"""烛龙交易分析：逐笔拆解盈利/亏损交易，找规律""" import pandas as pd, numpy as np df_raw = pd.read_parquet("data/btc_multidim.parquet") df_raw.columns = [c.lower() for c in df_raw.columns] d = df_raw.resample("1h").agg({"open":"first","high":"max","low":"min","close":"last","volume":"sum"}).dropna() O,H,L,C,V = d["open"].values,d["high"].values,d["low"].values,d["close"].values,d["volume"].values O1,H1,L1,C1 = np.roll(O,1),np.roll(H,1),np.roll(L,1),np.roll(C,1) n = len(d) dd = df_raw.resample("1D").agg({"close":"last"}).dropna() dd["ma20"] = dd["close"].rolling(20).mean() dd["trend_up"] = dd["close"] > dd["ma20"] d_ts = d.index trend_1h = np.array([dd["trend_up"].reindex([ts], method="ffill").values[0] if ts >= dd.index[0] else True for ts in d_ts]) prev_l20 = pd.Series(L).shift(1).rolling(20).min().values prev_h20 = pd.Series(H).shift(1).rolling(20).max().values near_s = abs(L - prev_l20) / (prev_l20 + 1e-9) < 0.008 near_r = abs(H - prev_h20) / (prev_h20 + 1e-9) < 0.008 bull_engulf = (C1O) & (O<=C1) & (C>=O1) nbull = np.roll(C,-1) > np.roll(O,-1) long_raw = bull_engulf & near_s & nbull swup = (H>pd.Series(H).shift(1).rolling(20).max().values) & (C=n: continue entry = C[i] # 开仓时的各种特征 vol_ratio = V[i] / vol_mean20[i] # 当前量/均值 k_line_range = H[i] - L[i] # K线振幅 body_size = abs(C[i] - O[i]) # K线实体 body_ratio = body_size / (k_line_range + 1e-9) # 实体占比 # 关键位距离（相对于入场价） sup_dist = abs(C[i] - prev_l20[i]) / C[i] # 距支撑 res_dist = abs(C[i] - prev_h20[i]) / C[i] # 距阻力 # 近期走势 ret_1 = C[i]/C[i-1]-1 ret_3 = C[i]/C[i-3]-1 if i>=3 else 0 ret_6 = C[i]/C[i-6]-1 if i>=6 else 0 # 前一根K线 prev_ret = C[i-1]/C[i-2]-1 if i>=2 else 0 prev_body = abs(C[i-1]-O[i-1]) prev_vol_ratio = V[i-1] / vol_mean20[i-1] if i>0 else 0 # 确认K线 confirm_ret = C[i+1]/C[i]-1 # 确认方向 confirm_body = abs(C[i+1]-O[i+1]) # ── 回测这笔交易 ── closed = False bar_count = 0 max_favor = 0 # 最大浮盈 max_against = 0 # 最大浮亏 for j in range(1, MB+1): if i+j>=n: break ret = (C[i+j]/entry-1)*dirc bar_count = j if ret > max_favor: max_favor = ret if ret < max_against: max_against = ret if ret>=TP: trades.append({ "period": period, "dirc": "long" if dirc==1 else "short", "time": d.index[i], "entry": entry, "exit": C[i+j], "pnl": TP-0.001, "bars": bar_count, "reason": "TP", "max_favor": max_favor, "max_against": max_against, "vol_ratio": vol_ratio, "body_ratio": body_ratio, "k_range": k_line_range, "sup_dist": sup_dist, "res_dist": res_dist, "ret_1": ret_1, "ret_3": ret_3, "ret_6": ret_6, "prev_ret": prev_ret, "prev_vol": prev_vol_ratio, "confirm_ret": confirm_ret, "trend": "up" if trend_1h[i] else "down", }) closed=True; break if ret<=-SL: trades.append({ "period": period, "dirc": "long" if dirc==1 else "short", "time": d.index[i], "entry": entry, "exit": C[i+j], "pnl": -SL-0.001, "bars": bar_count, "reason": "SL", "max_favor": max_favor, "max_against": max_against, "vol_ratio": vol_ratio, "body_ratio": body_ratio, "k_range": k_line_range, "sup_dist": sup_dist, "res_dist": res_dist, "ret_1": ret_1, "ret_3": ret_3, "ret_6": ret_6, "prev_ret": prev_ret, "prev_vol": prev_vol_ratio, "confirm_ret": confirm_ret, "trend": "up" if