AI for Financial Markets
Good afternoon ladies and gentlemen, I accept you obey. I sincerely hope all of you are in the best of your health. Today's panel discussion will be on the topic AI of financial markets organized by talentsprint firstname.lastname@example.org. But before we get started, I would take a couple of minutes to set the context and then bring in our speakers to get the discussion going for the day. Financial Markets are undergoing transformation fueled by AI and other cognitive technologies. As financial market firms work to digitize and finance from growth and operational efficiency. Advanced AI and ml techniques are helping create customized investment options, thereby improving customer experience and helping secure a larger share of assets and create new generation client experiences. In fact, it's estimated that by the year 2025, the financial service sector is expected to spend $19 billion on AI and cognitive technologies compared to only 5.6 billion in 2019, which is indeed a very big boost. So to build on this context for the entities session, we bring to you insights from some key industry veterans, who will give us a closer look at how AI is transforming financial markets not to by leveraging data on helping companies not just compete, but rather win in the market. First we have with us Mr. Mukesh Agarwal, CEO at NSE, data analytics limited analytics and decision makers. Mr. Agarwal is a business leader with about three decades of experience in the financial services industry. He also oversees the mF platform business act and previously he was at the president at the Chrysler limited a standard and poor's company and was leading that India research business with full pain and responsibility. Welcome Mr. Agarwal. Thank you. Next we have with us Mr. De combat but now I achieved economics and the National Stock Exchange of India that has close to 20 years of rich experience in Indian capital markets, macro and sector strategy, quantitative finance and consumer banking. He started his career as a researcher at ag IDR and worked on a number of academic and corporate projects in the area of econometrics and quantitative finance that are followed by a stint in consumer banking analysis. This focus area was equity strategy for institutional clients, initially with Citigroup global markets as equity India equity strategist, and then with relevant capital markets limited as to India strategies, as well as chief economist and assignment prior to joining NSE. He was the chief strategist and head of research for India and Japan based in Dubai. Welcome Mr. bagnaia. Thank you. Our next speaker is Mr. Alan J. Rahman, program director at talentsprint. Mr. German has 60 years of experience in the financial sector and nearly a decade of that was at a US base, quantitative hedge fund and support for your manager at the hedge fund with the systematic trading strategies and also treated in all four asset classes that is equities, commodities, currencies, as well as interest rates using data times. He has also taught at various universities like Duke and 10 state, he teaches AI data analytics, computational finance and algorithmic trading. He's also a chartered alternative investment analyst charterholder Welcome Mr. J. Rahman. At this actually Thank you. That's because the leaves definitely could be Hi, Mr. Shantanu. Paul, MD and CEO of tan spring, who shall also be our moderator for the session today. Under Mr. Paul's leadership talent spring has emerged as a leading player in the deep tech education industry, and is currently a part of the National Stock Exchange. He also serves as an independent director at nsdl, Cayman bands and national payment Corporation of India and also leads various both subcommittees. Welcome, Mr. Paul. Thank you so much. With that, I want to again extend a very warm welcome to all the gentlemen on board today, a quick announcement before we move on to our discussion. So as the session progresses, I would request all our audiences to keep sharing their questions with us in the q&a tab. Towards the end, we will pick up the question for our speakers to answer them. So without any further delay, I will now request Mr. Paul, to start up today's discussion over to you, Mr. Boyle.
Thank you for shocky. And thank you economic times for being a partner on this wonderful show that's hopefully unfolding in front of us in the next one hour or so. I'm delighted to have such eminent panelists that you've introduced already, so I won't spend time introducing them. But I will start off by setting a bit of context to say that, you know, this whole question of AI is obviously everywhere right now. I mean, we live in a world today where we know that AI can drive cars better than human drivers. We know that AI can detect tumors better than human doctors. And we also That AI can lend money better than human underwriters, right? It's been proven to so it's been shown that these examples I just cited are examples where technology has begun to outperform humans on a steady and regular basis. And that's been the power of AI. Clearly AI works very well in the context where there's lots of data, massive amounts of data available. It works in situations where the future is uncertain. And therefore making the right call in the future making the right decision about the future or predicting the future well, has a very high value and premium. So in that context, ai usually thrives. And the examples I gave you earlier are all examples of that if you think about it. So with that, without further ado, I would like to open up and ask the same question first to all my panelists. That is AI for real when it comes to financial and capital markets, or is AI hype? So the question I'd like to first pose to Mukesh, do you think, with regard to this particular sector, industry, financial markets and capital markets is AI as potentially powerful as it seems to be for other industries and sectors?
I saw, I think you've asked a relevant question, because nowadays everyone keeps talking about AI machine learning and you also hear that people ask you, is it a buzzword or is it a real thing, what I would like to say is AI and ml is having a disruptive impact on the financial markets and people are using AI across the board and disruption areas, which you would areas in which the disruption is happening happening, it includes like collecting structured and unstructured data. Now, as people keep talking about data is the future oil No, how does one collect and analyze the data. So that is one area which is happening there. Then also it is also happening helping in lowering trading costs, it is helping in modeling some other areas in financial markets, and it can be used across different functions, be it cybersecurity, be it or it will be on surveillance, compliances customer onboarding. So some of the areas in which AI, or ml is being used in the financial markets. It includes things like trading and portfolio management, capital optimization, regulatory compliance, then also client facing chatbots, Robo advisory, market analysis and forecasting. So some of the examples which we see are in real life, people are using AI ml for all these activities. So I would stop over. Thanks.
Thank you. Okay, so I think you've just rolled out a whole bunch of categories of activities and functions that are going to be disrupted or being disrupted as we speak. So I will turn this to the tanker, I'm sure to think as a trained economist, when you did your PhD, you were not exactly thinking about AI as a top of mind. But I guess now things may have changed, I would love to hear your thoughts as well.
