Screening-Off Information and Conditional Risk in Portfolio Choice

Battle Count: 78

Highly relevant to quantitative trading and portfolio management. The paper directly addresses how conditioning information affects portfolio construction, covariance estimation, and risk assessment. The screening-off criterion provides a falsifiable test for whether selected information adequately captures common cross-asset dependence before portfolio optimization. The exact covariance decomposition and perturbation bounds give practitioners a direct map from representation failure to portfolio instability. The empirical results on 150 U.S. equities and 17 hedge-fund indices demonstrate practical applicability. The residual-aware covariance estimator provides an implementable improvement over diagonal-residual restrictions.

High complexity due to: (1) theoretical framework requiring understanding of conditional independence, screening-off, and causal separation hierarchies; (2) exact covariance decomposition with four distinct terms requiring careful estimation; (3) constrained Markowitz optimization with sensitivity restrictions; (4) finite perturbation identities and non-asymptotic bounds; (5) multiple covariance estimation methods (Ledoit-Wolf, POET, PCA, structured estimator, residual-aware estimator); (6) rolling out-of-sample validation with frozen parameters; (7) driver selection from 123+ candidate economic/financial series. The Woodbury-based structured solve reduces computational cost to linear order in assets for fixed driver/constraint dimensions, but the overall pipeline is sophisticated.

Complete code, deterministic configurations, tests, and derived outputs are publicly available via Zenodo DOI (10.5281/zenodo.21224465). All empirical specifications use rolling, temporally ordered folds with frozen parameters. Data descriptions are thorough (asset universe, driver identifiers, transformation rules, sample periods, alignment procedures). However, market data requires licensed access to Tiingo, Bloomberg, and BarclayHedge, limiting full independent replication without subscriptions.