Stationarity Tests

Feeding non-stationary features (raw prices, trending volume) into ML models produces spurious correlations that vanish out of sample. Test stationarity first, transform if needed.

The Problem

Raw asset prices are non-stationary: their mean and variance change over time. A regression of one non-stationary series on another can produce high R-squared and significant t-statistics even when there is no causal relationship (spurious regression). In ML4T, this means a model trained on raw prices will appear to predict well in-sample but fail live. Returns and most financial ratios are stationary. The gap between "looks predictive" and "is predictive" often comes down to stationarity.

The Pattern

WRONG

import numpy as np
from sklearn.linear_model import Ridge