Ex-Balyasny PM: “Automation will increase demand for hedge fund talent.”

I think one of the best secret is fundamental investors are not good at sizing on average. Confunds are really good at sizing. Anything that's related to historical pattern matching, that's quite. Anything that is related to how is this time different than history? That's one thing. That's actually the beauty of Confund mental is you kind of can capture alpha that neither group is really looking at. If you're kind of running a few hours to few days of the point side of things, that's more like Blackjack with car county. Especially if you bring a Porsche on. Sometimes a game is not fundamental. Sometimes a game is a decision. You think AI will lead to more efficiency or more inefficiency in markets. I think AI will lead to more. Hey, when I look at the back end of our YouTube statistics, I can see that only around 25% of you guys are subscribed. And I'd like to ask a favor. If you like the content we create and you want to help us grow this show, if you could hit that subscribe button, I would really appreciate it. I promise that in return, I will do my best to give you guys the best content we can with the best guests and best interviews we can do. Thanks and back to the episode. Ying, thank you so much for coming on the pod. Thanks for having me. You were a quantum mental PM at Baliazni. How would you put a position on? I think it involves several steps. I think you're defying quantum mental a little bit. There's obvious piece of quantum mental, which is using our term data, using non-traditional data to be a factor. That's number one. I think there are two other pieces that's important. One is your infrastructure. How do you get the fundamental data faster? How would you update a model? This is before the loop of the world became very popular, but we changed our process so that every model can be updated within two minutes. So that's very important infrastructure. Then there's alternative data. Lastly, sizing. Quantum funds are really good at sizing. So we bring up a lot of the best practices from there. Signing should be determined by Baliazni, your conviction level on the fundamental side. Positioning how crowded this position is. How much upside you think there is. So I think there is a way to quantify and factor those in to determine what the position flies. So those three together is how we kind of put on a position under a quantum mental framework that we go out as a team. What parts of the investment process are fundamental and are quant? Because I think quantum mental is such an umbrella word. I recently had a guest on the podcast saying he hates that word. You think it's the worst thing ever. But you know, idea generation. You know, just walk me through what are the parts that you think or that you know, the quant approach goes great and the fundamental approach goes great. Yeah, that's a great question. I think to distinguish quant versus fundamental in an investment process. The only criteria is anything that can be determined by data only without knowing what company it is falls under quant. So that can be alternative data. So I don't really know what this company is. I just need to know that historically there's a correlation between this data set and the reporter's results that should be quant. Same similar to sizing, we're talking about sizing, right? I don't really need to know what this company is. I don't really need to know what situation this company is in. Then that should be quant. Fundamental are things, oh, set another way. Anything that is based on historical pattern matching. Can be quant. Anything that falls outside of historical pattern matching. So let's say fundamental, I know this company historically beat it, beat 90% of the time. But I know this quarter, something is different. They're working on a deal. The CEO is a new CEO. There is an incentive to kitchen sink, the guidance. All these factors that forces you to study about this particular company. And that is historically like agnostic to historical data patterns. That should be fundamental. So that's kind of how I distinguish it. I don't think any particular process like idea generation or you know, I'm tracking my thesis. I don't think that any part of that is fundamental versus quant. It's more just what information are you trying to derive from all the bits and pieces you have. Anything that's related to historical pattern matching, that's quant. That's fundamental. Is there any difficulty associated with having to think in those two different terms? I imagine if you're tracking the thesis that requires constant monitoring of everything. I'm how do you overlay kind of that discretionary judgment with, you know, really understanding sizing deeply, you know, understanding risk management in a quantitative way as well? Great question. When I think about how you pay attention to kind of both sides of coin, which I think every investment should have both sides, right? The good thing about quant is it's well studied, it's well explained, well for related. Right? So my actually view is doing quantum mental puts a lower burden on your brain because a portion of your process can be automated now based on well-defined framework. So sizing, I cannot tell you how much time we've spent thinking about sizing. And I was in most of that before we ran the quantum mental, right? On the pure fundamental basis, there's a lot of emotion that goes into it. A lot of art that goes into sizing. But if you just objectively look at the data, I think most fundamental investors are not backward sizing. But sizing is actually pretty well solved, I think, problem in the quantum world. So if you take that burden out, then you can focus on things that's truly art, right? We think management, this quarter is motivated differently, historically. We think they have a motivation to put up this kind of print and I think that's not what the market is expecting. That's art. That's much more well less studied. Well, you don't have enough data to do a back row working pattern matching. That's a great time to spend. I think that's what fundamental investors should spend time on. Is less about the pattern matching, but how is this different? If you're a student who wants to work at a great trading firm, listen up. Our brand partner, Onyx, the largest oil derivatives trading firm in the world, is hiring junior rust developers. They're opening this up to people who haven't written the line of rust in their lives because they'll train you from the ground up. From day one, you're on a small team working on real projects, learning from senior engineers who built the systems. If you're self-taught with serious projects to show for it, apply the link below. You give an example of a particular name or something that you were doing at the time that illustrates how the judgment portion came into how you mix the fundamental and quantum approaches together. Yeah, I think it's a little bit hard to tell stories while sizing because those are very small and stuff like that. I can tell you some interesting examples of using quant or quant data, but based on fundamental knowledge so that we can put everything together. I was covering insurance companies. One of the things that's always hard is how do you forecast a loss? And you sure as having different businesses. A Coring Insurance Company had pretty stable loss because there's a lot of numbers. It's pretty stable loss, but to predict that every quarter is very much a fingering the company exercise. I'm like, okay, this quarter, the temperature, you know, it's 4Q. The temperature is not that cold. It's less snow this year than last year. So probably it's, you know, five to ten basis of point better loss ratio than last year. That was kind of the fingering the wind process, generally speaking people use. And then on the other hand, you have that these one-off catastrophes, wildfires, hurricanes, and kind of use kind of a very generic way to figure out, okay, this is industry estimate losses. And this is how that translates to a loss ratio in my model. So we actually took a separate different approach for both of these. Number one, take a Coring Insurance Company. We actually scraped about 15 different states, highway patrol data of 15 different states. We're using that data then you can kind of model out where a frequency of accidents. We use a duration of car accidents as a proxy for severity. So how long did it take for that accident to clear out? The longer it takes the more severity it is. And because of each