1 00:00:05,960 --> 00:00:08,280 Speaker 1: Welcome to the Fear and Greed Business Interview. I'm sure 2 00:00:08,320 --> 00:00:11,920 Speaker 1: n Alma AI is everywhere and investing is no exception. 3 00:00:12,440 --> 00:00:15,080 Speaker 1: I wanted to look today at how artificial intelligence can 4 00:00:15,120 --> 00:00:17,880 Speaker 1: be used in the investment process by someone who actually 5 00:00:17,920 --> 00:00:20,280 Speaker 1: does it, and then the big question, if you can 6 00:00:20,400 --> 00:00:22,439 Speaker 1: use AI to pick stocks, does it actually mean the 7 00:00:22,480 --> 00:00:25,880 Speaker 1: results will be better? As always, this is general information 8 00:00:26,040 --> 00:00:28,560 Speaker 1: only and you should always see professional advice before making 9 00:00:28,600 --> 00:00:33,280 Speaker 1: investment decisions. Z chen is Senior vice president, portfolio manager 10 00:00:33,280 --> 00:00:36,640 Speaker 1: and research lead at Acadian Asset Management. Z Welcome to 11 00:00:36,680 --> 00:00:37,240 Speaker 1: Fear and Greed. 12 00:00:37,560 --> 00:00:38,880 Speaker 2: Thanks for having me on the show. 13 00:00:38,920 --> 00:00:43,040 Speaker 1: Sean, Okay, how are you using AI? Explain it to us. 14 00:00:43,680 --> 00:00:46,640 Speaker 2: Yeah, so we're using AI in a number of different 15 00:00:46,640 --> 00:00:49,680 Speaker 2: ways across our process, and we're really trying to leverage 16 00:00:49,720 --> 00:00:52,520 Speaker 2: the specific strengths of AI for each of these applications. 17 00:00:53,200 --> 00:00:55,120 Speaker 2: So I give you a couple of examples. You know, 18 00:00:55,200 --> 00:00:57,920 Speaker 2: one way we use AIS to find hidden patterns in 19 00:00:58,080 --> 00:01:00,480 Speaker 2: massive data sets. So you know, if you're looking at 20 00:01:00,520 --> 00:01:03,160 Speaker 2: market data and you're looking for something like declining short interest, 21 00:01:03,600 --> 00:01:06,520 Speaker 2: rising volumes, and price strength, you know that's usually symptomatic 22 00:01:06,560 --> 00:01:10,000 Speaker 2: for short squeeze happening. In the market. What the AI 23 00:01:10,120 --> 00:01:13,040 Speaker 2: allows us to do is basically to look for that 24 00:01:13,120 --> 00:01:15,840 Speaker 2: pattern over the forty three and a half thousand stocks 25 00:01:15,840 --> 00:01:18,640 Speaker 2: that we cover, and so what we're leveraging there is 26 00:01:18,680 --> 00:01:21,840 Speaker 2: the power of machine learning to basically see everything, right. 27 00:01:22,640 --> 00:01:26,039 Speaker 2: We also use AI, for example, to forecast fundamentals, so 28 00:01:26,200 --> 00:01:29,240 Speaker 2: you know, the earnings of a company. A human analyst 29 00:01:29,360 --> 00:01:32,880 Speaker 2: is going to rely on their intuition and historical experience 30 00:01:33,640 --> 00:01:36,280 Speaker 2: to do that. The AI is doing the same thing, 31 00:01:36,319 --> 00:01:39,160 Speaker 2: but it's learning from the data, right, So the AI 32 00:01:39,280 --> 00:01:44,360 Speaker 2: is looking for where analysts have consistently overestimated or underestimated earnings. 33 00:01:44,680 --> 00:01:46,679 Speaker 2: And so the advantage that AI brings there is that 34 00:01:46,720 --> 00:01:50,920 Speaker 2: ALI can also be perfectly objective because it knows what 35 00:01:50,960 --> 00:01:53,840 Speaker 2: the actual you know, future earnings have been in the past, 36 00:01:53,920 --> 00:01:58,320 Speaker 2: and it's always correcting itself to hit those particular numbers. 37 00:01:59,160 --> 00:02:00,840 Speaker 1: Can I just ask on that and Z. So the 38 00:02:00,880 --> 00:02:03,920 Speaker 1: first one I get in terms of the patterns in 39 00:02:03,960 --> 00:02:05,880 Speaker 1: the data, and you can actually see where short selling 40 00:02:05,960 --> 00:02:10,480 Speaker 1: is happening without I mean, it almost comes to automatically fantastic. 