NAVIGATING UNCERTAINTY WITH ACCURATE MACHINE LEARNING

James Johnston, Regional VP at Cloudera

 

2020 will undoubtedly prove to be an unforgettable year. The pandemic has been unforgiving, plunging the UK into a recession, and many industries have faced closure and untold disruption. In the Financial Services sector in particular, 86% of profit warnings in the first seven months of 2020 cited Covid-19. But Covid-19 is not the only thing on the sector’s mind – another sizable challenge looms large on the horizon: Brexit. Individually both are highly disruptive events, together they create a double shock wave with a long tail of unknowns: how long the COVID-19 pandemic will last? What the fallout from Brexit will be? How resilient is the UK economy in the longer term? A key topic for discussion is therefore, how will we adapt to these seismic events and how can technology help?

 

Predicting the unpredictable

When it comes to planning, Machine Learning (ML) models have become an integral part of how most financial institutions operate, because of its ability to improve the financial performance for both businesses, and their consumers, through data. United Overseas Bank is a key example of a business that has used ML to make it’s customers’ banking experience simpler, safer and more reliable. Through analysing the thousands of files that are uploaded to the platform everyday, the ML models have a more comprehensive view of customer and transaction data to optimize their business processes, design distinctive customer experiences, and to improve detection of financial crimes.

However, in these circumstances of heightened uncertainty, the accuracy of ML models come into question. This is because the majority of ML models that are in use today have been built using large volumes and long histories of extremely granular data. With the world being as unpredictable as it is right now, it will take some time for ML models to catch up and adjust to this year’s events. The most recent example of such complications and abnormalities, at a global scale, was the impact on risk and forecasting models during the 2008 financial crisis. Re-adjusting these models is by no means a simple task and there are a number of questions to be taken into consideration when trying to navigate this uncertainty.

 

Adjusting to the ‘new normal’

The first step is to determine whether the disruption we are facing right now can be defined as a ‘Structural Change’ or a once in a blue moon ‘Tail Risk Event’. A structural change would represent a situation where the COVID-19 pandemic has had a seismic impact on how the world as a whole, and financial institutions in particular, operates. This would result in the world settling into a ‘new normal’, one that is fundamentally different from the pre-COVID-19 world. This shift would require institutions to develop entirely new ML models that rely on sufficient data to capture this new and evolving environment. On the other hand, if the COVID-19 pandemic is perceived to be a one-off ‘tail risk’ event, then as the world recovers and businesses, financial markets and the global economy return to some sort of normality, they should operate in a similar way to the pre-COVID-19 days. The challenge for ML models in this situation is to avoid becoming influenced and biased by a rare, and hopefully, once-in-a-lifetime event.

 

Readjust and reinvest

There’s no one size fits all solution for businesses, however there are some key steps financial institutions can take to them navigate today’s current climate:

  • Modify existing models: This is where all data science teams should start. Modifying models can range from using the latest data elements while creating scenario-based projections adjusted for various levels of model bias. There are a range of alternative ML-based approaches that can be used to revamp existing 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. These are also some of the reasons why meta-learning is successful in applications that require data-efficient approaches; for example, robots are tasked with learning new skills in the real world, and are often faced with new environments.
  • Stress testing: This is a fundamental step as it helps businesses gain a clearer understanding of their vulnerabilities before it’s too late. This isn’t just the job for one team, cross collaboration from finance leaders to Chief Risk Officers is required to set up multiple, dynamic stress testing scenarios. The learnings from these tests should then be implemented and then retested, to ensure businesses are in the best position possible.
  • Industrialisation of ML: If businesses haven’t already done so, now is the perfect time 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. Nowadays, enterprises are faced with increasing amounts of data on their customers, entering the organisation from a range of different sources, from the customer service team to social media platforms. For ML models to work at their best, they need to take every stream of data into account, while being able to understand what the different data is saying, and quickly. This can only be achieved with a unified enterprise data cloud platform.
  • Prescriptive Analytics: This approach is complementary to ML and uses simulations for more accurate decision-making for different scenarios, brought on by shocks or market changes. One common approach is Agent-Based Modeling (ABM), a bottom-up simulation for modelling of complex and adaptive systems. ABMs help businesses project thousands of future scenarios without having to depend upon the limitations of historical data.

 

Businesses have had to cope with a lot this year and those that have survived have faced a steep learning curve. When faced with such a crisis, they need to look inwards, towards the technology they have invested in, review whether it’s working in the new circumstances, and whether crucial tools such as ML models are being deployed in the best way possible. Financial institutions shouldn’t look at the issue as a one-off, but instead as a chance to implement longer-term strategies that enable them to prepare and tackle the next crisis head on. Businesses that invest the time now to re-evaluate their ML models are the ones that will set themselves up for success, now and into the future.

spot_img
Ad Slider
Ad 1
Ad 2
Ad 3
Ad 4
Ad 5

Subscribe to our Newsletter