They certainly have changed on my PhD was in high frequency finance. And, you know, one of the things that you pointed out earlier on that is that the AI helps you in crunching massive amounts of data. So you know, 25 years and the world has gone, you see pages up in terms of the size of data. No, you also allowed us, let me just give you a quick anecdote here. Two years of data that I did, for one part of my thesis, had about 300 GB of data. Two years, we've had about 100 million observations, all the records the mind, no, you hire worthless piece of data today, that kind of data is there, one day of the futures and options file that we kind of use. So the extent to which data we, you know, talk about now, is no longer in megabytes and gigabytes, but in unfamiliar territory, petabytes. So, no longer even terabytes, we are going to be the wide world and you know, anywhere where one, this entire concept of computers helping you in repetative data, that has also been taken to an entirely new meaning today. So the massive amounts of data and especially so this, no, we are talking about financial markets today. This is one place where AI has a very, very firm role. There were no again, in financial markets, we will of course, this are probably in the know as we go forward. But we are here to say AI is not going anywhere. And in this technology AI actually is a self propagating mechanism, you have more data, you have machines to generate more data and therefore to be better AI to assimilate that.
So we entered some kind of virtuous cycle loop is what you're saying, which is a very interesting point, I think have to for two people saying that this is here to stay. And this is for real r&d or a kind of a rare distinction of being among us, somebody who's been a practitioner of both AI and financial markets. You've also been an academic in terms of teaching and research and you've been a consultant on both sides of this. So you have a rare perspective having looked at this from all Possible angles? How serious is the AI revolution in financial markets?
Absolutely, I have to agree with what location the banker said, I my background actually is in physics and the way physics you start off with things is that, you know, there is a clear model for things around you, right? You have a, you're hitting a ball on a billiard table, there is Newton's laws, and you know what those laws are, and you try to use the equations from there to fit in data. The thing is, once things start to get more complicated, where you don't know what the equations are, that is where the power of ml comes in, in ML, the way you think about it is not some preset ideas about how the world is, because we are not even sure whether financial markets, there are a set of equations which can be used to describe it accurately. So what we have is just this large set of data, and you use computers to figure out what kind of patterns might be there. And, and that's pretty much the, I think, their objective way of analyzing markets. So for me, definitely, this is not going anywhere, this is the way to move forward, this is absolutely the right thing to do. Looking at models, looking at data, and understanding financial markets from data perspective.
Thanks, Simon. So we have over 200 plus people watching this as we speak. And I think for the benefit of that audience, I feel responsible to at least clarify a few terms. When we talk about AI and machine learning and deep learning interchangeably in some fashion. In reality, from a computer science perspective, AI is the big super circle. And within that there is a smaller circle called machine learning. And then there's another small circle called Deep Learning, and the algorithm should get more and more interesting and more complex and more rich as you go into those smaller circles. So just for the benefit of all of that it may appear we are using this word loosely and we perhaps are, but by and large, we talking about machines that can make determinations based on past data and predict the future in some fashion. Just a little bit of a tutorial, one on one on the side. Now moving on to my questions. Let me start with Anand bow. And I think if you look at the life of a fund manager, let's say you're running a big pension fund or whatever, or any other fund, for that matter, a managing lots of money for other people in other institutions. It's kind of tempting to say that okay, in the near term in the next three months, where will the markets go? You know, I mean, just today, for example, I was coming out, I saw the Federal Reserve held its meeting and left everything untouched. So the question then becomes the next few months, what will happen based on these decisions happening globally, by central banks, US etc, China, Europe, and of course, India. So there's a, there's a real desire on the part of fund managers to kind of predict the near term future to a reasonable degree. And then frankly, we think of how we do this, you know, you can search on CNBC or any other channel for that matter, and you can suddenly hear, you know, financial experts Talking Heads giving us bearish bullish and everything in between opinions. And I would like to think that doesn't have to be such an emotional issue. I mean, this should be something that a machine can predict in the next few months, what's going to happen, love your thoughts on other technologies and AI today that can do this in terms of helping fund managers make better decisions about the near term in terms of allocation in terms of exposure management and therefore maximize returns?
Absolutely. chatroom, interestingly, this is about six months back, we did a consulting project exactly on this, there was one of the one of the funds, which advises for retail investment advisors on on on the in general, what on economy and what kind of signals one should take many of these retail investment advisors, you know, they are advising traditional retirement clients, most of these clients do not want to trade they want to stay invested in the market, but they are trying to see you know, if they're relying on the investment advisor to to guide them on whether the oncoming quarter, or the oncoming six months, the markets are going to be volatile, are the markets going to be relatively risk free, and so on. If the markets are watertight, you want to sort of cut down your equity exposure, if the markets are safe, you want to add to their equity, exposure and so on. And this kind of decisions, typically, the retail Investment Advisors were making in a discretionary way looking at their understanding of what's happening in the markets. The fund that came to us the ask the question, but can I can you use data? And can you make an more objective decision? Right. And what we ended up doing was exactly this. We looked at things that typical retail Investment Advisors look at, for example, what's the current federal funds rate, how much money is fed pumping in? What is the difference in interest rates that banks are lending at versus what the federal government is giving, so on and so forth. I think They took all of this into a machine learning algorithm. And we asked the machine learning algorithm to try and detect if it can see ahead periods of volatility or not, right, and algorithm and did beautifully. Well, the thing is, this algorithm was designed before the pandemic. Now, it didn't predict the pandemic, no amount of machine learning, no amount of AI can predict pandemic without, you know, looking at this federal funds data. But interestingly, the algorithm at the end of April predicted that we are entering a low volatility phase. So what had happened, what we saw was, when pandemic was there, the markets crashed dramatically. Everyone we didn't know what to do, we didn't know whether the world was going to end or not. But after the governments all around the world decided that we are going to support the markets and these are the efforts we are taking the market started to shoot up like a rocket, I mean, it was one of the biggest bull markets that we have ever seen. And none of it made any sense. Because here we are, we are talking about, you know, job losses, and so on and so forth. But the stock markets were going up crazy. Now, interestingly, the, the algorithm that we had designed, was able to correctly predict that this was a low risk period, how was it able to do because it had looked up data before. And it's found that whenever fed had pumped up a pumping in dollars, there was a response for the next three months or next six months, and so on and so forth. It's hard to tell exactly what the algorithm thought, because it's essentially a black box. But it was fascinating that it was actually able to predict some of these things. And this is a model that we had been tested for, we had been testing for a while. So six months back when a pension fund, when this fund manager had come to us, we provided to them now, there's about you know, 14 $15 million on it, where they are testing out these ideas. So it Yes, absolutely, machine learning or AI can definitely help with your investment decisions in these kinds of situations.
