The Fix Is In: Properly Backing out Backfill Bias
研究发现对冲基金数据库中常用的回填偏差处理方法（保留或截断部分数据）仍会导致偏差，建议删除上市日期前的收益数据，并提出推断缺失上市日期的新方法。
Abstract Researchers have long known about backfill bias in hedge fund databases. The most common treatments include either retaining all backfilled returns or truncating a fixed number of returns from each return series. However, we show that truncation largely preserves backfilled returns and document that either of these backfill treatments can lead to biased empirical findings, including cross-sectional results. Thus, our findings show that the best practice for empirical tests is to remove returns prior to the listing date. Because most databases do not have listing dates, we propose a novel method to infer unavailable listing dates. Received August 19, 2018; editorial decision December 4, 2018 by Editor Wei Jiang.