hmo
b9c68a83a7
- 新增 strategy_research_methodology.md（由果及因/12维/铁律/支撑压力规范） - 新增 predictive_oversold_strategy.md（v5定稿，年化18.57%） - 新增 deployment-plan-predictive-oversold.md（整合部署计划） - 归档 docs/research/（63份研究过程文档）+ scripts/research/（19个研究脚本） - 更新 docs/README.md 文档中心（策略研究章节）
202 lines
8.0 KiB
Python
202 lines
8.0 KiB
Python
#!/usr/bin/env python3
|
||
"""step46_final_v4.py — 最终模拟 v4（负面过滤 + 限流）
|
||
在 step41 基础上，用数据提取的负面因子过滤（step43-45）：
|
||
- 信号: 预测信号 + 大盘回撤<=-5%
|
||
- 负面过滤: 新闻年内>=中位 + 连跌<=2 + flow1>-1e7
|
||
- 单日限流5 + 10槽 + 15%仓位
|
||
对比: 年化/回撤 vs step37(v3基准)
|
||
"""
|
||
import numpy as np
|
||
import pandas as pd
|
||
import sqlite3
|
||
import sys
|
||
sys.path.insert(0, "/tmp")
|
||
from sr_calculator import SRCalculator
|
||
|
||
print("=== 加载 ===", flush=True)
|
||
panel = pd.read_pickle("/tmp/panel_12d.pkl")
|
||
panel = panel.sort_values(["code", "date"]).reset_index(drop=True)
|
||
panel["_key"] = panel["code"] + "_" + panel["date"]
|
||
pos_map = {k: i for i, k in enumerate(panel["_key"])}
|
||
dates = sorted(panel["date"].unique())
|
||
|
||
# 大盘/气氛
|
||
conn = sqlite3.connect("file:/home/hmo/MoFin/data/mofin.db?mode=ro", uri=True)
|
||
idx_df = pd.read_sql("SELECT date, close, high FROM stock_daily WHERE code='sh000001' ORDER BY date", conn)
|
||
idx_df["date"] = idx_df["date"].astype(str)
|
||
idx_df["hi60"] = idx_df["high"].rolling(60).max()
|
||
idx_df["mkt_dd60"] = (idx_df["close"] / idx_df["hi60"] - 1) * 100
|
||
close_arr = idx_df["close"].values
|
||
down_days, streak, prev = [], 0, None
|
||
for c in close_arr:
|
||
streak = streak + 1 if (prev is not None and c < prev) else 0
|
||
down_days.append(streak)
|
||
prev = c
|
||
idx_df["mkt_down_days"] = down_days
|
||
dd_map = dict(zip(idx_df["date"], idx_df["mkt_dd60"]))
|
||
down_map = dict(zip(idx_df["date"], idx_df["mkt_down_days"]))
|
||
panel["mkt_dd60"] = panel["date"].map(dd_map)
|
||
panel["mkt_down_days"] = panel["date"].map(down_map)
|
||
|
||
news_cnt = pd.read_sql("SELECT substr(date,1,10) d, COUNT(*) c FROM stock_news GROUP BY d", conn)
|
||
news_cnt["d"] = news_cnt["d"].astype(str)
|
||
news_cnt["mkt_news5"] = news_cnt["c"].rolling(5, min_periods=1).mean()
|
||
news_map = dict(zip(news_cnt["d"], news_cnt["mkt_news5"]))
|
||
panel["mkt_news5"] = panel["date"].map(lambda d: news_map.get(d, np.nan))
|
||
# 年内新闻分位
|
||
panel["year"] = panel["date"].str[:4]
|
||
panel["news_yq"] = panel.groupby("year")["mkt_news5"].transform(lambda x: x.rank(pct=True))
|
||
|
||
# 信号
|
||
sig_cond = (
|
||
(panel["mkt_rsi"] < 50) & (panel["mcap_q"] < 0.2) & (panel["pe_q"] < 0.2) &
|
||
(panel["news3"] >= 1) & (panel["sec_ret20"] < 0) & (panel["bias60"] < -20) &
|
||
