3.0

Average Course Rating

This course introduces dynamic machine learning techniques for forecasting and assessing uncertainty and risks in macroeconomics and finance, with an emphasis on methods designed to handle large datasets. The first part focuses on linear models, covering (i) univariate predictive regressions with many regressors; (ii) dynamic factor models, as a first example of popular multivariate models that can handle large datasets; and (iii) Bayesian VARs, as a second example of big data multivariate models that also serve as a bridge between reduced-form and structural models. Additional topics include state-space models, Monte Carlo methods, model comparison, and model selection. Applications will cover nowcasting and forecasting in macroeconomics and finance, portfolio selection, term structure models, scenario analysis, monetary policy transmission, and long-horizon forecasts. The second part extends the focus to monitoring and forecasting risk, which requires making inferences about the likelihood of all possible future contingencies. To conduct a meaningful analysis of risk, we will go beyond linearity, employing flexible machine learning methods that capture asymmetries and fat tails in predictive distributions, addressing the non-linear nature of risk dynamics. The workhorse method will be quantile regressions for assessing downside risks, covering and connecting other models such as distributional regression, nonparametric density estimation, mean-variance regression, and Markov-switching models. Applications will focus on assessing risks related to economic growth, inflation, the labor market, the housing market, and the financial market, with practical applications to macroeconomic and financial risk management for policy and industry.