Markov Fund
量化多策略 · 风险调整超额

# ─── 制度权重矩阵（3大核心策略）──────────────────── REGIME_WEIGHTS = { "bull_low": {"dual_mom":0.55, "resid_mom":0.25, "hrp":0.20}, "bull_high": {"dual_mom":0.25, "resid_mom":0.30, "hrp":0.45}, "bear_low": {"dual_mom":0.05, "resid_mom":0.20, "hrp":0.75}, "bear_high": {"dual_mom":0.00, "resid_mom":0.10, "hrp":0.90}, } # ─── Top-Down 大类资产目标配置 ────────────────── REGIME_ASSET_TARGETS = { "bull_low": {"equity_us":0.80, "bond":0.08, "gold":0.12}, "bull_high": {"equity_us":0.55, "bond":0.25, "gold":0.20}, "bear_low": {"equity_us":0.37, "bond":0.40, "gold":0.23}, "bear_high": {"equity_us":0.20, "bond":0.55, "gold":0.25}, } # ─── 权益三桶分层分配 ────────────────────────── EQUITY_BUCKET_SPLIT = {"idx": 0.15, "sector": 0.00, "stock": 0.85} BTC_MAX_WEIGHT = 0.05 # Markov Plus BTC硬上限

def _sigmoid(x, k=1.0): """sigmoid 平滑函数：将连续值映射到 (0,1) 概率""" return 1.0 / (1.0 + np.exp(-k * x)) def detect_regime(spy_prices, vix_prices, vix3m_prices=None): """返回 bull_prob, high_vol_prob 软概率 + 硬状态标签""" ma200 = spy_prices.rolling(200).mean() spy_last = float(spy_prices.iloc[-1]) ma200_last = float(ma200.iloc[-1]) dev = spy_last / ma200_last - 1 # SPY 偏离 MA200 # MA200 斜率（20日变化率） ma200_slope = (ma200_last / ma200.iloc[-21] - 1) if len(ma200) >= 21 else 0 # VIX 期限结构 vix_last = float(vix_prices.reindex(spy_prices.index).ffill().dropna().iloc[-1]) vix_term = vix_last / vix3m_last if vix3m else 1.0 # 软概率：sigmoid 平滑避免边界跳变 bull_prob = 0.70 * _sigmoid(dev, k=25) + 0.30 * _sigmoid(ma200_slope, k=500) high_vol_prob = 0.60 * _sigmoid(vix_last - 20, k=0.5) \ + 0.40 * _sigmoid(vix_term - 1.0, k=10) bull = bull_prob >= 0.5 high_vol = high_vol_prob >= 0.5 state = ("bull" if bull else "bear") + "_" + ("high" if high_vol else "low") return {"state": state, "bull_prob": bull_prob, "high_vol_prob": high_vol_prob, "vix": vix_last, "spy_vs_ma200": round(dev * 100, 2), "ma200_slope": ma200_slope, "vix_term_ratio": vix_term}

def compute_soft_weights(regime): """用 bull_prob × high_vol_prob 对 4 区制权重矩阵做加权插值""" bp, hp = regime["bull_prob"], regime["high_vol_prob"] probs = {"bull_low": bp*(1-hp), "bull_high": bp*hp, "bear_low": (1-bp)*(1-hp), "bear_high": (1-bp)*hp} blended = {} for state, prob in probs.items(): for strat, w in REGIME_WEIGHTS[state].items(): blended[strat] = blended.get(strat, 0) + prob * w return blended def strategy_regime_integration(ret, regime): """按软权重混合 3 大核心策略 → 最终个股权重向量""" wts = compute_soft_weights(regime) funcs = {"dual_mom": strategy_dual_momentum, "resid_mom": strategy_residual_momentum, "hrp": strategy_hrp} combined = np.zeros(ret.shape[1]) for name, bw in wts.items(): if bw > 0 and name in funcs: try: combined += bw * funcs[name](ret) except: combined += bw / ret.shape[1] combined = np.maximum(combined, 0) return combined / combined.sum()

def strategy_dual_momentum(ret, lookback=252, skip=21): """双重动量：绝对动量过滤（>0）+ 相对动量选Top N，等权""" if len(ret) < lookback + skip: return np.ones(n)/n abs_mom = (ret.iloc[-skip] / ret.iloc[-lookback] - 1).values # 跳过最近21天 mask = abs_mom > 0 if mask.sum() == 0: return strategy_hrp(ret) # fallback rel_mom = abs_mom.copy(); rel_mom[~mask] = -np.inf n_select = max(5, int(mask.sum() * 0.5)) top_idx = np.argsort(rel_mom)[-n_select:] w = np.zeros(n); w[top_idx] = 1.0 / n_select return w def strategy_residual_momentum(ret, lookback=252, skip=21): """残差动量：对市场回归取残差 → 残差动量最强的前N只，等权""" if len(ret) < lookback + skip: return np.ones(n)/n mkt = ret.mean(axis=1) resid_mom = np.zeros(n) for i in range(n): beta = np.polyfit(mkt.iloc[-lookback:-skip], ret.iloc[-lookback:-skip, i], 1)[0] resid = ret.iloc[-lookback:-skip, i] - beta * mkt.iloc[-lookback:-skip] resid_mom[i] = (1 + resid).prod() - 1 top_n = max(1, int(n * 0.2)) top = np.argsort(resid_mom)[-top_n:] w = np.zeros(n); w[top] = 1.0 / top_n return w def strategy_hrp(ret): """层次风险平价：用距离矩阵做层次聚类，自底向上分配等风险贡献权重""" corr = ret.corr().values dist = np.sqrt(0.5 * (1 - corr)) Z = linkage(squareform(dist), "ward") n = len(ret.columns) w = np.ones(n) clusters = {i: [i] for i in range(n)} for row in Z: a, b = int(row[0]), int(row[1]) merged = clusters[a] + clusters[b] clusters[len(clusters)] = merged cov_a = ret.iloc[:, clusters[a]].cov().values.sum() cov_b = ret.iloc[:, clusters[b]].cov().values.sum() alpha = cov_b / (cov_a + cov_b) if (cov_a + cov_b) > 0 else 0.5 for idx in clusters[a]: w[idx] *= alpha for idx in clusters[b]: w[idx] *= (1 - alpha) return w / w.sum()

