Summary
SUMMARY OF 2026 KANSAS ECONOMETRICS WORKSHOP
Section I (Theoretical Econometrics)
Chihwa Kao – “Robust Cross-Sectional Independence Testing in High-Dimensions Under Temporal Dependence”
Chihwa Kao is a Professor of Economics at the University of Connecticut. His research focuses on econometric theory, panel data econometrics, and time series econometrics, with particular emphasis on dependence structures in large-dimensional data. At the 2026 Kansas Econometric Workshop, he presented the paper entitled “Robust Cross-Sectional Independence Testing in High-Dimensions Under Temporal Dependence.” In his presentation, he introduced a robust LM-type test for cross-sectional independence in high-dimensional panel data that allows for temporal dependence. The main contribution of the paper is to extend classical cross-sectional independence testing by allowing a Toeplitz temporal covariance structure. The proposed statistic is based on the trace of the squared sample correlation matrix, but its centering and scaling are corrected using spectral moments of the temporal covariance matrix. The paper uses a Wick/Isserlis moment approach rather than standard random matrix theory, which yields explicit mean and variance corrections and improves finite-sample size control.
Zhijie Xiao – “Functional Quantile Regressions in Economics and Finance”
Zhijie Xiao is a Professor of Economics at Boston College. His research focuses on econometrics, empirical finance, quantile regression, and time series models. At the 2026 Kansas Econometrics Workshop, he presented the paper entitled “Functional Quantile Regressions in Economics and Finance.” In his presentation, he introduced a functional quantile regression framework for modeling dynamic relationships in which the future behavior of economic and financial variables depends on past distributional information. The main contribution of the paper is to develop a functional structural model that incorporates both forward-looking expectations and micro-level heterogeneity in consumption and income dynamics. The model treats consumption and income distributions as functional objects and links them with aggregate macroeconomic variables through a reduced-form functional VAR. The empirical application to U.S. macroeconomic and household data shows that micro distributional shocks generate heterogeneous macroeconomic effects and that explicitly modeling forward-looking expectations strengthens micro-to-macro transmission.
Yingying Fan – “LLM-Powered Prediction Inference with Online Text Time Series”
Yingying Fan is Centennial Chair in Business Administration and Professor in Data Sciences and Operations Department at the USE Marshall of Business. Her research focuses on statistics, data science, artificial intelligence, machine learning, and applications in economics and business. At the 2026 Kansas Econometrics Workshop, she presented the paper entitled “LLM-Powered Prediction Inference with Online Text Time Series.” In her presentation, she introduced a framework that uses high-frequency online text data to improve time series prediction inference while preserving the interpretability of standard economic models. The main contribution of the paper is not to replace existing interpretable economic models, but to augment them with LLM-generated high-frequency surrogates. The paper uses online text embedding and LLM-based labels to construct surrogate inflation measures and combines a text-embedding-augmented ARX model for official measurements with a VARX model for the LLM-generated surrogates. The key modeling idea is an error-coupling structure, which allows the surrogate information to shorten prediction intervals while maintaining valid inference. The method is applied to inflation forecasting using online text data, and the results show that the proposed approach can produce shorter and more informative forecasts than traditional AR and ARX models, especially in noisy data environments.
Section II (Financial Econometrics)
Marcelle Chauvet – “Revisiting the Relationship between Geopolitical Risk and Oil Price Realized Volatility: A Markov-Switching Analysis”
Marcelle Chauvet is a Professor of Economics and the Chair of the Department of Economics at the University of California, Riverside. She also serves as the Director and Executive Secretary of the International Association for Applied Econometrics (IAAE), the Director of the Rimini Centre for Economic Analysis (RCEA) Headquarter, the Director of the Center for Research on Economic and Financial Cycles (CREFC), and the Editor-in-Chief of the Journal of Business Cycle Research. Additionally, she is a member of the Brazilian Business Cycle Dating Committee (CODACE). At the 2026 Kansas Econometrics Workshop, she presented the paper entitled “Revisiting the Relationship between Geopolitical Risk and Oil Price Realized Volatility: A Markov-Switching Analysis.” In this work, she explores the impact of geopolitical risks on oil price volatility over the past three decades. She argued that traditional linear models relying solely on market fundamentals (supply and demand) are no longer sufficient to forecast oil price volatility because the market exhibits non-linear features like abrupt changes and structural breaks. By introducing advanced non-linear frameworks，specifically the Dynamic Factor Markov-Switching model with breaks (DFMSB)，the study successfully captures the asymmetric and state-dependent effects of geopolitical acts and threats on oil volatility. The empirical results show that incorporating geopolitical information into these non-linear models reduces prediction errors by about 50% and increases forecasting accuracy two- to five-fold compared to traditional linear benchmarks.
