qlib/backend/app/domain/entities/research.py

"""研究领域对象：Research Specification、标准化研究结果。

原则（AGENT.md §16/§21/§24、ARCHITECTURE §14）：
- 前端 / Agent / 后端统一经 Research Specification 描述任务，禁止直接拼引擎配置
- 回测结果一律标准化为 BacktestResult；未建模的成本/市场约束显式列在
  unimplemented，禁止默认「无成本 / 永远可成交」假设
"""

from __future__ import annotations

from datetime import date

from pydantic import BaseModel, Field, field_validator, model_validator

# ---------- Research Specification ----------


class UniverseSpec(BaseModel):
    """股票池口径。MVP：市场 + 过滤条件；指数成分等 Phase 3 扩展。"""

    market: str = Field(default="CN_A", description="CN_A / CN_B / ...")
    exclude_st: bool = True
    exclude_suspended: bool = True
    min_listing_days: int = Field(default=250, ge=0, description="上市至少 N 个自然日")


class FactorSpec(BaseModel):
    """引用一个已注册因子并给定权重。"""

    name: str
    weight: float = Field(default=1.0, gt=0)


class SelectionSpec(BaseModel):
    """选股方式。MVP：按加权因子得分取 Top N 等权。"""

    top_n: int = Field(default=30, ge=1, le=1000)


class CostSpec(BaseModel):
    """交易成本模型（单边比例）。

    buy = commission + slippage；sell = commission + stamp_tax + slippage。
    """

    commission_rate: float = Field(default=0.0003, ge=0, le=0.01)
    stamp_tax_rate: float = Field(default=0.0005, ge=0, le=0.01)
    slippage_rate: float = Field(default=0.001, ge=0, le=0.05)
    benchmark: str = Field(default="000300.SH", description="对照基准指数代码")


class ResearchSpec(BaseModel):
    """一次研究的完整描述。type 决定执行路径。"""

    type: str = Field(default="backtest", pattern="^(factor_test|backtest)$")
    universe: UniverseSpec = UniverseSpec()
    factors: list[FactorSpec] = Field(min_length=1)
    selection: SelectionSpec = SelectionSpec()
    rebalance: str = Field(default="monthly", pattern="^(weekly|monthly)$")
    period: tuple[date, date]
    costs: CostSpec = CostSpec()
    initial_capital: float = Field(default=1_000_000.0, gt=0)

    @field_validator("period")
    @classmethod
    def _period_ordered(cls, period: tuple[date, date]) -> tuple[date, date]:
        if period[0] >= period[1]:
            raise ValueError("period 必须满足 start < end")
        return period

    @model_validator(mode="after")
    def _no_duplicate_factors(self) -> ResearchSpec:
        names = [f.name for f in self.factors]
        if len(set(names)) != len(names):
            raise ValueError("factors 存在重复因子名")
        return self


# ---------- 回测结果 ----------


class CurvePoint(BaseModel):
    date: date
    value: float


class MonthlyReturn(BaseModel):
    year: int
    month: int
    return_pct: float  # 百分数，如 3.2 表示 +3.2%


class YearlyReturn(BaseModel):
    year: int
    return_pct: float


class BacktestSummary(BaseModel):
    start: date
    end: date
    initial_capital: float
    final_equity: float
    total_return_pct: float
    annual_return_pct: float
    sharpe: float
    max_drawdown_pct: float
    volatility_pct: float
    win_rate_pct: float
    total_trades: int
    avg_turnover_pct: float
    benchmark_return_pct: float | None = None


class Trade(BaseModel):
    entry_date: date
    exit_date: date
    symbol: str
    entry_price: float
    exit_price: float
    return_pct: float


class Position(BaseModel):
    date: date
    symbol: str
    weight: float


class BacktestResult(BaseModel):
    """标准化回测结果（ARCHITECTURE §14）。前端只依赖该结构。"""

    summary: BacktestSummary
    equity_curve: list[CurvePoint]
    drawdown: list[CurvePoint]
    monthly_returns: list[MonthlyReturn]
    yearly_returns: list[YearlyReturn]
    positions: list[Position]
    trades: list[Trade]
    turnover_pct: float
    unimplemented: list[str] = Field(
        default_factory=list,
        description="本结果中未建模的约束（AGENT §24：必须显式标注，禁止假装支持）",
    )
    config_snapshot: dict = Field(default_factory=dict, description="复现用完整配置快照")


# ---------- 因子测试结果 ----------


class QuantileReturn(BaseModel):
    """分层收益：按因子值升序分 N 层后各层等权组合的区间收益。"""

    quantile: int
    return_pct: float


class FactorTestReport(BaseModel):
    factor_name: str
    ic_mean: float
    icir: float
    rank_ic_mean: float
    positive_ratio_pct: float
    quantile_returns: list[QuantileReturn]
    spread_quantile: int | None = Field(
        default=None, description="分层价差 = 最高层收益 - 最低层收益（若多头/空头语义适用）"
    )
    sample_days: int
    unimplemented: list[str] = Field(default_factory=list)
    config_snapshot: dict = Field(default_factory=dict)