That was a fascinating anecdote. And thank you for sharing that with us. So clearly, machine learning and AI is good at predicting such short term events. In this case, it appears to have worked question for Mukesh, how about, you know, indexes or indices over time? If you look ahead 123 years, and I'm not talking about indexes that you personally are close to. But worldwide, there's so many indices out there that are tracking markets of various types. Do you think over a one to three year period horizon? Ai? Or have there been attempts made to use AI or machine learning or deep learning to predict these indexes on their movements? Or do you think that's a very immature science at this point?
Shantanu, I will add on to what Ana said, actually. So I just gave an example that they had built a predictive model using the machine learning, which was able to predict Yes, it could not predict pandemic, but post pandemic when the markets will start reviving and giving some signals. So I'll also talk about a very real time kind of real life example. So we are working with a Canada based firm, they are also into a quantum technology based predictive modeling. And so what they do is always have they're collecting the data. And then also saying that you need US federal aid, then you liquor you, because today, you're going globally, all the buckets are integrated. So you can't just look at data of one particular geography and do the prediction. So you're when you're talking about interest rates, you have to look at interest rates across different countries across the world. Similarly exchange rates across the world, then GDP across the world, then also, if you look at it, weather conditions, weather condition, also because commodities and agricultural things are in prices are interlinked because it's let's say, in India, India is one of the largest producer of sugar, if there is any adverse climatic impact, if that impacts the sugar production in India, that has an impact on the global prices. So that model captures prices of various things, and it is all stored. And then I think they use some of the models, they have 15 models, and then they come up with a conclusion. So you can use that model to predict anything you can use that model to predict equity stock like you can use like they're done an exam pilot for us, you can use that model for predicting prices of reliance, Tata Steel, already a series of stocks, you can also use that model for predicting the question that you asked for indices also. So like suppose now you have input the data of let's say, you want to predict nifty 50, the nifty 50 that will give you what is the outlook over the next 60 days. And also it gives what is the outlook for the next two weeks. That is the kind of pressure on prediction for predicting nifty 50 index or any other index and as you need to have all the data probably for predicting a 5050 You will also need that data Hold the underlying 50 stocks Because ultimately, the value of the underlying stocks determine the value of the nifty 50. So probably you need to track the underlying data trading prices, trading volumes of nifty 50. Plus also, I would say the news related to each of these stocks. So I think so beauty of this whole AI is that it is able to digest interest, large amount of data and quickly give you the output. And the beauty of that is people like Ana, they develop the models, and they give it to the users and users, they would not be AI and ml experts. So use this model, because all the data is already collected to string model. And based on that model, real time predictions, you can send out their predictions and your predictions you can get. And based on that you can take your trading decisions. But also at least I would put a disclaimer and Alan can add on to that. So I think one also needs to see yes, models will tell us and all all of us can't just blindly follow the models and start doing our trading I as an investor, I won't invest in realize that nifty 50 I also have to keep my eyes and ears open and see what's happening on the ground like pandemic happened, no model will be able to factor in pandemic, because something had gone by the model in March last March 20. And February 20, probably I would have gone gone wrong in the decision. So I have to keep my eyes and ears open and see what's the ground reality. So I have to mix and match what I see. And plus what the models are predicting. And then based on that take my training distance is able to use consume as fields and said, we're talking about terabytes of data, it has the ability of digest all those that terabyte process all the terabyte of data and give you output. Look at that. On top of that you look what's happening around you and take your investment decisions. I think that's why I would say
fascinating story. Mukesh. So I think the first thing that is obvious is that both near term and medium long term, there are models in place, AI is definitely at work, we can see
and Chandra, sorry, sorry for it. But now what I'm saying this model, which company I'm talking to, in fact, they have I can't take the names, exact names, but they are companies both in India as well as they are using those models to predict the prices of the commodities you will not forget, one company buys coal. So they're using that model to buy to predict the prices of coal and take the picture. company and us. Again, it's a fmcg. So they are also using that model to take a decision on some of the raw material purchases they do in those are equity products. So basically, they also using that model to take decisions on their agri products purchased. So these are actually using realize or whatever we are talking it's not that all of us are talking theoretically, people are putting those models to real life use. And this will keep improving. See this, again is as and when more and more data keeps coming in. Your models will learn from the data from the past, and they will keep
improving. Absolutely, I just want to make one more quick point. Because it this anecdote came to mind when he was speaking regarding that, you know, you keep your eyes and ears open to make your own additional value. And I remember many years ago, I was flying on a plane to the US and I was sitting next to somebody who was actually a Boeing 777 commander who was going on holiday. So he was kind of sitting in the passenger seat next to me. So I asked her as usual, you know, I'm kind of trained to be a computer scientist. So my question is always similar to any profession. I said, so how much of your flying is automated? And how much do you have to use your own client skills and brains? Is it oh, you know, Middle Eastern Airlines like Etihad, and Emirates is about 97% is automated. 3% is, you know, commanded Riven. And in American Airlines, it's about 93% automated and 7% driven. So I know the whole interesting conversation happened after that as to why that difference exists. But fundamentally, I say the 97% of your flying is under machine while the pay you so much. And they said okay, well, because for the remaining 3%. So that example came to mind, as you're talking about how trading works, you know that remaining three to 5% is where you actually make all the difference, perhaps, in this model. So moving on to the tanker, the things that we talked about markets specifically but I think as an economist and that was chief economist of NSE group, you must be really worried about macro economic issues and you must be worried about implications of things like FBI and FBI. And speaking of, you know, foreign institutional investments in India, you know, that's always been a bit of a hot button issue like hot money, right money can come money can go It has a direct impact on how you know, the stock markets will do how people perceive things to be going well or not right? I mean, there is this view. So my question to you is that can you actually use AI? Or is anybody using AI or data science or any of these different technologies to predict the future of funds flowing
in and out of our country? We The Economist and we bristle at the thought of my panelists, when they say market can be predicted no people who believe in efficiency market hypothesis say that market time but I for one, know that they are right and they Not enough and more attempts to predict the market. There is also a distinction to be made about trying to predict the first moment as we call versus the second moment. So when, when Alan was talking about he was talking about the second moment about volatility. Mukesh mutation from low when it's time to predict the index. And it's talking about the first movement there. one anecdote again here from no to investing was in fact, so we engaged with several academic institutions. One of the one of them is I'm about to move to another point about how their algorithm was able to notice that there will be a steep drop in volatility. Just to confirm this, there is a journal article on nomic letters, which we look at. And this basically segues into also what Mukesh was saying earlier, due to the integrated nature of market, so picture yourself in March 2020, everyone thought the world is gonna end and then you had in April, you had a massive stimulus, which came in globally across fiscal and monetary side. And then did you see that in, in, in, in terms of price discovery, when we are talking about market, we are talking about the sum total, once again, look at an index as a sum total of where multiple agents, millions upon millions of agents want to be giving you that blended indicator. So we'll be surprised. The implied volatilities in India, in India, not in the US, which is where you actually got a lot of the similar status, the implied volatility is in India, and our derivative market is where the world's most liquid one look came down equally low. Putting no I think mid April, on the third week of April, exactly, in line with what I mentioned, you know, and there's this beautiful paper, which shows you how implied volatility is shot up, you know, around the around the end of March, and how low IV basically started coming off me to be as evil as the similar measures were announced. And then they stayed low. So for for us exactly. Again, as you know, Alan pointed out, we were human Lee in the worst possible time of our life care, sitting at home working from home trying to adapt, and market were going up. It was baffling. But no and, and because
