Distributionally Robust Recovery of Omitted Factors from Forecast Residuals with Application to Interest Rate Risk Management

Battle Count: 71

Highly relevant for fixed income quantitative trading and risk management. The paper provides a principled method to identify systematic risk factors that a forecaster misses, directly applicable to duration positioning, DV01 risk management, and tail risk assessment. The covariance forcing statistic and neutralization test offer actionable tools for portfolio managers. The DRO framework for robustifying decisions against residual uncertainty is directly applicable to trading strategy deployment. However, the method is specialized to yield curve forecasting and the one-month horizon limits high-frequency applicability.

The framework involves multiple interconnected components: rolling DNS/FADNS estimation with VAR, joint multi-output random forest with cross-validated hyperparameters, Ledoit-Wolf shrinkage, two-layer DRO program (closed-form and concave optimization), covariance forcing computation, FarmSelect with SCAD+Huber, block-permutation testing with nested penalty selection, and the neutralization hedge. Each component is well-specified with pseudocode, but the full pipeline requires careful implementation of 11 algorithms with proper data handling, expanding windows, and multiple specifications (10 FADNS factor counts, 10 RF seeds). The optimization components are tractable (N=15) but the naming regression in p>n regime requires careful regularization.

The paper provides detailed pseudocode for all procedures (Algorithms 1-11), specifies all hyperparameters (window w=60, burn-in 36, lambda=0.0609, SCAD a=3.7, Huber delta=1.345, 199 permutation draws with fixed seed), and uses publicly available data (LSEG Reuters Workspace Treasury yields). The expanding training scheme and all calibration formulas are fully specified. However, no GitHub repository is mentioned, and the 111-indicator panel composition, while listed in Table 6, would require specific data access.