Two fundamental goals for any quantitative model are to forecast new observations accurately, and to correctly represent the structure of the system being studied—which variables are linked to which others, how. Forecasting allows for anticipating what will (probably) happen; getting the structure right allows for counterfactual prediction (that is, prediction of “what would happen if”). For both purposes, forecasting and counterfactuals, macroeconomists have, over the last three decades, increasingly relied on dynamic stochastic general equilibrium (DSGE) models, supposedly because of their predictive powers but especially for their ability to support counterfactual prediction. This faith is misplaced. It is by now well known that current DSGE models forecast very poorly. This however leaves open the question of what one should do instead, since not making predictions and sheer guesswork are not viable options. This project develops a suite of statistical modeling tools for macroeconomic forecasting which can serve either for prediction in their own right or as baselines against which to evaluate macro models with more economic content (“microfounded” or otherwise). This project also tackles the harder problem of structure and counterfactuals and draws on statistical methodology designed to evaluate whether models of causal structure are compatible with observations.
High-Dimensional Statistics for Macroeconomic Forecasting