feat: 策略研究Tab全流程回测框架 + API端点 + 前端渲染 + 盘中大跌分析日志

backtest_framework.py

@@ -0,0 +1,488 @@
+#!/usr/bin/env python3
+"""MoFin 回测框架 v2 — 从 stock_daily 原始行情计算全部指标，模拟全流程"""
+import sqlite3, json, math
+from datetime import datetime, timedelta
+from collections import defaultdict
+
+DB_PATH = "/home/hmo/MoFin/data/mofin.db"
+
+# ── 技术指标计算（纯 Python，不依赖第三方库）──
+
+def calc_ma(series, n):
+    """简单移动平均"""
+    result = []
+    for i in range(len(series)):
+        if i < n - 1:
+            result.append(None)
+        else:
+            result.append(sum(series[i-n+1:i+1]) / n)
+    return result
+
+def calc_ema(series, n):
+    """指数移动平均"""
+    result = []
+    k = 2 / (n + 1)
+    for i in range(len(series)):
+        if i == 0:
+            result.append(series[i])
+        else:
+            result.append(series[i] * k + result[-1] * (1 - k))
+    return result
+
+def calc_rsi(series, n=14):
+    """RSI"""
+    deltas = [series[i] - series[i-1] for i in range(1, len(series))]
+    gains = [d if d > 0 else 0 for d in deltas]
+    losses = [-d if d < 0 else 0 for d in deltas]
+    result = [None] * (n + 1)
+    avg_gain = sum(gains[:n]) / n
+    avg_loss = sum(losses[:n]) / n
+    if avg_loss == 0:
+        result.append(100)
+    else:
+        rs = avg_gain / avg_loss
+        result.append(100 - 100 / (1 + rs))
+    for i in range(n, len(gains)):
+        avg_gain = (avg_gain * (n-1) + gains[i]) / n
+        avg_loss = (avg_loss * (n-1) + losses[i]) / n
+        if avg_loss == 0:
+            result.append(100)
+        else:
+            rs = avg_gain / avg_loss
+            result.append(100 - 100 / (1 + rs))
+    # Pad front to match original length
+    while len(result) < len(series):
+        result.insert(0, None)
+    return result[:len(series)]
+
+def calc_macd(series, fast=12, slow=26, signal=9):
+    """MACD: 返回 macd_line, signal_line, histogram"""
+    ema_fast = calc_ema(series, fast)
+    ema_slow = calc_ema(series, slow)
+    macd_line = [ema_fast[i] - ema_slow[i] if ema_fast[i] is not None and ema_slow[i] is not None else None for i in range(len(series))]
+    # Recompute signal line from macd_line (only non-None values)
+    macd_clean = [v for v in macd_line if v is not None]
+    if not macd_clean:
+        return [None]*len(series), [None]*len(series), [None]*len(series)
+    signal_raw = calc_ema(macd_clean, signal)
+    signal_line = [None] * (len(series) - len(macd_clean)) + signal_raw
+    histogram = [macd_line[i] - signal_line[i] if macd_line[i] is not None and signal_line[i] is not None else None for i in range(len(series))]
+    return macd_line, signal_line, histogram
+
+def calc_tr(highs, lows, closes):
+    """True Range"""
+    tr = [highs[0] - lows[0]]
+    for i in range(1, len(highs)):
+        hl = highs[i] - lows[i]
+        hc = abs(highs[i] - closes[i-1])
+        lc = abs(lows[i] - closes[i-1])
+        tr.append(max(hl, hc, lc))
+    return tr
+
+def calc_atr(highs, lows, closes, n=14):
+    """ATR"""
+    tr = calc_tr(highs, lows, closes)
+    return calc_ema(tr, n)
+
+def calc_trend_strength(highs, lows, closes, n=14):
+    """简易趋势强度（替代 ADX）: (DI+ - DI-) / (DI+ + DI-) 归一化"""
+    tr = calc_tr(highs, lows, closes)
+    up = [highs[i] - highs[i-1] for i in range(1, len(highs))]
+    down = [lows[i-1] - lows[i] for i in range(1, len(lows))]
+    # Directional indicators (raw)
+    di_plus_raw = [0.0] * len(up)
+    di_minus_raw = [0.0] * len(up)
+    for i in range(len(up)):
+        if up[i] > down[i] and up[i] > 0:
+            di_plus_raw[i] = up[i]
+        if down[i] > up[i] and down[i] > 0:
+            di_minus_raw[i] = down[i]
+        # Normalize by TR
+        tr_val = tr[i+1] if (i+1) < len(tr) else tr[-1]
+        if tr_val > 0:
+            di_plus_raw[i] = di_plus_raw[i] / tr_val * 100
+            di_minus_raw[i] = di_minus_raw[i] / tr_val * 100
+    
+    # SMA of DMI
+    result = []
+    for i in range(len(di_plus_raw)):
+        if i < n - 1:
+            result.append(None)
+        else:
+            avg_plus = sum(di_plus_raw[i-n+1:i+1]) / n
+            avg_minus = sum(di_minus_raw[i-n+1:i+1]) / n
+            if avg_plus + avg_minus > 0:
+                dx = abs(avg_plus - avg_minus) / (avg_plus + avg_minus) * 100
+            else:
+                dx = 0
+            result.append(dx)
+    # Pad front to match original highs length
+    return [None] * (len(highs) - len(result)) + result
+
+def calc_obv(closes, volumes):
+    """OBV"""
+    obv = [volumes[0]]
+    for i in range(1, len(closes)):
+        if closes[i] > closes[i-1]:
+            obv.append(obv[-1] + volumes[i])
+        elif closes[i] < closes[i-1]:
+            obv.append(obv[-1] - volumes[i])
+        else:
+            obv.append(obv[-1])
+    return obv
+
+def calc_roc(series, n=10):
+    """Rate of Change"""
+    result = []
+    for i in range(len(series)):
+        if i < n:
+            result.append(None)
+        else:
+            result.append((series[i] - series[i-n]) / series[i-n] * 100 if series[i-n] != 0 else 0)
+    return result
+
+def prepare_bars(code, start_date, end_date):
+    """获取原始数据并计算全部指标"""
+    conn = sqlite3.connect(DB_PATH)
+    rows = conn.execute("""
+        SELECT date, open, close, high, low, volume, amount
+        FROM stock_daily WHERE code=? AND date>=? AND date<=?
+        ORDER BY date
+    """, (code, start_date, end_date)).fetchall()
+    conn.close()
+    if not rows or len(rows) < 30:
+        return None
+    dates = [r[0] for r in rows]
+    opens = [r[1] for r in rows]
+    closes = [r[2] for r in rows]
+    highs = [r[3] for r in rows]
+    lows = [r[4] for r in rows]
+    volumes = [r[5] for r in rows]
+    
+    # 计算全部指标
+    ma5 = calc_ma(closes, 5)
+    ma10 = calc_ma(closes, 10)
+    ma20 = calc_ma(closes, 20)
+    ma60 = calc_ma(closes, 60)
+    rsi = calc_rsi(closes)
+    macd_line, signal_line, macd_hist = calc_macd(closes)
+    atr = calc_atr(highs, lows, closes)
+    trend_strength = calc_trend_strength(highs, lows, closes)
+    obv = calc_obv(closes, volumes)
+    roc = calc_roc(closes)
+    
+    bars = []
+    for i in range(len(dates)):
+        bars.append({
+            'date': dates[i],
+            'open': opens[i],
+            'close': closes[i],
+            'high': highs[i],
+            'low': lows[i],
+            'volume': volumes[i],
+            'ma5': ma5[i] if i < len(ma5) else None,
+            'ma10': ma10[i] if i < len(ma10) else None,
+            'ma20': ma20[i] if i < len(ma20) else None,
+            'ma60': ma60[i] if i < len(ma60) else None,
+            'rsi': rsi[i] if i < len(rsi) else None,
+            'macd': macd_line[i] if i < len(macd_line) else None,
+            'macd_signal': signal_line[i] if i < len(signal_line) else None,
+            'macd_hist': macd_hist[i] if i < len(macd_hist) else None,
+            'atr': atr[i] if i < len(atr) else None,
+            'adx': trend_strength[i] if i < len(trend_strength) else None,
+            'obv': obv[i] if i < len(obv) else None,
+            'roc': roc[i] if i < len(roc) else None,
+        })
+    return bars
+
+
+def compute_single_score(bars):
+    """对单个股票计算多因子评分（与 MoFin 现有一致）"""
+    if not bars or len(bars) < 15:
+        return None
+    last = bars[-1]
+    prev5 = bars[-5] if len(bars) >= 5 else bars[-2] if len(bars) >= 2 else last
+    prev10 = bars[-10] if len(bars) >= 10 else bars[0]
+    
+    def v(val): return val if val is not None else 0
+    score = {'trend': 0, 'momentum': 0, 'volume': 0, 'volatility': 0, 'risk': 0}
+    
+    # 趋势分 (0-25)
+    ts = 0
+    adx = v(last.get('adx'))
+    if adx > 25: ts += 10
+    elif adx > 20: ts += 5
+    mh = v(last.get('macd_hist'))
+    pmh = v(prev5.get('macd_hist'))
+    if mh > 0 and mh > pmh: ts += 10
+    elif mh > 0: ts += 5
+    elif mh < 0 and mh < pmh: ts -= 5
+    ma5 = v(last.get('ma5'))
+    ma10 = v(last.get('ma10'))
+    ma20 = v(last.get('ma20'))
+    if ma5 > ma10 > ma20 and ma5 > 0 and ma10 > 0: ts += 5
+    elif ma5 < ma10 < ma20 and ma5 > 0 and ma10 > 0: ts -= 5
+    score['trend'] = max(0, min(25, ts))
+    
+    # 动量分 (0-20)
+    ms = 0
+    roc = v(last.get('roc'))
+    if roc > 5: ms += 5
+    elif roc > 2: ms += 2
+    elif roc < -5: ms -= 5
+    elif roc < -2: ms -= 2
+    rsi = v(last.get('rsi'))
+    if 30 <= rsi <= 70: ms += 5
+    if 50 < rsi <= 65: ms += 3
+    elif rsi > 70: ms -= 3
+    elif rsi < 30: ms -= 3
+    close = v(last.get('close'))
+    if ma20 > 0:
+        pct = (close - ma20) / ma20 * 100
+        if -3 <= pct <= 5: ms += 7
+        elif pct < -10: ms -= 3
+    score['momentum'] = max(0, min(20, ms))
+    
+    # 量能分 (0-20)
+    vs = 0
+    obv_cur = v(last.get('obv'))
+    obv_prev = v(prev5.get('obv'))
+    vol_cur = v(last.get('volume'))
+    vol_prev = v(prev5.get('volume'))
+    if obv_cur > obv_prev and vol_cur > 0 and vol_prev > 0:
+        vr = vol_cur / vol_prev
+        if vr > 1.5: vs += 10
+        elif vr > 1.2: vs += 5
+    if obv_cur > obv_prev: vs += 5
+    if vol_prev > 0 and vol_cur / vol_prev < 0.5: vs -= 5
+    score['volume'] = max(0, min(20, vs))
+    
+    # 波动率分 (0-15)
+    vls = 5
+    atr = v(last.get('atr'))
+    if close > 0 and atr > 0:
+        atr_pct = atr / close * 100
+        if 1.5 <= atr_pct <= 3.5: vls += 5
+        elif atr_pct < 1: vls += 3
+        elif atr_pct > 5: vls -= 5
+    if len(bars) >= 20:
+        cls = [v(b.get('close')) for b in bars[-20:]]
+        if min(cls) > 0:
+            vol_range = (max(cls) - min(cls)) / min(cls) * 100
+            if vol_range < 10: vls += 5
+            elif vol_range > 30: vls -= 5
+    score['volatility'] = max(0, min(15, vls))
+    
+    # 风险分 (0-20)
+    rs = 10
+    if len(bars) >= 10:
+        low5 = min(bars[-5:], key=lambda x: v(x.get('low')))
+        low10 = min(bars[-10:-5], key=lambda x: v(x.get('low'))) if len(bars) >= 10 else low5
+        if v(low5.get('close')) < v(low10.get('close')) and v(low5.get('macd_hist')) > v(low10.get('macd_hist')):
+            rs += 5
+        if v(low5.get('close')) > v(low10.get('close')) and v(low5.get('macd_hist')) < v(low10.get('macd_hist')):
+            rs -= 5
+    if ma20 > 0:
+        dist = (close - ma20) / ma20 * 100
+        if dist < 2: rs += 5
+        elif dist > 15: rs -= 5
+    score['risk'] = max(0, min(20, rs))
+    
+    total = sum(score.values())
+    return total, score
+
+
+def compute_kelly(total_score, target_pct, stop_pct):
+    """半 Kelly 仓位"""
+    p = min(0.6, total_score / 100)
+    if stop_pct <= 0:
+        return 0
+    b = target_pct / stop_pct
+    if b <= 1:
+        return 0
+    kelly = (p * b - (1 - p)) / b
+    return max(0, min(0.5, kelly)) * 0.5  # 半 Kelly
+
+
+def run_strategy_research(start_date, end_date, capital=1000000):
+    """主回测入口"""
+    conn = sqlite3.connect(DB_PATH)
+    
+    # 1. 获取所有有历史数据的股票