(panel["mkt_dd60"] <= -5)
|
||
)
|
||
# 负面过滤
|
||
neg_cond = sig_cond & (panel["news_yq"] >= 0.5) & (panel["mkt_down_days"] <= 2) & (panel["flow1"] > -1e7)
|
||
|
||
print("基础信号:", sig_cond.sum(), "负面过滤后:", neg_cond.sum(), flush=True)
|
||
|
||
def build_cand(cond):
|
||
cand = panel[cond][["code", "date"]].copy()
|
||
cand = cand.sort_values(["code", "date"])
|
||
cand["prev"] = cand.groupby("code")["date"].shift(1)
|
||
cand["gap"] = (pd.to_datetime(cand["date"]) - pd.to_datetime(cand["prev"])).dt.days
|
||
cand = cand[(cand["prev"].isna()) | (cand["gap"] > 30)]
|
||
return cand
|
||
|
||
def prep(sigdf):
|
||
codes = sigdf["code"].unique().tolist()
|
||
ph = ",".join("?" * len(codes))
|
||
df = pd.read_sql("SELECT code, date, close, high, low FROM stock_daily WHERE code IN ({}) ORDER BY code, date".format(ph), conn, params=codes)
|
||
df["date"] = df["date"].astype(str)
|
||
df["code"] = df["code"].astype(str).str.zfill(6)
|
||
df = df.sort_values(["code", "date"]).reset_index(drop=True)
|
||
df["_key"] = df["code"] + "_" + df["date"]
|
||
dpos = {k: i for i, k in enumerate(df["_key"])}
|
||
sigdf["kidx"] = (sigdf["code"] + "_" + sigdf["date"]).map(dpos)
|
||
sigdf = sigdf.dropna(subset=["kidx"]).copy()
|
||
sigdf["kidx"] = sigdf["kidx"].astype(int)
|
||
sr = SRCalculator()
|
||
out = []
|
||
for r in sigdf.itertuples():
|
||
bars_code = sr.get_bars(r.code)
|
||
d_idx = bars_code.index[bars_code["date"] == r.date]
|
||
if len(d_idx) == 0:
|
||
continue
|
||
sr_full = sr.sr_full(r.code, d_idx[0])
|
||
pv = sr_full["pivot"]
|
||
chip = sr_full["chip"]
|
||
if not pv:
|
||
continue
|
||
sig_close = df["close"].values[r.kidx]
|
||
support = pv["s2"]
|
||
if chip and chip["chip_ss"] < sig_close:
|
||
support = max(pv["s2"], chip["chip_ss"])
|
||
resist = pv["r2"]
|
||
if chip and chip["chip_sr"] > sig_close:
|
||
resist = min(pv["r2"], chip["chip_sr"])
|
||
if support >= sig_close or resist <= sig_close:
|
||
continue
|
||
out.append({"code": r.code, "date": r.date, "kidx": r.kidx,
|
||
"sig_close": sig_close, "support": support, "resist": resist})
|
||
return pd.DataFrame(out), df
|
||
|
||
def run_sim(sigdf, df, max_daily=5, slots=10, hold=40, pos_frac=0.15, stop_buf=0.05):
|
||
dpos = {k: i for i, k in enumerate(df["_key"])}
|
||
closes = df["close"].values
|
||
highs = df["high"].values
|
||
lows = df["low"].values
|
||
sig_by_date = {}
|
||
for r in sigdf.itertuples():
|
||
sig_by_date.setdefault(r.date, []).append(r)
|
||
INIT_CAP = 1_000_000
|
||
positions = {}
|
||
cash = INIT_CAP
|
||
navs = []
|
||
for di, d in enumerate(dates):
|
||
for code in list(positions.keys()):
|
||
pos = positions[code]
|
||
k = dpos.get(code + "_" + d)
