metrics.py

def max_drawdown(equity: pd.Series) -> float:
    peak = equity.cummax()
    dd = (equity / peak) - 1.0
    return float(dd.min())


def qs_metrics(equity: pd.Series) -> dict:
    """
    Quantstats-style metrics from an equity curve.
    """
    r = equity.pct_change().fillna(0.0)
    n = len(r)
    years = n / 252.0

    cagr = float(equity.iloc[-1] ** (1.0 / years) - 1.0) if years > 0 else float("nan")
    ann_vol = float(r.std(ddof=0) * math.sqrt(252))
    ann_ret = float(r.mean() * 252)

    sharpe = float(ann_ret / (ann_vol + 1e-12))

    downside = r.copy()
    downside[downside > 0] = 0.0
    downside_dev = float(downside.std(ddof=0) * math.sqrt(252))
    sortino = float(ann_ret / (downside_dev + 1e-12))

    mdd = max_drawdown(equity)
    calmar = float(cagr / (abs(mdd) + 1e-12))
    win_rate = float((r > 0).mean())

    return {
        "CAGR": cagr,
        "AnnVol": ann_vol,
        "Sharpe": sharpe,
        "Sortino": sortino,
        "Calmar": calmar,
        "MaxDD": mdd,
        "WinRate": win_rate,
    }


def print_metrics_table(non_m: dict, llm_m: dict | None) -> None:
    rows = {"Non-LLM": non_m}
    if llm_m is not None:
        rows["LLM"] = llm_m
    df = pd.DataFrame(rows).T
    for c in ["CAGR", "AnnVol", "MaxDD", "WinRate"]:
        df[c] = df[c] * 100.0
    df = df[["CAGR", "AnnVol", "Sharpe", "Sortino", "Calmar", "MaxDD", "WinRate"]]
    print("\nQuantstats-style metrics (OOS 2023+):")
    print(df.to_string(float_format=lambda x: f"{x:,.3f}"))