Program
2026 Kansas Econometrics Workshop
Date: May 2 (Saturday), 2026
Venue: Centennial Room in KU Union on the KU campus
07:30-08:15 Breakfast (Centennial Room in KU Union, the sixth floor)
08:20-08:30 Opening Remark (Centennial Room in KU Union, the sixth floor)
Jie Zhang, Associate Dean, College of Liberal Arts and Sciences, KU
Session I (Theoretical Econometrics): Chair, Zongwu Cai, University of Kansas
[1] 08:30-09:00 Chihwa Kao, University of Connecticut
“Robust Cross-Sectional Independence Testing in High-Dimensions Under Temporal Dependence”
[2] 09:00-09:30 Zhijie Xiao, Boston College
“Functional Quantile Regressions in Economics and Finance”
[3] 09:30-10:00Yingying Fan, University of Southern California
“LLM-Powered Prediction Inference with Online Text Time Series”
10:00-10:30 Coffee Break (Centennial Room in KU Union, the sixth floor)
Session II (Financial Econometrics): Chair, Zongwu Cai, University of Kansas
[4] 10:30-11:00 Marcelle Chauvet, University of California at Riverside
“Revisiting the Relationship between Geopolitical Risk and Oil Price Realized Volatility: A Markov-Switching Analysis”
[5] 11:0-11:30 Ethan Chiang, University of North Carolina at Charlotte
“A Portfolio-based Evaluation of Multivariate Conditional Volatility Models”
[6] 11:30-12:00 Zeqin Liu, Shanxi University of Finance and Economics
“An LLM Approach to Study Forward Guidance Frictions by Decoding China’s Central Bank Communications”
12:00-14:00 Lunch (Centennial Room in KU Union)
Session III (Regression Methods): Chair, John Keating, University of Kansas
[7] 14:00-14:30 David T. Jacho-Chavez, Emory University
“Regression with Observational Multilayered Network Data”
[8] 14:30-15:00 Amaze Lusompa, Federal Reserve Bank of Kansas City
“Regression Model Selection Under General Conditions”
[9] 15:00-15:30 Shahnaz Parsaeian, University of Kansas
“Mixed-Frequency Panel Regressions with Sparse and Heterogeneous
Structures”
15:30-16:00 Coffee Break (Centennial Room in KU Union)
Session IV (Time Series): Chair, John Keating, University of Kansas
[10] 16:00-16:30 Yiru Wang, University of Pittsburg
“Heterogeneous Local Projections”
[11] 16:30-17:00 Ali Mehrabani, University of Kansas
“Latent Group Structures in Functional Coefficient Panel Data with Time-Varying Interactive Fixed Effects”
17:30-20:30 Reception and Dinner (Kansas Room in KU Union, the sixth floor)
ABSTRACT:
Robust Cross-Sectional Independence Testing in High-Dimensions Under Temporal Dependence
Chihwa Kao, University of Connecticut
This paper develops a robust test of testing the cross-sectional independence that accommodates temporal dependence in high-dimensional panel data. Our approach derives explicit adjustments for the mean and variance of the trace statistic Tr(R²) under general Toeplitz dependence structures, where R is the sample correlation matrix. These adjustments restore the test's size control without compromising power, thereby enabling reliable inference in settings where traditional approaches fail. This is the first paper to establish a CLT for Tr(R²) under serially correlated errors, and to express the mean and variance adjustments.
Functional Quantile Regressions in Economics and Finance
Zhijie Xiao, Boston College
Empirical studies indicate that the future behavior of many economic and financial variables can be heavily dependent on the past distributional information of itself or other variables. Distributional information helps policymakers understand the distributional consequences of their actions, informs economic analysis, and can be used to design more equitable and effective policies. In finance, cumulated evidence indicates that tail distributional information of financial variables can affect future behavior of many financial and economic variables. Information of stock return distributions deliver important information for risk management and investment. In macroeconomic analysis, it is well-known that the income distribution in the past have important influence on future consumption and other economic activities in the economy. More generally, the dynamics of the distributions of consumption, income, and wealth are central to understanding macroeconomic dynamics and the transmission of monetary and fiscal policy. We believe that directly modeling such kind of dependence relationship based on the distribution/quantile functions can provide a useful alternative to address some of these issues. Time series models with functional relationships provide a useful approach in capturing dependence structure over distributions. In this paper, we study functional dynamic relationships using a functional quantile regression model which allows information from the whole previous distribution affecting future behavior of the time series. This is a join work with Guojun Wu.
