Three-Currency HJM for Brazilian Credit Markets

Battle Count: 51

The paper is primarily academic/theoretical in nature, developing a no-arbitrage framework and testing it empirically. However, the findings have direct practical implications for Brazilian fixed income trading: the 640bp within-issuer wedge between CDI and IPCA segments represents a persistent, quantifiable mispricing that could inform relative-value strategies. The tax clientele mechanism is structural and stable across monetary regimes, making it potentially exploitable. The framework provides a rigorous diagnostic for when cross-segment arbitrage is possible versus when it reflects genuine segmentation. However, the paper does not propose specific trading strategies, and the wedge is largely explained by tax structure rather than being a pure mispricing.

Implementation requires: (1) bootstrapping nominal and real curves from DI futures and NTN-B prices with specific Brazilian conventions (business-day compounding, VNA accrual); (2) PCA on weekly curve changes with block structure; (3) nonlinear least squares fitting of exponential shape functions; (4) Girsanov change-of-measure with 7-dimensional correlated Brownian motions; (5) Musiela-grid simulation with closed-form drift integrals; (6) block bootstrap for standard errors; (7) duration-bucketed spread curve construction from JGP index constituents. The mathematical machinery (three-currency HJM, exchange-rate dynamics, martingale conditions) is sophisticated. The simulation algorithm is well-specified but requires careful numerical implementation of the Musiela transport and exponential FX updates.

The paper provides detailed mathematical derivations, calibration procedures, and parameter tables. However, it relies on proprietary JGP Idex index data and specific ANBIMA indicative prices. The simulation algorithm (Algorithm 1) is fully specified with pseudocode. Data sources (B3 DI futures, ANBIMA NTN-B, JGP Idex) are publicly accessible but require subscriptions or specific access. No code repository is provided. The 15-issuer sample selection criteria are clearly stated.