41 00:02:11,360 --> 00:02:14,080 Speaker 1: The second one, though, is it that is more about 42 00:02:14,080 --> 00:02:16,760 Speaker 1: analyst behavior, So you're actually looking at a bunch of analysts. 43 00:02:16,840 --> 00:02:20,799 Speaker 1: You're saying person A continually is you know, four percent 44 00:02:20,919 --> 00:02:23,919 Speaker 1: over what comes in because they're optimist, Person BS four 45 00:02:23,960 --> 00:02:26,799 Speaker 1: percent under, and so you're using AI to basically bring 46 00:02:26,840 --> 00:02:32,079 Speaker 1: all that together and say so, based on their past performance, 47 00:02:33,000 --> 00:02:34,480 Speaker 1: this is where it's probably going to be. 48 00:02:34,760 --> 00:02:36,639 Speaker 2: Is that the idea that's right? So that is one 49 00:02:36,639 --> 00:02:38,120 Speaker 2: of the things that we can correct for. We can 50 00:02:38,200 --> 00:02:41,079 Speaker 2: look at the track record of every single analyst who 51 00:02:41,080 --> 00:02:44,640 Speaker 2: reports to see where they're consistently over and under. And 52 00:02:44,680 --> 00:02:47,080 Speaker 2: the other thing, in fact that happens is that, you know, 53 00:02:47,120 --> 00:02:50,240 Speaker 2: we find that analysts, as a cohort, they tend to, 54 00:02:50,280 --> 00:02:51,919 Speaker 2: you know, when you're a year out from when a 55 00:02:52,000 --> 00:02:54,960 Speaker 2: company actually announces the earnings, the group tends to be 56 00:02:55,000 --> 00:02:57,079 Speaker 2: overly optimistic. And what happens is, you know, as the 57 00:02:57,160 --> 00:02:59,720 Speaker 2: year progresses and your cost out isn't quite as successful 58 00:02:59,720 --> 00:03:01,840 Speaker 2: as what it would be, and you haven't grown revenue 59 00:03:01,919 --> 00:03:04,280 Speaker 2: quite as much as you think, you know, those estimates 60 00:03:04,320 --> 00:03:05,960 Speaker 2: tend to come down so that when you're, you know, 61 00:03:06,000 --> 00:03:09,600 Speaker 2: two or three days before the actual earnings are announced, 62 00:03:09,720 --> 00:03:12,720 Speaker 2: those analyst estimates are pretty spot on. Now the issue 63 00:03:12,760 --> 00:03:14,560 Speaker 2: is going to be, you know, when you're comparing company 64 00:03:14,600 --> 00:03:17,880 Speaker 2: A versus company B. Company A is reporting, you know, 65 00:03:18,040 --> 00:03:19,760 Speaker 2: a year out in the future, and company B is 66 00:03:19,800 --> 00:03:22,360 Speaker 2: reporting three days in the future. Company A is just 67 00:03:22,400 --> 00:03:25,960 Speaker 2: going to look too good from an analyst perspective. We 68 00:03:26,040 --> 00:03:27,200 Speaker 2: can correct for that effect. 69 00:03:27,600 --> 00:03:30,280 Speaker 1: Yeah, okay, So other uses are the AI uses? 70 00:03:30,919 --> 00:03:32,919 Speaker 2: Yeah? Absolutely. I mean one of the big uses right 71 00:03:32,919 --> 00:03:37,360 Speaker 2: now is obviously using AI to extract insights from unstructured data. 72 00:03:37,760 --> 00:03:39,760 Speaker 2: So I mean, for example, we can get an AI 73 00:03:39,880 --> 00:03:43,520 Speaker 2: to listen in on analyst earnings calls, and you know, 74 00:03:43,560 --> 00:03:46,720 Speaker 2: the AI learns that there's a difference between for example, 75 00:03:46,760 --> 00:03:49,480 Speaker 2: when a CEO says we expect revenues to grow year 76 00:03:49,480 --> 00:03:51,640 Speaker 2: on year by you know, ten and a half percent, 77 00:03:52,080 --> 00:03:56,320 Speaker 2: versus when that CEO says we expect potential earnings uplift. Right, 78 00:03:56,400 --> 00:04:00,800 Speaker 2: the former when they're very specific, that had a much 79 00:04:00,920 --> 00:04:05,320 Speaker 2: more positive relationship with future returns than when they're vague. 