company, we have their state premium market share. We can then wait those results to kind of end up in fact as to whether historical reporter results and all of that energy backtested. Now if you're running this as a court, it doesn't work. The reason is there are about like three auto insurance companies that you can do this backtests. It's not enough data. Plus, every insurance company in three quarter will adjust their reserves. So you have to back out that reserve adjustment thing. You cannot just look at the reporter results. So you kind of need the fundamental knowledge in that analysis and to use a ton of data that's more on the coincide to combine it to do this analysis. So for us, the beauty is we have a live tracking loss ratio into the quarter. We never have to like adjust it every quarter. We just look at a live every single day, how is that training, how is that versus consensus? Are we far enough on consensus that we can make a bet? So that's one example. A second example is those one off are catastrophes, right? I still remember this is one of the most fun projects I've done. is this was a, there was a, there was a wildfire in the Malibu area. So that is a lot of mentions. So it's a low frequency, high-stabary event. It's really hard to predict those losses. And this is before any industry consultants coming up with any industry losses that you can just take a market share of. So because this was around Malibu, there's a lot of movie stars that actually tweeted on X and Twitter at the time, that their house cover announced all like that. So I actually tracked every single tweet about someone saying their house burn out, I kind of stalk them online and basically no roughly where they live. And it kind of started drawing map. This circle, everybody said, they're house cover now. So you can assume everyone in the circle, their house cover now. This circle someone said, my house is half damaged and you can draw a ring. And then you overlay the property data in this ring, how much value the property is. And I remember we ran through the exercise. We were like very, very accurate. And of course, the expectations less about, oh, I'm going to get it to the last digit to be as close as it. It's more just like that exercise gave us a sense of how big of catastrophe this is very quickly and very reliably. So those are two kind of interesting examples. I think I call these quantum mental, because I actually think it falls outside of both fundamental and quantum analysis kind of route. It falls out of fundamental because it's a lot of data gathering, a lot of interesting data gathering, a lot of scraping, a lot of that. It kind of falls out of the quantum, too, because quantum historical can only work with data that is applicable for thousands of tickers so that they can do that back test. Something like this, well, a flying mellum would never happen. How do you back test that? So I think to me, that's actually the beauty of quantum mental is you kind of can capture alpha that meter group is really looking at. That was the part that stood out to me that it's not the quantum, you know, a pure quantum approach can't do that, because there's just not enough data. Plus, fundamental is it's hard to actually, you need tools to model that. You know, before we started the podcast, I asked you if you had any contrarian takes. You said there's, you should have a lot of contrarian takes. How do you think that the fundamental investing process can be automated by all this new technology, by agents, by the new models would love to hear your thoughts? - I think a lot of it can be automated. So let's take a step back about what does a fundamental investor's do? There are really three steps. You gather the data, you analyze the data, you process the data, and then you make a decision. Right, regardless of how complex your process is, it's usually those three steps. So take a step back first, gathering data. This can be like actually finding data, you know, alternative data, it can be even gathering reporter results updating our models, it can be gathering commentaries, management say, like gathering data self, I think should be mostly automated. Right, and especially with AI, I think, like historically, I feel like you can only really automate the number, but I think with AI, you can actually automate a lot of the qualitative data sets as well. So I don't know why I think that should be mostly done. Processing data. There's always so many ways to get process that data. Qualitative data, you know, do some regression, you do some analysis on how this is translated to this number. Like that is all problematic anyway. A lot of people just historically do that Excel. You can move that outside of Excel if you need to, or you can even do Excel as well. So that part is, I think, more or less doable. And then think about processing the qualitative data sets. This is, okay, management, these are things that I think people, fundamental investors have an idea in your head. They have a, like, oh, this management is very conservative. This management is, you know, like to kitchen sink. All of those intuition they build over time is really a result of a process data. So think about that. How likely is this management to beat the guidance it gives? That is their credibility score, right? I'll give you an interesting one. And how confident is management? So you often hear a management, a good investor saying, "Hmm, this, the CFO sounds a little bit different this time around on this topic, catbacks, whatever it is." It sounds a little different. That is really the process of data of how did management answer this particular question in this particular topic? That is different than historical how they've done it. So, I'm letting slide one of the things because we do a live transcription. We actually look at bill or words, by management, by topic. Then you can say, "Oh, this CFO, awesome, they historically, there's, you know, 8% of his words are fill or words, it jumped to 20% this quarter." Right? So my point is all the intuition that a fundamental investor has build over time. A lot of it can be kind of a reconstructed using a process data, use law data and kind of processing away and then to get that inside. So I think those two can be automated. The park cannot be automated. In my view is judgment call, right? The two judgment, okay, all these pieces gather together is that, should I buy for myself? And this is a very interesting question. I remember when we first started building the AI product, one of the questions we got from investors in my recovery is, okay, you're using public data sets. You're using close-source models. How can you differentiate? And my pushback is, well, every investor use the same amount of data that's available out there. But every investor come up with more or less a different view on what the issue by itself, that makes up with the market. And what makes a good investor good is they ask the right question and they get to the core of the problem, right? Like the most interesting thing about this quarter should be this metric. That is judgment. I think that part, I can see it day eventually, if alums have all the context it needs to make that judgment call, but I think we're pretty far away from it. So if you think about the three parts of an investor's process, I think the first two parts, data gallery, data processing, data gallery, almost 100% can be automated. Data processing, I think 80% can be automated. I think the hard part is not is how do you process that data? In order to get to this piece of information I want, how do you drug go these while data that you have? So I think that part is still I think human driven. And then the judgment call is still I think 100%. Do you think that the demand for investing talent expands or contracts with those first two parts of the process being automated? I have a contrary view, I think that it's fans. It's kind of like this. This is not first time there is a automation wave investing. You can argue that quant, quant was the first automation wave, like a lot of this is just automated. Oh, think about the other one, indexes. That's an automation way too. Or at least a non-fundamental wave. You'll agree that way, right? And I mentioned earlier the way I think about quant or anything is anything I think about non-fundamental is facing things that is a historical pattern matching. It's something that agnostic you what this company is. I think AI is kind of falling out track too, right? Like AI doesn't really know the company. AI is