207: Rishi Narang – Evolving Trends—How Quants Are Earning Alpha

Chat with Traders is brought to you by Trade the Pool. Did you know that every decade the market reinvents itself? Online brokers opened the doors, mobile apps made trading seamless, and commission free trading erased barriers. Now a new era has begun. Meet, trade the pool, limited risk trading. And now you also have unlimited time to reach the profit target. From now on, your trading risk is capped and your trading opportunities are limitless.

Trade the pool funds home-based stock traders with up to$200,000 in buying power. That means you can trade larger positions and scale your strategies without risking your own savings. It's time to trade with more capital, making it truly worth your time and effort. Ready to trade the pool? Click the link in the description and join the stock trading revolution today.

Hey traders, welcome. Thank you for joining me on episode 207. This week I had the pleasure of catching up with Rishi Narang. Longtime listeners of the show may recall I interviewed Rishi back in 2016 on episode 54. Either way, Rishi is the founding principal of T2AM, a fund of funds investment manager with a sole focus of allocating to quantitative trading strategies.

The fund was formed in two thousand and five. Dorishi's time in markets extends more than twenty years, as he co founded Distinguished Quant Shop TradeWorks in ninety nine. Rishi's also the author of an excellent book Inside the Black Box and an advisor to DARPA's Financial Markets Vulnerabilities Project. Yes, that is DARPA as in the government agency responsible for the robot dogs you've probably seen in those uh somewhat terrifying videos.

So the purpose of our catch up was mostly to go over prominent trends he's seen develop in recent times, specific to quant trading and investing. Without giving away too much, it involves talk of stat arb and index rebound strategies. an attraction to China's market, narrower trading models, use cases of alt data, and we wrap up discussing the rise of retail or DIY quants.

I can appreciate these topics won't be of interest to everyone, but as I've said before, I think it is highly beneficial to have some insight on how the larger players are moving in markets and how they are earning alpha. Finally, please excuse the odd bird sound. I believe Rishi was wandering outside his home as we were talking. And after this first question, you actually won't hear from me for about

twenty minutes or so. My suggestion, think of this first part as more of a talk or a an interesting presentation from Rishi and then an interview follows. Please enjoy, I present to you from Los Angeles, California, Rishi Narang.

You know, speaking here with regard to research and strategy, what has stood out to you as notable trends in the field of quantitative trading and investing? Sure. Yeah. So there's there's a couple of few things to highlight. One one is the oh man, there's a few things actually. So one is that as the landscape has shifted, meaning You try to earn alpha in a context. It's not something that happens in a vacuum.

And that context is sort of maybe best described as follows. Alpha comes from inefficiencies in asset prices. that are themselves caused by either inefficiencies or relative indifferences on the part of various market participants. So an old example would be that institutional traders, institutional investors were pretty sloppy in their handling of trading of single stock. And so they would do these large lumpy orders and and ignore market impact.

And that basically created a whole game called Stat Arb, statistical arbitrage. Uh and what would happen is there would be a price dislocation in some name. You would see that, but it's hard to know from the outside whether that came for good reason or not. And so rather than just buy a name that that fell or short a name that went up a bunch. You would find something that reduces your risk. And so you'd put on a hedge. So you could imagine hedging with the index.

Uh but the problem with hedging with the index is there's a lot of basis risk. You know, if you're if you're buying Tesla and shorting the S P five hundred, there's a lot of idiosyncratic risk in Tesla. Well, I maybe picked a really crappy example because I don't know who a good peer for Tesla is that's publicly listed today. But the point is that you would try to find something that reduces your risk. So at least an automaker reduces your risk, right? So maybe Ford or General Motors.

is a better hedge for Tesla than you know the S P five hundred index. So that game of finding related things and shorting the outperformer and buying the underperformer, which is also known as statistical arbitrage or relative mean reversion. Existed in the eighties, nineties, early two thousands because of the sloppiness of a certain class of investors, mostly for that reason. Well, as those folks evolved.

that inefficiency kind of went away. And it changed. So relative mean reversion still exists, but it exists for a very different reason. And that reason also causes it to manifest in a different way. means if you're pursuing statistical arbitrage today, your model better not look just like it did. And say nineteen ninety seven. Because otherwise you're like not gonna make any money and probably just lose money.

