Truth Triangulation: Advancing Robust Anomaly Detection and Macroeconomic Forecasting Using Machine Learning Model

The three stage framework whose main objective is to perform robust econometric estimation and inference for anomaly detection is proposed. The first stage focuses on rigorous econometric modeling to obtain a stable, theory, based benchmark against which we can measure the impact of our model. Thus, the econometric model has the double purpose of both setting a structurally grounded baseline and providing valid statistical relationships upon which the rest of the stages can be built. Next, a machine learning stage narrows down the search within the confirmed econometric space. Most importantly, the last stage uses Bayesian inference to rigorously evaluate the candidate anomalies in the context of the structural model, thus yielding interpretable posterior probabilities and credible intervals.