they really didn't really make the mistake of us. They did not go on sentiment, they actually went on what happened, When, when, but significant support is provided, what really is the end game on Corona? And the market meant again, as an algorithm, it takes an algorithm to go where you will sentiment will not and why am i i think that is an added incentive of using exactly what Mukesh said, you use your intuition. But you also use the algorithm as support. The algorithm was telling you exactly as it did on us that this is the time to invest and people who did we know what happened towards the end of 2020? I mean, the market gave us the best results. And I will again juxtapose this with what for India turned out to be a far worse situation in 2021. As an economist, I have to talk about market movements, right. So in 2021, our second wave in India was much worse than 2020. We thought it was the worst, but no, we were wrong. But what happened with market volatility, they didn't go anywhere. What happened with the market, it didn't go anywhere. We saw no oxygen shortages. We are lots of things happening in April and May, the market didn't go anywhere, what we do is we go anywhere, and that is where I believe there is room there is actually a room for dispassionate reasoning. And this is where algorithms are absolutely needed. It's not just a simple prediction, but it is it is a mix of where logic where you will might sometimes miss the logic assets, algorithms will tell you that you know going by history going by the integrated depth of market, looking at indicator hundreds of them across market, you have to be in the market. This is something you know absolutely interesting. And let me tell you one other example here very quickly, we had a there are anomaly there are anomalies in the market, which are not meant to be exploited because that markets are not behaving as they should and therefore, you know, try and make money. Now these are normally in markets like the United States has been found to go down once they're brought to life. know there are academic papers written about those anomalies and that the anomalies disappear in India and in several other markets, which are very different but you know, which are do not have formally algorithms like this Attacking the market. Now, these take much longer to be used to be to be, you know, to be taken out. So my point is, we are in a world we are already in a world. To my mind, there is no question of we are already in a world that are those who are doing a lot of our decision making. And in markets where there is a significant share, the anomalies are not there, there is no money on the table, in very simple sense. There is not much money left on the table market like India and several other emerging markets you do have, but not for long.
That's a lovely exposition. Good, thank you. Thank you for that. In fact, I'm going to just clear that into the next question for Arnold, you talked about how you know humans can bring emotion humans can bring sentiment humans can bring, you know, sort of his right brain quirky response. But the machines and algorithms of AI and others can actually give us a very high quality signal and like a left brain, right, if you use your technology, your left brain and your instincts as your right brain, then all of a sudden, you have a whole brain approach to markets, which might just be a fantastic collaboration between humans and machines. So that makes sense. But then again, humans also expressed sentiments in social media in the press, and all the stuff going on. And sentiment analysis of social media is a huge topic. You know, can you look at Twitter feeds and decide where the markets will go? For example, it's things like that, that people have been studying this for the last 10 years, at least in research. So the question to you from an academic perspective, technology perspective? What is the role of I mean, what is the role of AI and data science and sentiment analysis? And can that provide, like I said before, and unstructured data based view, which is completely not the markets, historical behavior, but perceptions of humans collectively put together a social media? Because that's like the world's largest collection of psychological? You know, I would say repository right. What does like collective psychology is? So any thoughts on how technology is helping us get insights about markets and companies and things like that, and indexes from sentiment analysis?
I'm absolutely shattered on so I, before I answer that question, I just want to add something to what Tucker said, I, I absolutely love the term that he used. dispassionate reasoning, right? That's wonderful. If you haven't copyrighted it, I'm going to be using that in my classes, which about AI, are one of the points you made about how, when this kind of anomaly or inefficiency is brought to light, suddenly, they after some time they disappeared? You mentioned that how in US markets, many of these anomalies have disappeared, because it's a lot closer to efficient markets. Whereas in India, it's still there. And I can attest to that. As a part of one of the courses that I'm teaching, we had, we were developing an algorithm as a as a class homework example, really, right, where we are trying to see, can I predict in the next month, which whether small caps are going to outperform mid caps are going to outperform, or large caps are going to perform? fundamental, right, this was just to show students how you set up the problem and how you frame and how you go about it. And what I was actually surprised by, I completely expected not to be able to get any predictability at all, because I've tried this in US markets. And in US markets, you can't get, you know, any out fireball. But interestingly, in Indian markets, it's there's still some amount of predictability left based on out of some of these other interesting macroeconomic variables. So it's, it was interesting that you brought that up, because just last week, we were doing this as a part of our course. Now, the question that Shantanu asked about sentiment analysis that I, so typical machine learning or AI courses, they teach you how to do sentiment analysis. Yeah, take a look at an article. And from that article, can we extract sentiments? And the goal is, if I analyze a whole bunch of articles, then can I get from this the general sentiment around any particular stock, or, or the market in general, there's a very famous paper from about 10 years back, which actually showed, you know, you don't really need to do
any high end tech stuff to get some advantage from sentiments. The this was a paper which looked at the Google's search results. This was a time when in in US people are talking about the financial cliff, you know, the the Congress or the senate could not agree upon the right level of budget deficits to have and so on and so forth. And there was a lot of uncertainty about whether markets are going to collapse or not. Now, one could just look at the search terms in Google and And Google gives them out free, right? There's no you don't need anything, you can look at the trend of search terms. And you can look at the trend of search terms, whether the search terms are negative or positive. And based on that trade on the market, and the amount of extra performance, you can get over a traditional buy and hold was just amazing. Right? Obviously, once the paper is published, everyone was trying to do sentiment analysis. And I don't know if there's still that much alpha left to extract. But what really, I thought was curious was that now every single institution when they teach machine learning, they show you how to do it sentiment extraction. I was recently doing a consulting project for one of these firms, where these guys, this firm was subscribing sentiment, data streaming sentiment data from a provider from a data provider. Bloomberg is one famous data provider that is Reuters famous data provider, which gives you pic level data. This data provider, however, was providing safety streaming sentiments on any particular stock based on articles that are published based on tweets that are put out and so on. And for this service, they were charging about $5,000 a month, this is one of those techniques, right? Right out of package, you know, you have a free pie, you download Python, and there are free packages available, which can quickly get you sentiments of articles, right, I know that I'm not minimizing the amount of engineering effort that would be needed to eventually get a sentiment on a streaming level. But to me, it was surprising that institutions were willing to pay up to $5,000 a month for streaming sentiments. And that tells me that, you know, for a young technologist, someone who's looking at AI, or should I get into data science, they should tell you the amount of opportunities that are available, right, these technologies are not as out of reach as it used to be before, it's easy to get an N Panther is free software, right? And you get it and you are able to actually code up these things, you're able to test out models, the kind of democratization of data science techniques, the way it does improve technology and and what we can do is just mind blowing. So I truly believe Yes, sentiment, can add alpha to your portfolio. And it's definitely institutions are using it quite regularly.
