What do we mean by financial crisis? What are some classic approaches to predicting such crises? How can machine learning algorithms contribute to its prediction?
Financial crises take a variety of forms: they range from sovereign defaults to bank runs to currency crises. What these events all have in common is that an internal vulnerability gets worse over time and, after an associated trigger, leads to a financial crisis.
Determining the exact trigger can be difficult, so you should monitor the development of internal vulnerabilities. What exactly are these internal weaknesses? Statistically speaking, they are the explanatory variables in crisis models. In periods of historical crises, it has often acted as a response variable.
While this is part of the classic approach to modeling financial crises, it is not the only way to model financial risks.
In the classic crisis model, the standard method is to use logistic regression to estimate the probability of a financial crisis. The explanatory variables are related to the response variable by a nonlinear correlation function. The dependent variable is 0 for no crisis and 1 for crisis. This approach hinges on the definition of a financial crisis. The a priori variables are modeled with the aid of maximum likelihood by varying the exposure to the explanatory variables of the response variable. In machine learning terms, this is a supervised learning technique or a single hidden layer logistic regression. It is also known as a shallow neural network.
Determining the odds of default or crisis from market prices is among other crisis modeling methods. For example, from credit default swaps (CDS), the implied probability of default can be calculated. Of course, this is fundamentally different from both the logistic regression described above and the application of machine learning algorithms described below.
So what can machine learning algorithms do to better estimate the odds of a financial crisis? First, unsupervised learning differs from supervised learning in that there is no response variable. Clustering is one technique worth highlighting. The goal of clustering is to group data points in a reasonable manner. These datasets will be linked to a cluster center to help define the structure within the datasets. Clustering can be applied to both the dependent and the independent variable. Instead of using a fixed threshold to define currency crisis, for example, we can divide the currency returns into different groups and extract a plausible meaning from each group.
Machine learning algorithms can add significant value this way. While clustering is just one example of the strength of encryption, these algorithms have a number of other useful applications
Of course, while machine learning is simply an umbrella term for many useful algorithms, whether a machine actually learns is an entirely different question.
However, the division of time series into the training and test set remains among the major weaknesses of machine learning. How do you define split? The decision is often arbitrary.
Whatever these shortcomings, they hardly detract from the great benefits that machine learning can bring. In fact, now is the time to invest in these capabilities.
If you liked this post, don’t forget to subscribe Venture investor
All posts are the opinion of the author. As such, it should not be construed as investment advice, nor do the opinions expressed necessarily reflect the views of the CFA Institute or the author’s employer.
Photo credit: © Getty Images / noLimit46
Professional learning for CFA Institute members
CFA Institute members are empowered to report self-earned and self-report Professional Learning (PL) credits, including content on Venture investor. Members can easily score credits using their online PL tracker.