 Hi, and welcome to Bright Minds from Tickmill. I'm your host, Patrick Munnally, and in this series we're setting out to answer some of the most commonly asked questions around investment and trading through entertaining and insightful conversations with seasoned insiders. Artificial Intelligence or AI has been hitting the headlines in recent years thanks to its disruptive effect on a wide range of industries, and that disruption has been felt particularly keenly in the world of investment and finance. Although data analysis has always played a part in trading and investment, in recent years advances in AI have revolutionised the way financial institutions and investors process data and make investment decisions. With the help of machine learning algorithms, AI is able to process vast amounts of financial data at lightning speed and identify patterns that may have been missed by human analysts. According to a recent report by Markets and Markets, the global market size for AI in the financial industry in terms of revenue is projected to grow from US$86.9 billion in 2022 at a compound annual growth rate of 36.2% until 2027 to a projected value of around US$407 billion. This rapid growth is driven by increasing demand for advanced analytics, rising adoption of cloud-based services and the growing need for a real-time fraud detection and prevention. In addition, a survey by Deloitte found that 50% of financial institutions have already implemented AI in some form, and 79% plan to increase their AI investment in the next three years. However, the use of AI in finance is not without its challenges and potential risks, such as data privacy concerns, algorithmic bias and regulatory compliance issues. In this episode, we'll be speaking to Harris-Dawn Christow. Harris-Dawn is head of quantitative research at Tickmill and has extensive experience of working in both the public and private sectors. Harris-Dawn works at the cutting edge of data science and thanks to his role at Tickmill is perfectly positioned to talk to us about the challenges and opportunities that AI presents to our industry. Harris-Dawn, thanks for joining us today. Could you kick things off by telling us a little more about your career so far and your interest in data science and AI? My career with data science goes back when I was doing my PhD in computational free dynamics and when I was doing my postdoc on the reset mission of the European Space Agency. The task there was to actually analyze data from the reset mission and try to find patterns and understand what are the rocks on the comet that we were studying. We were running thousands and thousands of simulations and trying to understand the rocks and through that it was actually the beginning of myself and into the data science field. After that, I spent another one year studying data science and actually machine learning but actually on the finance sector, which I always had an interest. These came actually quite natural to be a quant, especially in Tickmill, trying to find patterns in different types of area in the FX market and broadly in the financial sector. Interesting. Can you tell us what are the main ways in which AI is currently being used in investments and trading? There are definitely various ways that AI and machine learning is being used. If we go back in 1960s, actually when Renaissance technologies and major hedge funds, they were starting and trying to find patterns in the financial market, they were trying to find out what drives in some patterns. We have seen throughout the decades that data science has seen an enormous amount of attraction from the investment community. In our days, what we have seen is that there are various types of data. For example, there are people looking at how many people they have visited, a particular store today, how this could actually impact the price of an asset tomorrow in the next hour, in the next maybe quarter. We have seen that their data, that they track vessels in the sea, they carry oil. So how much oil has been in these vessels for how many days, why it's onshore, why it's offshore and all this kind of stuff, they have been analyzed by QANTS. We have seen credit card transactions, meaning that there are people trying to find out if the spending of consumers can drive the market. Also, we have seen how many coffees, for example, a community is taking every day and how we can find patterns in it, not only for a particular asset or for coffee company, but actually for an economy. If we see that there is a decrease in the coffee consumption of a community, it might mean something for the broader economy. So that's more or less where we see the last few years, the new trends. But we have also seen the more traditional one, which is how positive or how negative is an earning call for a company and how this can affect its pricing for a few quarters ahead and how this has been affected. We have also seen very, very recently, phase recognition, where by some phase recognition algorithms, the QANTS trying to find patterns between the phase of someone in the banking sector where they announce the interest rate decision is going to affect the market. That's more or less, if we want to cover the whole board, that's more or less what we have seen the last few years. Sure. So with that type of progress, it's staggering to think that phase recognition is studying central banks because they announce interest rates. I guess with that level of progress, we've seen recently influential tech figures calling for a halt to AI research so that safeguards can actually be put in place. What are some of the main challenges, do you think, called risks associated with using AI and investments and finance? And how do you think these will be addressed? There are various risks, to be honest. However, the major risk that comes and what we have seen actually from the banking sector, but also from the private sector that we have seen the last few years is when data scientists, QANTS, they take a model that is out of the shelf and in principle, they do not fully understand what's behind it, so they don't understand what's under the hook. So we have seen that this could actually cause millions and millions of losses. So that's the one that we have seen to be causing major problems in the sector. We have seen companies going bankrupt because of this, and especially in the real estate sector because of this. So we strongly believe that if a researcher or if a data scientist understands what's behind a model, what we call a machine learning model, that's quite crucial. And we see that that's where organizations are trying actually to go that direction, to have their own dedicated teams that they do understand what's behind the hook, what's under an algorithm, what it does, what it doesn't. And that's, we think, what should be done at least at the first stage. And so what kind of AI-based tools are being used at Tickmill? And what benefits do you think they will have as the technology progresses for retail traders? So in Tickmill, we have a huge amount of data that we do actually handle. At any given point in time, at every millisecond, we process tremendous amount of data to be sure that the company and especially our clients, they are in a safe environment. And we actually analyze pretty much everything. So for example, we do analyze the profile, if there's fraud or anything, we are actually responsible for that. However, if we take it from the retail client's perspective, what we are trying to do and we're in a certain extent, we have achieved that is that we act in their favor, meaning that we are actually next to them. And what we do is that we analyze the pricing in real time. And what we do is that we try at every single millisecond to provide them with the best price