Replication package for "Does Scale Erode Performance? Evidence from Large Capital Transfers Decided by Another Fund's Board"

<p>Construction and estimation code for a study of whether capital erodes mutual fundperformance, identified from transfers that arrive at a surviving portfolio because a<em>different</em> fund is wound up by its own board.</p> <p><strong>What the paper estimates.</strong> Performance is the covariance between changes inportfolio weights and subsequent returns, so no factor model mediates the result. Amongacross-family transfers the post-event change in the rate lies in [-16.14, +22.21] basis pointsa quarter. That interval excludes the magnitude the fund-merger literature reports and admitsthe smaller magnitude implied by the manager-promotion literature, against which the design isnot powered. The same events measured on a four-factor intercept with pre-event loadings return+12.67, and on a size, book-to-market and momentum characteristic benchmark -9.55.</p> <p><strong>Contents.</strong> Python scripts that build the estimation panel from the licensedsources and reproduce every table and figure, including: the HML and book-equity constructionfrom Compustat annual fundamentals and the CRSP-Compustat link history; the four-factor againstthree-factor horse race on identical portfolio-quarters, with its non-event placebo; the125-cell DGTW characteristic benchmark formed on size, then book-to-market within size, thenmomentum within both; the manager-continuity classification and the split of the estimate bywhether the named manager changed; the figure-building scripts; and a reference-verificationscript that checks every citation in the manuscript against its Crossref registration record.The LaTeX source of the manuscript and internet appendix is included.</p> <p><strong>Data.</strong> The analysis uses licensed data from the CRSP Survivor-Bias-Free USMutual Fund Database, the CRSP Monthly Stock File, the Thomson Reuters Mutual Fund Holdings(s12) file, the MFLINKS crosswalk, and Compustat Fundamentals Annual with the CRSP-Compustatlink history, all accessed through Wharton Research Data Services. <strong>The licence does notpermit redistribution, so no underlying data are deposited here.</strong> The manuscript statesevery sample restriction and every row count at each stage of construction, so a reader holdingthe same licences can rebuild the panel exactly: 257,556,508 holdings records read, 28,375,112holding rows entering the weight file, and an estimation panel of 154,415 portfolio-quarters on5,025 portfolios over 1991Q4-2024Q4, of which 794 carry a merger arrival on 626 portfolios.</p> <p><strong>One construction point that travels beyond this paper.</strong> Of 12,010 linkedmerger events, 3,554 - 29.6 per cent, carrying $357 billion of nominal assets - combine twoshare classes of a single fund and move no capital between portfolios. They cannot be toldapart at share-class level and identify themselves at portfolio level. Retaining them attenuatesthe dollar coefficient from +14.39 (t = 5.39) to +7.35 (t = 2.05). Any study built on theseevents inherits this.</p> <p>Requires Python 3.12 with numpy, pandas, pyarrow, matplotlib and openpyxl.</p>