HKUDS/Vibe-Trading · error · RuntimeError

{alpha_id}: IC series empty — insufficient overlap between f

Error message

{alpha_id}: IC series empty — insufficient overlap between factor and returns

What it means

After computing the IC series for an alpha, _bench_one_alpha requires a non-empty result. Empty means the factor DataFrame and the forward-return DataFrame share no overlapping timestamps (or symbols), so no rank correlation can be computed.

Source

Thrown at agent/src/tools/alpha_bench_tool.py:693

        raise ValueError("panel missing 'close' — cannot derive forward returns")
    # Next-period return aligned to current row (use t+1 close, shift back).
    fwd = close.pct_change(fill_method=None).shift(-1)
    return fwd


def _bench_one_alpha(
    registry: Any,
    alpha_id: str,
    panel: dict[str, pd.DataFrame],
    return_df: pd.DataFrame,
) -> dict[str, Any]:
    """Compute IC stats for one alpha. Returns a dict, may raise SkipAlpha / RegistryError."""
    from src.factors.factor_analysis_core import compute_ic_series  # local import

    factor_df = registry.compute(alpha_id, panel)
    ic_series = compute_ic_series(factor_df, return_df)
    if ic_series.empty:
        raise RuntimeError(
            f"{alpha_id}: IC series empty — insufficient overlap between factor and returns"
        )
    ic_mean = float(ic_series.mean())
    ic_std = float(ic_series.std())
    ir = ic_mean / ic_std if ic_std > 0 else 0.0
    ic_pos = float((ic_series > 0).mean())
    alpha = registry.get(alpha_id)
    meta = alpha.meta or {}
    return {
        "id": alpha_id,
        "zoo": alpha.zoo,
        "theme": meta.get("theme", []),
        "formula_latex": meta.get("formula_latex", ""),
        "ic_mean": round(ic_mean, 6),
        "ic_std": round(ic_std, 6),
        "ir": round(ir, 4),
        "ic_positive_ratio": round(ic_pos, 4),
        "ic_count": int(len(ic_series)),

View on GitHub (pinned to 80ffdda44c)

Solutions

1. Reindex/align factor_df and return_df to a common trading calendar before benching
2. Extend the period so factor and return dates overlap by at least a few bars
3. Verify the alpha's compute() output is not empty or all-NaN for the requested panel

common = factor_df.index.intersection(return_df.index)
assert len(common) >= 5, f'only {len(common)} overlapping dates'

def has_overlap(factor_df, return_df, min_bars: int = 5) -> bool:
    return len(factor_df.index.intersection(return_df.index)) >= min_bars

try:
    _bench_one_alpha(...)
except RuntimeError as e:
    if 'IC series empty' in str(e):
        reindex_factor_to_calendar(); retry or skip alpha

When it happens

Trigger: Factor timestamps that don't align with return timestamps (different calendars/timezones); a factor computed only on dates after the returns window ends; all-NaN factor columns causing row-wise drops in compute_ic_series.

Common situations: Factors built on a different trading calendar (e.g. crypto 24/7 vs A-share calendar); timezone misalignment between factor and close panels; too-short windows where next-bar shifting removes all overlap.

Related errors

• var_backtest needs at least 2 aligned observations, got {ret
• period must be string, got {type(period).__name__}
• period {period!r} must be YYYY-YYYY or YYYY-MM-DD/YYYY-MM-DD
• start_date ({start}) > end_date ({end})
• universe {universe!r} not recognized; expected one of {sorte

AI-assisted analysis of HKUDS/Vibe-Trading@80ffdda44c (2026-08-28). Data as JSON: /api/errors/1c243388b1651770. Report an issue: GitHub.