not even meeting with you to see management body and languages and stuff like that. You still have more context about things that AI cannot seek. AI is just still pattern matching. Like AI by definition next next token prediction is pattern matching. So I think it's more just like a it's like another wave of quant. But this time less on number data and more on like more data that's kind of high sea. So in that construct, you argue certain alpha is collapsed, certain alpha is collapsed. But meaning like things that everyone should be able to back test. Like I think of all the like you know, you know, old maybe the analogy would be like some of the technical analysis. Right? Everyone can see the same thing. Everyone can analyze very quickly. Those alpha collapse. But this when that alpha collapse, it does two things. Number one is the other things that's not pattern matching. Here is more weight. I don't know if anyone's ever mentioned like I feel like, I'm just smooth, but I feel like the volatility of earnings print has significantly increased. Right? Meaning like like when I draw, okay, when I started in English, in 2010, literally like to be fair, called a very boring English street. And sure it's like a two three percent movie insurance line is a huge. But I thought I lost, I mean like you see intraday 15% 20% move. The reason is I think the predictable thing forces everyone to be more or less in the same side of the boat. So if you can have a differentiated view, you'll get paid more. Here's more paying out there. So from that perspective, I actually think that issue yield more people to do more things that only human can do because there's actually more opportunity. Another way to think about it is credit card data. When credit card data first came on board, I think it was a very easy alpha. Like I think the Hatch one said like a five to six year lead of like just easy money. But if you talk to anyone who using credit card data to analyze, you see two things. No one want people generally would complain. It's a lot harder. Like a lot of people buy the credit card data, not to generate alpha, just to just to know what everyone expected. Right? You can just see it said the bar for everyone. The second thing is, you'll see. crazy moves for our earnings. Because usually those crazy moves are things that's not predicted by the credit card data, hence everyone missed it, hence everyone used to write that get to the right side. So I generally believe any type of automation, any type of pattern matching, collapse also in one area, but expense opportunity in others. And I think that should allow for more human forces to look for those opportunities. I find it fast today when you said that a differentiated view today pays more and talking about the volatility of earnings for the print. I mean, you see all these mega caps trade like penny stocks. Yeah, not even earnings, you know, one article as a trainee, you know, a sub-stacker, right? And then everything just moves. It's truly crazy. Give me your take on AI and investing. Why can't I just get cloud code to do everything? Why can't I just, you know, prompt cloud code, can you build some, you know, extract some data, do some research, you know, connect it to, I don't know, Carbon Arc or one of these data providers, data bento and Carbon Arc and just let it rip. Let me start my own quantum mental book for my bedroom, you know? I think, first, I'm a huge problem of AI, obviously I'm building an company right now. But I think there's a lot of limitation of AI people tend to overlook. And very first one I think is what is AI? I'll go back to my point earlier. AI just is, as I pawn for word data. And like AI doesn't truly understand company. It doesn't truly understand market. It can regurgitate things to make you think that understand and understand the market. But it does not truly understand the market. That's number one. It's like, you cannot treat AI as like an investment sub-stab. Even if it can't answer everything, Warren Buffett said, everything monger said, like it's not and it's not. That's number one. Second thing is, I personally think people overstate the knowledge, just even the knowledge it can't regurgitate. What do I mean by that? I don't know if you remember when Chattapet first came out. There's a lot of talk about every industry should fine tune their own model. And I don't know if you remember but Bloomberg trained a model using finance-specific languages. Finance is document. The result of that was a pretty large model that underperformed a smaller over-source model. So people ask questions like, how is that possible? How can you fine tune something with finance-specific knowledge and do worse than a generic model smaller? I think the challenge is thinking about what data are they trading the model on? They talk about finance-specific data. That's transcripts, filings. And they have all those data. Using those data to train doesn't actually teach you about the company. So what do I mean by that? Take any earnings release. Pick a company you don't know. As a human, pick a company you don't know. Pick their earnings release. For the last quarter, pretend you got that earnings release the day before earnings. And that's the only context you have that earnings release. Now predict how it will do the next day. What's your hit rate? If it gets less than 50%, or maybe 50%, I guess, running gas, right? So why? Because reading that earnings release, we had a great quarter. Revenue was record high. blah, blah, blah. It tells you nothing about how good or bad that quarter is. Why? Because you truly tell you how good that quarter is. You need to know this is what they reported. This is the expectation. This is what they reported. This is what they said last quarter. You need to a lot more context than just that data. But language models are not trained with all those contexts provided. So the second point, my point is, it never really learned about the market. That's the second thing. The third thing is, I generally think about horizontal models as a college grad. Horizontal models are training generic capabilities, such as reasoning, such as ability to use tools. It's very similar to a college grad. They come out. They can read the things, they understand things, they can read, they can use Excel, they can use PowerPoint. They can do all of that. But you have no domain knowledge. A college grad, regardless of how smart you are, you're not to be able to throw into a hedge fund and be like, oh, go make money. The only instructions go make money. You're not going to do that. Why? Because there's a lot of context, a lot of domain context. You have to gather over time. So I'll give you an example. You can hook up a cloud or anything, charge a community or whatever, to a conference call or earnings call. Say, hey, what did JP Morgan say last quarter? That's all you. That's a fair typical thing, right? PM asks you, what did they say on the call last quarter? And you'll have a chat ball respond to you. JP Morgan had a strong quarter because they said they had a strong quarter in the press release, right? And then say, total revenue is this. Margin is this. Well, I will stop you there because nobody cares about total revenue for a bank. That's not something people care about. Anything that's more like someone I say is relevant context. But how would it-- how would an element know? Elemental, I always think total revenue is most important thing for a pretty much everything company. So there's a lot of domain knowledge. You need to provide in order to have elements but out answer that's relevant. So I think those three things-- one, it's not really a context thing. And two, it was never trained to understand. And without that understanding, it cannot make judgment call. I think those three things added together fundamentally in the possibility where-- I'll just say, go, do it. Now, I do think there could be a-- I think there could be an experiment to be done. That's possible, maybe to nudge AI in that direction. Number one is provide that context. So we spent last two and a half years building a thicker level domain knowledge. So they know, OK, when I read JP Morgan, these are things I should care about. And by the way, this is a tricky problem of finance. And it's different than law or medicine, right? The truth changes over time. What do people care about this quarter? Well, this quarter might be different in the next quarter. So you always have to update that. That's number one. So if you have a domain knowledge, I assume we have