Skip to content
All library documents

Tracking Stock Clusters Over Time with Weekly Re-estimation

Article Quant Q&A · Author: Egodym
Read the original

Summary

The document raises a methodological question about monitoring how groups of stocks change over time. The author has stock price data and wants to cluster stocks with similar characteristics, then follow the evolution of those clusters. They report that stocks often change cluster from one week to the next, making it unclear how to interpret a sequence of independently estimated groupings.

One proposed procedure is to refit a clustering algorithm each week and look backward across the resulting assignments. The document does not specify the features, distance measure, clustering method, or any empirical comparison, and it provides no answer about whether weekly refitting is statistically sound. Those omissions matter because changing assignments could reflect real shifts in stock behavior, sensitivity to the sample, or differences in cluster labels between fits. The text is therefore a research question about temporal cluster stability, not a demonstrated strategy or validated workflow.

Key ideas

  • The author wants to group stocks by similar characteristics and monitor changes in group membership.
  • Weekly changes in stock assignments raise questions about whether observed cluster evolution is meaningful.
  • The proposed approach is to refit the clustering algorithm each week and compare assignments over time.
  • The document gives no method details or evidence establishing the statistical validity of that approach.

Tags

EquitiesStatisticsMachine learning
Full text
# Clusters evolution over time


# Clusters evolution over time












I have a dataset of stock prices and I want to group stocks that share similar characteristics together using cluster analysis. I'm interested in following the evolution of each cluster over time, but since stocks have very different behaviours (roughly 50% of the time a stock will change cluster the week after), I was wondering what would be a statistically sound approach. Is it a good idea to train a clustering algorithm every week and look backwards at the weekly evolution of each cluster?

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.