today. I want to honor that. I think he's absolutely right. One is, we should look at both social media analytics, as well as the news analytics because social media again, Authenticity, and all is not a tested. So US adults, we need to look at both. When you talk about analytics, we should look at both news analytics as well as social media analytics. Second thing is he talked about the examples of capital markets, I would say it's happening in the banking and financial institutions, because we have just invested in a startup building commercial, but the guy was telling me that he is talking to one of the banks are here in India, like they do lending. So they want to track sentiment analytics or news analytics on the portfolio of the companies to whom they have landed, because here they have like basically what happens is you would have a large portfolio and so many news etc, it keeps coming up very difficult for anyone to track it manually. And if you just automate the whole thing, you have someone who can again collect all the data all the news, which comes about the companies in your portfolio and gives you a summary of all the negative news coming about the companies in your portfolio. So based on which you can take further decisions, these kind of things are actually not only in your depression market bfsi segment, for example, which I gave you
some Okay, let me push that a little further. We talked about you know, numerical data, which is stock market performance, historical, we talked about sentiments, which is text based emotions, rolling around in social media, etc. But we can go a step further. I know people talk about alternate data records, satellite imagery, and if I look at the satellite image of a country and the monsoons are coming or not coming or I look at various parking lots of companies and say this guy is selling really well. There's a whole computer vision is one of the areas where AI is very, very powerful. So I'm just curious, what is your thought on this that will be on top of numerical data and textual data which is social media and they will also get a visual data or imagery data that we can analyze to predict stocks and markets.
Yeah, I think basically, as all of us know, the stock markets are determined by multiple factors and alternate data is becoming a big thing that is again is a big future because whatever conferences etc We attend on stock exchanges and data, data etc. People are talking about alternate data is using and see an alternate and let's get into the basics what exactly would be Alternatively it I would say that altering data is any data other than that trading or trading price or index value I think any data like now suppose you have Walmart now Walmart, yes all the retail retail outlets like DMR or reliance and all they give their monthly quarterly sales year over year growth month on one floor, what is the potential for people want to go one step ahead, as you correctly said satellite images, there are entities in us who would track the they would track the parking lots of Walmart's, another Kmart etc. So, they capture the images and see how many cars are parking over there. So based on that, now, people try to see how is the sales of Walmart or Kmart? So, that is one thing. Similarly, no weather forecasts weather forecasts people look at it very closely No, you have this in India this guy might so people know try to see politically forecast weather much more closely and regularly because we know that in India corporate case weather forecast calculator. Similarly now car sales are sales in India also companies declared the cost is monthly. Now Can people are trying to see if we can get the number on a daily basis because whichever investor has access to all this data before does get it because he Multan all the investors will get the data now if I can hire this investor, I'm smart enough and all this data if I can get it on a daily basis, then I can crunch this data come up with some analytics and my analysis of that. And based on that I can take my investment decision now Walmart car parking is area is heavily packed with good number of cars are coming every day. Why why to wait for the mountain data announcement by Walmart, I see okay, they use parking Good, good demand sales will be higher, we'll go and buy one those are the kind of stuffs are happening today people want to know more recently started giving data on a daily basis on a two day lag. Give an example people have started driving did this data from number of new carport insurance policies second car policies being taken or being sold by the insurance companies I think the people are finding innovative ways of getting data or having access to data before others can get it and use that data for your predictions and analytics and based on that you will take your investment decision. So I think alternate data is going is also going to take off in a big way people are looking at ways and means finding new data points like if you look at it and also what also people are trying to see as an economic Google unity and all the reports you read what he was doing. Every month they will come out and say okay, the mobility is so NCT is so and so City is 40% behind normal pre pandemic, but it isn't improving. So I think people are using this is all alternate data. So and people are trying to visit already Google and Apple are doing it but I'm sure going forward, people will eagerly look forward to such data points and alternatives anything whether it's related to economy, interest rates, car sales, anything you name it, steel imports, steel prices, now, what is the steel imports? What is the steel which is there lying in the vessels in the sea, how much steel will land in India, if you can make that before others actually get the data landed? You can make smarter decisions. That is what is going to be it will keep growing as we move forward. Thank you. Okay. It's
always fascinating, I suppose lovely discussion, we could have a whole session on altering data, but I'm running out of time. So I'll keep two more questions and they'll hand it back to shocky. So the tanker next question for you. This is regarding you know, those old saying in stock markets that there is greed and there is fear, right, the two prevailing motions that drive stock markets greed and fear. So whenever you look at this market, and of course, we talked about all the upside and how we can do better in terms of performance all this while but there's also the issue of governance compliance risk management, as a regulator and I know for example, nse is you know, quasi regulator and deeply coupled with the holes centrality and the stability of the Indian stock market system itself, or capital market system itself. So how can AI or data science be a major tool in improving our compliance and improving our risk management so that we don't have scams and frauds which traditionally hits every capital market from time to time how can we use technology to prevent them