that they will not fill tomorrow or in one week or today that the prices are not good. We try to do the best that there is available in the market without any unmeaning changes in the market. We try to have quite smooth pricing, not gaps. And we analyze all this kind of stuff and we know when it happens, why it happened and how to process it. That's where we stand right now. So one thing that more particularly we offer to our clients is the signals through TickMeal and the sentiment. So for example, even our QANTS team spends hours and hours in predicting and trying to identify the sentiment for some news where retail clients, they do actually have it ready from TickMeal. And we have seen quite good patterns that they can help them in their trading decisions. But not only trading decision, what we have seen is that it's one of the best tool in managing their risk. Addition to trading signals, which they are also available through TickMeal, which they can provide them some signals if they are blended and combined through sentiment that it's been provided for people that they don't know. It says how positive or how negative it is in news. It can help them create quite good trading strategies and quite good risk management for their assets. It's interesting that you mentioned sentiment analysis there, Haritone, because as we know in the markets, once sentiment becomes skewed in one direction, the market will often move in the other direction to, as always, make the fools of the most amount of participants in any one instance. So in terms of that theory, that market theory, how will that impact AI? Because at some stage, I guess, you're going to have AI trading against AI. And at what stage will they start to game each other to get better outcomes? Yeah, that's a good question, Patrick. It's one of the problems that we always actually discuss within the team every single day. No matter how good we are, we have this discussion. And it camps actually back to the user, to the people that they have designed this AI system. Even though we see that in a period where skewed sentiment drives certain decisions of AI systems, we have seen other AI systems and machine learning systems work quite differently. And the reason behind that is a human element. So because people design systems differently, they have different mindset, they have actually different holding periods, different horizons, this is affecting the AI. We don't think that 10 years from now, this will still be an issue because at the very, very end is how people react on their system. So if I design a system that takes purely the sentiment, it will more or less take the same decision that what would you do if you take the same data? However, if I start transforming the data, meaning that, okay, if instead of watching this sentiment, what if I take the rate of change of the sentiment, how this will react and more or less is going to be quite differently. So with this level of sophistication, I guess, how accessible are these tools to people who may not have previously had much experience with technology and can something as intuitive, I guess, as chat GPT be used to help with trading decisions? Yeah, definitely. I mean, we see that that's how the world is progressing. So we think tools like chat GPT, they will not create your trading strategy, but they will be actually to give you a hand when you will need it as a trader, as a discretionary trader, where you think that you want to switch to a more systematic one. We think that the world is going to this direction, where traders would go from discretionary ideas to try to systemize them and help themselves. We don't think that it's going to be needed too much time in terms of training and learning all this technology stuff. In our days, all these kind of things, they're quite easily be understood and actually work few days, few hours per day to learn some coding skills with the help of all these tools like chat GPT, where they will actually be next to you. So for example, if tomorrow you're writing a code about an idea that you had, but for XYZ reasons, it does not work instead of going in Google and trying to search why this doesn't work. You just copy paste it to one tool like chat GPT and what it does actually, we expect that it will correct your mistake and it will be next to you to provide an optimum solution for your trading ideas. So that's where we think the world is going. Right, that's interesting because I guess that symbiosis between the human and the algorithm or the AI brings to mind it's been shown that human chess players working with machine learning algorithms can consistently beat an unaccompanied algorithm. Do you see some version of this type of partnership as a way forward for AI and investors, human discretionary traders in the financial sector? Yeah, that's where we see the world actually, where we see the next step to be humans along with AI and machine learning to work together to find patterns and understand. We don't have to forget that AI is not always like a black box. There are various ways where you can understand a prediction that the model does. So for example, a model says that there is a fraud in this element, then you as the developer, you know more or less why this decision was made actually and you will be able to go more deep and understand why this decision was made, was it correct, was not correct. Even from a report from Bank of England, they were actually discussing about fraud detection with machine learning, with explainable actually machine learning. That's the thing that we believe is the next step, where you have explainable machine learning, where you understand their prediction and you have this human element blended in that. So that's the thing that we believe the world will go. Interesting. And then talking about progress, I mean obviously with the the rate of change that we're experiencing at the moment, I guess to even be thinking about five or 10 years out seems like an incredibly long period of time from the perspective of machine learning. But what do you see as the most promising applications of AI for investors in the financial arena? There are various, depending on the profile of the investors. So for example, we have seen investors interested in some low volatility returns and some investors interested in high volatility returns. We have seen people interested in some particular asset class, for example, FX or stocks, that we have to categorize them based on their profile. But even in that, we see and we believe that the financial sector and the investors will go into a more actually systematic way of investing. But on the same time, the ingredients for this machine learning or AI prediction will come from humans. That's what we think will be at least for the next few years ahead. I would say three to four years ahead. We believe that the human element is needed to take meaningful decisions for the AI. But the AI is actually responsible to pick some patterns within the dataset, which and it should actually in a certain extent differentiate itself from the noise, which we all know financial market have a lot of noise. Yeah, that age old difference between signal and noise still exists even in this modern era. Harrison, thank you so much for your time today and for joining us. Is there anywhere listeners can go to follow your work online? Yeah, we do have, they can follow on LinkedIn and Twitter and where we always post interesting things about finance, machine learning and also about Tick Meal and what we are trying to do at Tick Meal to revolutionary the way that financial markets work, especially in the retail space where we try to stand by the traders and help them achieve their goals. Superb. Thanks again for your time, Harrison. Thank you, Patrick.