A Three-Variable Benchmark for Post-GFC Covered Interest Parity Deviations

Battle Count: 80

The paper is primarily an academic benchmark contribution rather than a trading strategy. However, it is relevant to quantitative trading in several ways: (1) CIP deviations represent persistent cross-currency basis trades that can be exploited by FX swap desks and cross-currency basis traders; (2) the benchmark identifies which macrofinancial state variables (NFCI, dollar, yield curve) drive the persistent component of CIP deviations, useful for timing basis trades; (3) the out-of-sample R² of ~0.50 suggests meaningful predictive content for the background component; (4) the paper explicitly states it is NOT a high-frequency arbitrage-execution model, limiting direct trading applicability. The benchmark is more useful as a research hurdle than as a direct trading signal.

The core benchmark is extremely simple to implement: three lagged public variables regressed on CIP deviations with panel-specific intercepts. All data are from FRED and a public website. The main complexity lies in the robustness diagnostics (cointegration tests, aggregation-difference analysis, PCA rotation, expanding-window forecasts) and the careful data alignment (weekly NFCI carried forward, one-observation lags, non-overlapping block construction). A basic implementation requires only OLS with HAC standard errors. The full paper's analysis requires more econometric sophistication but remains within standard panel regression toolkit.

The benchmark is explicitly designed for reproducibility. All three regressors (NFCI, broad dollar index, Treasury 10Y-2Y slope) are available from FRED. The dependent variable (government-bond CIP deviations) is from Du et al. (2025b), publicly available on Jesse Schreger's website. The specification is intentionally simple: three lagged public variables, panel-specific intercepts, no proprietary data, no high-dimensional search. The paper provides detailed data alignment procedures (weekly NFCI carried forward, one-observation lags). No GitHub repository is mentioned, but all inputs are public.