80 00:04:05,320 --> 00:04:07,600 Speaker 2: So what the what the AI is doing there is 81 00:04:07,640 --> 00:04:11,120 Speaker 2: that it's also continuously learning from everything it sees, and 82 00:04:11,160 --> 00:04:15,840 Speaker 2: it's learned to distinguish between being vague and being quite specific. 83 00:04:16,560 --> 00:04:18,920 Speaker 1: It's an interesting one that because I have worked in 84 00:04:19,480 --> 00:04:24,400 Speaker 1: communications enlisted companies and every word in that media release 85 00:04:24,480 --> 00:04:28,720 Speaker 1: or every word in that AX release is looked at, considered, 86 00:04:29,400 --> 00:04:30,920 Speaker 1: And I mean, this is what you're kind of saying, 87 00:04:30,920 --> 00:04:32,800 Speaker 1: you're actually being able to pick up or AI is 88 00:04:32,839 --> 00:04:35,960 Speaker 1: being able to pick up the nuances that companies are making. 89 00:04:36,480 --> 00:04:38,839 Speaker 2: Yeah, that's an excellent point that you've made, which is, 90 00:04:39,360 --> 00:04:42,800 Speaker 2: you know, the official release has often been prepared weeks 91 00:04:42,839 --> 00:04:45,839 Speaker 2: months ahead by a team in the back end. What's 92 00:04:45,880 --> 00:04:48,440 Speaker 2: really interesting is actually when you get into the question 93 00:04:48,600 --> 00:04:52,120 Speaker 2: answer section, that unstructured bit, right, because that's where you 94 00:04:52,160 --> 00:04:54,960 Speaker 2: can really draw out insights that maybe the company did 95 00:04:55,000 --> 00:04:55,880 Speaker 2: not want to release. 96 00:04:56,440 --> 00:05:00,359 Speaker 1: Yes, yes, okay, how do you use that? So you 97 00:05:00,400 --> 00:05:04,480 Speaker 1: can draw out these insights? And then are you using 98 00:05:04,520 --> 00:05:07,679 Speaker 1: AI to make an investment decision? Are you using AI 99 00:05:07,760 --> 00:05:11,800 Speaker 1: to inform an individual making an investment decision? How does 100 00:05:11,839 --> 00:05:12,200 Speaker 1: that work? 101 00:05:12,760 --> 00:05:16,560 Speaker 2: Yeah, so we run basically an automated investment process, right, 102 00:05:16,960 --> 00:05:20,000 Speaker 2: the AI sort of plugs in so you know, we 103 00:05:20,080 --> 00:05:22,920 Speaker 2: have an alpha model. The role of the alpha model 104 00:05:23,000 --> 00:05:26,240 Speaker 2: is to predict the expected returns across every single stock 105 00:05:26,279 --> 00:05:29,200 Speaker 2: that we cover, with all forty three thousand of them. 106 00:05:29,800 --> 00:05:31,960 Speaker 2: And what that means is that you're able to produce 107 00:05:32,160 --> 00:05:35,839 Speaker 2: this friend of forecasts. They then get fed into a 108 00:05:35,880 --> 00:05:38,920 Speaker 2: portfolio construction process where the idea is we want to 109 00:05:38,960 --> 00:05:42,120 Speaker 2: pick an optimal set of stocks based on those our forecasts, 110 00:05:42,400 --> 00:05:45,640 Speaker 2: based on transaction costs, based on risk characteristics, to give 111 00:05:45,640 --> 00:05:49,880 Speaker 2: you a portfolio with strong positive expected returns, but that 112 00:05:49,920 --> 00:05:54,159 Speaker 2: are also within a certain risk framework. Right. And that 113 00:05:54,240 --> 00:05:58,840 Speaker 2: uses an optimization engine which also leverages AI in ords 114 00:05:59,120 --> 00:06:03,760 Speaker 2: to find the optimals. And then once we have that portfolio, 115 00:06:03,960 --> 00:06:07,120 Speaker 2: then it gets center trading and AI also comes through 116 00:06:07,160 --> 00:06:10,599 Speaker 2: that process in order to determine optimal order outing for example. 