Latent Variable Autoregression with Exogenous Inputs

The preprint operates within the dominant paradigm of econometrics and time series analysis, specifically focusing on autoregressive models with exogenous inputs (ARX). It seeks to refine and extend this established model by incorporating latent variables.

This preprint introduces (C)LARX, a novel variant of the ARX methodology. This represents an incremental advancement within the existing paradigm of time series analysis by addressing limitations of traditional ARX models, namely the assumption of perfectly measured variables. The authors propose a method for handling latent variables, which are unobserved but can be approximated from observed data, and incorporate them within the ARX framework. This refinement seeks to improve the accuracy and interpretability of existing models. The paper doesn't challenge fundamental assumptions or propose radically new ways of thinking about economic relationships but instead offers a sophisticated tool for enhancing current analytical practices. Therefore, it is best classified as "normal science" because it contributes to the existing paradigm by solving puzzles and improving the precision of existing models.