There’s no question that the asset management industry is making significant strides in implementing data and AI models for the screening and monitoring of investments and risks. The tools used to be in the hands of a limited group – top tier funds and hedge funds with deep pockets – but as data has become more accessible across organizations, and as more powerful models have become available for investors, it’s now much more feasible for other players to create and run live data models.
This progress is needed, given the huge volatility and unpredictability introduced by COVID-19. The ability to predict and manage risks, especially on the downside, has become more critical than ever.
Getting the operating model right
But what are the factors for success? Firstly, it’s important that asset managers devise the right target operating model for their data approach. There is no single operating model that works for all firms – it’s a case of aligning the approach with the organization’s capabilities, objectives, culture and maturity with respect to producing and delivering innovation.
Operating models for data and innovation can range widely, from highly centralized (all data scientists co-located in one team, separate from the rest of the business) right through to decentralized and independent (data scientists integrated directly into teams across the organization). The truth is that the most successful models are likely to be somewhere in between – a hybrid approach. The danger of a very centralized approach is that the data team is too removed from the portfolio managers who are making investment decisions on the ground; while a highly decentralized model can mean duplication of effort and inefficiency (data scientists working separately on the same things).
It’s also key to appreciate that an organization’s operating model is likely to evolve over time. It may look quite different in year three, for example, to how it looked in year one. Most likely, it will start off relatively centralized and then move further to the right over time.
Putting the tools into action
But how should an asset manager actually utilize these high-potential tools? Again, there are various possible models. A portfolio manager may simply read a report that has been generated through the application of data and AI and then make decisions accordingly; equally, a data signal may feed directly into an algo-trading model itself.
Models may be developed in-house, or they may be taken from a provider (such as KPMG Lighthouse). Or there may be a hybrid approach – working in partnership to develop and enhance a model that can then be applied across a portfolio.
Today’s models are becoming increasingly sophisticated, making use of a range of technologies such as AI, machine learning (ML) and natural language processing (NLP). They don’t stand still either – nowadays, there is not only ML and NLP but advanced variants that make use of techniques such as neural networking.
Whether it’s to screen investments and forecast M&A events ahead of time; monitor investments and forecast default risks ahead of them coming to pass; or news analytics and other external monitoring to pick up potential risk and reputational issues – data and AI techniques provide a powerful tool.
For example, a default predictor that we have developed at KPMG Lighthouse ingests about 20 million data points spanning the previous ten years, covering all public companies across North America. The model can be reviewed, validated and tuned across hundreds of iterations to achieve optimal performance. The output includes explanatory detail too – not just highlighting investments that are highest risk, but listing the factors behind that.
It is important to remember, however, that these tools are intended to augment human judgment, not replace it. They provide insights that would otherwise be difficult (or impossible) for a human being to capture – but ultimately, it remains with the portfolio manager to make final decisions.
War for talent
Another key factor for success is a difficult one to crack: obtaining the right talent to build the models. Competition is high – and the big tech giants tend to hold the upper hand in attracting people with the highly specialized skills needed. Remember, though, that it doesn’t come down to getting that one, perfect person: it’s about assembling a team with complementary skills across it. Working with external organizations can also be a route to success.
2020 has been a big year for data models to prove themselves. 2021 is likely to be just as big. Achieving the right approaches to data-driven innovation will be key to asset managers’ performance.