I-Hsuan Ethan Chiang – “A Portfolio-based Evaluation of Multivariate Conditional Volatility Models”
I-Hsuan Ethan Chiang is the Associate Dean for Graduate Programs and a Professor at the Belk College of Business, University of North Carolina at Charlotte. He previously served as the Chair of the Finance Department and holds a Ph.D. in Finance from Boston College. His research focuses on empirical asset pricing, portfolio management and performance evaluation, and fixed income securities, with publications in leading scholarly journals such as the Journal of Finance. At the 2026 Kansas Econometrics Workshop, he presented the paper entitled “A Portfolio-based Evaluation of Multivariate Conditional Volatility Models.” In his presentation, he provided a methodological critique of the common industry practice for evaluating and selecting multivariate conditional volatility models. He argued that the traditional plug-in approach is fundamentally flawed because it minimizes conditional variance but evaluates performance using the unconditional Sharpe ratio, creating a structural misalignment. To resolve this, he proposed using the Ferson-Siegel unconditionally efficient (UE) portfolio strategy, perfectly aligning the portfolio construction strategy with the final performance evaluation metric. Applied to real-world data, this dynamic model-timing framework generates substantial economic value, yielding a 5.01% risk-adjusted annualized excess return, with 85% of the gains directly attributed to correctly selecting the superior volatility model. The research highlights that investors would be willing to pay significant management fees (ranging from 2.90% to 13.71%) for the superior performance unlocked by this aligned evaluation method.
Zeqin Liu – “An LLM Approach to Study Forward Guidance Frictions by Decoding China’s Central Bank Communications”
Zeqin Liu is an Associate Professor at the School of Statistics, Shanxi University of Finance and Economics, and a postdoctoral visiting scholar at the University of Kansas. She holds a Ph.D. in Quantitative Economics from the Wang Yanan Institute for Studies in Economics (WISE) at Xiamen University. Her research primarily focuses on macroeconomic policy evaluation, econometric theory and applications, and the application of big data in economics and finance. Recognized as a "Shan Cai Scholar - Academic Leader", she has published in top-tier academic journals such as the Economic Research Journal and currently leads a National Natural Science Foundation of China (NSFC) Youth Project. At the 2026 Kansas Econometrics Workshop, she presented the paper entitled “An LLM Approach to Study Forward Guidance Frictions by Decoding China’s Central Bank Communications.” This research dissects the communication challenges of the People's Bank of China (PBOC) and explains the efficacy gap in its forward guidance. The study identifies two major institutional frictions: the dual-track monetary policy system and multi-policy objectives. To overcome the limitations of traditional text analysis, the researchers developed an innovative Large Language Model (LLM)-powered framework to decode PBOC communication and construct indices of net policy stance, quantity-price signal divergence, and multi-objective communication dispersion. Using local projection models, the findings reveal that while forward guidance successfully works through the credit channel, quantity-price signal conflicts completely offset this transmission. Furthermore, multi-target communication triggers defensive overshooting in the debt market and dampens the pricing sensitivity in the equity market. The study suggests that aligning quantity-price signals and focusing on a single policy mainline are crucial for enhancing policy efficacy.
Section III (Regression Methods)
David T. Jacho-Chavez - “Regression with Observational Multilayered Network Data”
David T. Jacho-Chavez is Professor and Director of Graduate Studies in the Department of Economics at Emory University. His research spans both theoretical and applied econometrics and statistics, with a current focus on network analysis. At the 2026 Kansas Econometrics Workshop, he presented the paper entitled “Regression with Observational Multilayered Network Data”, where he introduces a new econometric method to estimate peer effects in linear-in-means models using observational multilayer network data, addressing the challenge of endogenous social networks. The key idea is to exploit a two-layer (multiplex) network structure, where one network is endogenous and another is predetermined and exogenous, allowing identification without modeling network formation. He proposes a Generalized Three-Stage Least Squares (G3SLS) estimator, which is consistent, asymptotically normal, and easy to implement, and shows that identification relies on conditions such as exogeneity of the second network and sufficient variation between layers. Monte Carlo simulations demonstrate that G3SLS performs well compared to existing methods, especially in handling issues like measurement error, misclassification, and unobserved heterogeneity. An empirical application using economics journal articles finds positive and significant peer effects in citation patterns through co-authorship networks, highlighting the practical relevance of the approach.
