Overview

A significant paper spread is still not a book the mandate can hold. This topic takes the book-to-price specification — already insignificant on the reported information-coefficient row — and asks what implementation would have done to it if the row had passed, and what implementation cannot be asked to do given that it did not. Turnover, the long-only bound, and the difference between a persistent score and a live forecast are the three decisions.

By the end of this note you should be able to:

• pair quarterly score autocorrelation with quarterly turnover at a stated tracking error, and refuse an asset-level capacity cap the extracts do not give;
• translate a long-only mandate into $w_i \ge -b_i$ and describe what that does to a short decile of small names;
• separate score autocorrelation from a lagged information coefficient, including the bias from overlapping labels.

1. Contents
2. The big picture