117 00:06:11,080 --> 00:06:18,200 Speaker 1: Stay with me, z We'll be back in a minute. 118 00:06:19,720 --> 00:06:24,240 Speaker 1: My guest this morning is Zchen from a Kadian asset management. 119 00:06:26,120 --> 00:06:29,400 Speaker 1: You went through the process ahead of the break. How 120 00:06:29,480 --> 00:06:34,560 Speaker 1: much value does it add because at the end of 121 00:06:34,600 --> 00:06:40,240 Speaker 1: the day, it still relies on endless forecasts, It still 122 00:06:40,279 --> 00:06:44,760 Speaker 1: relies on words used by management in a release or 123 00:06:44,839 --> 00:06:49,440 Speaker 1: what they're talking to analysts about in the post results conference. 124 00:06:50,440 --> 00:06:52,640 Speaker 1: I think probably the shorting in that is a bit 125 00:06:52,640 --> 00:06:54,640 Speaker 1: different because it's kind of that's actually real money and 126 00:06:54,680 --> 00:06:57,680 Speaker 1: you can actually see what people are thinking. How much 127 00:06:57,880 --> 00:07:00,920 Speaker 1: I mean, have you got sort of data that can 128 00:07:00,960 --> 00:07:01,840 Speaker 1: show that it works? 129 00:07:02,600 --> 00:07:04,760 Speaker 2: Yeah, certainly, I mean I think to show that it 130 00:07:04,800 --> 00:07:07,679 Speaker 2: works you know, you need to look at our performance history, 131 00:07:08,200 --> 00:07:11,760 Speaker 2: and I think it has beed up, you know, throughout 132 00:07:12,160 --> 00:07:14,720 Speaker 2: the track record of the strategies. But you know, I 133 00:07:14,720 --> 00:07:18,000 Speaker 2: can talk more specifically around you know, where AI is 134 00:07:18,040 --> 00:07:21,320 Speaker 2: really useful compared to say, human analysts, and where perhaps 135 00:07:21,320 --> 00:07:23,720 Speaker 2: you know, the human hours still have an edge. So 136 00:07:23,760 --> 00:07:26,400 Speaker 2: if you think about an AI driven model and how 137 00:07:26,400 --> 00:07:28,600 Speaker 2: that compares to a human analyst, you know, the human 138 00:07:28,640 --> 00:07:31,360 Speaker 2: expert is going to know everything about the company, right, 139 00:07:31,440 --> 00:07:35,080 Speaker 2: the cash flow projections, the project pipeline, is the management structure, 140 00:07:35,280 --> 00:07:39,240 Speaker 2: et cetera. Where they may not be strong in is 141 00:07:39,320 --> 00:07:44,440 Speaker 2: actually determining whether any one of those characteristics is particularly 142 00:07:44,480 --> 00:07:47,360 Speaker 2: good or bad for predicting x amount of returns? Right, 143 00:07:47,440 --> 00:07:51,160 Speaker 2: so how good is having marginally better management in a 144 00:07:51,360 --> 00:07:55,600 Speaker 2: expect return sense. What the AI driven model does is 145 00:07:55,800 --> 00:07:58,960 Speaker 2: we're not actually talking about one you know, AI module. Typically, 146 00:07:58,960 --> 00:08:01,200 Speaker 2: what happens is we have a whole bunch of different 147 00:08:01,240 --> 00:08:04,720 Speaker 2: AI modules and each of those modules are assessing one 148 00:08:04,840 --> 00:08:08,320 Speaker 2: very specific aspect of a company. Right, So think about 149 00:08:08,360 --> 00:08:12,320 Speaker 2: you know, the AI that's forecasting earnings, that's its only job, 150 00:08:12,880 --> 00:08:14,680 Speaker 2: and what it's going to do is it's going to 151 00:08:14,680 --> 00:08:18,760 Speaker 2: look for information that is additive to the consensus estimate 152 00:08:19,120 --> 00:08:21,440 Speaker 2: for that company, and then it will do that by 153 00:08:21,480 --> 00:08:24,320 Speaker 2: you know, correcting for analyst biases, which you talked about 154 00:08:24,440 --> 00:08:26,520 Speaker 2: a little bit earlier. It can also bring in other 155 00:08:26,600 --> 00:08:30,000 Speaker 2: information like looking at you know, suppliers and customers of 156 00:08:30,000 --> 00:08:33,080 Speaker 2: that particular company to see if they're getting upgraded or downgraded. 