AI Swing Backtest
How the ARGPT AI Swing rule would have traded 2010 to 2026 prices, before it ever traded live. Every result is in R after trading costs: 1R is the loss of one full stop.
This is a backtest on past prices, not a live record and not a forecast. Live results are on the Stats page under the SWING model. See live results
Cumulative result after costs
The line is the running total in R. The red band below zero is the drawdown: how far it sat below its last high.
Result per year
Each bar adds up the trades that opened in that year.
What we tested and rejected
Each change was written down, with the bar it had to clear, before it was run on gold: pick the best version on 2010 to 2018, then it must beat the rule as it trades on 2019 to 2026 without a deeper drawdown.
| Change | Total | Max drawdown | 2010-2018 | 2019-2026 | Result |
|---|
Why waiting for a better price fails: almost every entry dips below its price within the hour, but the trades that dip the most are the ones that lose. The winners are the breakouts that never look back, and waiting misses them.
How the test was done
- The rule. A weekly breakout on the 1-hour chart, entered on the 5-minute close, with a stop of two times the hourly ATR, an exit when price closes back through the half-week range, and a stop that locks in profit as the trade runs. The same rule, unchanged, on every market.
- The data. Dukascopy 1-minute bid and ask prices: gold from January 2010, Nasdaq from September 2011, Bitcoin from May 2017, all to October 2026.
- The costs. Spread, slippage, commission and overnight swap are charged on every trade: 0.60 per trade on gold, 2 points on Nasdaq and 0.05% of the price on Bitcoin.
- R. 1R is the loss of one full stop. A +3R trade made three times what a stopped trade loses.
- Two halves. Each market is also split in two (gold and Nasdaq at 2019, Bitcoin at 2022). A real edge should show up in both.
- Limits. The rule was designed with gold's recent years already in view, so read gold's 2019-2026 half with some caution. Bitcoin and Nasdaq were never used to design anything. The S&P 500 was tested too and lost -72.9R, so it is not traded.