all the right data. And we talk about data too, because I actually don't think that's the same thing. So I think that's the same thing. And I think that's the same thing. You make that. And I go say I will buy JP Morgan. And I see how the stock reacts. When it's dripping your Azure reports, you go figure out like, OK, what did I get right? What did I get wrong? Is there anything in there I should have gotten right based on the data I had? And have all of them actually in the RL and RL and I'm actually learning. I think if we do it that way, potentially, eventually, you will get an AI that can't invest. But we're pretty far from there, because we're missing actually data and knowledge, right? That implied how do you think about providing that context for the models to understand the most important drivers for a particular name or sector? I think to teach AI about market, you need to build a framework on how to teach AI to learn about the market. It's no different than training an intern who has no context about the market. If you're like, hey, I need to learn about Tesla. How does an intern learn about Tesla? OK, start with reading some kind of, thank you, thank you, you understand what they do. No, no, no, no. OK, then you move to where important topics people care about. Then you can either go read the earnings call and say, OK, these are things that people talk about. These are like, management spent x% of time on this topic. Celci asked, five out of eight questions this, you can get clues on what matters. You can look at news, what drives stock. You can go to the forum, you can gather all of that. So the way I think about building that knowledge back, it's a data pipeline. It's a sequential data pipeline. Process of data once you have that knowledge, can you do that? So I would say we spend almost two years just doing that because the challenge is there's so much data you need to gather and it has to be updated in real life, right? There are some news break need to update that. So that's number one. I actually think we recently moved to what I call level two knowledge. Remember I mentioned like, raw data and process data. So all the initial knowledge is built on raw data. What they said in 10K, when they said 10K, or what they say earnings call. But there's a lot of things you can analyze. And those can be added to the knowledge base. So I mentioned about managing credibility. They sound more less confident or the menu credibility, management confidence, right? It can also be like simple things that qualitatively were quantifying a fundamental person's intuition, such as when JPMorgan reports, which stock moves with its results, right? Usually people know, oh, JPMorgan, if they beat an NNI, this ticker should move because that's the best way through. Well, all of that can be quantified. All of that can be mapped. So we can process those raw data to generate kind of derived knowledge that's verifiable, that's factual, that's factual innocence that's verifiable. And then we can store that in the knowledge too. So we just move down to this side. I think eventually there's another-- knowledge that's through RL. You make a decision, you get verified right or wrong from the market, and then you move to that. So that's how I think we built it. I think it's a very long process. I think that is a much harder question to do than building the AR hardest. And AR House is hard enough on its own. But I think that is a much, much harder process. And we're hopefully we have an early start, but I think our work is not going to be done. How are you making sure that the models don't commoditize your business? That's a great question. I think in the age of AI, two build has become easier. Once you have a direction where two build getting there becomes easier. But how two build has not. And I would believe that is a much more important question. So I'll give you some examples. And I think individuality none of them seem like unsurrupable for a model company. But I think collectively together is how you need some of who have done a job, but also have a technical sense to truly rethink about the process. About, okay, this is how you should do this product, right? So making me a very simple example. The way that model company have approached the application layer is they started with a chat and then hooking everything up in your chat. So chat is a main interface. And the first thing I'll say is I don't think chat should be the right interface for everything go thing in a investor's stack. I remember when a lot of like tools out there was doing demos about like, what was Apple's China sale over the last 20 quarters in Type-N and you see like all the answers that was supposed to be like the demo. And I always kind of chuckled at it because I was like, I can get that from Bloomberg a lot faster than you type that question, right? A terminal, you're getting certain data is all faster than a chat. Sometimes I want a dashboard. I want the data to be present when I look at it. I don't want to ask a question. Sometimes I, here's another question. Okay, chat box is great. What if I just want to look at Tesla's last 10k, 10k? I promise you AI has not reduced number clicks. You have to get it. It's not a problem itself. But to me, I think these are all the things one should solve when you're building an AI system for investors. Should it just be chat? Chat right now, I was say 80% of investors are still using chat as a accelerator search, right? Like a kind of replace a maybe digging through the filings and stuff like that. But it should not stop there, right? And investors process involves a lot more than to search. They have to listen to earnings calls. They have to do models. They have to pull alternative data. You have to be supporting every step of that process. So the investor ideally never have to leave your platform. Everything can be done. And everything can be done connected. We'll do that. Well, in the dream scenario, when an earnings release hit, it should update my quarter. When the call is ongoing, it should be updating my next quarter and the next, I'll your number. At least tell me which line I should change, right? You have to feel kind of an integrated process that's not just driven by chat. So I think there's a lot of things, a lot of a kind of taste type of thing that you do consider. And second thing is I think you're to build a right product. It truly requires someone who understand both sides of what is the investor's process and what is the technical capability and the technical limitations. So I have a very simple question, right? Like think about document search. Document search is not a technically not a hard process to build. So here's a question. Why hasn't any large fun replaced that before I came along, right? Before I came along, why didn't anyone try to replace document search? Because everyone's spending a lot of money on vendors about document search and stuff like that. And it's not for lack of trying either. But I think it was really hard to figure out the right workflow that clicked with people. They're like, okay, this is very easy to use. This is actually, I'm going to actually use that as part of my process. So I think like they have to know the workflow process. But there's another flip side of things. If you go around in 2020, even actually now I would say, if you go around to ask 100 fundamental PM analyst, what would you like AI to do? Most of the things they talk about will still be kind of things like, okay, I want you to automate that process. I want like that process revolves searching instances. Or maybe I want you to automate earnings previews. But there's a lot of capability they cannot think about because they don't know that it's possible. They don't know you can actually update model during live calls while it's ongoing. They don't know that there is capabilities to tie things together. They cannot think about it. So you need someone who knows the process. So you know, what is the pain point? But then some will also understand technical. And I think that's really hard. Last thing is I actually think because the model companies are trying to be as horizontal as possible to do integration as easily as possible. They're coming up with these shortcuts that is going to hit a ceiling pretty soon. Some examples, MCPs. I think MCP make collecting data very easy. But what happened when you have 10,000 data to connect? Your MCP is going to eat up all your contacts available. You still have practical to connect 10,000 data using MCPs. Second example is looking