Statistical arbitrage

• ✓Runs in any market; steadier but smaller returns
• ⚠Spreads and funding can blow out in extreme conditions

Backtest curve is an example on a deterministic simulated price path — not real returns.

"""统计套利（Statistical arbitrage）—— 配对比价 z-score 均值回归。

思路：对一组相关资产（如 ETH 与其配对腿）计算比价序列的 z-score：主腿相对配对腿显著便宜（z ≤ -k）
做多主腿、显著偏贵（z ≥ +k）做空主腿，比价回归（|z| 收敛）后平仓。经典 pairs trading 的单腿化简：
配对腿仅作信号、不下单，避免单 connector 下的双腿撮合复杂度，同时保留统计套利的核心信号结构。
参数：window（z-score 窗口）、k（进出场偏离倍数）、size（目标数量）。

降级行为：feed 未提供配对行情（ctx.pair_symbol() 为 None）时保持空仓——拒绝退化成裸方向敞口；
连接器不支持做空时只做多腿（z ≤ -k 进、回归出）。
"""
from __future__ import annotations

from ..base import StrategyBase
from ..context import StrategyContext
from ..indicators import zscore
from ..registry import register


@register("统计套利")
class StatArb(StrategyBase):
    description = "配对比价 z-score 偏离入场、回归平仓（配对腿作信号，主腿交易）。"
    params = {"window": 40, "k": 1.5, "size": 4.0}

    def __init__(self, window: int = 40, k: float = 1.5, size: float = 4.0) -> None:
        self.window = int(window)
        self.k = float(k)
        self.size = float(size)

    async def on_tick(self, ctx: StrategyContext) -> None:
        sym = ctx.conn_symbol()
        pair = ctx.pair_symbol()
        if pair is None:
            return  # 无配对行情：不建仓
        n = self.window + 1
        a = await ctx.history(sym, n)
        b = await ctx.history(pair, n)
        m = min(len(a), len(b))
        if m < self.window:
            return
        ratio = [x / y for x, y in zip(a[-m:], b[-m:]) if y]
        z = zscore(ratio, self.window)
        if z is None:
            return
        allow_short = bool(getattr(ctx.conn, "allow_short", True))
        band = self.size * 0.1
        if z <= -self.k:
            await ctx.target(sym, self.size, band=band)       # 主腿便宜 → 做多
        elif z >= self.k:
            await ctx.target(sym, -self.size if allow_short else 0.0, band=band)
        elif abs(z) < 0.3:
            await ctx.target(sym, 0.0, band=band)             # 回归 → 平仓

The full SDK and all strategies are MIT-licensed open source — backtest, paper-trade, or fork them directly.