XAS Cross-Asset Signals

import numpy as np

def cross_asset_features(leader_returns, follower_returns, lags=[1, 4, 8, 24], window=60):
    """Build cross-asset features from a leader-follower pair."""
    n = len(follower_returns)
    features = {}
    
    for lag in lags:
        if lag >= n:
            continue
        
        # Lagged leader return as a feature
        lagged_leader = np.zeros(n)
        lagged_leader[lag:] = leader_returns[:-lag] if lag > 0 else leader_returns
        
        # Z-score the lagged feature using rolling window
        z_scored = np.zeros(n)
        for i in range(window + lag, n):
            window_vals = lagged_leader[i - window:i]
            mean = np.mean(window_vals)
            std = np.std(window_vals)
            z_scored[i] = (lagged_leader[i] - mean) / std if std > 0 else 0
        
        features[f'leader_lag_{lag}_zscore'] = z_scored
    
    # Rolling cross-correlation (stability monitor)
    rolling_corr = np.zeros(n)
    for i in range(window, n):
        x = leader_returns[i - window:i]
        y = follower_returns[i - window:i]
        corr = np.corrcoef(x, y)[0, 1]
        rolling_corr[i] = corr
    features['rolling_correlation'] = rolling_corr
    
    return features