at how AI is doing Excel. AI is doing Excel by writing Python code to operate Excel. That creates two issues. One, it is still impacted by limitation of Excel. There's a lot of limitation Excel that you're like, why would you do build it this way? And then two, it's not using the good things about Excel. There's a lot of good things about Excel that you're also not using when you're turning into Python code. So I think there's a lot of shortcuts that the horizontal model layer are doing. That is a very beneficial to bring people on board. But what people quickly realize is one, they cannot truly build something 100%. Maybe you can get the 60% there, 70% there. And then two, I think very soon people will suddenly see the limitations of just like a generic model hucking up with different data and doesn't have any domain contacts. Efficiency. I think AI will lead to more efficiencies in market. I think AI will shrink the delta between a large fund and a small fund. Most of it is data access and resources. I think AI will shrink the delta between a call fund and a fund and a call. And I think I actually think that right now, like you see a lot of use case of, you know, fundamental people trying to accelerate a process. It seems like a more obvious thing. I actually think there's a lot of benefit of AI. That's kind of been utilized by quant. And I know every one of you might be like, oh my god, no, quant has been using AI for the last 20 years. Yes, they've been using machine learning for 20 years or whatever more. I think the challenge is, if you look at the plot, right, again, we talked about this earlier. Quant is a pattern matching. So it's great when your current environment is following a historical trap. So you know, you hit 60% betting ratio and you can make a lot of money. Quant fund loses money with regime changes. It's no longer the same environments before. I actually think AI is very good at identifying those regime changes. AI is a very, it's a narrative. AI is good at narratives, right? And they get pick up changes in narratives. But you don't hear a lot of quant talk about it. I actually think like AI can help quant in a different way than AI helps fundamental. But I think eventually they'll all converge and the gap between the two is going to be a lot smaller. So I think AI will increase efficiencies. Now, I think it's the good thing about equity market is that it is an unsolvable puzzle. Whenever you eliminate some inefficiencies, different forms of inefficiencies will occur. Right? This is what we'll kind of talk about. I think like indexes create certain level of inefficiencies that that an investor can explore. I think quant actually solves certain inefficiencies. They quant has materially increased efficiencies in market. But they also create certain type of inefficiencies that I think actually a fundamental person can explore. Right? And I think AI is going to be the same. It's going to be overall making the market more efficient. But it's going to create certain type of inefficiencies that a PM who can rationalize about, can't understand how that works. How AI is driving those decisions because it becomes a very predictable and can think about why is this time different, can can explore those opportunities as well. So I think overall, it should be more increased. Yeah. Let's say I'm a fundamental PM. What are the most important things you would get me to do right now to, I guess, enhance my process that Gen AI or you've seen Gen AI be able to do through workflows like the ones you're running at employed? If you are a fundamental PM, one thing I actually highly recommend is just sit down with your team to rethink about how you would build your process. knowing what AI can automate today. So it's a little bit less about like, this is what we do currently. Let's find how AI can accelerate different parts because what you end up doing is gonna be a very fragmented process. Like this block are accelerated, but then you end up just copy-pasting things around like, I don't think that is, it really should be okay. Knowing what AI can do today, how can we accelerate? How can we rethink about the process, for scratch, and maybe certain things we shouldn't even touch, maybe certain things we need to redesign, how things are interested. That's number one. Number two is I highly recommend every fundamental team actually allowing someone on the team to learn the technical side. I know there's a view that, oh, you don't need to learn Python out with AI. I actually think someone should at least know the basics of Python, know the basics of quantitative, kind of skill sets. Because you, in order to audit, in order to make sure AI doesn't go down a rabbit hole that you can't fix, you kind of need to still know what the right way of doing things is. So I highly recommend it. So back at the band, my team, we hired kind of fundamental analysts to that is that we're coming from traditional background, thinkers, ex-sensile side, all that. But we made sure everyone learned Python. And I told them, you don't need to be the best programmer, but you need to at least be able to tell me, knowing Python, I would design this process this way, so that it's fully automated. And I know enough to go back to the code that engineer gives me to tell you know this part is wrong, this part is wrong, this is something is wrong. I think that's important. And I think that that ability is still valuable in today with AI, especially given AI, I can turn out things that's so convincingly right, or it looks like so convincingly right. So one small assumption can be completely wrong, right? So I think those are very important. So those two things, I think like rethink up on your process of scratch. And then two, have someone on the team that truly understand your process and the quantitative side of things. - You said that you think the gap between the large funds and small funds shrinks a little bit. What modes do you think expand in this new era? - I think quality of talent is going to be the single most important thing at different funds. And I think the definition of a high quality talent will change in the AI world that before. I think like before, what does it take to be a good career person? You can build models, you can grind, you can learn things. You know, I think some of those are less important now. You can gather information, you can do things very quickly. Like some of those things are less important now. So I think about is important in the AI era is number one, you have to be able to ask good questions and you have to be able to understand the core of business model. Like there is very first thing is that you have to understand core business, how does this business model work? And from career first principle, is this a good business model? All right, and a lot of those, I think it's very interesting. I a lot of this, I think it is size and art. Like it's sizing in a sense that, you know, there's usually a template on these questions, you know, the SWOT analysis, whatever. Like all the business school come up with different ways for you to analyze it. But honestly, it's kind of like just like, who can't ask questions from first principle basis? And I think that skill set, like some of this, it's hard to train, I think. So I think that's number one. Number two is I always said, like it is interesting when I was hiring for BAM, even actually now for For Ancient Years, I had this view that you have to test people's logic. Can someone call from point eight to logically to point B? Is that logic right? That's also something I find very hard to train. Like some people, it's just really hard. Like if you cannot bridge that, it's really hard for me to tell you this is how you draw the conclusion. So I think that's very important too. Because in AI, you can, like, I would say print AI, a junior's person's job is to figure out what A is, what B is, and then the PM draw conclusion about A to B. I think in the A to A, AI will do, what is A, what is B? And anyone actually should be able to figure out, okay, if it's A, then it's highly likely the next step is B, because of these things. Because I think that's the second thing. And then I think the third thing is someone who is willing to learn the technical side of things. I think there's actually varying degrees of how well you can utilize AI. Like we see it in our user too. Like some people are truly pro. Like it's like AI whisper. They know exactly how to ask a question to get what they want. And they know exactly why when they ask this way, it