bagelquant-bt

bagelquant-bt provides typed signal composition, evaluation, portfolio-policy application, performance metrics, and Plotly visualization.

The public investment flow is:

AlphaValue Panel -> SignalComposer -> SignalPanel -> SignalDatePolicy
-> ScheduledSignal -> ExecutionPolicy -> PortfolioPolicy -> Weights Panel

Public backtests require a core SignalPanel. Ordinary panels, raw DataFrames, and direct weights cannot enter the backtest API. Prices remain explicit long-form Polars data with time, asset_id, and price.

from bagelquant_core import IdentitySignalComposer, Panel
from bagelquant_bt import (
    BacktestConfig,
    MissingSnapshotAction,
    SignalAnchor,
    SignalDatePolicy,
    compose_signal,
    run_signal_backtest,
)

alpha_value = Panel.from_domain(alpha_frame, domain, name="quality")
policy = SignalDatePolicy(
    id="month_end",
    frequency="monthly",
    anchor=SignalAnchor.LAST_TRADING_DAY,
    missing_snapshot=MissingSnapshotAction.PREVIOUS_IN_PERIOD,
)
signal = compose_signal(
    {"quality": alpha_value},
    IdentitySignalComposer(),
    calendar,
    policy,
)
result = run_signal_backtest(
    signal,
    prices,
    calendar,
    policy,
    config=BacktestConfig(initial_capital=1_000_000),
)

run_signal_evaluation computes execution-to-execution IC, quantiles, spread, TOP N, lag, and IC-decay diagnostics from ScheduledSignal. Sparse or monthly signals rebalance only on their snapshot dates; positions are held between snapshots. OrderSizingPolicy reserves the later target-weight-to-order boundary, but quantity sizing is not implemented yet.

During an asset-specific price gap, its holding is frozen at the last observed price. The gap sessions have zero return and the cumulative move is recognized when a new price appears.

Development

uv run ruff check .
uv run python -m pytest