From Our Briefings Newsletter
Artificial intelligence (AI) – the science of teaching computers to think like humans – and machine learning (ML), which is a subset of AI that includes the computer models that analyze data, are often used to describe the technological forces reshaping the finance industry. But according to practitioners who shared their thoughts at a Goldman Sachs European Quantitative Investing Conference in London earlier this month, the current reality is slightly more restrained. We sat down to speak with Goldman Sachs' Jo Hannaford, Ezra Nahum and Xavier Menguy about the impact that AI and ML are having in technology, fixed income and equities, respectively.
The article below is from our BRIEFINGS newsletter of 25 September 2017:
Briefly . . . on Artificial Intelligence and Machine Learning in Finance
Jo, as head of the Technology Division for EMEA, how are AI and ML used in finance today?
Jo Hannaford: Many people are under the perception that computers are running the world and replacing human beings. But on the trading floor, humans are still making the final decisions. The reality is that we're using machine learning to analyze different data sets to come up with the recommendations to make richer and more informed decisions.
Fixed income is an area where we're seeing more ML activity. Ezra, as the global head of the firm's Fixed Income Currencies and Commodities (FICC) Strats, how is machine learning informing the business?
Ezra Nahum: In fixed income, there is a lot of data but the data covers incredibly wide areas. If you think about the examples where AI has been widely publicized – such as Google's AI computer program beating the top player in the board game Go – those are situations where the constraints are narrowly defined and well established. In financial markets, the constraints aren't well established. ML models provide statistical tools that help identify patterns of relevance from the data, and in doing so, can guide practitioners to a place that would have otherwise taken them much longer to get to. In our FICC business, for example, we use ML primarily for business intelligence, i.e., to help us understand our business and our clients. Through the process of gaining a deeper understanding of both, we can deliver services and products that are tailor-made for clients. Indeed, we can use the information provided by the data and the models trained on it to enhance our understanding of sparse data sets, therefore getting a better sense of our own performance with clients as well as being able to identify some of their preferences. Meanwhile, we're organizing our own infrastructure so that all of our internal and publically available data – which is usually collected anecdotally, or is based on someone's intuition and years of experience – is collected systematically. By streamlining the processes by which our traders and salespeople work with their clients, we're improving the firm's ability to service and even anticipate clients' needs.
Xavier, as global head of Equities Desk Strategists, how is ML being applied in equities?
Xavier Menguy: In equities, ML models are being used across wide areas – from business intelligence and flow analysis to inventory management and derivative pricing, among other things. We're also looking to incorporate natural language processing – an AI technique that translates human language into structured data sets – to automate all sort of tasks that used to require human intervention like chat based quote requests. Because trading equities in the capital markets is mostly electronic, there's a massive amount of quality data available on which ML models can be applied. Another key factor is regulation which enforces transparency and data reporting. With MiFID II – a package of rules that is set to take effect in January across the European Union – financial services firms will soon be required to capture all over-the-counter trading activity. That will, in turn, produce more data which we can analyze and leverage. With that in mind, we're investing heavily in ML and this is directly reflected in the backgrounds of the people we're hiring.
So it's important to have a strong background in data science. What are other factors that can affect the success of AI and ML?
Jo Hannaford: In finance, ML can be used to create different signals, but the quality of those signals depends entirely on the quality of the data. The models themselves are very sensitive to the data that is used to define its recommendations. If you have the wrong signals or the data is not relevant, you can distort the outcome. For us, the way you approach the inclusion of the data and the diligence you apply has a strong correspondence to how good the recommendations are.