doesn't work that well. And then some people don't are not like that, right? Like, like, sometimes when we encore people, we have to tell them like, no, you cannot just putting a ticker and it's like you have to read your mind about what you want to know, right? So I think there's kind of a sense of like, you need to know what the limitation of AI is. And very much like how you would need to know your junior person. You need to know the member of your team, where are they good at, where are they bad at, how to incentivize them to give you the right answer. They're, I think that's a necessary skill. And last thing I think it's actually very related. People reading skill. Ultimately, I think the only thing that human truly have a advantage of meaning like, it's really hard. I just, I don't even know how to build an AI to replicate that is reading someone. What you're saying, we're in with people and that this goes both like reading your PM, reading your junior, bringing your men to the event and reading the situation, like reading people, understanding that the dynamics and the market about who's incentivized to do work. And because of these different forces of incentives, this is how you should think about it. And that's, hey, this one, I don't know. And this one, I really do think that, I don't know if you hear this comment a lot at all, but I do think it depends on what sector you cover. Depending on what area of hatch luck for it, especially if you're being in a posh shop. Sometimes a game is not forming at all. So sometimes the game is a position you get. I've always kind of compared certain parts of market to be much more similar like a poker game. The who wins the hand is non-necessary, who has the best cards. In this case, who get the fundamental most right? It's understanding the different positioning. This is especially true if you're covering a sector that is not experiencing inflows from retail or mutual funds. So, okay, if you're not in savings, how do you make money in market? Is you need people to follow you, right? As long as there's someone who's willing to buy your shares at a higher price, that would mean someone needs to follow you in order to do that. But if you're in a sector, and by the way, the reason why I can say this is, I cover the sector literally has been abandoned by mutual funds for a long time. So we see this very, very often. If you cover a sector where it's not very popular, no new money is coming in. Then what you end up doing, you're just competing with a bunch of posh shops. And by the way, because everyone's process more or less the same, because they build more or less the same data, your position can be the same. Then if anything happens, a wind, draw downs, and then they'll circle through the circle, right, circle through the whole pot, the different pots. So I think sometimes you also do the recognize what you're playing. You might not be playing a fundamental game. You might be playing a position game. So think about your poker position. And by the way, what matters in a poker game? Your hand sort of the perception of your hand to other people. The position of where you sit, the stack of different players at the table. Those are, I actually think that, actually one day I would love to build a poker game, a poker simulation that I resemble the market. But these are all things actually is very relevant for a fundamental investor that has, that has like a slightly shorter time horizon. A lot of time that matters. So those, I think those are things that matter for a good talent. And some of it requires intuition. And I think maybe some of it intuition really depends on how early do you start training those intuitions, actually, really? Yeah, you can train it a lot earlier. So those are, I would define like, in the age of AI, what kind of talent I'll be looking at. How much of a game like poker do you think translates to fundamental investing? I think poker is extremely similar to investing when your time horizon is a few weeks to a few months, which is kind of what most multi-manager wants to equity fund balls, hedge funds balls, right? Think about, I think about like the super shortening, if you're time-hurry, then a few hours to few days the quant side of things, that's more like blackjack with carcami. It's purely a arts, you just need a slight edge, you keep a betting enough time, you will make money. On the longer end, like the, I'm gonna buy and hold for three years, five years, then you're truly betting on the industry trend, right? Like, I know this will play out, I know this is gonna happen. I think the harder part is, this is the kind of the middle part, is I cannot purely bet on arts, but nor do I necessarily have long enough time horizon for my thesis, my fundamental thesis is truly play out. That I think is truly like poker. So what character resticles does poker half? It is not about your hand. It is about other people's perception about your hand. It is about how people react to your actions based on where they sit and how much stack they have. But can't enough for you to call your boss or not. And it depends on how many people are playing the game. So I go back to like, if you're in a sector that long only are not coming in, it's just their general alphas. Index is not coming in. Retail is not coming in. And your players, who sits at your table are the other multi-manager shops like Prashar's. Now, then you have to look across when the card is laid out of the table because you have similar process, because you use similar data. You more or less know what everyone's trying to present. Or everyone's trying to present. So you more or less do the same thing. What that end up being is everyone's on the same side of boat. So let's go back to our example earlier about credit card data. Everyone more or less on the same side of boat. So that one time, yes, you're right. So the bummer on that is you're probably right as a group. Maybe 60, 70% of time. But because everyone on one side of boat, when you're right, you don't get paid as much as you think because you need to sell to pay and then everyone's selling. But that, not even maybe 20% chance you're wrong, you really get screwed because everyone's getting out. Anyone has to go to the other side. And it's tough, right? So I think being a good investor sometimes have to understand what gave you a claim. And I think the other analogy I've used before is it's kind of like surfing. Ideally, you want to write the big waves. These are the true fundamental waves that's going on in the sector that you have certainty. But you need to know they're coming just in time for you to actually write it. But you can also try to catch the little waves. And I think a lot of the medium term investing is about catching the little waves. And that you have to be careful because it's not as predictable as a fundamental trend nor is it as predictable as like, oh, that I need to be 51% right and I can capture the alpha. So that's kind of how I think about it. - I had a hedge fund manager on the podcast and he was saying that when he looks at the analysts who started around the time he was starting, the ones who ended up in tech were worth 50, ended up being worth 50X, whatever else was worth. Purely by merit, not actually not even by merit, but purely by luck of being in the right sector. Do you think for young analysts today, should you try to select the hottest sector and try to let it run? What are your thoughts? - I think choice matters more in their heart work. So will you choose or will you not choose whatever it is? Like where you end up, which track you're on? Do you matter a lot more than a heart work? So it's kind of like what Munger say, right, go fish where there is fish, right? So there's one. So I would say don't try to go in to a sector that doesn't have alpha. And by the way, you can have a very different year to be. You can have a view that most people don't think there's alpha here, but I know there is alpha here. That's totally fine. All right, I think that's totally fine. That said, I don't generally recommend you picking the hottest sector because everything is cyclical. And unless you have such clear view on where the market going, you cannot predict which sector is gonna be the next hot sector when you become an analyst with capital. I think the worst thing that can happen is actually an analyst at a junior level is gonna ride the wave with their PM. I'm such a great analyst. I know exactly how to do this. This is, I'm very, very good. They attribute like kind of like the luck