Ichimoku MTF v3.7
Our main system, run by an automated bot. Finds the trend on D1/H4, then enters on H1.
1. Confirmed on TradingView
TradingView Strategy Tester results (OANDA H1, 12 Bar Replay windows × 2 years, $10,000 per pair, not compounded, v3.6 default risk)
| Market | PF | DD | Longest losing streak |
|---|---|---|---|
| 4-pair portfolio | 1.970 | 4.12% | 9 |
| XAUUSD | — | 5.68% | 4 |
| USDJPY | 1.721 | 6.05% | 7 |
| EURJPY | 1.765 | 3.33% | 6 |
| GBPJPY | — | 1.98% | 5 |
Total profit $58,976 on $40,000 total capital (not compounded) · Python engine entries match TradingView 99.7% (USDJPY) and 99.8% (EURJPY) after aligning the D1/H4 bar close to 17:00 New York with DST
2. Split in half (v3.7)
If the system only looked good because it happened to fit one period, the two halves would differ a lot.
| Period | Per year | PF | DD |
|---|---|---|---|
| 2006–2015 | 29.8% | 1.98 | 8.3% |
| 2016–2026 | 30.4% | 1.98 | 9.1% |
3. Markets the system never saw
Same rules, no re-tuning. Passing means the edge does not come from fitting a single data set.
| Market | PF | Result |
|---|---|---|
| XAGUSD (gold rules, unchanged) | 1.61 | Pass |
| AUDJPY (EURJPY rules) | 1.35 | Pass |
| CHFJPY (EURJPY rules) | 1.25 | Pass |
| CADJPY | 1.06 | Fail |
| NZDJPY | 1.05 | Fail |
| EURUSD / GBPUSD / AUDUSD / NZDUSD / USDCAD / USDCHF | 0.79–1.26 | Fail |
| S&P 500 | 1.04 | Fail |
4. Monte Carlo (150 paths)
The real history is reshuffled in 3-month blocks to see how bad results could get if events came in a different order.
| Set | Per year (worst 1 in 10 / median / best 1 in 10) | DD (median / worst 1 in 20) |
|---|---|---|
| v3.7 | 24.3% / 29.6% / 35.9% | 10% / 13.7% |
| v3.6 | 22.8% / 27.9% / 33.8% | 10.2% / 14.1% |
5. Added costs (v3.7)
Extra cost per trade in R (spread, slippage) is deducted from the backtest to see when the edge disappears.
| Extra cost | Per year | DD | PF |
|---|---|---|---|
| As tested | 30.1% | 9.1% | 2.00 |
| +0.03R | 26.4% | 10.6% | 1.85 |
| +0.06R | 22.9% | 12.5% | 1.71 |
| +0.10R | 18.2% | 16.3% | 1.54 |
Not proven yet
- Everything is a backtest; there is no long real-money record yet (Demo run started Oct 2026)
- The real cost per trade at the broker has to be measured from the live run
- The longest drawdown in testing lasted about a year, so expect to wait