while they are already being used to track non standard behavior so to speak, as brocade pointed out earlier, a few commands operate at LSP also, there are several areas where the until now, I must say that we have been looking at only the direction and part of markets we will be talking about how can we talk about the future that is Navigate. But see, if you look at a more general sense of making sense of enormous amounts of data, trying to identify patterns where that are more easily visible to human eyes. And this is where alternate data also comes in. Where I mean, that is where there's a, there's an enormous on set secondary use of AI and AI techniques as well, I can give you several examples, just look very quickly, you will be surprised that you have orders in our market, and and then orders become trade, the number of orders for everything has used to be about five, six, now it is for some stocks with over 100, sometimes even 1000 orders, not likely to one day, all of this is only because there are algorithms sitting on both sides. So you're not dealing with each other. Now what happens is, this is where the role for AI becomes more important. You want to understand non standard behavior, you want to understand behavior that is falling afoul of the rules in a market where a second is eternity today. Second, no one second of the day, literally eternity, when you have, you have 20,000 seconds in the trading day, the equity markets 20,000 or the trading day, the number of capital markets orders are about 270 million, you can do the math about how many new orders come in per second. And we only the moment we think that's a huge number 270 300 million, you know orders in a day, you have to look at the derivative market, where it ranges between 2.72 sometimes 3.7 billion orders per day. So when you have markets that are so far, identifying non standard behavior is not something humans are equipped to do. And this is where no, and there is there is a there's an enormous scope, I think it's already being used in a meaningful sense. But scope is near infinite in this point. So it's not about tracking is no longer about looking at the directional aspects of market. But looking at factors where they are, they are separate from what regulations would allow us to do. I can give again, there are strategies methods, you pump and dump strategies, circular trading strategies. So these are strategies that will be very difficult. If you are likely buying and selling to each other, it's relatively easy. But if the same strategies done between 10 people with different levels of you know, sharing, then it becomes something that only an algorithm will be able to identify. And we are talking market, which are at the pace that I said with a couple of billion orders per day. So I think I've given a flavor of where things are going in this fabulous chapter. I will add over here, I think just what you said, actually. So AI is being used by the exchanges for what we call it on the market surveillance activity. I think that's what it is. So it can be used for detecting any if it's anyone is doing any manipulative, active activity as Ivanka said, Because today, if any stock prices go up, you have there are so many stocks, so many orders happen. Now, if I could track what happened. So I find I have news analytics, sentiment, analytics, etc. So all these things put together, I can do my market surveillance and make sure that any kind of multiplayer activities which are taking we are able to stop that and prohibit that from happening, and this is actually happening in real life. It's not that AI is being used by our market market.
Yes, this whole area is called reg tech, I think and it's becoming very significant and mature. My last question to Anand, before I go to audience questions, and I know I'm way over time, but I did want to ask this question too. And so if you look at one of the challenges that the retail investors face today, in India, as an example, as to worldwide, there are all these, you know, various consumer tech consumer internet companies coming up with IPOs. And these companies have, you know, they're trading on things that people don't really understand why the valuations are so high just to cut the story short. Now, is there a way for example, if you're a good investor, you're tempted again, greed and fear, you're tempted to kind of go up to the shiny objects on one side, and there's a fear of missing out if you don't, on the other hand, you know, you really don't know what you're getting into. And there are plenty of examples around us. It's been a very, I would say, very, very fertile IPO season, so to speak, are the technologies that are out there helping investors or retail investors in particular, who don't have the resources to make those kinds of research, investments and so forth? Are there tools that can help people make better decisions as to whether to enter with or do not enter whether to wait for a bait etc, etc.
I my answer is going to be very much from Traders perspective. That's where I learned finance today. That's how I grew up in finance. So that's what is going to be my perspective. Now, regarding actual valuation of companies, this very well established, clear literature on how people go about valuing companies. And there's a lot of guidance available on what is the fundamental value of these companies and so on. From a traders perspective, however, it's very clear from simple data analysis, you can one can download data of how what are the number of IPOs that are gone in the past two years? And what you can look at what was the average first, again, what is the average secondary gain, and so on, and so forth. And typically, what you find is that the average gain over the past year or so has been in the range of 20 to 25%. In a in a in a period of one month. Now, the thing is, right now, we are in the midst of very positive sentiment and the markets are happy. And I'm so it's not surprising that the average IPO is getting that high numbers, you can go back and look at data. And what you will see is that this is not new in that sense. Whenever markets have been have had this very bullish mindset, the IPOs typically have done very well. Now, for a retail investor, The Color of Money is exactly the same, regardless of whether it comes from a shorter term bet or a month, or from a longer term investment. And so as a trader, what we do is we generally look at these trends. And you would, on the top of the trends, you apply some reasonable amount of intuition, some amount of subjective judgment, all these are really bad ones. But on average, when you hold, you know, 80% of the stuff that comes up during these kinds of bullish periods, you do well, right. And that's just a fact. And this is not it, won't you, I won't even call this AI. This is data science, you look at data, you look at what has happened in these particular periods. And, and this is the reason why one would expect this to continue happening is there are fundamental economic theories which talk about existence of momentum in markets with things which are stocks to move up, continue to move up, you know, that kind of idea happens, and which is why sort of bubbles continue for a while. And as long as you're you realize that you're not making a one year commitment or five year commitment, and you are in for the short term, right? I would say why not. But again, I as I said, I come from a trading perspective rather than a fundamental investor perspective. And I
agree with my opinion. In summary, I think what you're really saying is, when in Rome do as Romans do. So with that we will audience questions, especially what you
are moderating such an insightful session, also a huge amount of time. So all our panelists for sharing some very thought provoking and interesting learnings with all of us today. So with that, we now move on to a q&a session, I'll keep the questions open to the panel, and anyone can take it up as deemed suitable. The first question that we have from our audience today is, is quantitative trading and algorithmic trading the same? Also how can individual investors benefit by building knowledge in this domain, since not every citizen has access to Advanced Computing systems?
