quantitative-stationarity-toolbox

A Python feature engineering pipeline to transform non-stationary financial time-series data into model-ready features for machine learning algorithms.

Quantitative Stationarity & Feature Engineering Toolbox

Machine Learning | Time-Series Preprocessing | Python

🧠 Bridging Financial Data & Machine Learning

Standard machine learning algorithms (like XGBoost, Random Forests, and LSTMs) assume that input data is independent and identically distributed (IID). Financial time-series data naturally violates this assumption due to non-stationarity, shifting means, and volatility clustering.

This repository serves as an Advanced Preprocessing and Feature Engineering Pipeline. It automates the extraction of stationary features from raw financial data so they can be safely ingested by predictive ML models without causing data leakage or spurious regressions.

⚙️ Core Pipeline Components

• Automated Stationarity Testing: Pipeline integration for Augmented Dickey-Fuller (ADF) and KPSS tests to verify the integration order of assets.
• Memory-Preserving Transformations: Implementation of fractional differencing to stabilize variance while retaining long-term signal momentum.
• Volatility Scaling: Automated rolling $z$-score normalizations and Log-Return conversions for heteroskedastic datasets.
• Walk-Forward Validation: Custom cross-validation iterators designed strictly for time-series, ensuring zero look-ahead bias during model training.

📐 Mathematical Foundation

To ensure data is model-ready, the pipeline tests for the presence of a unit root using the ADF test: $$\Delta y_t=\alpha+\beta t+\gamma y_{t-1}+\sum_{j=1}^{p}\delta_j\Delta y_{t-j}+\varepsilon_t$$

To achieve stationarity without losing market memory (which often happens with standard integer differencing), we apply fractional differencing using the binomial series expansion: $$(1-B)^d=\sum_{k=0}^{\infty}\binom{d}{k}(-B)^k$$

🚀 Quick Start (Example Usage)

from quant_toolbox.adf_tests import check_stationarity
from quant_toolbox.differencing import fractional_difference
import pandas as pd

# 1. Load raw, non-stationary financial data
df = pd.read_csv('data/asset_prices.csv')

# 2. Check initial stationarity for ML ingestion
is_stationary, p_value = check_stationarity(df['Close'])

# 3. Apply fractional differencing to create a stationary, model-ready feature
if not is_stationary:
    df['Stationary_Feature'] = fractional_difference(df['Close'], d=0.45)