Building a Cross-Sectional Cryptocurrency Momentum Factor

Summary

The document describes a thesis design for testing cross-sectional momentum in cryptocurrencies. Each week, coins are ranked by their past value-weighted log returns over several lookback windows, then divided into five equal-weighted portfolios. The following week’s returns are used to measure the performance of each ranked portfolio, with the highest past-return group treated as winners and the lowest as losers.

The central question is how to form a momentum factor in the style of Kenneth French: whether the factor should use the ranking-period returns or the subsequent portfolio returns. The setup points toward using past returns to assign assets to portfolios and subsequent returns to calculate portfolio performance, though the document itself poses this as an unresolved question. It provides no empirical results or full factor-construction specification, so details such as the winner-minus-loser definition, rebalancing conventions, and treatment of overlapping holding periods remain open.

Key ideas

• Rank cryptocurrencies using returns observed before portfolio formation.
• Use subsequent returns to evaluate the portfolios formed from those rankings.
• The proposed design uses several weekly lookback windows and five equal-weighted groups.
• The document raises, but does not resolve, how to translate the portfolio sorts into a factor.

Tags

# Constructing a cryptocurrency momentum factor


# Constructing a cryptocurrency momentum factor












This might be a very trivial question, but I simply can't find a practical answer anywhere.

I am writing my Master's Thesis, and in this regard, i'm investigating the cross-sectional momentum effect in the cryptocurrency market, using the methodologies of Jegadeesh & Titman (1993). I am using past J (1, 2, 3 & 4) value-weighted weekly log returns in order to sort the currencies' subsequent 1-week log return in 5 equal-weighted quantile portfolios in ascending order, such that P5 is the winner portfolio and P1 is the loser portfolio.

Example: I calculate each coin's log return from t-1 to t (past 1-week log return), and then sort each of these past returns (every week) in ascending order. Then, for every week, I find each of the coin's subsequent 1-week log return from t to t+1. I will then sort the coins in 5 portfolios (based on the past 1-week return), in which each of the portfolios will contain an equal-weighted average of the subsequent 1-week return of the coins contained in the specific portfolio.

My question is how I can construct a momentum factor in lines with how Kenneth R. French does it? Do I use the past 1-week returns, or do I use the subsequent 1-week returns to construct the momentum factor?

Shown in full with attribution under the source's licence. Licence: CC BY-SA 4.0 (Stack Exchange)

This summary was written by Stratmill's research agent from the original; it is not a copy of the source.