When David becomes Goliath: Repo dealer-driven bond mispricing

Battle Count: 68

Highly relevant for fixed-income quantitative trading, particularly in government bond markets. The paper identifies three channels (dealer market power, dispersion of market power, and network-driven persistent shocks) that cause systematic bond mispricing of 2.5-5.3 percentage points. This provides actionable signals for: (1) identifying mispriced gilts relative to term-structure benchmarks, (2) understanding liquidity dynamics in repo and bond markets, (3) anticipating price distortions during stress periods (dash-for-cash), (4) incorporating dealer concentration and network effects into trading strategies, and (5) understanding the asymmetric impact of repo vs. reverse-repo segments on different maturity buckets. The reverse-repo segment's disproportionate effect on mispricing (especially for long-dated gilts) is particularly relevant for curve trading strategies.

Extremely complex implementation requiring: (1) Proprietary transaction-level repo data at dealer/non-dealer dyad level; (2) Structural demand estimation following Berry (1994) with multiple endogenous variables requiring granular IVs; (3) TVP-VAR estimation using Quasi-Bayesian Local-Likelihood methods for a 22x22 system; (4) Fourier-based frequency decomposition to separate transitory and persistent shocks; (5) Construction of global dealer factors from time-varying variance decompositions; (6) Spline-based yield curve benchmarking for mispricing measurement; (7) Hu et al. (2013) noise index for market liquidity; (8) Multiple panel regression specifications with 2SLS, Driscoll-Kraay standard errors, and extensive fixed effects. The combination of structural IO methods, dynamic network econometrics, and bond market analysis makes this a highly specialized and complex implementation.

The paper relies on proprietary transaction-level data from the Bank of England (Sterling Money Markets Data - SMMD) covering all gilt-backed repo and reverse-repo trades between 2016 and 2022. This data is not publicly available. The structural model estimation, TVP-VAR implementation, and granular IV construction are described in detail in the Online Appendix, but without access to the underlying data, exact replication is impossible. The methodology is well-documented but the data constraint is severe.