Portfolio Overlap and the Effective Variation in Granular Instruments: Replication Package

This record contains the replication materials for “Portfolio Overlap and the Effective Variation in Granular Instruments.”

The package includes analysis code, analysis-ready panel data, frozen empirical outputs, manager-classification sensitivity results, a complete sample-attrition audit, threshold and power diagnostics, and a weak-identification-robust Anderson–Rubin confidence set for the linear no-threshold specification.

The fixed universe contains 300 securities selected by total Form 13F reported value in 2016Q1. CUSIP mapping produces a 214-security proxy panel, while the final granular-instrument panel contains 212 securities, 39 quarters, and 8,017 stock-quarter observations.

The exact granular-instrument decomposition attributes approximately 76.3% of instrument variance to common manager-quarter exposure, approximately 23.6% to heterogeneous portfolio weights and eligibility, and approximately 0.023% to the stock-specific leave-one-out adjustment. Removing four clearly misclassified passive managers leaves the common component essentially unchanged at approximately 76.2%.

The linear no-threshold specification produces a 2SLS point estimate of 1.782604. The stock-clustered CR1 Anderson–Rubin 95% confidence set is [-0.557390, 4.932063], which includes zero.

CUSIP mapping attrition was not neutral: retained securities were held by more institutional managers and had less concentrated institutional ownership. The results should therefore be interpreted primarily as applying to large, broadly held securities with relatively dispersed institutional ownership.

The full archive includes the manager-level cache required to reconstruct the instrument and manager-classification sensitivities. The compact archive contains analysis-ready data and frozen results but excludes the large manager-level cache. Package-wide and archive-level SHA-256 checksums are provided.