Trend factors are a phenomenon in finance/economics that refer to quantitative metrics that are believed to have some sort of statistically significant relationship with future returns. The most traditional example of a factor model is the Fama-French 3 Factor (FF3) Model, which assumes that returns can be explained by a portfolio’s factor loadings on “market”, “size”, and “value”. In the decades since, many other factors have been proposed, each with varying amount of success in terms of their predictive power.

A natural approach to factor investing is to simply test models with any combination of factors that one might believe to be indicative of performance. However, this approach comes with several issues, one of which being the hidden correlation between factors. With so many different factors, it becomes easier to obscure what factors actually provide an additional source of returns. Returns from “new” factor models are often subsumed by existing models when analysed thoroughly.

The way Fieberg et al. addressed this issue, in their paper “A Trend Factor for the Cross Section of Cryptocurrency Returns”, was to use a machine learning technique: elastic net regression. Elastic net regression is a regression method that utilises a linear combination of the penalties observed in LASSO and ridge regression, shrinking irrelevant predictors to 0 while overcoming the difficulty LASSO regression faces with correlation between predictors.

The paper found this new factor to be supposedly robust and not subsumable by existing factors, while producing greater returns than any of the comparable indicators. These results, however, are from a testing period from 2015 to 2022, which would indicate that it caught all of the explosive growth of crypto and very little of the proceeding drawdown. I wanted to evaluate the performance of this