wbt Python Package

Python API for the wbt Rust backtesting engine.

中文文档

Development Objectives

This Python subproject aims to provide a practical research-facing interface for weight-based backtesting while keeping the heavy computation in Rust.

Design priorities:

1. Keep data input flexible for common research formats.
2. Return analysis-friendly outputs as pandas objects.
3. Preserve one consistent metric schema across stats outputs.
4. Provide plotting utilities that work directly on backtest outputs.

Project Layout

This directory is an independent Python subproject.

python/
|-- pyproject.toml
|-- README.md
|-- scripts/
|-- tests/
`-- wbt/

The Rust crate remains one level up at ../Cargo.toml. maturin builds the extension module from there.

Installation And Local Setup

Requirements:

• Rust toolchain
• Python 3.10+
• uv

Setup:

cd python
uv sync --extra dev
uv run maturin develop --release

Quick Start

import pandas as pd
from wbt import WeightBacktest

df = pd.DataFrame(
    {
        "dt": [
            "2024-01-02 09:01:00",
            "2024-01-02 09:02:00",
            "2024-01-02 09:03:00",
            "2024-01-02 09:04:00",
        ],
        "symbol": ["AAPL", "AAPL", "AAPL", "AAPL"],
        "weight": [0.5, 0.2, 0.0, -0.3],
        "price": [185.0, 186.0, 186.5, 184.5],
    }
)

wb = WeightBacktest(
    df,
    digits=2,
    fee_rate=0.0002,
    n_jobs=4,
    weight_type="ts",  # "ts" or "cs"
    yearly_days=252,
)

print("all:", wb.stats)
print("long:", wb.long_stats)
print("short:", wb.short_stats)

print(wb.daily_return.head())
print(wb.dailys.head())
print(wb.pairs.head())

print(wb.segment_stats("2024-01-01", "2024-12-31", kind="多空"))
print(wb.long_alpha_stats)

Accepted Inputs

The data argument accepts:

• pandas.DataFrame
• polars.DataFrame
• polars.LazyFrame
• file path as str or Path

Supported file formats from path input:

• csv
• parquet
• feather
• arrow

Required columns:

Notes:

• Null values are not allowed.
• Weight normalization is performed once by the Rust engine using digits and half-away-from-zero rounding. The first BAR of each symbol is excluded from return rows and fees; later price returns and same-BAR position-change costs belong to that BAR's date.

Main API Surface

Top-level imports (all reachable from import wbt):

from wbt import (
    # Backtest engine
    WeightBacktest,
    backtest,
    # Performance metrics (Rust-backed)
    daily_performance,
    top_drawdowns,
    rolling_daily_performance,
    cal_yearly_days,
    # Strategy utilities (pure Python)
    weights_simple_ensemble,
    cal_trade_price,
    log_strategy_info,
    # Reporting
    generate_backtest_report,
    # Test data
    mock_symbol_kline,
    mock_weights,
)

Primary class and helpers:

• WeightBacktest(...): main backtest engine entry.
• backtest(...): convenience wrapper returning a WeightBacktest.
• daily_performance(returns, yearly_days=252): standalone metric utility on a daily-return array.
• top_drawdowns(returns, top=10): top-N drawdown windows.
• rolling_daily_performance(df, ret_col, window=252, min_periods=100, yearly_days=None): rolling-window daily performance.
• cal_yearly_days(dts): infer yearly trading-day count from a date series.
• weights_simple_ensemble(df, weight_cols, method="mean", only_long=False, **kwargs): ensemble multiple strategy weights (mean / vote / sum_clip).
• cal_trade_price(df, digits=None, windows=(5, 10, 15, 20, 30, 60)): TWAP / VWAP and next-bar trade-price table grouped by symbol.
• log_strategy_info(strategy, df): pretty-print per-symbol weight summaries via loguru.
• generate_backtest_report(wb, output_path): render a self-contained HTML report.
• mock_symbol_kline(...) / mock_weights(...): generators for quick experiments.

