The panellists were Francois Mercier, Vice President, Senior Data Scientist, Citi; Victor Daniel Casas Hernandez, Global Head of Data Science, Machine Learning & Artificial Intelligence – GTI, JPMorgan Chase & Co, Michael Natusch, Chief Science Officer, Prudential and Alexei Tchernitser, Director, Analytic Solutions, Quantifi.
The views and opinions expressed in this blog are those of the individual and not of the companies they represent.
How is data science changing decision-making in the context of investment securities? Francois, if we can come to you first with that one. How do you define data science in this context?
Francois: This is a very interesting question and actually a very hard question to answer. I think there are different ways to define data science, and I guess it depends on where we are coming from. So one way could be to have a more scientific or data driven approach for decision making, and it could be a way to try to find, for example, causality, which could be very useful especially in finance, since it’s very hard to know what caused what. Another perspective, from a business point of view, is to try to apply science, but for business needs and it is less formal than the science that you may do for fundamental research and academia and so on. And the last one, maybe from a technology point of view, could be a more sophisticated way or approach to automate some processes. So there are three different perspectives, and I don’t think there is a single one to define data science.
Michael: I agree with most of what Francois just said. I think the real value we can exploit here comes from the fact that data science is, as Francois mentioned, bottom up, its data driven rather than a top down model or human hypothesis driven. That gives us a completely different point of view about the kind of various problems we’re looking at. So, if we say how do we define data science in this context? That’s really a bottom up, data driven, probabilistic approach that captures, not just the traditional, numerical variables, but also the categorical variables or any other input data that we’re looking at. And obviously, there’s a connection – data science is not in isolation. We’re also looking at the relationship with things like alternative data now, where we wouldn’t otherwise have a handle on these kinds of things if it wasn’t for the tools that data science provides for us.
Victor: I hope that people understand that it’s a science, meaning that if you don’t have a hypothesis, you cannot test anything. So, first it is how do you not define it. It is not a magic button that you click and it solves your problems. You have to state the hypotheses. And as Michael was saying, in a data driven way you can demystify or not. Data is widely available, so whatever you can measure is the data you can work with. So, this goes along with what are the sensorial tools i.e. what are you measuring in order to enable data science? That will hopefully cover what is the tech stack you actually need to deploy these things really efficiently into any production application/business.
Alexei: I would only add that I would define data science in a very broad way – a whole set of new technologies which we are using to analyse large data sets. In practical terms, in finance, data science found its way into almost every aspect of the industry. Some of the examples would include the analysis of trading signals, decision making, algorithmic trading, systematic market making, optimisation of derivatives calculations. All of these use cases would use data science in one form or another.
In practical terms, in finance, data science found its way into almost every aspect of the industry. Some of the examples would include the analysis of trading signals, decision making, algorithmic trading, systematic market making, optimisation of derivatives calculations.
Alexei Tchernitser, Director, Analytic Solutions, Quantifi
Do we actually need data science in decision-making? And Michael, perhaps I could come to you first on this. What value does data science add to decision-making in this context?
Michael: I think it’s as if you’re the pilot of an airplane and you just have the speedometer in front of you, that’s probably not going to be terribly helpful. You want the altimeter as well and you want, not just one pressure gauge, but you want a whole bunch of them around the airplane. So, I think the use of data science for us is really in having different angles on the same problem. If you look at a particular business or finance problem that you’re interested in just with your traditional toolset, then you will just see that one angle that you’ve already traditionally looked at. But if you can look at the same problem from lots of different angles, you can all of a sudden see this is how it actually looks in the full light of day. And I think none of us can afford to miss that point of view. I think that really describes the value, as far as I’m concerned, of what we’re trying to bring to ourselves and our colleagues.
Alexei: It brings a huge amount of value and in ways which weren’t possible before. While it is true that a lot of the models and theory behind data science has existed for a long time and we didn’t used to call it data science before, for example with regression models, these have been around for a long time, but it is only recently that we combined them with the computational power which is exponentially growing, and this is a relatively new phenomenon that is growing very fast. And so now combining these theories, some new and some traditional, with this new technology, what we have now is the ability to analyse structured and unstructured data, which is limited only by the computational capacity. So, our decisions are better informed when using data science and machine learning enabled frameworks.
Victor: For me, there are two parts. One is the actual problem that you want to solve, like taking Michael’s approach with an airplane. But imagine, you want to buy groceries from your grocery store two blocks away. Do you need an airplane to go there? Probably not. So if you are developing huge tools and using them to just travel one block, you’re overcomplicating the problem. Now, if you actually think of portfolio management, imagine that I want my Sharpe ratio to be infinite. I basically need a money-making formula. Since the problem is so open ended, that’s when you throw everything you have at that problem and that’s when you use data science, artificial intelligence, machine learning. There are techniques where no theoretical solution is possible, but approximation solutions are the ones that you need to help you drive the value. My Sharpe ratio from exit strategy went from 2 to 2.5. We know that’s a huge amount of alpha that was captured just by doing machine learning. And now those are the things that do materialise as value because the problem has inherent value also to achieve. So, I hope that people can think of first the business problem open-ended enough to be machine learning applicable and then apply the techniques, because without a business problem being clearly defined, you can throw all the machine learning at it and you literally won’t have any results from it.
And one last thing, don’t think this would replace us. We are human, assisted by AI, meaning that AI is here to help us do better things instead of replacing us. Imagine it’s like a cell phone, it’s just another tool that you will use to do your work a little bit better. That’s the value added.
Francois: To me, I think we need to think the best intelligence that we’ve got right now is still human intelligence by far, despite what we say about AI and so on. But it doesn’t mean that it’s the best at everything, there are some weaknesses of the human intelligence. The first one is that we cannot handle large amounts of data. So, if we come back to finance and decision making, a human would be very good at following 2 securities and that’s it. But if we want to go through a whole universe of securities, you cannot have a human tracking everything. You need to have an automated process, and a data science approach is usually the most advanced one in terms of the best performance. So, that is one added value. Also we need to make progress and to make progress we need to measure it. The measure is always coming from the data and so we have to get some kind of data driven approach for that.
The last one, to come back to the human weakness, is that our intelligence is not as rational as reinforcement learning and so on. We do have some bias due to evolution or whatever it is. Those biases make it so that we cannot trust ourselves 100%, especially in finance. Behavioural finance, for example, will exploit this idea that actually people are doing something which is not rational. So having a data science approach can be a way to potentially detect the bias we may have and complement what humans think.