of a being, the right sector at the right time to their scale. And then they take control of capital, then they invest. And all of sudden they start crashing, right? It could be because the game changed. By the way, the game had changed a lot in the last decade or so. Like I think like the game people were playing 15 years ago, it was very different. And five years go very different than you know now. So I think it's really hard to predict. I think the criteria should just be like, one, do I think there is alpha in the sector? What do I think the alpha come from? Do I have an advantage in getting the alpha? Like for certain sector, actually think like alpha come from you knowing the right people, you know who to talk to. Like there are sectors like that. Then maybe as a new person, you don't have that network. Maybe it's a little bit hard, right? There are a sector that come from, I just need to do differentiated analysis. I can get data and then I can, I just need to be half a step ahead of everyone else. If you think that's what you're good at, then you should do that. But I think chasing the hottest sector, like I think before AI, did people know that Nvidia is going to be the hottest company? Like before AI, did people know people would have guessed I'd say I'm so I'm just going to be the most profitable company? Like it's really hard to predict that. And I think because you're career, especially if you're just starting out, your career is going to be next 10, 20 years. You better be able to predict that. So I think don't go purposely choose a sector that is, you know, I definitely think they're a sector that's just kind of very little alpha. You have to work extra hard to get you like half of where people are. But obviously, I think you can be choosy about those. And then also pick a sector, know where that alpha come from in that sector, and make sure it matches with your strength. My strength is talking to people, figure out these. My strength is to do very deep analysis. My strength is to kind of think continually. Like by the way, if you're in the sector that is not popular, but you're a very contrarian person, you always find things that's different everywhere else. You might be able to make a lot of money in those sectors, just be different everywhere else. So you have to kind of find some track that has alpha in your view. And it's alpha that fits your personality, fits your style, fits your, fits your kind of value system. So I think that's more important. If you were starting today as an analyst with your skill set, what sector would you pick and why? If I'm starting today, I think my advantage is gathering data other people don't have. And I think my advantage is also generally contrarian. I think a good short gives me so much more happiness than a galon. I don't want to be honest with you. I think even though I started a tech company, I never thought I would start a tech company. I've never been a very tech aware person, like me being like a mobile, like, oh, that's going to be a wave. Oh, cloud, that's going to be like, I think I would have missed every single tech wave, maybe a other than AI. Like I truly understand AI. I think I do do well with a sector that has complexity. That's not very obvious to people. So if you're nearly all of those, some sector that you can gather differentially data, some sector that has some complexity, whether it's a regulation, whether it's things that people don't really want to dig. And then some sector that do have alpha. I can't say tech because realistically, I would have been a terrible tech. Like I just even being AI, I feel like I'm not super into like all the things that we work you're currently varying to right now. So I can also tech, unfortunately. But like something that's not like simple consumer, I think simple consumer, I think you're like, you know, the the, the, the chip holiday bell, like every single thing is a credit card data. I just don't think I will have an edge. But something maybe like real estate, I think there's a lot of and tap data potential. Some other like consumer adjacent, like, even like gaming of the world, I think probably more of those. I personally love insurance. Like I think insurance is one of those things that's, it's boring to everyone else, but I should truly like it. That said, I don't think insurance is a little tough just given I think it's just like it's a sector that there's no new money coming in. I think that's that would be a challenge. - I think fundamentally what a hedge fund manager does or PM does at the highest level is understand wealth creation very deeply. What do you think is the best philosophy for creating wealth today? - I think the best philosophy for creating a wealth depending on someone's personality and belt and system. So if you just take a look at two extreme examples, you have Charlie Munger, you have Elon Musk, both are extremely wealthy, but they got there differently, right? Charlie Munger is more about kind of preserved my capital, don't die. And with each one of them, I expect reasonable return, not crazy, crazy return. I expect reasonable return. And I think I have their good here ratio, right? So they're combination of that. And then you have Elon Musk, is like, I'm gonna create my own way, I'm gonna be contrarian, I'm gonna push all the trips I have with each move I do. So it's either hit a home run or I fail. I think those both systems works, right? I really think that it depends on kind of what your personality is. Are you more of a home run type of person? I'm either home run or I'm fine losing the game or are you the single's doubles I want to be consistent type of person. So that's not why I think it really depends on the personality. But regardless of your personality, I do think one of the most important thing a young person does and regardless what the other is, is figure out who you are, figure out what your balance is. You kind of have to learn about yourself before you can kind of see what track you belong to. And I think like investing is a very good self-discovery process. Nothing forces you to look inwardly, then a crap. I'm getting beaten by a market every single day. And I don't even know why I'm wrong. But what do I do here? Do I stick with my gut? Do I trim? Like what do I do? I think everyone should get in a habit of writing down every single decision point. Mostly to figure out what is your tendency, what is need your reaction. At what point of drawdown do you feel so much pain, you feel compelled to act? That level is different for everyone. When you're compelled to act, or you're compelled to double down, or you're compelled to shrink. And I think in that, in by recording a list of actions you do, you can slowly figure out what your natural tendency is. And then I think you work backwards. Okay, given this natural tendency, it could be different investment jobs. I think it can be different type of hedge funds. It can be even with a different PMS or different philosophies and different frameworks. I think you have to know who you are before you can decide that. One of my mentors asked in this question one time, and he gave an answer that I was surprised by using a correlation between self understanding and skill in investing. And then second thing, do you think there's a correlation between personal life decisions and being a great investor? That's a very interesting question. Let me think about it each. Is there a correlation between self awareness to investing? Go to a bad. Can I give an answer? I think it's a buy model. I think there are two types of people who do really well in a hedge fund game. One is you're super self aware. You're super self aware in a sense that you evaluate everything you do. You learn about your tendencies. You try to correct anything that's nalogical. And I think those people are very disciplined. I can do very well. On the other hand, if you're super self aware, you always think you're 100% right. You might actually do very well too. Because again, I think most people in the hedge fund world will get the fundamental right. It's just duration. Can you last until the time you get proven right? So if you're truly self unaware, a lot of times you get to hold that pain. Your pain tolerance is a lot higher than someone who's aware because someone who's aware but like, "Oh, crap, I'm probably wrong." Market is telling me, "I'm wrong for so long, I'm probably wrong." So someone who's not aware at all, I actually think that it might work out. So I think for your first question is buy model. So I