Random forests and macro analysis
Philippe Goulet Coulombe develops a Macroeconomic Random Forest (MRF) algorithm which adapts the canonical Machine Learning (ML) tool to flexibly model evolving parameters in a linear macro equation. The output of the model, Generalized Time-Varying Parameters (GTVPs) is the result of nesting many popular specifications (structural breaks/change, threshold effects, regime-switching, etc.) and letting the data decide whichever combination of them is most suitable. His results suggest:
- In most instances, MRF forecasts are better than off-the-shelf ML algorithms and traditional econometric approaches;
- The output, namely the GTVPs, can be interpreted. For instance, the successful unemployment forecast is due to the influence of forward-looking variables (e.g., term spreads, housing starts) nearly doubling before every recession. Interestingly, the Phillips curve has indeed flattened, and its might is highly cyclical.
Ultimately, we get a new methodology leveraging the power of ML and big data to provide a modern take on the decades-old challenge of estimating latent states driving linear macroeconomic equations.
The Macroeconomy as a Random Forest
Author: Philippe Goulet Coulombe
From: University of Pennsylvania
Random forests and bank crises
In this paper Suss et al. use confidential UK regulatory data covering the period 2006-2012 to predict bank distress one year out, comparing and contrasting a number of machine learning techniques (random forest, boosted decision trees, K nearest neighbors, and support vector machines) with classical statistical approaches (logistic regression and random effects logistic regression). They find:
The random forest significantly and substantively outperforms the classical models and other machine learning techniques based on standard performance metrics: the AUC and Brier Score.
Predicting bank distress with machine learning
By: Joel Suss, Henry Treitel – Bank of England
Random forest and financial crises
In this paper, Fouliard et al. use online machine learning to run a horse race between very eclectic predictive models and aggregate them in order to produce an optimal forecast. The advantage of this methodology is summarized in:
Unlike in the classical statistical theory of sequential predictions, where the sequence of outcomes is assumed to be a realization of a stationary stochastic process, in our framework, pre-crises are the product of some unknown and unspecified mechanism, which could be deterministic, stochastic, or even adversarially adaptative to our own behavior. is allows us to make no assumptions on how the data are generated, which is a big advantage as there is no consensus on a theory of financial crisis. Indeed online machine learning is specifically geared at real-time prediction in situations where the true models driving outcomes are not known and can be different over time.
Through the use of a mix of 26 experts, some of them being central bank financial crises models, some of them being machine learning models, they are able to predict a systemic financial crisis 3 year ahead out-of-sample with relatively low signal-to-noise ratios for France, UK, Germany, and Italy.
Answering the Queen: Machine Learning and Financial Crises
Authors: Jeremy Fouliard, Michael Howell, Hélène Rey
From: Paris School of Economics, CrossBorder Capital, London Business School
