optz

labs/analysis/etf/analysis.py

@@ -0,0 +1,376 @@
+"""ETF 网格策略回测分析：轮次统计、资金占用、逐月分布、参数敏感性。
+
+直接调用 ``backtest.py`` 的模拟内核，不复制策略逻辑。
+
+用法：
+    py -3.14 -B analysis/etf/analysis.py
+"""
+
+from dataclasses import dataclass, replace
+from datetime import date, datetime
+import json
+import math
+from pathlib import Path
+import statistics
+import sys
+
+HERE = Path(__file__).resolve().parent
+sys.path.insert(0, str(HERE))
+
+from backtest import (  # noqa: E402
+    CACHE, OUT, REPO_DEFAULTS, START_CASH, SYMBOLS, SYMBOL_PARAMS, EtfDefaults,
+    EtfSymbolConfig, analyze, fetch_daily, simulate,
+)
+
+
+@dataclass(slots=True)
+class RoundTrip:
+    """一次"建网 → 整仓清空"的完整轮次。"""
+
+    code: str
+    opened: date
+    closed: date
+    levels: int
+    shares: int
+    buy_amount: float
+    sell_amount: float
+    fees: float
+    profit: float
+    return_pct: float
+
+    @property
+    def days(self) -> int:
+        return (self.closed - self.opened).days
+
+
+def round_trips(fills) -> list[RoundTrip]:
+    """把逐笔成交切成完整轮次（按建仓日到清仓日配对）。"""
+    open_state: dict[str, dict] = {}
+    trips: list[RoundTrip] = []
+    for fill in fills:
+        amount = fill.price * fill.volume
+        if fill.side == "BUY":
+            state = open_state.setdefault(
+                fill.code,
+                dict(opened=fill.day, buys=0.0, sells=0.0, fees=0.0, volume=0,
+                     buys_volume=0, levels=0),
+            )
+            state["buys"] += amount
+            state["fees"] += fill.fee
+            state["volume"] += fill.volume
+            state["buys_volume"] += fill.volume
+            if fill.kind == "base":
+                state["opened"] = fill.day
+                state["levels"] = 1
+            else:
+                state["levels"] += 1
+        else:
+            state = open_state.get(fill.code)
+            if state is None:
+                continue
+            state["sells"] += amount
+            state["fees"] += fill.fee
+            state["volume"] -= fill.volume
+            if state["volume"] <= 0:
+                profit = state["sells"] - state["buys"] - state["fees"]
+                trips.append(
+                    RoundTrip(
+                        code=fill.code,
+                        opened=state["opened"],
+                        closed=fill.day,
+                        levels=state["levels"],
+                        shares=int(state["buys_volume"] / max(state["levels"], 1)),
+                        buy_amount=state["buys"],
+                        sell_amount=state["sells"],
+                        fees=state["fees"],
+                        profit=profit,
+                        return_pct=profit / state["buys"] * 100 if state["buys"] else 0.0,
+                    )
+                )
+                open_state.pop(fill.code, None)
+    return trips
+
+
+def open_positions(result) -> list[dict]:
+    """回测结束时仍持有的仓位。"""
+    rows = []
+    for code, book in result["books"].items():
+        if book.volume <= 0:
+            continue
+        rows.append(
+            {
+                "code": code,
+                "volume": book.volume,
+                "avg_cost": book.avg_cost,
+                "anchor": book.anchor,
+                "max_level": book.max_level,
+            }
+        )
+    return rows
+
+
+def monthly(result) -> dict[str, float]:
+    """按自然月统计净现金流与期末权益变化。"""
+    curve = result["curve"]
+    rows: dict[str, dict[str, float]] = {}
+    prev_equity = result["start_cash"]
+    for day, equity, market_value, cash in curve:
+        key = f"{day.year}-{day.month:02d}"
+        row = rows.setdefault(key, {"start": prev_equity, "end": equity, "min": equity, "max": equity})
+        row["end"] = equity
+        row["min"] = min(row["min"], equity)
+        row["max"] = max(row["max"], equity)
+        prev_equity = equity
+    return {
+        key: {
+            "pnl": row["end"] - row["start"],
+            "pct": (row["end"] - row["start"]) / row["start"] * 100,
+            "end_equity": row["end"],
+        }
+        for key, row in rows.items()
+    }
+
+
+def exposure(result) -> dict:
+    """资金占用与在场时间。"""
+    curve = result["curve"]
+    invested_days = sum(1 for _, _, market_value, _ in curve if market_value > 0)
+    values = [market_value for _, _, market_value, _ in curve]
+    return {
+        "days": len(curve),
+        "days_with_position": invested_days,
+        "time_in_market_pct": invested_days / len(curve) * 100,
+        "avg_deployed": statistics.fmean(values),
+        "avg_util_pct": statistics.fmean(values) / result["start_cash"] * 100,
+        "max_deployed": max(values),
+        "max_util_pct": max(values) / result["start_cash"] * 100,