157 00:08:33,360 --> 00:08:35,840 Speaker 2: It can look at you know, the reported fundamentals of peers, 158 00:08:35,880 --> 00:08:37,800 Speaker 2: for example, to see if they're getting better or worse. 159 00:08:38,200 --> 00:08:41,480 Speaker 2: But it has, for example, no idea about the management quality. Right, 160 00:08:41,720 --> 00:08:43,960 Speaker 2: we'll have another AI module that will look at that 161 00:08:44,200 --> 00:08:46,880 Speaker 2: or you know, it's looking for, for example, technical training patterns 162 00:08:47,400 --> 00:08:49,800 Speaker 2: for that particular stock. The advantage that you have there 163 00:08:49,960 --> 00:08:51,920 Speaker 2: is that you know, each of these modules can also 164 00:08:52,080 --> 00:08:56,200 Speaker 2: measure their own effectiveness. Now, what is its strength and 165 00:08:56,440 --> 00:09:00,000 Speaker 2: how good is that ability in actually predicting future returns? 166 00:09:00,360 --> 00:09:03,000 Speaker 2: And that allows us to basically bring together all these 167 00:09:03,000 --> 00:09:05,480 Speaker 2: different insights in a very measured way to give a 168 00:09:05,559 --> 00:09:08,720 Speaker 2: very precise expected outcome for a stock. Right, that's very 169 00:09:08,760 --> 00:09:11,160 Speaker 2: hard to do from a fundamental point of view, Where 170 00:09:11,320 --> 00:09:15,120 Speaker 2: where does the human analyst help? Well, look, AI at 171 00:09:15,120 --> 00:09:19,680 Speaker 2: the moment, you know, can only consume media. What it 172 00:09:19,760 --> 00:09:22,440 Speaker 2: can't do is actually probe right. So you know, we 173 00:09:22,520 --> 00:09:25,240 Speaker 2: talked a little bit about sitting in on that earning scores. 174 00:09:25,480 --> 00:09:27,320 Speaker 2: We're listening to the Q and A, but we're not 175 00:09:27,400 --> 00:09:31,240 Speaker 2: posing the questions and so you know, as a human analyst, 176 00:09:31,320 --> 00:09:33,640 Speaker 2: that's where you can add additional value. And you know, 177 00:09:33,679 --> 00:09:36,679 Speaker 2: another example is, you know you have activist investing right 178 00:09:37,200 --> 00:09:38,920 Speaker 2: from our point of the investing is that it's a 179 00:09:39,000 --> 00:09:41,960 Speaker 2: passive exercise that we're consuming the data and we're picking 180 00:09:41,960 --> 00:09:44,240 Speaker 2: stocks that we think will better than other stocks. An 181 00:09:44,280 --> 00:09:46,720 Speaker 2: activist investor will come in and they will look for 182 00:09:46,760 --> 00:09:50,520 Speaker 2: a turnaround opportunity where their own actions can lead to 183 00:09:50,840 --> 00:09:54,040 Speaker 2: that company then out performing down the line. We're not 184 00:09:54,120 --> 00:09:57,040 Speaker 2: quite there yet with our suite of AI tools. 185 00:09:57,360 --> 00:09:58,959 Speaker 1: Do you think you get? I mean, another example I'm 186 00:09:58,960 --> 00:10:01,880 Speaker 1: guessing is like legislat change. Like, so AI probably can't 187 00:10:01,920 --> 00:10:06,360 Speaker 1: predict legislative change yet, and if there's big legislative change 188 00:10:06,400 --> 00:10:08,920 Speaker 1: that can make big difference to equity prices, et cetera, 189 00:10:09,400 --> 00:10:12,400 Speaker 1: these sorts of things that will it get there? Do 190 00:10:12,440 --> 00:10:12,760 Speaker 1: you think? 