A network-based strategy of price correlations for optimal cryptocurrency portfolios

Research report

TLDR

The paper proposes a modified Markowitz portfolio optimization model which utilizes a network based approach to calculating asset correlations via the construction of a minimum spanning tree. The paper’s model, testing, and results are mediocre.

Minimum spanning trees (MSTs) are well-suited to analyzing correlations between high-dimensional analytics.
The paper’s approach, testing methodology, and results were substandard, it is not worth looking into further.

This paper proposes an alternative calculation methodology for the inter-asset correlation values by utilizing a graph/network based approach. The correlations are modeled as distances between nodes in the graph.
A minimum spanning tree (MST) is constructed to generate the optimal hierarchical structure maximizing the correlation between cryptocurrencies (i.e., minimizing the total distance correlation).
The portfolios generated by this modified MPT model were tested on a range of individual cryptocurrencies and indices. Most portfolios comprised of altcoins, and only a few total portfolio constructions, performed markedly better than benchmark models.

While directly relevant to the work Concrete does, the methodology proposed in this paper are substandard and are akin to an undergraduate applied math/computer science project.
Putting aside the simplistic nature of this particular paper, applying a graph based approach has the potential to provide non-trivial relationships between difficult-to-parse time series features.