+    }
+
+
+def entry_context(data, symbol_params, result) -> list[dict]:
+    """每笔建网当天的位置：现价在近一年/近 60 日区间里的分位。"""
+    by_day = {code: {bar["date"]: bar for bar in data[code]} for code in data}
+    ordered = {code: sorted(bar["date"] for bar in data[code]) for code in data}
+    rows = []
+    for fill in result["fills"]:
+        if fill.kind != "base":
+            continue
+        stamp = fill.day.strftime("%Y%m%d")
+        dates = ordered[fill.code]
+        index = dates.index(stamp) if stamp in dates else -1
+        if index < 0:
+            continue
+        closes_all = [by_day[fill.code][d]["close"] for d in dates[: index + 1]]
+        window60 = closes_all[-60:]
+        price = fill.price
+        rows.append(
+            {
+                "code": fill.code,
+                "day": fill.day.isoformat(),
+                "price": price,
+                "pct_in_year": sum(1 for c in closes_all if c <= price) / len(closes_all) * 100,
+                "pct_in_60d": sum(1 for c in window60 if c <= price) / len(window60) * 100,
+            }
+        )
+    return rows
+
+
+def grid_span(result) -> list[dict]:
+    """每个标的的格距与阶梯跨度（对照 docs/etf.md §3.4）。"""
+    rows = []
+    for code, book in result["books"].items():
+        symbol = book.symbol
+        entries = [
+            (fill.day, fill.price)
+            for fill in result["fills"]
+            if fill.code == code and fill.kind == "base"
+        ]
+        rows.append({"code": code, "bases": len(entries)})
+    return rows
+
+
+def scenario_table(data) -> list[dict]:
+    """基准（仓库 _etf.yaml）+ 单变量敏感性。"""
+    base = REPO_DEFAULTS
+    runs: list[tuple[str, dict]] = []
+
+    runs.append(("基准（当前 _etf.yaml）", {}))
+    for value in (2.0, 4.0, 5.0):
+        runs.append((f"add_pct={value}", {"defaults": replace(base, add_pct=value)}))
+    for value in (0.5, 0.8, 1.5, 2.0):
+        runs.append((f"min_profit_pct={value}", {"defaults": replace(base, min_profit_pct=value)}))
+    for value in (10.0, 20.0, 30.0):
+        runs.append((f"channel_pct={value}", {"defaults": replace(base, channel_pct=value)}))
+    for value in (3, 5, 15):
+        runs.append((f"max_adds={value}", {"defaults": replace(base, max_adds=value)}))
+    for value in (0.0, 0.0003, 0.001):
+        runs.append((f"佣金率={value}", {
+            "defaults": replace(base, commission_rate=value),
+            "commission_rate": value,
+        }))
+    runs.append(("无副出口", {"secondary_exit": False}))
+    runs.append(("成交=反弹确认价（贴近实盘）", {"fill_mode": "bounce"}))
+    runs.append(("成交=当日收盘价（悲观）", {"fill_mode": "close"}))
+    runs.append(("无 T+1 限制（min_hold_days=0）", {"min_hold_days": 0}))
+    # 副出口可达性：inner_step 必须小于 min_profit_pct（见 REPORT §5.2）
+    for step in (0.2, 0.4):
+        runs.append((f"inner_step={step}（副出口可达）", {
+            "symbol_params": {code: {**SYMBOL_PARAMS[code], "inner_step": step} for code in SYMBOLS},
+        }))
+    # 仓位规模：逐标的每档股数同乘一个系数（保持 10 档容量）
+    for factor in (0.25, 0.5, 2.0):
+        runs.append((f"buy_shares×{factor}", {
+            "symbol_params": {
+                code: {**SYMBOL_PARAMS[code],
+                       "buy_shares": max(100, int(SYMBOL_PARAMS[code]["buy_shares"] * factor)),
+                       "max_shares": max(1000, int(SYMBOL_PARAMS[code]["max_shares"] * factor))}
+                for code in SYMBOLS
+            },
+        }))
+    # ATR 倍数整体缩放（逐标的同乘）：只影响格距与跨度，不影响任何触发价位
+    for factor in (0.5, 2.0):
+        runs.append((f"atr_multiplier×{factor}（仅格距）", {
+            "symbol_params": {
+                code: {**SYMBOL_PARAMS[code],
+                       "atr_multiplier": SYMBOL_PARAMS[code]["atr_multiplier"] * factor}
+                for code in SYMBOLS
+            },
+        }))
+    runs.append(("channel_pct=1（贴近区间下沿）", {"defaults": replace(base, channel_pct=1.0)}))
+
+    rows = []
+    for label, kwargs in runs:
+        defaults = kwargs.pop("defaults", base)
+        result = simulate(data, defaults=defaults, **kwargs)
+        stats = analyze(result)
+        rows.append(
+            {
+                "label": label,
+                "net": stats["net"],
+                "equity_delta": stats["final_equity"] - result["start_cash"],
+                "return_pct": stats["return_pct"],
+                "max_dd_pct": stats["max_dd_pct"],
+                "bases": stats["buy_count"] - sum(
+                    1 for f in result["fills"] if f.kind == "add"
+                ),
+                "adds": sum(1 for f in result["fills"] if f.kind == "add"),
+                "exits": sum(1 for f in result["fills"] if f.kind == "exit"),
