By Wei Ma, Head of Model Risk Management, Sumitomo Mitsui Banking Corp
Can you please tell the Risk Insights readers a little bit about yourself, your experiences and what your current professional focus is?
I started my financial career as a structurer in the securitization industry. During the great recession, I structured and managed the Maiden Lane II and Maiden Lane III transactions FRBNY set up to rescue AIG from ABS CDO and subprime mortgage bonds. Also during the same period, I started to get deeper into the risk analytics field.
I have been on the current role of head of model risk management at SMBC for over three years and have built the model risk management framework and team from the ground up. Now we have become a more mature team and expanded our reach within the bank globally, by providing model validation and model risk governance services.
What, for you, are the benefits of attending a conference like the Risk Americas and what can attendees expect to learn from your session?
Through the process of developing an MRM function from zero to becoming a core piece of SMBC Group’s global risk management practice, I have learned a lot, from technical, risk management, and people management perspectives. Through our interactions, I hope that you can learn a few things from me while sharing your experience.
A big discussion point across the industry is the use of technology to drive efficiency, how do you foresee technology impacting model risk?
Technology is changing the industry. Model risk management should adapt to it and take proactive actions to understand the benefit, the risk, and the limitations of the technology. The skillsets of the MRM team need to be updated to provide effective challenge.
In addition, MRM team should seek to take advantage of technology to improve efficiency, quality, and transparency of various MRM activities.
How does the treatment of supervised vs. non supervised machine learning models differ?
Supervised machine learning is heavily influenced by business intuitions and usually result in models better understood by business and more predictable in their performance. However, it is not common for this type of models to bring much new insights and sometimes the model can be biased due to the human intervention.
Unsupervised machine learning can sometimes result in models that are not very intuitive and hard to explain. It Is also very data dependent so anomaly in data can heavily impact the model. The flip side of it is that it can bring insights out of the data that are usually too deep for even experienced business manager to see.
When looking at end to end management of model risk, how can institutions demonstrate effective challenge of assumptions?
Assumptions need to be proven to be true. If not, rationales need to be provided.
It is also important to assess the impact of the uncertainty of assumptions. Usually that entails sensitivities analysis and benchmarking.
Another aspect of effective challenge is ongoing monitoring. The assumptions should be constantly monitored and tested to ensure that they hold true or the uncertainty will not exceed the tolerance level.
How do you see model risk management evolving over the next 6-12 months?
The industry continues to look into streamlining and enhancing governance;
The industry continues to look into improving efficiency in MRM activities while improving quality of work;
More technological solutions will be developed to support the drive to achieve higher efficiency and effectiveness.