+    stocks = conn.execute("""
+        SELECT DISTINCT sd.code, COALESCE(s.name, sd.code) as name
+        FROM stock_daily sd
+        LEFT JOIN stocks s ON sd.code = s.code
+        WHERE sd.date>=? AND sd.date<=?
+    """, (start_date, end_date)).fetchall()
+    
+    results = {
+        'period': f"{start_date} ~ {end_date}",
+        'capital': capital,
+        'total_stocks_screened': len(stocks),
+        'scored_stocks': 0,
+        'buy_signals': 0,
+        'trades': [],
+        'summary': {}
+    }
+    
+    trades = []
+    code_names = {}
+    eval_dates = []  # collect evaluation dates
+    
+    for code, name in stocks:
+        code_names[code] = name
+        bars = prepare_bars(code, start_date, end_date)
+        if not bars or len(bars) < 20:
+            continue
+        
+        # 2. 滚动评估：每周评分一次
+        window_start = 20  # need at least 20 bars for indicators
+        while window_start < len(bars):
+            window = bars[:window_start]
+            score_result = compute_single_score(window)
+            if score_result is None:
+                window_start += 5
+                continue
+            total_score, score_components = score_result
+            results['scored_stocks'] += 1
+            
+            last = window[-1]
+            close = v_close = last.get('close', 0) or 0
+            ma20 = last.get('ma20', 0) or 0
+            
+            # 3. 买入条件：总评分 >= 45 且 momentum >= 8
+            buy_signal = total_score >= 45 and score_components['momentum'] >= 8
+            
+            if buy_signal:
+                results['buy_signals'] += 1
+                entry_price = close
+                atr_val = last.get('atr', 0) or 0
+                if atr_val > 0:
+                    target_price = entry_price * 1.10
+                    stop_price = entry_price - atr_val * 2.0
+                else:
+                    target_price = entry_price * 1.10
+                    stop_price = entry_price * 0.93
+                
+                kelly = compute_kelly(total_score, 0.10, (entry_price - stop_price) / entry_price)
+                
+                # 4. 模拟未来走势
+                future = bars[window_start:min(window_start+20, len(bars))]
+                exit_price = None
+                exit_reason = None
+                max_high = entry_price
+                min_low = entry_price
+                
+                for fb in future:
+                    fb_high = fb.get('high', 0) or 0
+                    fb_low = fb.get('low', 0) or 0
+                    fb_close = fb.get('close', 0) or 0
+                    max_high = max(max_high, fb_high)
+                    min_low = min(min_low, fb_low)
+                    
+                    if fb_high >= target_price:
+                        exit_price = target_price
+                        exit_reason = 'target'
+                        break
+                    elif fb_low <= stop_price:
+                        exit_price = fb_close
+                        exit_reason = 'stop'
+                        break
+                
+                if exit_price is None:
+                    if future:
+                        exit_price = future[-1].get('close', entry_price)
+                    else:
+                        exit_price = entry_price
+                    exit_reason = 'keep'
+                
+                hold_days = 0
+                if exit_reason and exit_reason != 'keep' and future:
+                    for idx, fb in enumerate(future):
+                        fb_high = fb.get('high', 0) or 0