Core WeightBacktest properties and methods:

• stats, long_stats, short_stats
• daily_return, long_daily_return, short_daily_return
• dailys, pairs
• alpha, alpha_stats, bench_stats
• segment_stats(sdt, edt, kind)
• long_alpha_stats
• get_symbol_daily(symbol), get_symbol_pairs(symbol)

Logging Note

cal_yearly_days and rolling_daily_performance emit warnings from Rust (e.g. short-span fallback) via the log crate. The package initializes pyo3-log at module load, so those warnings show up through Python's standard logging. If you use loguru, install an InterceptHandler once to route them into your loguru sinks.

Plotting Utilities

All plotting functions are single-purpose figures that consume a BacktestResult (from wb.to_result()) with zero data transformation — each field maps straight to a plotly trace. There are no composite (subplot) charts; the HTML report composes single figures into a CSS grid instead.

from wbt.plotting import (
    plot_colored_table,  # stats as a colored table
    plot_cumulative_returns,  # cumulative curves (voladj=True for vol-normalized)
    plot_daily_return_dist,  # daily-return histogram
    plot_drawdown,  # drawdown + cumulative (dual-axis single figure)
    plot_drawdowns_table,  # top-drawdowns detail table
    plot_key_trades,  # yearly best/worst key trades
    plot_monthly_heatmap,  # monthly-return heatmap
    plot_pairs_hold_dist,  # holding-bars distribution by direction
    plot_pairs_pnl_dist,  # pnl-ratio distribution by direction
    plot_rolling_metrics,  # rolling sharpe/return/vol over time (252d window)
    plot_segment_comparison,  # recent-1y vs full-sample metric table
    plot_stats_comparison,  # 多空/多头/空头/基准/超额 metric comparison table
    plot_symbol_returns,  # per-symbol cumulative returns
    plot_verdict,  # history (yearly) + recent-window verdict
    plot_yearly_returns,  # yearly absolute vs excess returns (grouped bars)
)

Typical usage:

result = wb.to_result()

fig1 = plot_cumulative_returns(result, keys=["多空", "多头", "空头", "基准"])
fig2 = plot_cumulative_returns(result, keys=["多空", "多头", "空头", "基准", "多头超额", "空头超额"], voladj=True)
fig3 = plot_drawdown(result)
fig4 = plot_pairs_pnl_dist(result)

# Optional HTML export
html = plot_cumulative_returns(result, to_html=True)

# Full HTML report file (composes single figures into a tabbed CSS grid)
generate_backtest_report(df, "report.html")

Quality And Testing

Run checks from python/:

uv run pytest -v
uv run ruff format --check .
uv run ruff check . --no-fix
uv run basedpyright

Architecture Snapshot

repo-root/
|-- Cargo.toml
|-- src/
|   |-- lib.rs                       # pure Rust crate entry; Python bindings are off by default
|   |-- python.rs                    # PyO3 bindings (enabled by the python feature)
|   `-- core/
|       |-- cal_yearly_days.rs       # Rust core for cal_yearly_days
|       |-- daily_performance.rs
|       |-- rolling_daily_performance.rs
|       |-- top_drawdowns.rs
|       `-- ...                      # backtest engine internals
`-- python/
    `-- wbt/
        |-- __init__.py              # top-level exports
        |-- _df_convert.py           # pandas <-> Arrow IPC helpers
        |-- _wbt.pyi                 # Rust extension stubs
        |-- backtest.py              # WeightBacktest class
        |-- mock.py                  # mock_symbol_kline / mock_weights
        |-- top_drawdowns.py         # adapter for _wbt.top_drawdowns
        |-- utils/                   # adapters + pure-Python utilities
        |   |-- __init__.py
        |   |-- cal_yearly_days.py
        |   |-- rolling_daily_performance.py
        |   |-- weights_simple_ensemble.py
        |   |-- cal_trade_price.py
        |   `-- log_strategy_info.py
        |-- plotting/                # single-purpose plotly charts
        |   |-- __init__.py
        |   |-- _common.py
        |   |-- returns.py
        |   |-- risk.py
        |   |-- trades.py
        |   `-- overview.py
        `-- report/                  # HTML report + composite charts
            |-- __init__.py
            |-- _generator.py
            |-- _plot_backtest.py
            `-- html_builder.py

License

MIT