trend_1h[i] else "down", }) closed=True; break if not closed: fr = (C[min(i+MB,n-1)]/entry-1)*dirc trades.append({ "period": period, "dirc": "long" if dirc==1 else "short", "time": d.index[i], "entry": entry, "exit": C[min(i+MB,n-1)], "pnl": fr-0.001, "bars": MB, "reason": "timeout", "max_favor": max_favor, "max_against": max_against, "vol_ratio": vol_ratio, "body_ratio": body_ratio, "k_range": k_line_range, "sup_dist": sup_dist, "res_dist": res_dist, "ret_1": ret_1, "ret_3": ret_3, "ret_6": ret_6, "prev_ret": prev_ret, "prev_vol": prev_vol_ratio, "confirm_ret": confirm_ret, "trend": "up" if trend_1h[i] else "down", }) df_trades = pd.DataFrame(trades) print(f"共 {len(df_trades)} 笔交易") print(f"\n各原因分布:") print(df_trades.groupby("reason").agg( 笔数=("pnl","count"), 胜率=("pnl",lambda x: (x>0).mean()), 均收益=("pnl",lambda x: f"{x.mean()*100:.2f}%"), 累计=("pnl",lambda x: f"{np.prod(1+x):.3f}") )) print(f"\n{'='*70}") print(f"📊 盈利 vs 亏损 对比（只看样本外）") print(f"{'='*70}") out = df_trades[df_trades["period"]=="样本外"] wins = out[out["pnl"]>0] losses = out[out["pnl"]<0] print(f"\n盈利: {len(wins)}笔 亏损: {len(losses)}笔") features = ["vol_ratio","body_ratio","k_range","sup_dist","res_dist", "ret_1","ret_3","ret_6","prev_ret","confirm_ret","bars"] for feat in features: w_mean = wins[feat].mean() if feat in wins.columns else 0 l_mean = losses[feat].mean() if feat in losses.columns else 0 w_median = wins[feat].median() if feat in wins.columns else 0 l_median = losses[feat].median() if feat in losses.columns else 0 diff_pct = ((w_mean - l_mean) / (abs(l_mean)+1e-9)) * 100 star = " ⭐" if abs(diff_pct) > 15 else "" print(f" {feat:15s}: 盈利均值={w_mean:.4f} 亏损均值={l_mean:.4f} 差异={diff_pct:+.0f}%{star}") print(f" {'':15s} 盈利中位={w_median:.4f} 亏损中位={l_median:.4f}") # 找出最有区分度的特征 print(f"\n{'='*70}") print(f"🔍 特征分析——按特征分组看胜率") print(f"{'='*70}") # 放量维度 for threshold in [0.8, 1.0, 1.2, 1.5, 2.0]: sub = out[out["vol_ratio"]>threshold] if len(sub)>=3: wr = (sub["pnl"]>0).mean() print(f" 放量>{threshold:.1f}x均值: {len(sub)}笔 胜率={wr:.1%}") print() # 前一根K线收益 for threshold in [-0.005, 0, 0.005, 0.01]: sub = out[out["prev_ret"]>threshold] if threshold>=0 else out[out["prev_ret"]=0 else "<" if len(sub)>=3: wr = (sub["pnl"]>0).mean() print(f" 前K收益{dir_str}{threshold:+.4f}: {len(sub)}笔 胜率={wr:.1%}") print() # 入场前短期走势 for threshold in [-0.01, -0.005, 0, 0.005, 0.01]: sub = out[out["ret_3"]>threshold] if threshold>=0 else out[out["ret_3"]=0 else "<" if len(sub)>=3: wr = (sub["pnl"]>0).mean() print(f" 3h涨幅{dir_str}{threshold:+.4f}: {len(sub)}笔 胜率={wr:.1%}") # 近因分析：最近几个月 print(f"\n{'='*70}") print(f"📅 时间维度分析") print(f"{'='*70}") out["month"] = pd.to_datetime(out["time"]).dt.to_period("M") monthly = out.groupby("month").agg(笔数=("pnl","count"), 胜率=("pnl",lambda x: (x>0).mean()), 累计=("pnl",lambda x: np.prod(1+x))) print(monthly) # 交易时长分析 print(f"\n{'='*70}") print(f"⏱ 持仓时间分析") print(f"{'='*70}") for b in [6, 12, 24, 36, 48]: sub = out[out["bars"]<=b] if len(sub)>=3: wr = (sub["pnl"]>0).mean() cum = np.prod(1+sub["pnl"]) print(f" 持仓<={b}h: {len(sub)}笔 胜率={wr:.1%} 累计={cum:.3f}")