What ML techniques can be used in digital transformation and risk integration? Alla Gil, CEO, Straterix explores
New technologies, many of them AI and ML based, are revolutionizing risk management. They are already successfully applied to consumer finance, compliance, anti-money laundering (AML), and cyber risk. However, application of these techniques to strategic risk tasks, such as allocating capital and defining risk appetite, have been lagging behind. Many senior risk management professionals are against applying ML-based models.
There are good reasons for such cautiousness.
Navigating risks in unprecedented market conditions is like playing a chess game where the rules, the number of pieces and the size of the board are changing frequently and without warning.
No existing AI methods can adapt to such environment without human intervention. But with tremendous increase of data volume, the human mind cannot possibly process all the potential combinations of dependencies, changing correlations and their consequences. Thus, the hybrid approach integrating computer power and human judgement is a must.
This is where transparent, explainable and adaptable ML methods come to the rescue.
The most efficient use of these techniques is not to replace the decision makers and model developers but to expand their intuition to cover situations that have never been encountered in the past.
Navigating risks in unprecedented market conditions is like playing a chess game where the rules, the number of pieces and the size of the board are changing frequently and without warning. No existing AI methods can adapt to such environment without human intervention.
It doesn’t mean they must be simple. Since the problems they are addressing are extremely complex, the answers cannot possibly be too easy. Addressing such complex issues can be done in two possible ways. One is a black-box solution trained on big data and suitable for algorithmic or high frequency trading, fraud detection and other split second-based decisions. It doesn’t work for decision making. This is the main reason for not using ML-techniques in strategic risk management. The other solution is the modular approach where the inputs and outputs of each module are observable and intuitive. They can be verified, challenged, and controlled (or corrected) by modelers and business experts.
This means using the data prior to a crisis, calibrate models and project model outcomes through the crisis period. Then compare these results against the actual historical performance as shown in Figure 1.
Figure 1. Backtesting Model Results Against Historical OutcomesSource: Straterix
And finally, the modeler and business expert can verify the risk drivers suggested as explanatory ones using ML-based methods such as regression with regularization.
They can reject the variables that they view as spurious and review the results. If goodness of fit substantially declines after such rejection, it tells model developers that they might have overlooked the variables explaining the outliers and other major deviation from the historical trends. At the same time, exhaustive cross-validation embedded in such ML methodologies helps reduce dimensionality and avoid overfitting.
This approach expands the modelers’ limited experience in the new market regime, while still offering them the flexibility to test, accept or reject the drivers suggested by the ML algorithm.
At the same time, exhaustive cross-validation embedded in such ML methodologies helps reduce dimensionality and avoid overfitting. This approach expands the modelers’ limited experience in the new market regime, while still offering them the flexibility to test, accept or reject the drivers suggested by the ML algorithm.
Once the risk drivers have been identified and equations driving revenue segments, loan levels and deposit volumes have been estimated, the full range scenario analysis can be applied.
It consists of a combination of advanced simulation techniques that incorporate market, credit, climate and operational shocks and the ripple effects of these shocks that naturally produce historically unprecedented scenarios.
The next steps include reverse scenario analysis and the use of clustering techniques to identify early warning indicators (EWIs).
This means identifying the variables that are highly correlated with the specific KPI outcomes (like stressed capital and liquidity ratios) a few quarters prior to the realization of such outcomes.
This enables the risk managers to analyze these EWIs and proactively develop contingency plans. Figure 2 demonstrates an example of a heatmap that depicts correlations with the 99th percentile worst-case tier I capital ratios of a bank.
Figure 2. Heatmap with the Correlations between KPIs and Risk Drivers
These explainable and transparent ML technologies can have great immediate impact on risk management processes and help adapting to extreme economic and market uncertainty.
Alla Gil is co-founder and CEO of Straterix, which provides unique scenario tools for strategic planning and risk management. Prior to forming Straterix, Gil was the global head of Strategic Advisory at Goldman Sachs, Citigroup, and Nomura, where she advised financial institutions and corporations on stress testing, economic capital, ALM, long-term risk projections and optimal capital allocation.
The current environment of unprecedented uncertainty presents a perfect opportunity for accelerating digital risk initiatives without overreliance on technology. Risk and finance functions of organizations must reconsider what factors are driving their revenues and what scenarios they should select for stress testing their capital and liquidity positions. In the process, they need to answer the following questions: 1) What are the proper data mining and machine learning techniques that can be used by risk managers to support the development of new models? 2) How to identify the outdated models? 3) How to combine the modeler’s expertise, experience, and intuition with ML-based outcomes?