Telephone service for the AEA administrative office is temporarily unavailable. Please direct all AEA and ASSA inquiries to the publications office at 412-432-2300.
« Back to Results

High-Dimensional Problems in Financial Econometrics

Paper Session

Monday, Jan. 4, 2027 10:15 AM - 12:15 PM (EST)

Marriott Marquis Washington DC
Hosted By: Econometric Society
  • Chair: Allan Timmermann, University of California-San Diego

Large and Deep Factor Models

Bryan Kelly
,
Yale University
Boris Kuznetsov
,
AQR Capital Management
Semyon Malamud
,
Swiss Finance Institute
Yuan Zhang
,
Shanghai University of Finance and Economics

Abstract

We show that a deep neural network (DNN) trained to construct a stochastic discount factor (SDF) admits an additive decomposition separating nonlinear characteristic discovery from the pricing rule that aggregates them. This decomposition yields a linear factor representation governed by the Portfolio Tangent Kernel (PTK), which summarizes the network’s learned features. In population, the implied SDF converges to a ridge-regularized version of the true SDF, with the degree of regularization determined by spectral complexity. Empirically, using U.S. equity data, the PTK representation delivers economically and statistically significant performance gains, while rising spectral complexity imposes tighter limits on finite-sample pricing.

Imputation-Powered Inference for Missing Covariates

Junting Duan
,
Stanford University
Markus Pelger
,
Stanford University

Abstract

Missing covariate data is a prevalent problem in empirical research. We provide a novel framework for handling missing covariate data for estimation and inference in downstream tasks. Our general framework provides an automatic and easy-to-use pipeline for empirical researchers: First, missing values are imputed using a broad class of imputation methods under general observation patterns. Second, we automatically correct for the imputation bias and adaptively weight the imputed values according to their quality. Third, we use all available data, including imputed observations, to obtain more precise point estimates for the downstream task with valid confidence intervals. Our approach ensures valid inference while improving statistical efficiency by leveraging all available data. Through simulations, we demonstrate the superior performance of our approach over natural benchmarks, as it achieves both lower bias and variance while being robust to imputation quality. In a comprehensive empirical study of the dependence of equity markets on carbon emissions, we show that properly accounting for missing emissions data yields no evidence of correlation between stock returns and emissions directly produced by companies, but a negative correlation with value chain emissions.

Focus or Spread? Attention Allocation under High-Dimensional Uncertainty

Hayong Yun
,
Michigan State University

Abstract

This paper introduces the rotate-then-quantize (RTQ) strategy of Zandieh et al. (2025) to economics and finance as a general method for decision problems under high-dimensional uncertainty. RTQ carries a distribution-free performance guarantee that does not depend on the number of variables, and its signals are discrete Lloyd-Max quantizers, a direct realization of the conjecture Sims (2006) raised for the non-Gaussian case. The classical rational-inattention solution of Van Nieuwerburgh and Veldkamp (2010) is the optimum within a Gaussian-restricted signal class; RTQ handles the general problem. We develop the theorems applied work needs and illustrate the method on rational inattention. Under a broad class of priors, the RTQ optimum spreads attention across composite signals of every variable rather than concentrating on a few. On S&P 500 returns over 2000 to 2024, RTQ delivers welfare gains of roughly 74 basis points per year out of sample over the VNV concentration rule. Across 17,500 manager-quarters from the LSEG 13F panel from 1980 to 2024, observed attention is broad rather than corner-concentrated. A cross-domain test on the Fisman et al. (2006) speed-dating data confirms the scaling identity the theory predicts, which supports RTQ as a tool that travels beyond portfolio choice.

Designing Agentic AI-Based Screening for Portfolio Investment

Mehmet Caner
,
North Carolina State university
Agostino Capponi
,
Columbia University
Nathan Sun
,
Columbia University
Jonathan Tan
,
Columbia University

Abstract

"We introduce a new agentic artificial intelligence (AI) platform for portfolio management.

Our architecture consists of three layers. First, two large language model (LLM) agents are

assigned specialized tasks: one agent screens for firms with desirable fundamentals, while a

sentiment analysis agent screens for firms with desirable news. Second, these agents deliberate

to generate and agree upon buy and sell signals from a large portfolio, substantially narrowing

the pool of candidate assets. Finally, we apply a high-dimensional precision matrix estimation

procedure to determine optimal portfolio weights. A defining theoretical feature of our frame-

work is that the number of assets in the portfolio is itself a random variable, realized through

the screening process. We introduce the concept of sensible screening and establish that, under

mild screening errors, the squared Sharpe ratio of the screened portfolio consistently estimates

its target. Empirically, our method achieves superior Sharpe ratios relative to an unscreened

baseline portfolio and to conventional screening approaches, evaluated on S&P 500 data over

the period 2020–2024."

Discussant(s)
Shrihari  Santosh
,
University of Maryland
Jungjun  Choi
,
University of Rhode Island
Rui Da
,
Indiana University
Leland Bybee
,
University of Chicago
JEL Classifications
  • C5 - Econometric Modeling
  • D8 - Information, Knowledge, and Uncertainty