no trades. The one-minute run used a seven-day lookback and returned the results shown above. Results are historical, educational, and specific to these settings. They exclude real-world factors that may affect execution. This is not financial advice, a recommendation, or a promise of future performance.