+                "levels": sum(1 for f in result["fills"] if f.kind == "level"),
+                "fees": stats["fees"],
+                "avg_util_pct": stats["avg_util_pct"],
+                "max_deployed": stats["max_deployed"],
+            }
+        )
+    return rows
+
+
+def main() -> int:
+    data = {code: fetch_daily(code) for code in SYMBOLS}
+    result = simulate(data)
+    stats = analyze(result)
+    stats["fee_pct_of_buy"] = stats["fees"] / stats["buy_amount"] * 100 if stats["buy_amount"] else 0.0
+    trips = round_trips(result["fills"])
+    expo = exposure(result)
+    months = monthly(result)
+    entries = entry_context(data, SYMBOL_PARAMS, result)
+    scenarios = scenario_table(data)
+
+    # 未实现盈亏：主出口只兑现盈利，亏损全部留在持仓里。
+    open_rows = []
+    for code, book in result["books"].items():
+        if book.volume <= 0:
+            continue
+        close = data[code][-1]["close"]
+        market = close * book.volume
+        cost = book.avg_cost * book.volume
+        open_rows.append(
+            {
+                "code": code,
+                "volume": book.volume,
+                "avg_cost": book.avg_cost,
+                "last_close": close,
+                "cost_amount": cost,
+                "market_amount": market,
+                "unrealized": market - cost,
+                "unrealized_pct": (close / book.avg_cost - 1) * 100,
+                "max_level": book.max_level,
+                "anchor": book.anchor,
+            }
+        )
+    unrealized = sum(row["unrealized"] for row in open_rows)
+    open_cost = sum(row["cost_amount"] for row in open_rows)
+    realized = stats["net"]
+    # 轮次口径：只统计"建网 → 整仓清空"的完整轮次，不含仍在持仓里的仓位。
+    all_in_net = realized - open_cost
+    per_symbol_trips = {}
+    for code in SYMBOLS:
+        rows = [t for t in trips if t.code == code]
+        book = result["books"][code]
+        held_cost = book.avg_cost * book.volume
+        per_symbol_trips[code] = {
+            "rounds": len(rows),
+            "wins": sum(1 for t in rows if t.profit > 0),
+            "net_realized_closed": sum(t.profit for t in rows),
+            "held_cost": held_cost,
+            "net_incl_open": sum(t.profit for t in rows) - held_cost,
+            "avg_days": statistics.fmean([t.days for t in rows]) if rows else 0.0,
+            "max_days": max([t.days for t in rows], default=0),
+            "worst": min([t.profit for t in rows], default=0.0),
+        }
+
+    report = {
+        "period": {
+            "first": result["curve"][0][0].isoformat(),
+            "last": result["curve"][-1][0].isoformat(),
+            "days": len(result["curve"]),
+        },
+        "base": stats,
+        "exposure": expo,
+        "pnl_bridge": {
+            "realized_net": realized,
+            "unrealized_net": unrealized,
+            "total": realized + unrealized,
+            "total_pct": (realized + unrealized) / result["start_cash"] * 100,
+            "open_positions": open_rows,
+        },
+        "per_symbol_rounds": per_symbol_trips,
+        "round_trips": {
+            "count": len(trips),
+            "wins": sum(1 for t in trips if t.profit > 0),
+            "losses": sum(1 for t in trips if t.profit <= 0),
+            "avg_days": statistics.fmean([t.days for t in trips]) if trips else 0.0,
+            "max_days": max([t.days for t in trips], default=0),
+            "avg_levels": statistics.fmean([t.levels for t in trips]) if trips else 0.0,
+            "max_levels": max([t.levels for t in trips], default=0),
+            "avg_profit": statistics.fmean([t.profit for t in trips]) if trips else 0.0,
+            "best": max([t.profit for t in trips], default=0.0),
+            "worst": min([t.profit for t in trips], default=0.0),
+            "detail": [
+                {
+                    "code": t.code,
+                    "opened": t.opened.isoformat(),
+                    "closed": t.closed.isoformat(),
+                    "days": t.days,
+                    "levels": t.levels,
+                    "buy": round(t.buy_amount, 2),
+                    "sell": round(t.sell_amount, 2),
+                    "profit": round(t.profit, 2),
+                    "return_pct": round(t.return_pct, 3),
+                }
+                for t in trips
+            ],
+        },
+        "open_positions": open_positions(result),
+        "monthly": months,
+        "entries": entries,
+        "scenarios": scenarios,
+    }
+    (OUT / "results.json").write_text(
+        json.dumps(report, ensure_ascii=False, indent=2), encoding="utf-8"
+    )
+    print(json.dumps(report, ensure_ascii=False, indent=2))
+    return 0
+
+
+if __name__ == "__main__":
+    raise SystemExit(main())