Hello. My name is Joonho (Phil) Hwang (황준호; 黃俊晧), and I am a fourth-year PhD student in economics at Seoul National University, where I am fortunate to be advised by Professor Seojeong Lee. My research interests are in econometrics, with a particular focus on

• panel data models
• causal inference

Working papers

Online Updating for Linear Panel Regressions
joint with Seojeong Lee

Abstract

In this paper, we develop online updating methods for linear panel regression models. Online updating refers to procedures for sequentially updating parameter estimates as new data become available. In practice, the potential size of the dataset or data confidentiality constraints may preclude researchers from storing or accessing the entire dataset. We propose an online updating procedure for widely used linear regression models in panel data, where data expansion can occur through either (1) the arrival of new units or (2) the arrival of additional time periods for existing units. The proposed procedure yields closed-form expressions for updating both the point estimates and associated standard errors in each scenario.

What is online updating? See the example below:

presented at: SNU Econometrics Workshop, SETA 2025, University of Sydney, KERIC 2025, SNU Workshop on Recent Advances in Econometrics

Summary

Big picture. Many panel datasets grow over time as new units or new time periods are added. Re-estimating fixed effects regressions from scratch can be slow and memory-intensive, especially when the raw microdata cannot be permanently stored. This paper develops online updating methods for linear panel regressions that update the estimator sequentially using only low-dimensional summary statistics.

Method. We study static fixed effects models and other linear panel regressions with unbalanced panels. We derive closed-form updating formulas for coefficients and cluster-robust variances when (i) new units arrive or (ii) additional time periods are appended for existing units. For TWFE with many time dummies, we adapt the incremental SVD algorithm of Brand (2006) to maintain an economy SVD of the normal matrix and recover the updated coefficients without direct inversion.

Findings. In large simulated panels, the online SVD-based estimator matches the batch fixed effects estimator (from plm) up to machine precision, while using orders of magnitude less memory and much smaller per-unit update time. This makes recursive, numerically stable fixed effects estimation feasible even when the full panel cannot be stored.

Work in progress

Dynamic Misspecification in Extended Two-Way Fixed Effects Regression

Abstract

It is often unclear for applied researchers whether they should include lagged outcomes as regressors. When the outcome of interest is dynamically persistent but lagged outcomes are omitted, the regression model is misspecified and the resulting estimates can be biased. Following Angrist & Pischke (2009), many applied researchers rely on a so-called bracketing relationship, treating specifications with and without lagged outcomes as forming an informal robustness check on the magnitude of the causal effect. In this paper, we first show that this practical relationship breaks down in modern staggered two-way fixed effects (TWFE) settings. Instead, we propose a new method for estimating staggered TWFE models that explicitly accounts for the dynamic evolution of the outcome. </div> </div>