Amaze Lusompa - “Regression Model Selection Under General Conditions”
Amaze Lusompa is an Economist in the Economic Research Department of the Federal Reserve Bank of Kansas City. His primary research interests are in econometrics and macroeconomics. At the 2026 Kansas Econometrics Workshop, he presented the paper entitled “Conditional Rank-Rank Regression”. In his presentation, he studies model selection criteria, such as AIC, BIC, and cross-validation, and addresses the lack of general proofs establishing their asymptotic optimality. He proposes a simple and general proof framework that applies to a wide range of regression models, estimation methods, and data types, showing that under standard assumptions, like exogeneity and a law of large numbers, these criteria will asymptotically select models with the same conditional mean. The key idea is to decompose the model’s mean squared error into components and prove that the cross-term converges to zero, overcoming difficulties in prior literature where dependence and complex structures made proofs challenging. Overall, the main takeaway is that widely used model selection methods are theoretically justified in large samples, as they consistently identify the best approximating model.
Shahnaz Parsaeian – “Mixed-Frequency Panel Regressions with Sparse and Heterogeneous Structures”
Shahnaz Parsaeian is an Associate professor of economics at the university of Kansas. She specializes in econometrics and is widely published in several top-tier field journals, including Journal of Applied Econometrics, Oxford Bulletin of Economics and Statistics, and Advances in Econometrics. At the 2026 Kansas Econometrics Workshop, she presented the paper entitled “Mixed-Frequency Panel Regressions with Sparse and Heterogeneous Structures”. In this work, she proposes a flexible panel MIDAS framework to handle mixed-frequency data (e.g., monthly predictors and quarterly outcomes) while addressing heterogeneity and high dimensionality. Instead of imposing restrictive lag structures, the paper proposes an unrestricted model combined with a doubly penalized least squares (DPLS) estimator that simultaneously selects relevant variables and lags and identifies latent group structures across units. The method achieves strong theoretical properties, including consistent recovery of group structure and an oracle property, and performs well in simulations. An empirical application to U.S. metropolitan housing prices shows that the approach significantly improves nowcasting and forecasting accuracy by effectively leveraging high-frequency information and cross-sectional heterogeneity, outperforming traditional aggregation and standard MIDAS models.
Section IV (Time Series)
Yiru Wang – “Heterogenous Local Projection”
Yiru Wang is an assistant professor of economics at the University of Pittsburgh. Her research interests include econometrics, time series, forecasting, and empirical macroeconomics. At the 2026 Kansas Econometrics Workshop, she presented the paper entitled “Heterogeneous Local Projections,” coauthored with Atsushi Inoue and Barbara Rossi. In this paper, they studied how to estimate dynamic effects of aggregate structural shocks on individual-level outcomes when individuals belong to unobserved latent groups with heterogeneous responses. Their method uses a two-step procedure: first, a grouped fixed-effects or k-means-type clustering step estimates latent group membership and group-time effects; second, a local projection step estimates group-specific impulse response. The paper shows that the classification error vanishes asymptotically, so the estimator behaves as if the true groups were known. It also establishes consistency, oracle equivalence, and asymptotic normality for the impulse response estimators, allowing valid inference for IRFs and multipliers. Monte Carlo simulations suggest that the method selects the correct number of groups with high accuracy, achieves near-oracle classification when the time dimension is large, and delivers confidence intervals with coverage close to nominal levels.
Ali Mehrabani – “Latent Group Structures in Functional Coefficient Panel Data with Time-Varying Interactive Fixed Effects”
Ali Mehrabani is an assistant professor of economics at the University of Kansas. His research interests include econometrics, high-dimensional statistics, data analytics, latent structures, model averaging in panel data models, and systems of equations with applications in business and economics. At the 2026 Kansas Econometrics Workshop, he presented the paper entitled “Latent Group Structures in Functional Coefficient Panel Data with Time-Varying Interactive Fixed Effects.” In this paper, he considered a panel data model in which slope coefficients vary both across individuals and over time, while factor loadings are also allowed to be time-varying. The paper proposes a two-step estimation framework to recover the time-varying coefficients, estimate latent group structures, and handle interactive fixed effects. The first stage uses nuclear norm penalized sieve estimation to estimate unknown parameters and the number of common factors; the second stage applies a post-NNPS procedure with pairwise shrinkage to identify latent groups, followed by post-Lasso estimation. The main theoretical results establish consistency, uniform classification consistency, and the oracle property, meaning that the estimator can asymptotically recover the true number of groups and true group membership. Overall, the paper contributes to the literature on heterogeneous panel data by combining smooth time variation, latent group structures, and time-varying interactive fixed effects in one estimation framework.