I am a researcher combining economics and machine learning, on the job market 2025–26. My focus is on developing a Transformer framework for macroeconomic analysis.

My job market paper proposes a Transformer architecture for estimating non-linear dynamic factors from time series data, requiring minimal strict identifying assumption. The results are improved substantially on small datasets by using a conventional factor model as prior information to guide the training. Prior information is supplied into the loss function through a regularization term. The results are interpreted with Attention matrices, which show the relative importance of each variable's every lag on the output. Changes in the Attention patterns over time can help identify regime switches, analyze shocks and evaluate narratives. A Monte Carlo experiment suggests that the Transformer is more accurate than the linear factor model, when the data deviates from linear-Gaussian. An empirical application uses the Transformer to construct a coincident index for the real economic conditions of the United States.