191 00:10:12,880 --> 00:10:16,280 Speaker 2: Yeah, that's a great question. So in terms of legislative change, 192 00:10:16,360 --> 00:10:20,440 Speaker 2: you know that kind of information already flows into the 193 00:10:20,520 --> 00:10:22,840 Speaker 2: model through a couple of different avenues. So one would 194 00:10:22,880 --> 00:10:26,840 Speaker 2: be the newsflow. Right, Often people are writing about the 195 00:10:26,880 --> 00:10:30,600 Speaker 2: impact of legislative change on different companies. Now we consume 196 00:10:30,679 --> 00:10:34,520 Speaker 2: that news flow through using AI to read the news, 197 00:10:34,960 --> 00:10:38,080 Speaker 2: and that will impact what our forecasts are going to be. 198 00:10:38,240 --> 00:10:40,120 Speaker 2: And of course the other avenue that it comes through 199 00:10:40,240 --> 00:10:43,520 Speaker 2: is through you know, broker analysts themselves. They will be 200 00:10:43,559 --> 00:10:45,079 Speaker 2: looking at what the impacts are and they will be 201 00:10:45,160 --> 00:10:49,439 Speaker 2: upgrading and downrunning their numbers with respect to prospective legislative change. 202 00:10:49,800 --> 00:10:52,400 Speaker 2: And again, you know, we will draw from that information 203 00:10:52,480 --> 00:10:54,120 Speaker 2: source to make an investment decision. 204 00:10:54,679 --> 00:10:57,240 Speaker 1: Yeah, fascinating. Is there a job for an analyst in 205 00:10:57,280 --> 00:11:04,480 Speaker 1: ten years? Look, I thinks. 206 00:11:03,760 --> 00:11:07,600 Speaker 2: I think from our perspective, it's still extremely important to 207 00:11:07,600 --> 00:11:11,760 Speaker 2: have humans in the loop, right. You know, AI does 208 00:11:11,880 --> 00:11:15,920 Speaker 2: need an immense amount of guidance in order to do 209 00:11:16,000 --> 00:11:18,280 Speaker 2: its job properly within an investment context. So, you know, 210 00:11:18,440 --> 00:11:20,880 Speaker 2: let me give an example of using AI poorly. Right, 211 00:11:21,520 --> 00:11:24,720 Speaker 2: I log into chat GBT and I asked chatgbt to 212 00:11:24,720 --> 00:11:27,240 Speaker 2: build me a portfolio of stocks. They will outperform the market. 213 00:11:27,720 --> 00:11:31,600 Speaker 2: It will probably do a terrible job of that. And 214 00:11:31,800 --> 00:11:33,440 Speaker 2: you know the reason for that is because you know, 215 00:11:34,040 --> 00:11:36,559 Speaker 2: it does not have access to market data, it doesn't 216 00:11:36,559 --> 00:11:38,920 Speaker 2: have access to any podcasts, it doesn't have back in 217 00:11:39,000 --> 00:11:41,600 Speaker 2: financial models. But the main thing is that like actually, 218 00:11:41,920 --> 00:11:46,160 Speaker 2: it doesn't have that underlying mechanism to translate insights into 219 00:11:46,200 --> 00:11:49,960 Speaker 2: return forecasts. Yeah, that's what where the humans still come in. 220 00:11:50,000 --> 00:11:53,000 Speaker 2: It needs to tell the AI how to make that connection. 221 00:11:53,679 --> 00:11:55,240 Speaker 1: Ze, thank you for talking to Fear and Greed. 222 00:11:55,480 --> 00:11:56,640 Speaker 2: Thanks John, thanks for having me. 223 00:11:57,040 --> 00:12:00,400 Speaker 1: That was the chain of Senior Vice president, Portfolio and 224 00:12:00,480 --> 00:12:04,280 Speaker 1: Research Lead at Arcadian Asset Management. This is the Fear 225 00:12:04,280 --> 00:12:06,880 Speaker 1: and Greed Business Interview. Remember this is general information only. 226 00:12:06,920 --> 00:12:09,960 Speaker 1: You should seek professional advice before making investment decisions. Join 227 00:12:10,080 --> 00:12:12,440 Speaker 1: us every morning for the full episode of Fear and 228 00:12:12,480 --> 00:12:16,120 Speaker 1: Greed business news you can use. I'm Chane Elmer. Enjoy 229 00:12:16,160 --> 00:12:16,480 Speaker 1: you today.