By Richard Harmon, Managing Director, Financial Services at Cloudera
It is a well known fact that the UK is going into a recession and it can feel like everywhere you look there are predictions of doom and gloom when it comes to the financial implications of the COVID-19 pandemic. In the Financial Services sector in particular, 86% of profit warnings in the first seven months of 2020 cited Covid-19. Then if you add on Brexit, the uncertainties surrounding the future of the UK economy intensify. Individually both of these are highly disruptive events, however, together they leave a number of unknowns: the pandemic has already almost lasted a year, when will it end? What will the financial fallout from the many sectors that have been hit hardest be? What will the final Brexit deal be and how will that impact trade costs? How resilient is the UK economy in the longer term? There are a lot of questions that need to be answered.
However, it's not all bad news. We have also seen massive strides in the advancement of technology in the past decade, and more than ever we have tools at our disposal to help weather this storm. One of the key advancements has been in machine learning models, which have been used to predict and prepare for change. However, the unprecedented nature of COVID-19 and its global impact, means that these models are facing their biggest challenge yet – predicting the unpredictable.
Preparing for the unknown
For financial institutions, Machine Learning (ML) models have become a critical element in how they operate, particularly in relation to preparing for what's to come. This is primarily due to the ML model's ability to enhance the financial performance, through data, for both businesses and their consumers. To illustrate this, consider the example of the United Overseas Bank. By analysing the myriad of data points from multiple sources, such as daily file uploads, the ML models have a more complete overview of transaction and customer data. As a result, ML helps United Overseas Bank to optimise fraud detection, map out distinctive customer experiences and improve businesses processes for the bank. It is precisely because of this use of ML that the United Overseas Bank can now ensure its customers' banking experience is more reliable, simpler and safer.
The challenge that many organisations face, however, is that the ML models that are used today have been created using huge amounts of very granular, historical data. While this is effective in predicting outcomes that follow a similar pattern, it does not account for the extremely uncertain environment businesses find themselves in today. However, to alter these models is not straight-forward and there are a number of considerations to take into account before diving in.
Adapting in the face of change
Before any action is taken, it is of the utmost importance to determine whether the current disruption can be defined as a one-off 'Tail Risk Event' or a 'Structural Change'. If the COVID-19 pandemic is a tail risk event, it's basically an exception to the rule. This would mean that as and when the world recovers — financial markets, businesses and the wider economy would bounce-back to normal and should function in much the same way as they did before the pandemic. In this case, ML models face the challenge of avoiding being too heavily adjusted, influenced and biased in reaction to this one off, once-in-a-lifetime, event. In contrast, if COVID-19 is perceived as a systemic, structural change, then it is going to have a significant impact on how financial institutions, and the world at large, operates. There will be no 'return to normal' and the adjustment required will involve businesses needing to develop completely new ML models that can account for this new and evolving landscape.
Adjust, invest, progress
There is not one simple solution for businesses. However, there are some key actions financial institutions can follow in order to best utilise their machine learning models in order to navigate this challenge:
- Modify existing models: This will be the starting point for all data science teams. Actions can range from using the latest data elements to modify current models, or creating scenario-based projections adjusted for various levels of model bias. There are a variety of techniques that can be utilised including a Bayesian approach to capture expert judgement into the models. One of the more innovative approaches to the lack of rich relevant data is a meta-learning approach. From a deep learning perspective, meta-learning is particularly exciting and adoptable for three reasons: the ability to learn from a handful of examples, learning or adapting to novel tasks quickly, and the capability to build more generalizable systems.
- Industrialisation of ML: 2020 has created the opportune moment for businesses to invest in a platform that supports the entire ML lifecycle, from building and validating processes, to managing and monitoring all of their models across the entire enterprise. The data storm that occurred this year, means enterprises have been faced with ever-growing amounts of data on their customers, business and wider market. These data sets have been entering the organisation from a variety of different sources, from the customer service team to social media platforms. For ML models to operate at their optimum level, they need to be able to quickly understand what all this different data is saying, while simultaneously taking every stream of data into account. Having a unified enterprise cloud platform is the only way to ensure this can be achieved.
- Stress testing: This step is crucial in helping businesses gather more clarity on their vulnerabilities before it's too late. Importantly here, the responsibility cannot fall on one team. In order to set up multiple, dynamic stress testing scenarios, cross collaboration from finance leaders to Chief Risk Officers is essential. The findings from these tests should then be implemented and retested, to ensure businesses are in the best position possible to face the challenges that Covid-19 and Brexit will bring
- Prescriptive Analytics: This is a complementary approach to ML and utilises simulations, brought on by shocks or market changes, to enable more accurate decision-making for different scenarios. A common approach to this is a bottom-up simulation for modelling of complex and adaptive systems called Agent-Based Modeling (ABM). ABMs help businesses put into play numerous future scenarios without having to depend on the limitations of historical data.
There have been an inordinate amount of challenges thrown at businesses this year and lessons needed to be learnt quickly. In these times of unpredictability and uncertainty, financial institutions shouldn't count on this type of wholesale disruption being a one-time occurance. In fact, the 2008 market crash proved this kind of sudden disruption and instability can happen. Instead, 2020 needs to be seen as an opportunity to ensure more long-term initiatives are put in place that enable financial institutions to be prepared to successfully react to the next crisis. This is the time for businesses to really evaluate what technological investments they have made, assess whether they are working well in the new circumstances and whether advanced tools such as ML models are truly being utilised to the best of their ability. The players that take the time now to assess how well their ML models are performing and make the necessary adjustments ,will be the ones that ready themselves for success beyond the pandemic.