Macro Regime Forecast — 2026-02-27

Data

22 weekly macro features are loaded from S3 Parquet files (FMP API + FRED): GDP, Unemployment Rate, CPI, Federal Funds Rate, Initial Jobless Claims, Retail Sales, Consumer Sentiment, Recession Probability, Industrial Production Index, 3-Month Treasury Yield, 2-Year Treasury Yield, 10-Year Treasury Yield, 2s10s Spread, 3m10y Spread, VIX, HY OAS, IG OAS, NFCI, 5-Year Breakeven Inflation, Fed Balance Sheet (WALCL), M2 Money Supply, and Capacity Utilization. All series are resampled to week-ending-Friday frequency; monthly/quarterly series are forward-filled up to 92 days.

Standardization

All features are standardized to zero mean and unit variance using sklearn.preprocessing.StandardScaler fit on the full history. All subsequent calculations (trend fitting, distance computation, projection) operate in this z-score space.

Clustering

K-Means is fit with k=12, random_state=42, n_init=20, max_iter=500. Each week is assigned to its nearest centroid. Cluster labels are auto-generated from the three features with the largest absolute centroid z-scores.

OLS Trend Analysis

For each feature, the last 12 z-score observations are fit with a first-degree polynomial (numpy.polyfit, degree=1) to estimate a linear slope and R². Features with R² < 0.15 are treated as having a noisy/unreliable trend; their slope is set to zero so the projection falls back to the current z-score rather than extrapolating noise.

4-Week Projection

The projected z-vector at +4 weeks is computed by adding slope × 4 to each feature's current z-score (with zero slope for noisy features). Euclidean distance from this projected vector to each regime centroid provides a distance-based regime affinity score.

Transition Probability Matrix

From the full cluster assignment history, P[i, j] = P(regime at t+4 == j | regime at t == i) is estimated by counting observed transitions. Rows are normalized to sum to 1; rows with no observations receive a uniform distribution.

Score Blending

Distance weights = softmax(−projected_distances). Combined score = 0.5 × transition_probability + 0.5 × distance_weight, normalized to sum to 1. The blending gives equal weight to momentum continuation (where the macro vector is heading) and historical regime succession patterns.

Limitations

• OLS assumes linear continuation of recent trends; mean-reverting or non-linear dynamics will be missed.
• Quarterly features (GDP, Capacity Utilization) update infrequently and are forward-filled, so their z-score may lag reality.
• K-Means assumes roughly spherical, equally sized clusters; regime boundaries may be non-convex in 22-dimensional space.
• Regime labels are data-driven abbreviations and may not fully capture all macro nuance.
• The novelty check uses a fixed 85% threshold of median inter-centroid distance; this is a heuristic, not a statistical test.