3 months paper trading two crypto swing bots, one is still…
0 reactions
3 months paper trading two crypto swing bots, one is still losing, sharing everything
Been paper trading two bots on Bitget perpetual futures since April. One runs on a 2H timeframe, the other on 4H. Same basic idea behind both (gated entry with a scoring system, multi phase trailing stop that locks in profit as it goes), just different speeds.
3 months later the 4H one is barely green, +9.47 on a 100 starting balance, 68% winrate over 91 trades. The 2H one is still red, -7.62, 43% winrate over 94 trades, even after several rounds of tweaking.
The thing that stood out when I actually looked at the data: a huge chunk of the losing trades were in profit at some point before flipping to a loss. 46% of them on the 2H bot, 76% on the 4H one. And on the 2H bot specifically almost half of all trades (47%) close within 2 hours, which feels wrong for something that's supposed to be a 2 hour strategy.
This isn't our first attempt at fixing it either. Over the past few months we went through several rounds of changes (redesigned trailing stop, removed a filter that was blocking good trades, added a profit lock mechanism, tactical circuit breakers after a bad day), and tested some of them before pushing live rather than just guessing blind. But looking back, none of that testing ever specifically targeted the exit/stop loss logic against this give back pattern. That whole history, version by version, what we tried and what the data says about whether it worked, is written up in full.
Before posting here I also tried simulating a scaling out approach (lock 50% of the position once it hits a certain profit threshold, let the rest ride). Tested a bunch of thresholds on both bots. Honestly it made the 2H bot worse across every single threshold I tried, and on the 4H bot it only looked better in a narrow band of thresholds which smells like overfitting on a small sample rather than a real edge. So that's not it either, or at least not in this simple form.
Full writeup, code, and the raw CSVs for both bots (every trade, entry/exit price, max/min pnl reached, why it closed, how long it lasted) are in the first comment, didn't want to clutter the post itself.
Not trying to sell anything, no product, just stuck and would rather ask than keep guessing. Has anyone run into this exact give back problem on shorter timeframe swing strategies? Trying to figure out if it's a stop placement thing, a timeframe/volatility mismatch, or something more fundamental I'm missing. All paper trading, no real money on the line, happy to answer questions about the exact rules if anyone wants details.
#cryptocurrency
Been paper trading two bots on Bitget perpetual futures since April. One runs on a 2H timeframe, the other on 4H. Same basic idea behind both (gated entry with a scoring system, multi phase trailing stop that locks in profit as it goes), just different speeds.
3 months later the 4H one is barely green, +9.47 on a 100 starting balance, 68% winrate over 91 trades. The 2H one is still red, -7.62, 43% winrate over 94 trades, even after several rounds of tweaking.
The thing that stood out when I actually looked at the data: a huge chunk of the losing trades were in profit at some point before flipping to a loss. 46% of them on the 2H bot, 76% on the 4H one. And on the 2H bot specifically almost half of all trades (47%) close within 2 hours, which feels wrong for something that's supposed to be a 2 hour strategy.
This isn't our first attempt at fixing it either. Over the past few months we went through several rounds of changes (redesigned trailing stop, removed a filter that was blocking good trades, added a profit lock mechanism, tactical circuit breakers after a bad day), and tested some of them before pushing live rather than just guessing blind. But looking back, none of that testing ever specifically targeted the exit/stop loss logic against this give back pattern. That whole history, version by version, what we tried and what the data says about whether it worked, is written up in full.
Before posting here I also tried simulating a scaling out approach (lock 50% of the position once it hits a certain profit threshold, let the rest ride). Tested a bunch of thresholds on both bots. Honestly it made the 2H bot worse across every single threshold I tried, and on the 4H bot it only looked better in a narrow band of thresholds which smells like overfitting on a small sample rather than a real edge. So that's not it either, or at least not in this simple form.
Full writeup, code, and the raw CSVs for both bots (every trade, entry/exit price, max/min pnl reached, why it closed, how long it lasted) are in the first comment, didn't want to clutter the post itself.
Not trying to sell anything, no product, just stuck and would rather ask than keep guessing. Has anyone run into this exact give back problem on shorter timeframe swing strategies? Trying to figure out if it's a stop placement thing, a timeframe/volatility mismatch, or something more fundamental I'm missing. All paper trading, no real money on the line, happy to answer questions about the exact rules if anyone wants details.
#cryptocurrency
| 0 mlx | |
| 3 | |
| 0 |