+                        fb_low = fb.get('low', 0) or 0
+                        if (exit_reason == 'target' and fb_high >= target_price) or \
+                           (exit_reason == 'stop' and fb_low <= stop_price):
+                            hold_days = idx + 1
+                            break
+                elif exit_reason == 'keep':
+                    hold_days = len(future)
+                
+                profit_pct = (exit_price - entry_price) / entry_price * 100 if entry_price > 0 else 0
+                
+                trades.append({
+                    'code': code,
+                    'name': name,
+                    'entry_date': last.get('date'),
+                    'entry_price': round(entry_price, 2),
+                    'exit_price': round(exit_price, 2),
+                    'profit_pct': round(profit_pct, 2),
+                    'exit_reason': exit_reason,
+                    'hold_days': hold_days,
+                    'score': total_score,
+                    'kelly': round(kelly, 3),
+                    'stop_loss': round(stop_price, 2),
+                    'target': round(target_price, 2),
+                    'max_high': round(max_high, 2),
+                    'min_low': round(min_low, 2)
+                })
+            
+            window_start += 5  # 每 5 个交易日评估一次
+    
+    conn.close()
+    
+    # 5. 汇总
+    if trades:
+        profits = [t['profit_pct'] for t in trades]
+        wins = [t for t in trades if t['profit_pct'] > 0]
+        losses = [t for t in trades if t['profit_pct'] <= 0]
+        
+        win_rate = len(wins) / len(trades) * 100 if trades else 0
+        avg_profit = sum(profits) / len(profits) if profits else 0
+        avg_win = sum(t['profit_pct'] for t in wins) / len(wins) if wins else 0
+        avg_loss = sum(t['profit_pct'] for t in losses) / len(losses) if losses else 0
+        
+        mean_ret = avg_profit / 100
+        if len(profits) > 1:
+            variance = sum((p/100 - mean_ret)**2 for p in profits) / (len(profits) - 1)
+            std_ret = math.sqrt(variance)
+        else:
+            std_ret = 0
+        sharpe = mean_ret / std_ret * math.sqrt(252) if std_ret > 0 else 0
+        
+        # 最大回撤：用实际资本曲线计算
+        capital_curve = [capital]
+        for t in trades:
+            new_cap = capital_curve[-1] * (1 + t['profit_pct'] / 100)
+            capital_curve.append(new_cap)
+        peak_cap = capital
+        max_dd = 0
+        for c in capital_curve:
+            peak_cap = max(peak_cap, c)
+            dd = (peak_cap - c) / peak_cap * 100
+            max_dd = max(max_dd, dd)
+        
+        results['trades'] = sorted(trades, key=lambda x: abs(x['profit_pct']), reverse=True)[:50]
+        results['summary'] = {
+            'total_trades': len(trades),
+            'win_rate': round(win_rate, 1),
+            'avg_profit_pct': round(avg_profit, 2),
+            'avg_win_pct': round(avg_win, 2),
+            'avg_loss_pct': round(avg_loss, 2),
+            'sharpe_ratio': round(sharpe, 2),
+            'max_drawdown_pct': round(max_dd, 2),
+            'profit_factor': round(abs(avg_win / avg_loss), 2) if avg_loss != 0 else float('inf'),
+            'wins': len(wins),
+            'losses': len(losses),
+            'capital_end': round(capital * (1 + avg_profit/100), 2)
+        }
+    
+    return results
+
+
+if __name__ == '__main__':
+    end = datetime.now().strftime('%Y-%m-%d')
+    start = (datetime.now() - timedelta(days=180)).strftime('%Y-%m-%d')
+    r = run_strategy_research(start, end)
+    print(json.dumps(r, indent=2, ensure_ascii=False))