Credit risk

What machine learning can bring to credit risk management

As credit markets continue to evolve, banks can take advantage of products that use machine learning – software that allows banks to anticipate risk more effectively. But should banks revise their credit risk management processes accordingly and use these new solutions?

AI and machine learning for credit risk management

According to McKinsey, artificial intelligence and machine learning technologies could add up to $1 trillion in additional value to global banking services each year.

Financial institutions are using machine learning to make more accurate and consistent credit decisions while reducing risk, fraud, and costs. For example, Citi Bank recently transformed its critical internal audit using machine learning, which contributed to high-quality credit decisions.

On the other hand, the more complex and nuanced applications of these technologies have remained, until now, largely in the academic realm. Today, however, quants and risk managers are bringing these technologies to real-world applications, paving the way for the simplification of their daily routines.

Artificial neural network model

Artificial neural networks are an effective tool for modeling and analyzing complex systems. They have been widely used in many scientific fields, such as pattern recognition, signal processing, forecasting and system control.

In recent years, the artificial neural network model for credit risk has attracted more and more attention from researchers due to the advantages conferred by its nonlinearity, parallel computing, high fault tolerance and good generalization performance.

How does the artificial neural network model work?

Training the artificial neural network classifier requires the category label of the sample data to be known. This requires determining the actual credit rating of each company in the given year.

A new solution to this problem is the cluster analysis method, where all companies are grouped into several categories. Thinking that the credit risk of all companies is normally distributed, the dimension is reduced by the factor analysis method and the total factor score of each company is obtained.

The actual credit risk score for each category can then be determined based on the extent to which the total average score for each factor category deviates from the total average score for the entire factor. Next, the accuracy of commonly used traditional credit risk forecasting models is tested.

Ultimately, the prediction effect is compared to the pattern generated by the artificial neural network model.

With its significantly improved non-performing loan prediction accuracy, commercial banks can use the perceptron neural network model to make risk predictions for credit risk assessment, achieving good results.

Machine learning market generators

With pre-pandemic historical data no longer accurately representing current risk levels, the ability of market makers to gauge risk from a shorter time series is invaluable.

How do market generators work?

Risk models are calibrated on historical data. The longer the time horizon of a model, the longer the time series needed to calibrate the model.

With traditional risk models, the short length of pandemic-era time series data does not allow for accurate model calibration. The time series for a given currency, stock or credit name is too short to gain statistical confidence in the estimate. Because standard market models for credit risk, limits, insurance reserves, and macroeconomic investing measure risk years in advance, they require a long time series that extends to pre-existing data. -pandemics that are no longer representative of the current level of risk.

Market generators are machine learning algorithms to generate additional samples of market data when historical time series are of insufficient length without relying on preconceived notions about the data. They can generate the data for the time horizons between 1 and 30 years required by the risk models, allowing accurate measurement of credit risk in the era of the pandemic, limits, insurance reserves (generation economic scenarios) and the performance of the macro strategy.

Using unsupervised machine learning, market generators rigorously aggregate statistical data from multiple currencies, stocks, or credit names and then generate sample data for each name. This reduces the statistical uncertainty inherent in short time series while preserving the differences between the names and incorporating them into the model.

This unique capability of the model has been validated by comparing it with out-of-sample data – the gold standard of model validation.

Eliminate the risks of AI and machine learning

McKinsey partner Derek Waldron says while artificial intelligence and advanced analytics offer significant opportunities for banks to seize, it needs to be done in a way that puts risk management at the forefront of people’s minds as well. As in statistical modeling, it is important to focus on the following six areas when validating a machine learning model:

  • Interpretability
  • Bias
  • Feature Engineering
  • Hyperparameter Tuning
  • Preparation for production
  • Dynamic Model Calibration

The risk of machine learning models being biased is real because the models can overfit the data if not processed correctly. Overfitting occurs when a model appears to fit the data very well because it has been tuned to replicate the data very efficiently. In reality, it will not stand the test of time when the model goes into production and is exposed to situations it has not been exposed to before. A significant deterioration in performance will be observed.

Another example is feature engineering. In statistical model development, a model developer typically begins with several assumptions about the characteristics that determine the predictive performance of the model. These features can be provided by subject matter expertise or domain expertise.

In artificial intelligence, the process is a little different. The developer feeds a large amount of data into the AI ​​algorithm and the model learns the features that describe this data. The challenge in doing this is that the model can learn features that are quite counter-intuitive and in some cases the model can over-fit the data. In this case, the model validator must be able to examine the types of predictor variables that appear in the AI ​​model and ensure that they are consistent with intuition and are, in fact , predictors of the output.

Ultimately, we believe machine learning will continue to play an important role in identifying patterns and trends that can help financial institutions thrive.