The Year-End Toll: Frictions Embedded in Option-Implied Rates

Battle Count: 78

Highly relevant for quantitative traders and researchers who use option-implied rates as risk-free rate proxies or for convenience yield estimation. The 2-3 bp unannualized year-end wedge translates to ~10 annualized bp at 3-month maturity, which is economically material for short-dated strategies. The finding that put-call parity identifies discount rates precisely but not their economic purity challenges assumptions in box-spread arbitrage, CIP-based strategies, and any model treating option-implied rates as frictionless. The mid-2010s regime shift and time-varying nature make simple constant corrections inadequate. Relevant for derivatives desk risk management, regulatory capital planning, and understanding limits to arbitrage in relative-value trades.

The empirical design is sophisticated: constructing option-implied discount factors from one-minute NBBO quotes via synthetic-forward regressions, building multiple benchmark discount curves (DGS bootstrap, DTB interpolation, OIS bootstrap), implementing strict same-date local RD with triangular kernels and trade-date fixed effects, running full-panel regressions with multiple controls, conducting functional-form horse races, break-point searches with scan-adjusted inference, and external validation across government-bond CIP and independent option panels. The data processing pipeline (quote validation, call-put matching, quality screens, intraday aggregation) is non-trivial. However, the core econometric techniques (OLS, RD, HAC) are standard.

The paper provides extremely detailed construction of option-implied discount factors, benchmark curves (DGS, DTB, OIS), and all regression specifications. Data sources are explicitly named (ThetaData for options, FRED for Treasury rates, NFCI, ABCP, SOFR, TGCR, BGCR). The Diamond-Van Tassel panel data is publicly available. However, the raw one-minute NBBO option quotes from ThetaData are commercially licensed. MATLAB code is referenced but not publicly linked. The author acknowledges AI assistance for debugging and drafting but states all code was reviewed and executed by the author.