|
||
if k is None:
|
||
continue
|
||
hi, lo, cl = highs[k], lows[k], closes[k]
|
||
exited = False
|
||
if hi >= pos["tp"]:
|
||
sell_qty = pos["qty"] // 2
|
||
if sell_qty > 0:
|
||
cash += sell_qty * pos["tp"]
|
||
pos["qty"] -= sell_qty
|
||
if pos["qty"] <= 0:
|
||
del positions[code]
|
||
exited = True
|
||
if not exited and lo <= pos["stop"]:
|
||
cash += pos["qty"] * lo
|
||
del positions[code]
|
||
exited = True
|
||
if not exited and di - pos["entry_di"] >= hold:
|
||
cash += pos["qty"] * cl
|
||
del positions[code]
|
||
if d in sig_by_date:
|
||
bought = 0
|
||
for r in sig_by_date[d]:
|
||
if bought >= max_daily or len(positions) >= slots:
|
||
break
|
||
if r.code in positions:
|
||
continue
|
||
cur_nav = cash
|
||
for c, p in positions.items():
|
||
k = dpos.get(c + "_" + d)
|
||
px = closes[k] if k is not None else p["avg_cost"]
|
||
cur_nav += p["qty"] * px
|
||
pos_val = cur_nav * pos_frac
|
||
price = r.sig_close
|
||
if price <= 0:
|
||
continue
|
||
qty = int(pos_val / price)
|
||
if qty <= 0 or qty * price > cash:
|
||
continue
|
||
cash -= qty * price
|
||
positions[r.code] = {
|
||
"entry_di": di, "qty": qty, "avg_cost": price,
|
||
"stop": r.support * (1 - stop_buf), "tp": r.resist,
|
||
}
|
||
bought += 1
|
||
nav = cash
|
||
for c, p in positions.items():
|
||
k = dpos.get(c + "_" + d)
|
||
px = closes[k] if k is not None else p["avg_cost"]
|
||
nav += p["qty"] * px
|
||
navs.append(nav)
|
||
nav_arr = np.array(navs)
|
||
final = nav_arr[-1]
|
||
years = len(navs) / 250
|
||
cagr = ((final / INIT_CAP) ** (1 / years) - 1) * 100 if final > 0 else -100
|
||
cummax = np.maximum.accumulate(nav_arr)
|
||
dd = ((nav_arr - cummax) / cummax).min() * 100
|
||
return cagr, dd, nav_arr
|
||
|
||
print("\n=== 基准（无负面过滤，限流5）===", flush=True)
|
||
c0, d0, _ = run_sim(*prep(build_cand(sig_cond)))
|
||
print("基准: 年化={:.2f}% 回撤={:.1f}%".format(c0, d0), flush=True)
|
||
|
||
print("\n=== 负面过滤（限流5）===", flush=True)
|
||
s1, df1 = prep(build_cand(neg_cond))
|
||
c1, d1, nav1 = run_sim(s1, df1)
|
||
print("负面过滤: 年化={:.2f}% 回撤={:.1f}% 信号={}".format(c1, d1, len(s1)), flush=True)
|
||
|
||
print("\n=== 负面过滤 限流参数 ===", flush=True)
|
||
for md in [3, 5, 8, 99]:
|
||
c, d, _ = run_sim(s1, df1, max_daily=md)
|
||
print("限{}: 年化={:.2f}% 回撤={:.1f}%".format(md, c, d), flush=True)
|
||
|
||
print("\n=== 分年净值（负面过滤+限流5）===", flush=True)
|
||
nav_df = pd.DataFrame({"date": dates, "nav": nav1})
|
||
nav_df["year"] = nav_df["date"].str[:4]
|
||
yr = nav_df.groupby("year")["nav"].last()
|
||
for y, v in yr.items():
|
||
print(" {}: {:.0f} ({}%)".format(y, v, (v/1000000-1)*100))
|
||
|
||
print("\n=== 完成 ===", flush=True)
|