Rules
- Signal on the close of a candle, entry at the open of the next one.
- Regular divergence between price and RSI(14): pivots 5/5 bars, 5–60 bars apart. The signal fires when the last pivot is confirmed, 5 bars after it.
- Only when the first pivot is outside 30–70: RSI below 30 for a long (oversold), above 70 for a short (overbought).
- Long on bullish divergence (price lower low, oscillator higher low); short on bearish.
- Stop-loss beyond the last pivot by 0.5 × ATR(14).
- Take-profit at 2 R.
- Exit on the close of bar 30 if neither level is hit.
- An opposite signal closes the trade at the next open and reverses.
- If stop and target are both touched in one candle, the stop counts.
- Fees 0.1% per side, slippage 0.05% on market fills.
The idea
The same divergence as D01, with one filter: the first low of a bullish divergence must be printed with RSI below 30, the first high of a bearish one with RSI above 70. The market has to be genuinely stretched before the second, weaker push counts.
The filter removes divergences that happen in the middle of the range, where RSI swings mean little. It leaves fewer signals — the question the backtest answers is whether the ones left are better.
Chart
Pick a trade in the table to centre the chart on it. Hollow circles are pivots where they formed; filled ones mark the bar the pivot became known — the signal can only fire there.
Results
| Side | Trades | Win rate | Profit factor | Avg trade | Net P&L | Max drawdown |
|---|---|---|---|---|---|---|
| All | 5 | 40% | 1.165 | +1.87% | -6.47% | −47.08% |
| Long | 1 | 100% | N/D | +26.88% | +26.88% | −0.00% |
| Short | 4 | 25% | 0.692 | -4.38% | -26.29% | −47.08% |
Monthly P&L
| Year | jan | fev | mar | abr | mai | jun | jul | ago | set | out | nov | dez | Year |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| 2026 | · | · | · | · | · | +39.30% | · | +26.88% | · | · | · | · | +76.74% |
| 2025 | · | · | · | · | · | · | -21.77% | · | · | · | · | · | -21.77% |
| 2024 | · | · | · | · | · | · | · | · | · | · | -23.71% | · | -23.71% |
| 2021 | · | · | · | -11.34% | · | · | · | · | · | · | · | · | -11.34% |
Three real trades from the backtest
Best winner · #5
Worst loser · #1
Borderline case · #2
By timeframe
Where it breaks
On 1h the strategy did worst (-93.97% over 654 trades); the best timeframe was 1d (+70.30%). Buy & hold over the same history: +1,804.93%.
Oversold can stay oversold. In a crash RSI sits below 30 for weeks and prints several divergences on the way down; the filter does not stop the stop-loss from being hit. With fewer trades the sample is smaller, so results swing more between coins and timeframes.
Common beginner mistakes
- Treating "RSI below 30" as a buy signal on its own: oversold is a state, not a trigger. The divergence and its confirmation are the trigger.
- Checking the zone on the second pivot: by definition the second RSI low is higher, so it often is already back above 30. The rule looks at the first one.
- Reading too much into a handful of trades: a stricter filter means a smaller sample — check the trade count before the win rate.
Open now
No open position on the last closed candle.
Trades · 5
| # | Side | Entry | Stop | Target | Exit | Result | Net | R | Bars | Show |
|---|---|---|---|---|---|---|---|---|---|---|
| 5 | Long | 2026-08-03 63,601.80 |
54,971.52 | 80,862.34 | 2026-08-24 80,862.34 |
Take-profit | +26.88% | +1.98R | 4 | |
| 4 | Short | 2025-11-10 104,670.59 |
130,473.18 | 53,065.41 | 2026-06-01 63,363.68 |
Timeout | +39.30% | +1.59R | 30 | |
| 3 | Short | 2025-03-03 94,222.86 |
114,466.78 | 53,735.01 | 2025-07-07 114,524.01 |
Stop-loss | -21.77% | -1.01R | 19 | |
| 2 | Short | 2024-07-15 60,767.51 |
75,004.31 | 32,293.90 | 2024-11-04 75,041.82 |
Stop-loss | -23.71% | -1.01R | 17 | |
| 1 | Short | 2021-03-29 55,749.76 |
61,920.60 | 43,408.09 | 2021-04-12 61,951.56 |
Stop-loss | -11.34% | -1.02R | 3 |
Spec v1 · engine 1.1.0 · 475 candles 2017-08-14 — 2026-09-14 · data hash 852bb5484e69
Binance spot data. The coin list is today's, so delisted coins are missing (survivorship bias is not corrected).
Educational material, not investment advice. Past results of a mechanical strategy do not predict future results.