BQuant — Quantitative Research Toolkit

BQuant is a toolkit for quantitative research of financial markets. Its core is a universal zone-analysis pipeline: it is not tied to any particular indicator and works with any oscillator.

• Documentation: https://bquant.readthedocs.io/
• Source: https://github.com/kogriv/bquant

Install

pip install bquant

Python 3.12+.

Quick start

from bquant.data.samples import get_sample_data
from bquant.analysis.zones import analyze_zones

data = get_sample_data('tv_xauusd_1h')

# The pipeline is indicator-agnostic: swap the `.with_indicator()` call,
# everything else stays the same.
result = (
    analyze_zones(data)
    .with_indicator('pandas_ta', 'rsi', length=14)
    .detect_zones('threshold', indicator_role='value',
                  upper_threshold=70, lower_threshold=30)
    .analyze(clustering=True)
    .build()
)

print(len(result.zones))                              # 64
print(sorted({zone.type for zone in result.zones}))   # ['neutral', 'overbought', 'oversold']

A zone is addressed by role (indicator_role='value'), not by column name. Column names depend on the library and on the call parameters and change with them; roles do not.

MACD zones in one line — a preset over the same pipeline:

from bquant.analysis.zones import analyze_macd_zones
from bquant.data.samples import get_sample_data

result = analyze_macd_zones(get_sample_data('tv_xauusd_1h'))

print(len(result.zones))                              # 30
print(sorted({zone.type for zone in result.zones}))   # ['bear', 'bull']

Note that the zone vocabulary follows the indicator: an oscillator crossing zero yields bull/bear, a bounded one yields overbought/neutral/oversold.

What is in the box

Zone analysis. Five detection strategies (zero_crossing, threshold, line_crossing, preloaded, combined) and five metric families (swing, shape, divergence, volatility, volume), plus hypothesis testing and clustering over the resulting zones.

Indicators. Built-in implementations (SMA, EMA, RSI, MACD, Bollinger Bands) and anything from pandas-ta or TA-Lib through one factory:

from bquant.data.samples import get_sample_data
from bquant.indicators import LibraryManager

data = get_sample_data('tv_xauusd_1h')

LibraryManager.load_all_libraries()
rsi = LibraryManager.create_indicator('pandas_ta', 'rsi', length=14)

print(rsi.calculate(data).data.columns.tolist())   # ['RSI_14']

Data. OHLCV loading, processing and validation, with sample datasets embedded in the package so that every example runs without external files.

Visualization. Interactive financial charts and statistical plots (Plotly, Matplotlib).

Performance. Vectorized computation and a two-level cache (memory + disk).

Command line

bquant list                                  # available sample datasets
bquant analyze tv_xauusd_1h                  # zones, MACD by default
bquant analyze tv_xauusd_1h --indicator rsi  # any supported oscillator
bquant analyze --json --no-chart             # structured output, for programs
bquant analyze mt_xauusd_m15 -o chart.html   # save the chart

Every flag and the JSON schema: CLI guide.

Documentation

Development

git clone https://github.com/kogriv/bquant.git
cd bquant
python -m venv .venv && source .venv/bin/activate   # Windows: .venv\Scripts\activate
pip install -e ".[dev]"
pytest

Extras: dev, docs, notebooks, research, and full (= dev + notebooks).

Repository layout: bquant/ is the package itself; tests/, docs/ and examples/ support it; research/ and scripts/ hold notebook-style studies and automation.

Status

Beta. The public API changes between releases without deprecation windows — renames are carried through in one change. Pin an exact version: bquant>=0.0.18,<0.0.19.

Breaking changes ship in the patch number, and the dangerous ones do not raise: when a defect in a calculation is fixed, nothing fails to import — your numbers change, and the old ones were wrong. Every breaking change is marked in CHANGELOG.md on a line naming its replacement, so one command lists them:

grep -nE "Ломает:|\(breaking\)" CHANGELOG.md

The full policy — what counts as breaking here, how to move a pin, what the project does not promise — is docs/versioning.md.

Not in the package: machine learning. The bquant.ml placeholder was removed in 0.0.7 because both of its public functions only ever raised NotImplementedError.

License

MIT — see LICENSE.

Contact

Author: kogriv · kogriv@gmail.com