A naukri.com initiative
Medium
1w
250
Image Credit: Medium
From Fundamentals to Factor Buckets : A K-Means Clustering Approach to Stock Styles
- Investors often face information overload, leading to analysis paralysis and sub-optimal decisions.
- The mismatch between individual investors' thinking and how data is presented by screeners is a significant issue.
- A K-Means clustering approach was used to categorize S&P 500 companies into intuitive 'styles' like Undervalued, High Growth, etc.
- The project focuses on capturing five core factor themes: Value, Growth, Quality, Defensive, and Momentum & Risk.
- Data cleaning steps involved imputing sector medians, transforming skewed distributions, sector-neutral robust scaling, and winsorizing outliers.
- Feature selection was based on Spread Ratio, and metrics with spread ratio > 4% were kept for clustering.
- The number of clusters for each factor bucket was determined using elbow analysis, Silhouette score, and Davies–Bouldin index.
- Clusters were labeled based on their characteristics, making the model output actionable and easy to interpret for investors.
- The model's performance was tested using hypothetical investor personas and equal-weighted mini-portfolios.
- The results suggested that the factor buckets captured distinct investment styles with varying performance.
Read Full Article
15 Likes
For uninterrupted reading, download the app