don't know if it's cheating answer. But okay. The second question is personal decision and success in picking stuff. I don't know about that. I'll tell you also another thing is how do you determine someone is a good investor or not? How many years of data do you? I actually think there's not that many investor who has a long enough attract record to determine whether that investor has been lucky or good. So without that data, it's really hard to correlate with, I think, personal decisions. I think from my personal organization, I don't think there is a correlation, but I think that's more a function of it's harder to tell if an investor is good or lucky. Can you tell automatically if an investor is atrocious, just horrible? I don't know. Let's say you meet some guy or girl at the bar and I don't know. This is a, I mean, it's a crazy analogy, but at the bar, are there any tells where you could say these tendencies correlate to poor investment process or decision making? I think someone will have very strong emotion. Can't be worse. Like, I think all the other factors I can think of both examples like someone's good at some of that. Emotion probably the only thing I think consistently. I think the reason is everything else can either help with luck or true fundamentals. Like understanding the company is every other quality can usually help with one of those. Emotion the only one doesn't. Like, being more emotional does it help your understanding about a company? Being more emotional does not improve your luck or paint tolerance or anything. So I think that probably is the easiest way. I think it's easier to identify a bad investor and a good investor. I'm trying to think I think some of the quality of a very good investor is pretty neat. Like I have never, I have not seen anyone, well maybe it's because like it's just the nature of the business model. Like, you know, you can only feel so many times say this circle like before you, you're no longer have a shot, right? I haven't really seen someone who improves from a bad investor to a good investor. I think that maybe maybe some of the skill gaining is more on the junior side, meaning like going from not taking risk to taking risk and then you learn from your PM how to take risk. Maybe there's some ramp there. But once you start taking taking risk, I think I haven't really seen that much like improvement, meaning for improvement, like going from really bad to really good. I second thing I think is it's really hard because the game is changing every single year. So I think, oh, this is against another quality I was like identifying a bad investor. Someone was very setting their way. I have a process. This is what I do. This is how they got all of that. Because I think every, it used to be like every five to seven years thing changed. I think right now it's more like every two to three years, something changes. So I think that mentality of this works. This is only way that works. I think that's part of harmful. Yeah, those part of the two I can do at BAM at Simulio at the top pod shops. How high is the talent bar? Because we see on the news all the time they hire the best people. You see the packages, wow, intern, it's acceptance rate, sub one percent. Right. How high truly is the talent bar at the top shops? I think it's definitely high, but I also have a view that I think the talent bar is high because there are certain hurdles you have to overcome in order to even be a consideration for these bonds. That said, I think it's true for every kind of elite career, you can think of it. It can be the elite class, the elite school. It can be elite like tech world. I think within that, the distribution is pretty wide. I think there's certain minimal thing you have to have to meet. But once you clear that hurdle, I do think the talent distribution is very wide. I will say it's not a good talent versus bad talent. I will say it as the difference in talent. I will say maybe this is a reason. Maybe that reason is different personalities, different strengths can win at the investing game. It doesn't have to fit one type of personality, one profile, and you can automatically win. I actually truly believe that. Investing is one of those things. You can truly be yourself. You can figure out an investment style that works for you and adapt with it. I think because of that, it tends to attract the filter down to people with different personality. There's a lot of bias. People who survived are people who made money in their own little way. Because of that, I think the distribution is wide. It's not that people are smart and smart. It's more like they're very different styles. That said, I do think that at a hedge fund, and I think there's only one other job that has a similar qualification. There's only one thing that matters, your P&L at the end of the year. It's a place where all the superficial on-paper stuff do not matter. Which school you graduate. Yes, you have to clear a certain hurdle to get there. But honestly, I've seen other people get to a good P&L track record before they got to a top fund and they may not come from target school, whatever it is and be able to make it. But my point is, hedge fund is one of those places where it's truly meritocracy. Truly, there's one thing that matters where everything else actually doesn't matter as much, that any other industry of the same. Even the other parts of whinies, I think like banking. A lot of things that's not truly P&L driven, it's much more like about who do you know, what school did you graduate? How do you carry yourself? All of that, none of that really matters at hedge fund. You just have to be right. You just have to make money. There's only one other job I can think of, that's like that. That's sales. There's only one number that matters. And a different personality can do sales differently and everyone can get to that same goal in some different routes. I think bar is definitely there. But I think it's very tolerant of people with different backgrounds. You can throw out this conversation we've talked about, quantum, mental investing, about AI and how that's changing the game. And about how the game is constantly changing. On that last point, the game's changed a lot and is changing even faster. You mentioned the one to two-year cycles might even speed up. What is one skill that is a must for the current regime and for future regimes that you've seen in working as a PM and in helping a lot of hedge fund managers today? I think the most important skill in any environment is being able to think from first principle. You can start with first principle. These are things that has to be true. And then you layer on, these are the current environment. These are things that could impact how other people see the current environment. And these are the fact that could influence how other people react to this environment. So I think everything else I mentioned after the first principle one is what everyone else sees in the market. The first principle is what allows you to see you peel out all the onions of things that you reconstruct the truth. There's a single truth that may change and may not change. But the point is you kind of have to dig through to that the very, very core of a business model of a company of the market. I think that's probably the most important thing. I think like all the other things is like, I feel like there's obviously you have to learn you have to write, ask over a question, but all of that really comes down to you. Can you see the core of, can you see the core truth of anything, a company business model, a person, a, the different players in the market? What game is everyone playing? Can you see through all the layers that's needed to reconstruct the narrative in market to that core piece that is true? So I think that's the problem of the most important thing, probably the hardest thing to and I think when you're able to do that, then you can figure out what should I do given this environment right now? I might have a view that the market is available, but I think the bubble is on a burst for another two years. I'm going to write the bubble, right? Like you have to be adaptable to to basically the market, but you have to, you have to have a very strong conviction on what is a truth right now. And I love that see through the noise, find the signal and you reconstruct the narrative, but you have to reconstruct the narrative. Reconstruct the narrative and your way. Thank you so much for coming around the pod. This was lovely. Thank you for having me. This was fun.