I can take that. So generally, again, in popular press, these terms, quantitative trading, algorithmic trading, are sort of used as if they are equivalent. There is subtle difference. More funds, I don't think there are any mutual funds or any wealth management firms which don't use data, they all use data, and they make judgments based on data. So in that sense, pretty much all funds are quantitative funds to some extent. But when you talk about agronomic what it means is that you do not let any emotions come in between at all there you Everything is coded into an algorithm. So a decision is made on whether to buy a particular stock or not. And the fund manager does not overlook that and it automatically gets placed into the into the market. So that would be called algorithmic where every single process every single step in the decision making process is completely automated and completely algorithm wise, that is typically called us algorithmic trading.
We'll move on to the next question. The next question is from audiences How can regulators and financial institutions use AI for the prevention of economic offences and financial crimes,
we did discuss this point actually. There is there is there is need regulators will be handled without using artificial intelligence, machine learning and data science in general, given what they are up against, and it is, it's a perpetual game of playing catch up, in some sense. People know the market attract the smartest set of people in the world. And no established theories don't always work. We have several examples, especially, you know, during the pandemic, see what happened with Robin Hood, see what's happening with GameStop? What happened to GameStop? What what is happening with Robin Hood, even IPOs that, you know, analysts for what them you know, in Indian concept, see what's happening with the Robin Hood IPO, even in placement. So, we are in an open world and, you know, in a better world, when there is massive amount of interest, then at that point, regulators cannot really do without artificial intelligence. So they do actually use and, and as again, rightly pointed out, it is being used, and and they get better and better. And all, you know, all the time, I would also like to take one part of your first the first question, because it was about how do you use and learn algorithmic trading? You know, I, maybe I know, the basic, maybe you might want to talk about that is that the company side did not know, how do we learn them without using expensive computer systems? So I don't think No, it is, it is costly to learn these things. There's a lot of democratization for us. And there's a lot of democratization today in terms of access to knowledge, in terms of access to computing power. You don't even have to own your own, you know, expensive hardware, you can you have access to cloud computing underneath, you might want to speak about it in greater detail, maybe for the benefit of the audience.
That's right. Thanks for reminding about that part of the question. It's absolutely true that you don't need very expensive hardware to do this. The Python is a free package, free language, the plenty of packages that are available, which will allow you to connect to your any of the standard brokerages. There's an IP API that's available for you to connect to zero dot, for example, or interactive brokers or any of the brokers that you are that you use, that are free API's that are available. And so all you need is that I got the idea. And that is where the real money is. The ideas are yours, if you have the ability to take that idea and code it up in a clear algorithmic way. Yes, you can do algorithmic trading, you might not be able to do millisecond level trading. But that's something that's those are details if you are able to find something where you can trade within hours or couple of days or whatever timeframe algorithmic trading is certainly possible for for a retail investor
to genomic data and move on to the third question. So the third question is how artificial intelligence and machine learning benefit in arbitrage trading online data.
I can talk about experience. We're one of the funds that we had consulted on, they do use AI to do all kinds of strategies, trend following mean reversion and arbitrage strategies. Last year, in fact, this particular fund God was ranked number one for the AI based strategy that they use, although it goes on volatility space, but it's very, very clear that in order to identify arbitrage opportunities, you need data science, you need to understand data. It might not be AI per se, but certainly it's a statistical decision you are making as to whether two stocks are mis valued or a given stock is relatively cheaper compared to something else. And data science absolutely is a key ingredient in order to in in any arbitrage strategies. Absolutely.
Moving on to the next question, the next question that we have is what is the current regulatory position in India and outside on applicability of registration and conduct requirements on AI MLB and manage investment advisory?
Basically, you have this liquid take this question you have this ri is registered investment advisors because in India, clearly their CVS segregated one is distribution distributors. And second is advisors. Advisors can't do distribution and distributors can't do advisory jobs. So you will arise and there is I don't think there is any regulation that says that you can use EMI AI ml or to Christian. So there are people who are using this. This is the jargon, keep using the hearing board. So people are using robo advisor to advise people to their application portfolio optimization But there is no regulation as such, which states that AI ml can be used for this thing and can't be used for that thing. I don't remember that. But clearly you have two models, advisors, control distribution distributors can do advisory advisors, depending on what their comfort level they use different models for advising their investors.
Have an add to that just a little bit to that question. And the answer is Well, I mean, one of the things that I've noticed over the last 1015 years, both in banking and generally markets is that, you know, the regulation, innovation trade off is a big challenge for regulators, right. I mean, regulators don't want to clamp down so much on by over regulating that innovation gets killed. And of course, when innovation runs amok and regulations too far behind, you have some bad accidents. I mean, that's the nature of this game. So I think one of the things that I think getting it right for regulators to say that how much regulation and no further and that's of course, a moving target based on what's happening in a broad landscape. So, I will say that most progressive regulators today are not trying to kill innovation, I think one of the ways to not get innovation and therefore not over regulate or overreach is to allow AI and data science experiments to happen and I think we will learn from this and things will get more sophisticated and if there's too much regulation too soon, and all you do is cut off innovation. So I think innovate regulators are getting it to large extent right,
I said earlier allocation model for non advisory customers, the model that is driven by machines, is it considered as financial planning from a regulatory angle in India,
you will see basically no financial planning will start right from top top basically, what is financial planning, financial planning will say, Okay, now it starts, okay, what is my age? Now? What is my future requirements? Now, what is my what were kids, what is my money from what situation of my own compared to my kids and all this kind of stuff, and then you also do a risk profiling of yours, then look at the trending of that, based on that, they will tell you where to invest what not to invest, and in what asset whether equity in commodities, gold, silver, whatever it is, and there is no regulations as such on that. So Robo ra is registered investment advisors, they will allow, allow and do that, they will tell you what to do what not to do. So distributors, as I said, they can't do anything. They just are distributors who take your money and you tell them where do you have to invest? And they will invest on your behalf? and advisors, they tell you what is that they do financial planning for you? And they will tell you what is your asset?
Our next question is, as retail investors or traders, we are more dependent on fundamental and technical charts. So will AI help to predict the future chart movements or can decide to predictions of price? If so, what might be the accuracy?
