Young Kwang KIM will defend his thesis on Friday 25 September 2026 at 03:00pm (online via Teams)
Title: Essays on Econometric Methods for Functional Data
Supervisors: Professeurs Nour Meddahi and Jean-Pierre Florens
To participate, please contact the doctoral school
Memberships are:
- Nour MEDDAHI : Professor in Economics, University Toulouse Capitole Supervisor
- Jean-Pierre FLORENS : Professor Emeritus, University Toulouse Capitole Co-Supervisor
- Marine CARRASCO : Professor in Economics, University of Montréal Rapporteure
- Senay SOKULLU : Professor in Economics, University of Bristol Rapporteure
- Jihyun KIM : Professor in Economics, Sungkyunkwan University Examinateur
- Anne VANHEMS : Professor of Statistics, Toulouse Business School Examinatrice
Abstract :
Economic data are increasingly observed as curves, trajectories, or high-dimensional objects rather than scalars: electricity prices over the hours of a day, temperature and precipitation across time and space, or cross-sections of asset returns. Reducing such data to scalar summaries discards the shape, timing, and heterogeneity that carry the economic content of interest. To exploit this structure, econometric modelling has moved toward an infinite-dimensional approach in which observations are elements of function spaces and parameters are operators. Functional data analysis provides the natural framework, treating observations as random elements of separable Hilbert spaces. Yet covariance operators are typically compact with unbounded inverses, making estimation ill-posed and requiring regularization, which introduces bias. Functional data analysis is, however, mainly reduced-form. Economists seek structural parameters with causal or counterfactual content, requiring identification, consistent estimation, and valid inference at the operator level. This thesis extends three classical econometric toolkits — factor models, instrumental variables, and simultaneous equations — to the functional setting.
Chapter 1: Instrumental Factor Models for High-Dimensional Functional Data. This chapter develops a factor model for panels of function-valued observations, with scalar latent factors and functional loadings depending on observed characteristics. The proposed Functional Projected-PCA projects the panel onto the span of the characteristics before applying PCA, removing idiosyncratic noise asymptotically and recovering latent factors consistently in large N, fixed T settings. The method is applied to European cereal markets through a factor-augmented VAR.
Chapter 2: Inference for Functional IV Regression with Possibly Weak Instruments. This chapter studies weak-instrument issues in functional IV regression, where the structural parameter is an integral operator. It analyzes the Tikhonov IV estimator under strong, semi-strong, weak, and unidentified regimes, and proposes the Functional Anderson–Rubin test, which remains robust under weak identification. The method is applied to intraday price elasticity of supply in Alberta’s electricity market.
Chapter 3: Identification of Functional Simultaneous Equations Models. This chapter extends simultaneous equations models to settings where endogenous and exogenous variables are function-valued and structural parameters are bounded linear operators. It develops identification and estimation methods using exclusion and recursive restrictions on operator-valued structural parameters. Together, the three chapters reconstruct the structural toolkit of econometrics at the operator level. The functional setting raises problems with no finite-dimensional analogue — ill-posed inverses, identification regimes tied to operator smoothness, and rank conditions expressed through ranges and null spaces — and the thesis shows that these problems admit tractable solutions.