There are a number of ways in which the landscape has shifted, really more than we have time to get into now, but you can just think about it for yourself, how much the world has changed, say in the last 15, 20 years. Um, there's been a gigantic move from passive to active. Uh, other way around. Wow. From active to passive. Sorry. and a gigantic move from discretionary trading, uh interaction with the markets to algorithmic interaction with the markets.

At least within US equities, for example, almost all the volume is well not almost all, a giant portion of the volume is now done market on close as opposed to throughout the day. And so there's a bunch of these dynamics that are all intersecting with one another and changing the approach to making alpha. So that's one thing. Uh that uh that we've seen that's changing the game around around us and uh forcing us to adapt and evolve.

So one example uh is the relative increase in prominence of index rebalance strategies. So these are strategies that try to anticipate what names are going to be added to various indices. Why are we doing that? Well Or uh added and deleted, by the way. Um why are we doing that? Well, we're doing that because there's a ton of money that gets invested on the back of what say Standard and Poor's puts in the SP 500.

And that's true for MSCI indices and all kinds of other indices as well. Um so all this passive money isn't exactly passive, it's just sort of cheap like following the the the dictums, I guess that would be dicta, of various index construction companies. And so those index construction companies uh say, okay,

Uh we're including company X in the S P five hundred. And that announcement means that now all of the vanguards and fidelities and all of the other index managers and the and the folks who administer the SPY ETF. And so on. They all have to go out and buy Company X in the right proportion. And it of course with a fixed number of names type of index like the S P 500, if a name is in, then a name is also out. And so those folks then also have to sell that other name.

So that's interesting too, right? So you have something where there's gonna be enormous buying pressure and enormous selling pressure. Pretty useful to know about that in advance if you could. You kind of can, because a lot of them publish indicators of how or or guide guidebooks or guidelines for how they think about adding and removing names for inclusion. So uh that game really exists because of inefficiencies in the way passive folks invest. There's an indifference there.

Uh, you know, if you're the S P five hundred, you only care a little bit about the performance of the S P five hundred. You're you know, certainly for an ETF manager, you don't really care. by definition, you get those stocks in those proportions and you just sit and hold them. Guess what return you're gonna get? You're gonna get the S P five hundred return. If you have adverse market impact.

Meh, it doesn't really matter. I mean, it goes into the price and whatever. So there's not that much incentive to be that clever about. Managing market impact. I mean, there is to some extent, but it's limited. And so there are a whole bunch of strategies that exist. Well, not a whole bunch, but there's a whole bunch of money being managed around this strategy of index rebalancing.

And it's not just the SP, of course, it's all these indices globally. So that's one big example of how the world has changed is this active to passive thing. And by the way, even if you're a stat ARB manager. uh you now kind of need to start accounting for index rebounds in your price action. So as an example If your model is ignorant of the fact that company X has just been added and company Y has just been deleted from the index. And you're just watching price action relative to peers.

then suddenly company X is soaring and company Y is reeling and you may get in front of that mountain of flow that's coming your way. So you kind of have to start to have almost like a risk factor or a uh an adverse selection avoidance technique. For index rebal. So even if you're not trading it as an alpha, which you would do if you're doing index rebal, you still have to have index rebal. Yeah, sir. try to be good at the game.

