Probability mean-reversion

• ✓When your view of the odds differs from the market's pricing
• ⚠Liquidity thins near resolution; binary outcomes can go to zero

Binary prediction-market simulation (YES/NO, settles YES after 240 steps) — not a promise of future returns.

"""概率均值回归（Probability mean-reversion）—— logit 空间的 z-score 反向下注。

思路：事件概率（YES 价格）常在短期过度反应。把 YES 价格映射到 logit（log-odds）空间计算滚动 z-score：
概率被显著高估（z ≥ +k）时买入 NO（等价于看空 YES，因 token 不能做空）、被显著低估（z ≤ -k）时买入
YES，回归（|z| 收敛）后两腿平仓。临近结算的窗口内停止加仓——此时结算跳变主导，均值回归假设失效。

适用性：instrument="event"、venues=["POLYMARKET"]。做空由"买对侧 token"实现。
"""
from __future__ import annotations

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


@register("概率均值回归")
class PmProbMeanRev(StrategyBase):
    description = "事件概率在 logit 空间偏离滚动均值时反向下注（高估买 NO、低估买 YES），回归平仓。"
    params = {"window": 30, "k": 1.2, "size": 15.0}
    venues = ["POLYMARKET"]
    symbols: list = []
    instrument = "event"

    def __init__(self, window: int = 30, k: float = 1.2, size: float = 15.0) -> 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()
        no = ctx.complement_token(sym)
        hist = await ctx.history(sym, self.window + 1)
        if len(hist) < self.window:
            return
        z = zscore([logit(p) for p in hist], self.window)
        if z is None:
            return
        band = self.size * 0.15

        # 临近结算：停止均值回归，平掉两腿（结算跳变主导）
        ttr = ctx.time_to_resolution()
        if ttr is not None and ttr < self.window:
            await ctx.target(sym, 0.0, band=band)
            if no is not None:
                await ctx.target(no, 0.0, band=band)
            return

        if z >= self.k:  # 概率高估 → 买 NO 看空
            await ctx.target(sym, 0.0, band=band)
            if no is not None:
                await ctx.target(no, self.size, band=band)
        elif z <= -self.k:  # 概率低估 → 买 YES
            if no is not None:
                await ctx.target(no, 0.0, band=band)
            await ctx.target(sym, self.size, band=band)
        elif abs(z) < 0.3:  # 回归 → 平两腿
            await ctx.target(sym, 0.0, band=band)
            if no is not None:
                await ctx.target(no, 0.0, band=band)

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