{"title":"迭代互补聚类PCA:揭示股票收益中的潜在行业结构","authors":"Daning Bi , Le Chang , Yanrong Yang","doi":"10.1016/j.econlet.2025.112611","DOIUrl":null,"url":null,"abstract":"<div><div>Principal component analysis (PCA) is a widely implemented statistical method for dimension reduction, but struggles to identify group-specific patterns (sub-homogeneity), such as the latent industry structures in high-dimensional stock return data. We propose an Iterative Complement-clustering PCA (ICcPCA) that jointly estimates homogeneity (market-wide effects) and sub-homogeneity (industry-specific risks), where a Leave-one-out principal component regression (LOO-PCR) clustering approach is developed to iteratively cluster variables (stocks) into disjoint multidimensional subspaces (groups). Simulations show that the ICcPCA outperforms the conventional PCA in both estimating the number of principal components and recovering the data. In analyzing stock returns of 160 firms across 8 industries, ICcPCA with LOO-PCR can separate market-wide effects from industry-specific risks, achieving higher clustering accuracy and lower recovering errors. Applications in portfolio optimization demonstrate that ICcPCA-based minimum variance portfolios can attain lower volatility and higher profitability than PCA-based portfolios.</div></div>","PeriodicalId":11468,"journal":{"name":"Economics Letters","volume":"256 ","pages":"Article 112611"},"PeriodicalIF":1.8000,"publicationDate":"2025-09-16","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"0","resultStr":"{\"title\":\"Iterative Complement-clustering PCA: Uncovering latent industry structures in stock returns\",\"authors\":\"Daning Bi , Le Chang , Yanrong Yang\",\"doi\":\"10.1016/j.econlet.2025.112611\",\"DOIUrl\":null,\"url\":null,\"abstract\":\"<div><div>Principal component analysis (PCA) is a widely implemented statistical method for dimension reduction, but struggles to identify group-specific patterns (sub-homogeneity), such as the latent industry structures in high-dimensional stock return data. We propose an Iterative Complement-clustering PCA (ICcPCA) that jointly estimates homogeneity (market-wide effects) and sub-homogeneity (industry-specific risks), where a Leave-one-out principal component regression (LOO-PCR) clustering approach is developed to iteratively cluster variables (stocks) into disjoint multidimensional subspaces (groups). Simulations show that the ICcPCA outperforms the conventional PCA in both estimating the number of principal components and recovering the data. In analyzing stock returns of 160 firms across 8 industries, ICcPCA with LOO-PCR can separate market-wide effects from industry-specific risks, achieving higher clustering accuracy and lower recovering errors. Applications in portfolio optimization demonstrate that ICcPCA-based minimum variance portfolios can attain lower volatility and higher profitability than PCA-based portfolios.</div></div>\",\"PeriodicalId\":11468,\"journal\":{\"name\":\"Economics Letters\",\"volume\":\"256 \",\"pages\":\"Article 112611\"},\"PeriodicalIF\":1.8000,\"publicationDate\":\"2025-09-16\",\"publicationTypes\":\"Journal Article\",\"fieldsOfStudy\":null,\"isOpenAccess\":false,\"openAccessPdf\":\"\",\"citationCount\":\"0\",\"resultStr\":null,\"platform\":\"Semanticscholar\",\"paperid\":null,\"PeriodicalName\":\"Economics Letters\",\"FirstCategoryId\":\"96\",\"ListUrlMain\":\"https://www.sciencedirect.com/science/article/pii/S0165176525004483\",\"RegionNum\":4,\"RegionCategory\":\"经济学\",\"ArticlePicture\":[],\"TitleCN\":null,\"AbstractTextCN\":null,\"PMCID\":null,\"EPubDate\":\"\",\"PubModel\":\"\",\"JCR\":\"Q2\",\"JCRName\":\"ECONOMICS\",\"Score\":null,\"Total\":0}","platform":"Semanticscholar","paperid":null,"PeriodicalName":"Economics Letters","FirstCategoryId":"96","ListUrlMain":"https://www.sciencedirect.com/science/article/pii/S0165176525004483","RegionNum":4,"RegionCategory":"经济学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"Q2","JCRName":"ECONOMICS","Score":null,"Total":0}
Iterative Complement-clustering PCA: Uncovering latent industry structures in stock returns
Principal component analysis (PCA) is a widely implemented statistical method for dimension reduction, but struggles to identify group-specific patterns (sub-homogeneity), such as the latent industry structures in high-dimensional stock return data. We propose an Iterative Complement-clustering PCA (ICcPCA) that jointly estimates homogeneity (market-wide effects) and sub-homogeneity (industry-specific risks), where a Leave-one-out principal component regression (LOO-PCR) clustering approach is developed to iteratively cluster variables (stocks) into disjoint multidimensional subspaces (groups). Simulations show that the ICcPCA outperforms the conventional PCA in both estimating the number of principal components and recovering the data. In analyzing stock returns of 160 firms across 8 industries, ICcPCA with LOO-PCR can separate market-wide effects from industry-specific risks, achieving higher clustering accuracy and lower recovering errors. Applications in portfolio optimization demonstrate that ICcPCA-based minimum variance portfolios can attain lower volatility and higher profitability than PCA-based portfolios.
期刊介绍:
Many economists today are concerned by the proliferation of journals and the concomitant labyrinth of research to be conquered in order to reach the specific information they require. To combat this tendency, Economics Letters has been conceived and designed outside the realm of the traditional economics journal. As a Letters Journal, it consists of concise communications (letters) that provide a means of rapid and efficient dissemination of new results, models and methods in all fields of economic research.