So that's one thing. Another thing is uh like a big dispersion or um diaspora of the places and markets that quant strategies are are applied in. So, you know, Indian stat arb and emerging market stat arb and China Stat Ar China Stat ARB is like a huge business right now. I I don't even know the numbers offhand, but the amount of money going into both onshore money and offshore money going into quantitative

Equities and futures trading of mainland Chinese assets is just stupendous. Billions and billions and billions of dollars. It's a huge number. Um, and the number of firms is is also pretty big. You know, there are a lot of folks who, for example, worked at US firms and then realize well there's this more or less frontier market where there's a lot more alpha, a lot more retail participation, and a lot more inefficiencies in China. It's also harder in a bunch of ways to trade offshore.

or if you're an offshore entity, especially up until quite recently. And so Do maybe you know, if you were someone who was here on H one B visa in the US and worked for Two Sigma or D Shaw or whatever quant firm, maybe just go home and set up a shop there. uh in China and have a bunch of local advantages over the US and and UK firms that are trying to play that game.

So that's definitely been happening. Um is oh and and the application of quant techniques into all kinds of other areas, by the way. Venture capital, uh crypto. Credit, peer-to-peer lending, um, municipal bond market making. I mean all kinds. I've seen so many interesting things that are not just basic equity stat ARB in the developed markets or basic futures strategies in the developed markets.

And I guess the third big trend is maybe the most interesting, which is the There's been a shift in the balance between taking overfitting risk and underfitting risk. And just philosophically, I think it's really interesting. In the old days, well, a pretty dirty insult for a quant was that they've overfit. But now we're seeing folks do a variety of things that their actions are telling you that they're at least as afraid of underfitting as of overfitting.

That looks like a couple of different things. So, first of all, in the old days, for the purposes of largely cross-validation and statistical significance. models were typically universal, meaning for some universe of many securities, basically without exclusion. uh this model applies to all of them. So if there are uh all US stocks, but then realistically you can only trade the top, say, three thousand of them if if that's what you determine from a liquidity standpoint.

Then you have one model with one set of parameters, one set of factors. uh and one set of weights for all 3,000 stocks, meaning the drivers of your forecast for Walmart are the same as for Pfizer, are the same as for Amazon, are the same as for Google and Facebook. And Tesla. All those companies are forecasted with the exact same set of factors, the same weights, the same parameters.

We're now seeing industry-specific and even company-specific models, or if you're talking about futures, individual market models. And what you're seeing with those is a decision that your prior

might be a lot stronger on a single company forecast forecasting model. Like I can know what the KPRs KPIs are for some specific company. But um I might not have a lot of data history to prove that out in a back test, but I'm willing to trust my prior more uh and not worry as much about having statistical significance.

Uh another easy example would be something like inventory data. So, you know, on US companies quarterly reports, there's a standard line item for inventory. Well, if you're a bank, what is inventory? if you're a uh you know, an an audit and consulting firm like Pricewaterhouse Cooper's or Ernston Young or whatever. What is that? You know? Uh, what is inventory? But if you're a a manufacturer of television sets, well, now we know what inventory is.

So but that is important interest information for the latter kind of company and not for the former. And so You don't want to throw out that data potentially. But that does mean you have a different model for those industries where inventory is a thing, and a different model for those sectors and industries where it's not. So that's one expression of this shift in the balance between over and underfitting risk. The second is the increased use of alternative data sources.

And this is related to the first thing, which we'll call narrower models as opposed to more broad models. um models are applied and applied on a more narrow set of um of security. For this second thing, um, this is now models that are driven by data that are only available for certain kinds of companies. So a great example is uh is credit card data. So I'm guessing that your listeners will know that their credit card spending is being tracked by various companies.

And whether individualized or aggregated, that data is being sold to hedge funds. uh, who use it mostly quantitatively, but not always, uh, to make decisions about what to buy and sell. And basically what they're doing is looking for a faster or more accurate estimation of say revenues for uh certain kinds of companies. Well, if it's credit card data.

It's interesting'cause it's you can't buy everything and you don't buy everything on credit cards. And in any case there is only a subset of things that go into Uh the category of things that humans like regular people buy. So as an example, it doesn't tell you anything about uh how Boeing is doing on its aerospace sales. Like credit card data does not apply to Boeing.

probably doesn't apply much to auto manufacturers either, because most people are not buying cars on their credit card. They're either sending a wire or writing a check or writing a check and then having a financing line. And that's a different thing. It tells you a lot about what they're doing with Amazon and what they're doing with Macy's and whatever else. Like so it's really useful for retail and consumer and hospitality type companies.

