CV RMSE

CV MAE

QLIKE

Predictions
Interactive FPCA GLMM Walkthrough
Predicted vs Actual Future Volatility
Future Bucket Calibration

Rows scanned

Stocks

Avg spread

Avg RV
WAP, Spread, and Imbalance
Order Book Depth Snapshot
30 Second Realized Volatility Curve
Return Distribution
Cluster Summary
Cluster Mean Future Volatility Trajectories
Final Pipeline Cluster Tuning
Supporting Clustering Methods
What FPCA Does Here

Each stock-time window becomes a short volatility curve: 20 realized-volatility values, one for each 30-second bucket.

FPCA reduces the observed part of that curve, buckets 1-16, into a small number of principal component scores. Those scores summarize the level, shape, and movement of the volatility curve without feeding all bucket values directly into the model.

The final FPCA GLMM uses these scores, cluster labels, future-bucket effects, and stock-level effects to predict buckets 17-20.

FPCA GLMM Configuration
Residual Error Bands
Model Configuration
Data Preview
Pipeline Coverage Map