LLM-Powered Prediction Inference with Online Text Time Series
Yingying Fan, University of Southern California
Time series prediction inference is an important yet challenging task in economics and business, where existing approaches often rely on low-frequency, survey-based data. With the recent advances of large language models (LLMs), there is growing potential to leverage high-frequency online text data for improved time series prediction, an area still largely unexplored. This paper proposes LLM-TS, an LLM-based approach for time series prediction inference incorporating online text data. The LLM-TS is based on a joint time series framework that combines survey-based low-frequency data with LLM-generated high-frequency surrogates. The framework relies only on an error correlation assumption, combining a text-embedding-augmented ARX model for the observed gold-standard measurements with a VARX model for the LLM-generated surrogates. LLM-TS employs LLMs such as ChatGPT and the trained BERT models to construct LLM surrogates. Online text embeddings are extracted via LDA and BERT. We establish the asymptotic properties of the method and provide two forms of constructed prediction intervals. To demonstrate the practical power of LLM-TS, we apply it to a critical real-world example: inflation forecast. We collect a large set of high-frequency online texts from a widely used Chinese social media platform and employ LLMs to construct inflation labels for posts that are related to inflation. The finite-sample performance and practical advantages of LLM-TS are illustrated through simulations and this noisy real data example, highlighting its potential to improve time series prediction in economic applications. This is a joint work with Jinchi Lv, Ao Sun and Yurou Wang.
Revisiting the Relationship between Geopolitical Risk and Oil Price Realized Volatility: A Markov-Switching Analysis
Marcelle Chauvet. University of California at Riverside.
This paper studies the impact of geopolitical risks on oil price volatility over the past three decades, allowing us to specify this relationship in the long run during both calm and turbulent times. Specifically, we assess the reaction of oil price volatility to geopolitical risks and track how this reaction has evolved over time, considering various shocks and geopolitical tensions (such as oil price shocks, political revolutions, wars, trade conflicts, economic sanctions, etc.). We also test whether geopolitical risks (acts and threats) can improve the forecast of oil price volatility. Our results show that oil price fluctuations are significantly sensitive to geopolitical risks. Furthermore, when considering structural breaks and nonlinearity, the proposed Dynamic Factor Markov Switching model with breaks outperforms linear specifications in forecasting oil price volatility. This is a joint work with Fredj Jawadi.
A Portfolio-based Evaluation of Multivariate Conditional Volatility Models
Ethan Chiang, University of North Carolina at Charlotte
This paper proposes a portfolio-based evaluation of multivariate conditional volatility models. The proposed method aligns the portfolio strategy (Ferson-Siegel unconditionally efficient portfolio) with the out-of-sample evaluation criteria (unconditional Sharpe ratio). Simulation results suggest that our approach is more effective in identifying the true data-generating model than alternative combinations in which the strategy and the evaluation measure are misaligned, as is often the case in the existing literature. When implemented in actual data, our method leads to superior out-of-sample volatility timing performance, and clients are willing to pay 2.90% to 13.71% to switch from traditional methods to our method. A performance attribution analysis shows that the superior performance is largely due to the use of the correct conditional covariance model identified by our method, highlighting the economic significance of covariance model selection. This is a joint work with Pierluigi Balduzzi, Nancy Mo and Zoe Nie.
An LLM Approach to Study Forward Guidance Frictions by Decoding China’s Central Bank Communications
Zeqin Liu, Shanxi University of Finance and Economics
Expectation management is an important yet challenging task for central banks, particularly under China's dual-track regulation and multi-objective constraints. To decode complex policy signals, this paper proposes a large language model (LLM)-powered analytical framework for central bank communication. We employ large language models to perform dynamic semantic segmentation across tense and topic dimensions, constructing indices for forward-looking net policy intention, price-quantity signal divergence, and multi-objective communication dispersion. Incorporating these LLM-generated signals into local projections, we systematically identify the transmission of forward guidance. We find that forward communication guides credit adjustments, with the Loan Prime Rate dynamic multiplier reaching 0.258 after one year. However, institutional frictions severely disrupt transmission. Price-quantity signal divergence reduces the LPR response by 0.210, neutralizing the public signal's coordination function. Furthermore, multi-objective communication causes short-term funding rates to overshoot by 1.364 and widens credit spreads by 0.9. Enhancing tool coordination and reducing communication dispersion are critical for restoring true pricing capacity. This is a joint work with Zongwu Cai and Ying Fang.