Um and a few other categories, but you know, credit card data is really interesting for a few hundred companies in the US. Not very interesting for the other two and a half, three and a half thousand easily tradable companies. So that's the second thing is this sort of trend towards alternative data, which is very much hand in hand with the trend towards narrower models. The third thing is the increased use of machine learning

or artificial intelligence uh strategies. So that's the other really big piece in all of this. So uh you know, these things by definition are looking at data to tell you how what the forecasting variables should be and how much they should matter. Uh and we can talk about the different use cases of ML and AI if you want, but um there are some firms that claim to be making forecasts without priors, without saying, oh, I have this economic model of how companies

asset prices work, what drives a stock to go up, what drives a stock to go down. I'm just gonna let the machine figure it out. And that's inherently a a more fitting kind of an exercise. Have you ever watched a stock explode and thought, if only I had the capital, or sat on the sidelines because your account balance felt too small to matter? Good news.

With Trade the Pool's limited risk platform, you don't need millions or even thousands to start trading the U.S. stock market. Bypass the PDT and tap into over 12,000 U.S. listed equities. From penny stocks to big caps, ETFs, even the newest IPOs, and short anything you like, with zero locate or hard to borrow fees.

Start your evaluation, get funded with up to$200,000 in buying power so you can go big without risking your own savings. And now you can also have unlimited time to reach the profit target. It's a game changer. Not ready to trade yet? Trade the pool offers a free demo and educational resources. practice on live data, master the platform, and build confidence risk-free before you even pay a cent. Click the link in the show notes to start trading with Trade the Pools Capital.

So here you've obviously outlined three different trends that you've that you've noticed uh in space of quant trading and investing. Going right back to the start and I just have a few questions about each of these You spoke about knowing in advance when a symbol may be added to an index. Can you maybe go into that a little bit more about how you may know in advance?

Sure. Yeah. So in many cases, and this will just be one example for the sake of time, in many cases, most of the determination is simply a market capitalization based cutoff. So we're gonna take like the Russell three thous uh Russell two thousand. There's a certain ranking of market caps that happens and if it Company one thousand one through three thousand. in terms of the ranking by market cap, kind of presumably that company is supposed to be in the Russell two thousand.

And they have certain dates that they announce these um inclusions and exclusions. They announce the new composition of the index, and then dates at which those changes are made effective. And there's some time in between there. So there's a range of strategies that you can do. Similar to the way that people treat merger arb, you can

look at like the history of all mergers and say, well, what kinds of companies get bought? And then I'm gonna go out and screen the world for those kinds of companies and maybe buy some or all of them. And then some of them are gonna be targets for for takeovers. So there you're trying to get it ahead of the uh the acquirer. Here you're trying to get ahead of the market cap um based index.

solution or index provider and then all the people and firms that have to follow suit and either bring in the new names or get rid of the old one. It's not exactly front running, but all alpha is in a sense probabilistic front running. You're trying to know before it happens that other people are going to buy this stock in aggregate or sell this stock or this asset.

in in advance. Um and you're doing that with a bunch of hard work and not with any kind of actual knowledge of their trades. But you do have to get there before they do. Otherwise it's not alpha. Then it's Something else that doesn't work, you know. I wonder what the name for that is. Doing uh doing all the right trades, but doing them after the fact. I mean I guess that's trend following, but

This information, I mean, is that publicly available to everybody? Yeah. The information about what's going to be included into an index? Some of it is so as an example, um the index providers have gotten smart about as as all capitalist companies generally are, but how to extract money for everything. So just as exchanges are now charging a huge amount for data.

that, you know, is a major part of their revenue streams, even though they're really not in the business of selling data, they're in the business of processing transactions of securities and providing a safe environment for the transaction of securities. And so they charge fees for that too. But then they they know you have to have that data. If you're like a high frequency trader, you have to have the data of every single exchange.

for various reasons in the US. And so um those exchanges have you know extract a toll for for that data. Well, just so the index providers are charging for index composition data and they are and it's some of it's not cheap. I mean to get like Russell S P M SCI That is some hundreds of thousands of dollars a year, just for those three, and not even for all regions. Well, really?