The retail investors, they depend on fundamentals and technicals? Right? So one of the points that we've been discussing all through is how AI can make sense of enormous amounts of data. So again, an anecdote you all technical analysis can be considered a set of trading rules. Do you see an RSI, you see Bollinger Bands? No, where numbers are in terms of, you know, 50, DMA, 100 DMA. There are simple rules. And then they get more and more complicated. There is there is an established set of papers actually, in the literature that that takes 10s of 1000s of the waiting room, think of a technical analyst who knows so many of these building rules and then tries to see how one can use to get statistical arbitrage. How can one use these technical trading rules in statistical arbitrage, to make you know, to generate alpha? So to answer your question in a simple way, yet, no technical analysis is being used already. In fact, they act as dipoles for for many AI models, because then AI can see your 10,000 patterns. You know, in a jiffy, where humans generally cannot and humans are limited to while they do not replace humans altogether. But there is a significant amount of use of technical training in in knowing data science in general and AI in particular.
Our next question is more or less related with Mr. Paul. So the question I'll read out, take power, the commandments example given by Mr. Paul, do you think the same holds true for financial market resulting in gradual reduction of human involvement, eventually leading to zero human intervention in the coming times?
Yes, I mean, the bad news is that, you know, definitely, AI is used to improve efficiency and reduce errors. And I think that's actually one of the reasons why AI is so popular in every field. I used examples from medicine, from aviation, from automobiles, to factory floors, AI is coming in everywhere for the simple reason that improves productivity and efficiency on one level, we can get more done And we can do it with more accuracy. Because we under certain point humans don't really perform well, to give a classic example, I mean, why has computer vision today become so good that automated self driven cars are even possible and they have lower accident profiles. In fact, in California, an insurance company will give you a better premium if you're a self driving car than an automated car rather than a human driven car. Right? That means that the trust is that the car that drives itself is going to have less accidents. And the simple reason is that somewhere along the way, the ability of human vision, I mean, human vision is about has a 45% error rate. So if I show you 100 pictures, quite likely, the average human will get 95 of them, right in terms of what object they're looking at, and 5%, they will be wrong. So there comes a time when the algorithms become so good that they outperform the human capability on vision. And same thing happened with chess, something happened with the game Jeopardy, right, this keeps going on. And now what's happening with medicine? So yes, short answer is that the future lies and more and more people who understand how to use the technology well, and make better quality decisions. So yes, more technology means fewer people making better decisions and having more supervisory role of the machine. I mean, if you look at the commander example I was giving 3% of the time he has the commander is flying 97%, they're not sleeping, they're actually supervising the automation, the technology that is running, right. So supervision itself becomes a real job, right? Another example is in automotive factories. You know, if you look at a picture of assembly line, if a car manufacturer from 50 years ago, there will be this huge assembly line, and there are humans leaning over and doing a task. Today, you will see the robots leaning over and doing the same task. And the humans have gone back one level to a control room from which they can see the whole factory floor and all the automation, right. So in that sense, I think this is no different. Better algorithms will lead to better quality decisions, fewer people smarter people doing it. And yes, I think wherever technology enters the nature of jobs change whether it leads to net gain or at loss of jobs, a different question. I mean, there are studies to show that in the next five years, there will be X number of x million jobs lost in the world, and there'll be 1 million jobs gained. And the question is, is why bigger than x? Or how much bigger? Is it? So short answer is that the jobs will change, whether they will net gain or net loss, we'll see. But certainly we'll have smarter people using smarter technology all the time. That's the trend.
time I'll pick up one last question. So the last question is, could you throw a light on the future of companies, secretaries and Chartered Accountants in an AI driven world?
I was thinking asked me a question which I didn't have time for, which was, we talked a lot about AI today, we didn't talk about blockchain, which is the other disruptive technology in the whole space of financial markets. So the whole idea of blockchain is that, you know, you start to see a model in which auditing becomes really, really automated, right? Today, if you look at auditing, still a function that, you know, compliance and auditing the function that again, especially his job comm, chartered accountants, company secretaries, lawyers do this for a very high rate of billing, let's put it that way. So therefore, the future lies in blockchain promise of blockchain is that is, you know, you just like AI automates intelligence, blockchain automates trust or trusted system. So if I can have an automation of trust, then I don't have to really have three days, four days or three weeks of auditor sitting in my office, they can just pull data from systems and the system will, you know, Rob you on a blockchain and therefore the tendency to have the information will be validated by the system itself. That's one example but not AI. It's a blockchain example. So okay, you can take this forward the rest of the question, I
think, you talked about blockchain, but human AI can be used for converting unstructured data into structured data. You will not not have hordes and loads of reports, company filings like we are doing it ourselves. I will give a classical example. In our index business for index rebalancing. We read all corporate filings, corporate filings, corporate announcements, which they do with exchanges on an average mighty, used to go through 500 corporate filings on a daily basis or manually, one person has to sit, read those 500 documents what they have filed and manually extract the data from those 500 filings. Now look at it 500 filings human one person looking at it, I'm consumption, secondary counsel, here's what we did, when I talked about this company, CQ s which is based on good commercial credit. So they have automated the whole process of doing a daily basis, they download those 500 600 circulars extract we have told them we have trained the machine what we need, what they need to look for, they get into the document, extract the data, give it to us and our team of people, they just take the data and do the processing. Similarly, if you look at any exchange is document heavy, like if you look at it exchanges, corporates keep filing documents with us. So our broker, they have filed a network certificate, they will file compliance certificate and all this kind of stuff. early adoption was people will look at all the data from the data and put it in Excel or software or whatever it is now B or C will display the major difference. Permission VM automated this process for all the corporate filings members following network file is compliances etc, we have automated it is all automatically extracted, put it into the wherever the data is has to be stored, and then our people don't spend time on extracting data but they are spending the time in much more productive way of analyzing the data isn't the same thing chartered accountants, seers, they also deal with document pages and pages of documents. So it will help them in reducing those manual work with you to keep doing it on a daily basis.
Once again, I would like to thank all our speakers for taking our time and being with us today. And with that, we come to the end of today's webinar. So I hope all our audiences found this event to be insightful and engaging as much as we enjoyed on putting it together. But before we wrap up, Tim economic times wishes all audiences safety and the Minister of Health Have a great evening ahead, everyone.
Watch the entire interview here https://www.youtube.com/watch?v=cOILh87KVTE