Finding Weakly Correlated Nodes in Random Variable Networks

Operations Research Forum. 2024. Vol. 5. Article 116.

Koldanov P., Alexander Koldanov, Semenov D.

The issue of identifying sets of weakly correlated stocks is explored. Four distinct

methods for constructing these sets are compared: the traditional approach using

Pearson correlation, the traditional approach using Kendall correlation, and multi-

ple hypothesis testing methods, which apply both Pearson and Kendall correlations.

To derive specific findings, we analyze daily returns of a selection of stocks listed on

the Frankfurt (FWB), London (LSE), and Paris (Euronext Paris) stock exchanges. Our

results reveal a significant difference between the identified sets of weakly correlated

stocks in Pearson and Kendall correlation networks. Notably, this difference is more

substantial in the statistically significant sets of weakly correlated stocks derived from

multiple hypothesis testing methods than in those obtained through traditional pro-

cedures. We recommend for the use of multiple hypothesis testing methods based on

Kendall correlation for analyzing market data.

Keywords: statistical significance, Pearson correlation, Stock returns, Kendall correlation, Random variable network, Multiple hypotheses testing procedure, Weakly correlated stocks, Family-wise error rate (FWER)