Yeah. But anyone like it's not like you have to know somebody who knows somebody to get the data, but you do have to have hundreds of thousands of dollars of data uh of uh budget for the data. Yeah. And that's true for a lot of stuff, man. And this it you're talking about the data around index rebalancing. Specifically that, yes.

And all the no no, just to be clear, this is not index rebound data. This is index composition data. This is to tell you just what is in the SP 500 right now, what tickers and at what weights. That's it. That's all that hundreds of thousands of dollars get. But you can imagine if you're doing index rebal, it's kind of an important ingredient in And doing the trade right.

So what what are some of the trades you see around this? Like is it I mean, I presume that they're probably a little bit more sophisticated than just buying a symbol which you think is going to be uh added into an index. Why w why would you presume that? It's the that's it. That's the trade. Now the sophistication may be in why do you think that it's going to be added?

But um and there are a range of strategies around that. So there are folks who traffic in like the lower probability names in index rebound versus those that traffic in the names that have already been announced, but it hasn't completely been implemented yet.

So there's like a range of certainty about whether or not this name is going to be included. And as a result of that range of certainty, there are different players with different risk appetites and different models. So let's say for example, That you're talking about the Russell one thousand, which if I if memory serves and if my knowledge is correct, is simply the one thousand largest market caps in the US equity market.

So now let's say that um there's a rebal coming up, an announcement coming up about the rebal. Uh and when and when I say coming up, like how far in advance You trade is an open question here. So, but let's just pretend that the answer is three months before the rebound. I'm going to start putting on positions.

Okay. So it's uh I think their rebounds are announced in March, but I don't remember. It could be November. I have no idea. Whatever. And I don't remember the frequency either. But let's say it's March. So I'm gonna start putting on my positions in December, January. At that time, there's a bunch of companies that are say ranked from nine hundred or eight hundred to twelve hundred in market cap, which over the course of three months

might move up or down enough in market cap to either get them excluded from the top thousand or included in the top thousand. So this is interesting because this isn't just about is this company going to be good? It's is it going to be good at it's is it going to be better or worse? than the companies around it in this specific weird peer group of companies that may have nothing to do with each other otherwise.

But they're all not really competing, but you can think of them as competing like horse racing to be in that top thousand so that they get included. Now again, maybe the companies don't care if they're included in the Russell 1000 or not, but their stock prices care.

Now let's say that I'm really good at forecasting three-month price outcomes for companies. Well, I better use that information and then I'll have a little bit more conviction to buy bigger those names that I'm gonna hold for three months-ish and are likely to be included because then when that announcement comes, there's a big pop. And as the certainty builds.

there's a big pop. So there's this kind of weird short term momentum thing that happens in those in that category of names that has nothing to do with, say, the top ten companies in the Russell one thousand because there's almost a zero probability. that in three months those companies are uh are gonna drop out of the index.

Right. So it's a little bit like options theory. If you're the number one ranked market cap in the US or in any market, and you know, you're far and away the biggest, which happens, it would be like Um, what is the value of a deep out of the money put on that company? With a three-month expiry. And the value of that's like very close to zero. It's basically zero. If you're talking about something where it's a three month expiry for very near the money options.

Well now that's fun, right? And if you have information that gives you a clue, then you should use that information. Does that make sense? It does make sense, yeah. And what's more, there's a there's an asymmetry here. If your company number nine hundred ninety nine The last time or 1000, the last time the Russell 1000 was constituted. And you're still kind of 999, 1000. There's a chance you get excluded, which means there's a whole ton of money that has to come out of you.

