Technical Signal Backtest: What These Indicators Actually Did

6,867 US stocks  ·  3,876,380 daily bars  ·  sample period 2024-01-01 to present  ·  computed 2026-11-04

Every site that describes technical indicators will tell you what a golden cross is. Almost none will tell you how often it actually worked. This page is our attempt to answer that second question honestly, using our own price history rather than a textbook.

How the test was run

For each stock in our universe we computed the signals below from end-of-day closes, took every occurrence, then measured the forward return at 5, 10 and 20 trading days. Indicator math was cross-checked against TA-Lib (maximum deviation under 0.0002). No parameter optimisation was performed — the periods are the conventional ones, precisely because tuning them would produce flattering results that would not survive out of sample.

Results — 20 trading day horizon

All horizons

What stands out

The golden cross underperforms. MA20 crossing above MA60 is the single most recommended entry signal in retail technical analysis. In this sample it produced a 48.8% win rate and a slightly negative median return over 20 days. Buying the golden cross was, historically, a coin flip that landed on the wrong side more often than not.

The death cross is not a sell signal. The MACD death cross — the one tutorials say means "get out" — was followed by a positive median return and a 53.6% win rate. Selling on it would have systematically cut positions just before further gains.

Oversold recovery is the one signal with a real edge, and it is small. RSI climbing back above 30 produced a 54.1% win rate and a +0.79% median 20-day return. That is not nothing, but with typical round-trip costs of 0.1% or more, most of it disappears.

Combining signals did not rescue them. The combined bullish setup — MACD cross, RSI filter and price above MA20 — had a 49.0% win rate. Adding conditions reduced sample size without improving the outcome.

Limitations you should weigh

• Sample period is roughly two years. It does not span a full bull/bear cycle.
• Returns are gross. Commission, spread and slippage are not deducted; at 0.1% round trip most median returns here fall to zero or below.
• No stop-loss or take-profit rules were applied — exits are purely time-based.
• Small, illiquid names are over-represented and could not realistically be filled at the closing print.
• Signal parameter choices are conventional but arbitrary; other periods would give different numbers.
• No survivorship correction was applied to delisted names.

Reproduce it

The backtest is deterministic and the indicator math is validated against TA-Lib. The methodology — exact signal definitions, periods, and statistical treatment — is documented at methodology.php. To see how these signals read on a specific company today, open any ticker from the home page.