Research

Publications

• Credit Rating Prediction Through Supply Chains: A Machine Learning Approach (with Jing Wu and Sean X. Zhou) link
  Production and Operations Management, Vol. 31, No. 4, April 2022, pp. 1613–1629.
• Impact of Primary Care Market Mergers on Quality: Evidence from the English NHS (with Yuan Lyu) link
  Journal of Health Economics, Forthcoming.

Working Papers

(Updated drafts and additional materials are available upon request)

• Tensor Factor Analysis of Global Supply Chains link
  Abstract

  This paper develops and estimates a semi-parametric tensor factor model for high-dimensional data where latent factor loadings are flexible, non-parametric functions of time-varying covariates. We introduce this framework to analyze a pressing economic problem: the identification of systematic risks in global supply chains. Applying the model to a five-dimensional country-industry level dataset (2000-2014), the empirical analysis reveals that a small number of latent factors—interpreted as major trade hubs like China and the U.S.—explain the vast majority of the network's variation. The results demonstrate that exposures to these systematic risk factors are strongly driven by observable economic characteristics such as economic size and productive capability. This research provides both a novel econometric tool for dynamic, multi-dimensional data and new insights into the evolving structure of the global economy.

• Tensor Factor Models, Missing Financial Data, and Stock Returns (with Martin Lettau) link
  Abstract

  We propose a tensor factor model for high-dimensional panels of stock characteristics that accommodates arbitrary patterns of missing data and uncovers the low-rank structure inherent in multi-way financial data. The model employs a weighted version of the Tucker decomposition, estimated via higher-order orthogonal iteration adjusted for unbalanced observation patterns. We apply this framework to a monthly panel of 153 firm-level characteristics for over 5,000 U.S. stocks from 1983 to 2022. Despite over 50% of entries being missing, the estimated model achieves over 90% out-of-sample explanatory power and compresses the data by more than 97%. Sorting stocks by these latent factor scores yields long-short portfolios with strong performance in asset pricing tests. The resulting tensor-based pricing factors earn high Sharpe ratios and exhibit lower pricing errors than traditional anomaly, PCA, and Fama–French factors. Our results demonstrate that modeling the characteristic tensor directly can offer significant empirical advantages in explaining the cross-section of stock returns.

• Determining the Number of Factors with Missing Observations (with Weiguang Liu)
  Abstract

  This paper proposes eigenvalue ratio-based estimators for determining the number of factors in latent factor models with missing observations. Our framework accommodates general missing data patterns and does not require the Missing Completely at Random (MCAR) assumption. Our approach is based on the spectral behavior of adjusted second-moment matrices under flexible overlap conditions, requiring only that cross-sectional pairs share a positive fraction of observations. Simulation studies demonstrate the robustness of our estimators across a wide range of missing data patterns and idiosyncratic structures. We further illustrate the practical applicability of our method with an empirical application to monthly excess returns (1995-2024).

• Detection of Weak Granulars in Large Panels (with Yinfeng Zeng)
  Abstract

  We propose a new framework to detect weak granular units in large panels—cross-sectional units that affect only a fraction of the data. Extending existing granular and factor models, our method accommodates both weak common factors and weak granular shocks. We derive consistency and asymptotic properties under relaxed loading assumptions, and develop a sequential testing procedure based on thresholding idiosyncratic variances. Our results bridge granular theory and weak factor analysis, and empirical applications show that weak granular structures are common in economic and financial panels.

• Impact of Horizontal Mergers on the Value of Suppliers and Customers
  Abstract

  This paper examines the impact of horizontal mergers on the value of merging firms’ suppliers and customers using newly constructed supply chain data. I analyze both general and horizontal mergers and test for cumulative abnormal returns to related firms. The findings show that horizontal mergers lead to significant negative announcement returns for both suppliers and customers of the merging firms. I show empirical evidence that this adverse effect is likely driven by increased industry concentration following the merger.

Selected Work in Progress

• Non-negative Tensor Factor Models
• Tensor Factor Analysis of U.S. County-County Commuting Flows (with River Chen)
• Tensor Factor Analysis of Institutional Stock Holdings