If your company one thousand and one And you remain company 1001, there's not a lot of downside for that price because it's already excluded and it's still excluded. So for those companies that are outside the index looking in. There is more upside than downside. And for those companies that are barely in the index, there's more downside than upside. So it's just very interesting that dynamic and what it means when you're building a model.

So ultimately you're trying to buy early enough predicting that a company may be added to an index um in order to essentially front run all the flow which is going to be coming into that stock over the coming weeks and months. Yeah. And again, this word front run has like actual legal meaning and we don't mean it that way. But yes, I call it that.

You want to get in front of it, which is a lot like front running, but but the legal definition of front running is about, you know, having knowledge of someone like actual knowledge of someone's order and then going in front of it. Like if you call your broker and say, I want to buy a billion shares of IBM. And then your broker's like, cool, let me go buy a million shares of IBM right before that, and then I'll do your order because that way I'll profit from all your market impact.

Let's just say you're you're trying to be the first to act. Or or yeah. Or get in front of. You're trying to be early to act. Early, yeah.

Yeah. The second point you made was about China, which I thought was quite interesting. You said there's a lot of stat arb opportunities there right now. Um, how come that is the case? It's a combination of factors. So there's onshore money and offshore money. There's onshore firms and offshore firms. Like you can't have onshore money if you're an offshore firm. That's like not a thing.

Uh, you can have offshore money if you're an onshore firm. So there's a supply thing and a demand thing that's happening. From the supply side, the ability to trade. China as an offshore investor, that's recently been a little bit um that's been made a little bit easier recently. And so that's that's one thing, right? So if you're like two Sigma or PDT or some other US or whatever non Chinese firm, you can now access China relatively easily actually.

There's a bunch of like international brokers like Morgan Stanley and UBS and so on that can give you access to that market, uh, simply, very simply. Um, and there are rules around shorting and in shorting individual names and shorting the index and so on. So you can even be market neutral. If I understand correctly, and I haven't studied this too too much, but if I understand correctly, if you're an onshore firm managing onshore money, uh you cannot short Yeah.

I'm not even sure if you can short the index, but maybe you can. But that's now Chinese money. Um, that's been going on a little bit of time, but there's also a different supply dynamic there, which is that you As I said, have a lot of new supply of plants who used to work at non Chinese firm. and have decided let's set up shop in China where we live where there's a a less efficient market.

That second thing where there's a less efficient market, that statement, is what's driving the demand side of this, which is to say, again, as some of the more obvious and straightforward markets to access. and bigger markets have filled up. If you look at China, it's a really big market. It's a really big economy, only set to get bigger.

And, you know, there's a rush to get in there before it gets too crowded and be early and gather data and all those things that give you an early mover advantage. Um access to talent, access to data, storing your own data is super valuable, but this, you know, you can't do that until you start doing it. So like go start doing it. So that there's been like a lot of that. Have you allocated to any managers who are trading Chinese markets, like in order for for your fund to have exposure there?

Yep, we have offshore two offshore Chinese well Money with two firms that are both based in mainland China, but are offering offshore strategies to offshore investors. And yeah, we've we've picked a couple of them and we've looked at dozens.

Okay. Is that just a recent thing? Pretty recent. Um both this year. Uh right. I wanna say the first one went live mid year, but we've probably know that means we've been talking to him for a little while. Yeah, okay. You also brought up the point about I I might get this slightly wrong, but it was something to the extent of things which are used in quant being applied to

other areas outside of financial markets. I know you mentioned venture capital and there were a couple other things as well. Could you speak to that a little bit more? Yeah, and it's again the same drive for diversification and and access to markets that are a little

weirder and more complicated to to so there are some more barriers to entry. And if you think about this like just from a basic economic principle standpoint, it's the same thing as saying, well there are barriers to entry to making this kind of widget. But if I happen to have some edge at making that widget, then I should probably go do that because I'm going to enjoy a competitive advantage for a while.

