Across finance literature, portfolio construction of cross-sectional long-short risk premia portfolios generally follows a similar framework as summarized here:

1. Measure the risk premia signal of each asset within a defined universe.
2. Sort and rank the assets by value of the signal.
3. Assign top (bottom) x% of the assets to long (short) leg.
   • Within a long or short leg, assets can be weighted in various ways (e.g., equal weighted, rank weighted, etc.).
4. Long and short legs are sized equally so that all weights sum to zero (dollar-neutral). For example, long (short) assets sum to 100% (-100%).

For the purposes of this article, because the focus is not on identifying the best signal or assets, we use a set of well-cited risk premia signals (carry, value, and momentum) across familiar asset classes (equities, bonds, commodities, and FX), building on Brightman and Shepherd (2016). Assets include 15 equity index futures, 17 government bond futures with maturities ranging from 2 years to 10 years, 27 commodity futures, and 15 FX forwards denominated against the U.S. dollar across both developed and emerging markets. To avoid data-mining concerns, we have chosen signals based on definitions that are relatively simple and have been found to be robust.