def build_topdown_portfolio(ret, regime): """区制驱动大类配置 + 权益三桶分层 + 桶内策略混合 → 最终权重向量""" targets = get_topdown_asset_targets(regime) # 软插值大类目标 buckets = classify_tickers(ret.columns) w = np.zeros(ret.shape[1]) regime_wts = compute_soft_weights(regime) for bucket, tgt_w in targets.items(): bk_tickers = buckets[bucket] if not bk_tickers or tgt_w <= 0: continue if bucket == "equity_us": # 三桶分层：A=指数(HRP) / B=行业(跳过) / C=个股(制度混合) idx_tks = [t for t in bk_tickers if t in INDEX_ETF_SET] # QQQ, SPMO stock_tks = [t for t in bk_tickers if t not in INDEX_ETF_SET and t not in SECTOR_ETF_SET] if idx_tks: w_idx = strategy_hrp(ret[idx_tks]) w[idx_tks] = tgt_w * EQUITY_BUCKET_SPLIT["idx"] * w_idx if stock_tks: w_stock = strategy_regime_integration(ret[stock_tks], regime) w[stock_tks] = tgt_w * EQUITY_BUCKET_SPLIT["stock"] * w_stock else: # bond → 6M风险平价 / gold → 等权 w_bk = _intra_bucket_weights(bucket, ret[bk_tickers], regime_wts) w[bk_tickers] = tgt_w * w_bk return w / w.sum() def compute_top20(weights_dict, ticker_names): """从权重字典取 Top 20，含大类资产分类""" ranked = sorted(weights_dict.items(), key=lambda x: x[1], reverse=True)[:20] result = [] for tk, w in ranked: ac = "黄金" if tk in GOLD_ETFS else "债券" if tk in BOND_ETFS \ else "比特币" if tk in CRYPTO_SET else "港股" if tk.endswith(".HK") \ else "美股" result.append({"ticker": tk, "name": ticker_names.get(tk, tk), "weight": round(w, 4), "asset_class": ac}) return result

def compute_entry_signal(spy_prices, vix_last, spy_vs_ma200_pct): """4维度评分：VIX恐慌、MA200偏离、短期动量、MA200斜率""" score = 0 dims = {"dim1": 0, "dim2": 0, "dim3": 0, "dim4": 0} # 死亡地带拦截：SPY偏离MA200在[-12%, -4%]时强制否决 dead_zone = -12 <= spy_vs_ma200_pct <= -4 if dead_zone: return {"score": 0, "dead_zone": True, "level": "死亡地带"} # 维度1：VIX恐慌（0-3分）→ VIX越高越恐慌=越接近底部 dims["dim1"] = min(3, max(0, int((vix_last - 15) / 5))) score += dims["dim1"] # 维度2：MA200偏离（0-3分）→ 偏离越深越超卖 dims["dim2"] = min(3, max(0, int(abs(spy_vs_ma200_pct) / 5))) score += dims["dim2"] # 维度3：短期动量（0-3分）→ 近20日跌幅越大信号越强 mom20 = (spy_prices.iloc[-1] / spy_prices.iloc[-21] - 1) dims["dim3"] = min(3, max(0, int(abs(mom20) / 0.05))) score += dims["dim3"] # 维度4：MA200斜率（0-3分）→ 斜率越低（趋势越弱）分越高 slope = (spy_prices.rolling(200).mean().iloc[-1] / spy_prices.rolling(200).mean().iloc[-21] - 1) dims["dim4"] = min(3, max(0, int(max(0, -slope) / 0.002))) score += dims["dim4"] stars = "★" * (score // 3) + "☆" * (4 - score // 3) level = "强烈买入" if score >= 9 else "买入" if score >= 6 \ else "中性" if score >= 3 else "观望" return {"score": score, "stars": stars, "level": level, "dead_zone": False, **dims}

# 安装依赖 pip install yfinance pandas numpy scipy requests # 手动运行 python3 fetch_data.py # 本地预览 python3 -m http.server 8080 # 访问 http://localhost:8080 # 部署到 Cloudflare Pages npx wrangler@3 pages deploy . --project-name=markovf --commit-dirty=true