跳转到内容

Phase 3 优化(归档)

Strategy Optimization Plan — Phase 3 (COMPLETED)

Section titled “Strategy Optimization Plan — Phase 3 (COMPLETED)”
StepChangePFWin%DDStatus
Baselinedefaults2.4737%25%
Step 1Remove LTC/ATOM/UNI/NEAR2.6143%18%✅ Kept
Step 2Faster exit1.8139%❌ Rejected (worse)
Step 3BTC reference pair2.6846%18%✅ Kept
Step 4Hyperopt EMA 30/61 ADX 163.1953%17%✅ Kept

Final optimized: PF 3.19, Win 53%, DD 17% — all significantly better than baseline.

  • 43 trades, +221%, PF 2.47, Max DD 25%, Win Rate 37%
  • Winners avg +88%, Losers avg -13%
  • Winners hold 109 days, Losers hold 36 days
亏损来源分析:
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
1. 垃圾币对 (占总亏损 ~40%):
LTC: -52% (0/3 wins) ← 从不赚钱
ATOM: -30% (0/3 wins) ← 从不赚钱
UNI: -24% (0/2 wins) ← 从不赚钱
XRP: -31% (1/4 wins) ← 极少赚钱
2. 出场太慢 (占总亏损 ~35%):
7 笔交易亏 -20% 到 -30%
4 笔交易亏 -15% 到 -20%
→ EMA 21/55 死叉太慢,价格已跌很多才出场
3. 参数未优化 (潜在收益):
EMA 21/55 是默认值,从没跑过 hyperopt
ADX 20 是默认值
→ 可能有更好的参数组合
4. Altcoin 逆势入场 (占总亏损 ~25%):
BTC 下跌时 altcoin 仍在触发入场信号
→ 需要 BTC 方向过滤器

Goal: Remove pairs that historically never profit Action: Blacklist LTC, ATOM, UNI, NEAR from config Expected: DD减少5-10%, PF提升0.2-0.3 Risk: 可能错过未来某次大行情,但历史证明这些币在我们策略下不赚钱

Goal: Cut losers faster, keep winners running Action: Add custom_exit with:

  • Time-based: if trade unprofitable after 30 days → exit
  • EMA slope: if EMA 21 slope turns negative while in trade → exit
  • Faster cross: use EMA 13/34 for exit (faster than 21/55) Expected: Avg loser from -13% to -8%, DD减少5% Risk: 可能过早退出最终会盈利的交易

Goal: Only enter altcoins when BTC trend agrees Action: Add BTC/USDT as informative pair, require BTC EMA uptrend for altcoin entries Expected: 减少逆势 altcoin 入场,win rate提升3-5% Risk: 可能错过 altcoin 独立行情

Step 4: Hyperopt Parameter Tuning ⏱️ 1 hour

Section titled “Step 4: Hyperopt Parameter Tuning ⏱️ 1 hour”

Goal: Find optimal EMA periods, ADX threshold Action: Run hyperopt with SharpeHyperOptLoss, 300 epochs Optimize: ema_fast (10-30), ema_slow (40-80), adx_threshold (15-30) Expected: PF可能提升20-30% Risk: 过拟合。用 walk-forward 验证。

After each step, run full backtest + walk-forward to confirm improvement. Only keep changes that improve BOTH total profit AND profit factor.