Investment Base Pairs

Abstract

Modern finance remains constrained by legacy portfolio construction techniques. While the conventional quantile approaches (e.g., long top 30%/short bottom 30%) and linear weighting schemes dominate cross-sectional strategies, we show that these methods discard crucial cross-asset information. We offer a new approach by decomposing common signals such as value, momentum, and carry, into “base pair” portfolios. Each signal-driven long-short position is shaped by five key drivers: own-asset predictability, cross-asset predictability, signal correlation, and signal mean and variance imbalances. Using 1,710 futures pair portfolios spanning equities, bonds, currencies, and commodities formed from common signal types, we show that targeting top pairs can triple average returns at fixed leverage over 20 years: the aggregate "All" portfolio rises from 3.4% to 10.4% annualized. Equity Value climbs from 3.6% to 14.3% and Currency Momentum reverses a -3.0% loss to a 10.3% gain. By harvesting cross-asset information and eliminating "junk" pairs, this approach offers a robust improvement over the status quo across diverse asset classes and time periods. 

Keywords: cross-signal correlation, signal bias, predictability, pairs trading, futures and forwards, signal sorting, investments, portfolio construction, value, momentum, carry, own-asset effects, cross-asset effects

JEL Classification: G10, G11, G12, G15