Predicting Changes in the U.S. Treasury Futures Spread During the Roll Period

Abstract

The goal of this thesis is to discover an accurate model to predict how the difference between the prices of the active and deferred treasury future contracts move during the roll period. In the first half of this paper, we apply iterative regression to a large set of features to screen for initial candidates, resulting in the discovery of reversion as an accurate predictor of how treasury futures spreads move. In the second half, we spend more time investigating reversion, and find that its effects are limited to the 10 year contract, it works best for roll events with low volume, and it demonstrates consistent accuracy throughout the time period considered. Future extensions of this work would investigate alternative features, different ways of trimming down the initial list of features considered, and an investigation into the macroeconomic factors that cause the underlying movement in treasury spreads.