Regression with Observational Multilayered Network Data
David T. Jacho-Chavez, Emory University
A novel method to estimate social effect coefficients in the popular so-called linear-in-means regression model in the Social Sciences is presented here that utilizes non-experimental multidimensional network data. The procedure can accommodate social interactions that correlate with the error in the model by making use of a different set of networks links among the same observations that are exogenous in the traditional sense. In particular, the full observability of a two-layered multiplex network data structure is assumed here to propose a new Generalized 3-Stage Least Squares (G3SLS) estimator that is consistent, asymptotically normally distributed and also easy to implement using widely used existing statistical software because of its closed-form definition. The underlying assumptions are general enough to accommodate common problems with observational data such as measurement error, simultaneity, and unobserved heterogeneity. Monte Carlo exercises confirm the good small sample performance of the proposed G3SLS estimator in these scenarios. An empirical application finds a positive and significant peer effects in citations among research articles published in top general-interest journals in economics. This is a joint work with Juan Estada, Kim P. Huynh and Leonard Sanchez-Aragon.
Regression Model Selection Under General Conditions
Amaze Lusompa, Federal Reserve Bank of Kansas City
Model selection criteria are one of the most important tools in statistics. Proofs showing a model selection criterion is asymptotically optimal are tailored to the type of model (linear regression, quantile regression, penalized regression, etc.), the estimation method (linear smoothers, maximum likelihood, generalized method of moments, etc.), the type of data (i.i.d., dependent, high dimensional, etc.), and the type of model selection criterion. Moreover, assumptions are often restrictive and unrealistic, making it a slow and winding process for researchers to determine if a model selection criterion is selecting an optimal model. This paper provides general proofs showing asymptotic optimality for a wide range of model selection criteria under general conditions. This paper not only asymptotically justifies model selection criteria for most situations, but it also unifies and extends a range of previously disparate results.
Mixed-Frequency Panel Regressions with Sparse and Heterogeneous
Structures
Shahnaz Parsaeian, University of Kansas
This paper develops a mixed frequency panel regression framework for nowcasting and
forecasting a low frequency outcome using a large set of high frequency predictors. We
propose a method that captures both sparsity in distributed lag predictors and heterogeneity across cross sectional units through latent group structures. In this setting, slope coefficients are homogeneous within groups but heterogeneous across them. To estimate the model, we introduce a doubly penalized least squares estimator that simultaneously selects the relevant high frequency predictors and uncovers the underlying group structure without prior knowledge of the number of groups or sparsity patterns. We establish oracle properties for the estimator and show that it consistently identifies both the relevant predictors and the group memberships in large samples. Monte Carlo experiments demonstrate strong finite sample performance. An empirical application to U.S. metropolitan statistical area housing prices illustrates the gains in mixed frequency nowcasting and forecasting achieved by incorporating sparsity and group heterogeneity.
Heterogeneous Local Projections
Yiru Wang, University of Pittsburg
In this paper, we propose a novel methodology to estimate the dynamic effects of structural aggregate shocks on individual outcomes in the presence of a latent group structure. The methodology is based on a panel local projection estimator that allows researchers to assign individuals to the correct group and estimate group-specific dynamic responses, thus enabling researchers to estimate heterogeneous local projections. Monte Carlo simulations show that, in small samples, the methodology successfully identifies the group structure with high probability and delivers reliable inference. This is a joint work with Atsushi Inoue and Barbara Rossi.
Latent Group Structures in Functional Coefficient Panel Data with Time-Varying Interactive Fixed Effects
Ali Mehrabani, University of Kansas
This paper proposes a framework for the joint estimation and identification of latent group structures within a time-varying functional coefficient panel data model that incorporates interactive fixed effects where the factor loadings are allowed to change over time. The model’s latent group structure enables the classification of individuals into distinct groups, where both the number of groups and individual group memberships are unknown. The smoothly time-varying coefficient functions and factor loadings are approximated using B-splines. A two-step penalized estimation procedure, employing a pairwise fusion penalized approach, is introduced to detect the latent group structure and estimate the functional coefficients. An Alternating Direction Method of Multipliers algorithm is developed to facilitate implementation. We establish the uniform consistency, and uniform classification consistency of the estimator, as well as the oracle property of the post-classification estimator. The asymptotic distribution of the post-classification estimator is derived. The efficacy of the proposed method is demonstrated through simulation studies and an empirical application.