Characteristics Similarity of the Return-Sorted Portfolios
Log in to collectAcademic paper
Seyed Mohammad Sina Seyfi
- FIAalto University
- ?Aalto University, School of business
Strategy in a nutshell
=This strategy focuses on large U.S. stocks (above the NYSE median market capitalization). Stocks are ranked into quintile portfolios based on expected returns, calculated using a 10-year rolling window. Portfolio 1 contains the lowest expected returns, and Portfolio 5 contains the highest. To ensure consistency, future portfolios are matched to the in-sample portfolios using 206 firm characteristics, measured via Euclidean distance. A long-short H–L portfolio is formed by going long on Portfolio 5 (high expected returns) and short on Portfolio 1 (low expected returns). Portfolios are value-weighted and rebalanced monthly.
Economic rationale
The strategy leverages big data and machine learning in asset pricing, addressing the challenge of high-dimensional factor space (“factor zoo”). By matching out-of-sample stocks to in-sample portfolios with similar characteristics, it captures predictable cross-sectional variations in returns. Key drivers include idiosyncratic risk, momentum, maximum past returns, proximity to historical highs, and cash-based operating profitability. This methodology enhances predictive power while systematically exploiting patterns in the cross-section of stock returns.