The Reconfiguration Premium: Co-movement Structure as an Unspanned Dimension of the Variance Risk Premium

Battle Count: 64

Directly relevant to variance risk premium analysis, correlation structure monitoring, and understanding what the VRP compensates for. The finding that rotation is an unspanned state variable (only 5-6.7% spanned by traded correlation instruments) has implications for derivative pricing and hedging. The prepayment mechanism (implied variance rises on impact, realized volatility follows 2-3 quarters later) informs timing of variance-selling strategies. However, the paper explicitly establishes NO timing alpha, NO crash protection, and the attribution is not extractable as a trading rule. The index is a conditioning/attribution variable, not a signal. Relevant for understanding the economic content behind VRP rather than for generating tradeable strategies.

Requires: (1) monthly S&P 500 return panel with delisted names and coverage filtering; (2) dual-space spectral decomposition of correlation matrices with Marchenko-Pastur edge refitting; (3) principal angle computation between consecutive subdominant eigenspaces restricted to common cross-sections; (4) three-month moving average smoothing; (5) Newey-West regression with level controls; (6) frequency decomposition (persistent vs. high-frequency); (7) simulated-null bootstrap with AR(2) DGP and i.i.d. innovation-pair resampling preserving contemporaneous correlation; (8) forward-looking regressions with proper lag controls; (9) Davis-Kahan mechanical channel bounding; (10) downside semicovariance Gram construction; (11) sector labeling of subdominant modes. The spectral geometry and inference design are the most complex components.

Full replication code and data available at https://github.com/lucas-p-carvalho/rec-anatomy. Script [01] rebuilds the index from the panel and reproduces the stored series at correlation 1.0000. Detailed construction steps, spectral placement tests, dose-response analysis, and simulated-null inference are all documented. One limitation: the daily Bloomberg panel used for the volatility-rank churn test (Section 8.2) is licensed and not reproducible from the public repository. The ticker_inventory.csv with all 430 retained names is distributed as an ancillary file.