So and being early is its own competitive advantage because that means I have a head start on others. But you know, there's something you have to risk, you have to put up a bunch of upfront costs to get there. So credit, um, you know, this is like debt issued by companies that So it's not government debt, it's usually uh you know it's it's it's debt from private companies. That debt trade.

So it's not just that the company issues debt and you decide I will lend to this company, you can buy and sell that debt. And there is a market for it. It's in the US. corporate credit market, not um fully exchange uh traded and in fact it's mostly not. Um, but you know, some of the credit default swaps are exchange traded and it's a changing market. It's an evolving marketplace. Um It it has very different challenges and one of the most obvious is is data.

Um like it's not that complicated to think about what would cause a bond to be more attractive relatively and what would cause it to be less attractive. What's complicated is asking like How do I even track the history of bonds? So unlike stocks, bonds expire. And on top of that, they're issued at various coupons at various times and various durations. And so managing that data set, it's like it's just a different problem.

from uh managing normal equity data set or even futures data set. You know, futures expire too, but there's far fewer of them than there are companies and credit. So yeah, there's like this crazy data wrangling problem. There's systematic strategies. uh in peer to peer lending. So this is like lending club and lending tree where if you're just some random person and you want to borrow five thousand dollars for whatever random reason. You can apply for a loan.

and other random people who have five thousand extra dollars sitting around can lend that to you. and earn on it at like their local savings account, but then they're taking credit risk on you as an individual, the random borrower. And so um there are now quant strategies out there. that are pouring over credit data and pouring over characteristics of borrowers.

and saying, Well, this borrower might not have the greatest credit score and I don't know if you guys have the equivalent in Australia or elsewhere, wherever listeners may be, but in the US You know, you have a credit score that a credit agency uh has a list of factors that uh describe your credit worthiness. And, you know, like a in a very popular such scheme, a really high credit score, I think the maximum is eight fifty.

850. I don't know why it's not a thousand or why it's not a hundred, but it's eight hundred and goddamn fifty. So that's what it is. And um, you know, you're considered as having good credit. I think if your credit's above seven fifty, let's call it. And so, you know, if you're a an 820 credit score person, it's very easy to get money fairly cheaply. If you're a 600 credit score person, it's a lot less easy. But it might be that that's just because of something stupid.

Because these credit a these credit ratings are not like God telling you this is a good lender or good borrower, I mean. This is like a a flawed, universally applied model telling you that. Maybe some random circumstance that happened three years ago. beat up your credit score, but it turns out you're actually a really good borrower who pays everything back. So they look for those kinds of people.

to try to find the ones that are better borrowers. And but because their credit score sucks, they know they're gonna have a higher interest rate. Does that make sense? So you try to find like these good value loans to make where you get a high yield, but the risk isn't that bad.

not not in line how high the yield is. Um so yeah, there are systematic strategies for that. And you can tell there's a whole different set of data there and a whole different kind of problem there than there is when you're trying to forecast the price of IBM. Gotcha. That's really uh quite fascinating. And some of the stuff isn't new. Some of it's been going on for a while. Um, you know, systematic credit trading has been going on

10, 15 years now. But at the same time, the prominence of it is much higher. Like there's at least two standalone quant funds that are doing just that. And then the magnitude of that activity within some of the larger multistrats is much larger than it used to be. And it's the same for quant trading in China. Like that's been happening for a minute, but it's just it's now like on the map, you know.

So the third point, uh, I wanted to ask you a question on that, uh, with regard to underfitting and overfitting the model. Do you have any tips for how to get the balance right?

Oof, that's actually like legitimately really hard. I mean there's so many factors And The reason that this change is happening is once again, both from a supply perspective and from a demand perspective. And I know I'm mapping things to supply and demand. This isn't like a clean mapping, but I'll explain what I mean. So on the supply side. We have an increasing number of people who are

In jobs in quant firms that have training in machine learning and artificial intelligence. And some of those folks used to work at places like Facebook and they're just sick of selling advertisements to people. Right. Like they thought, oh cool, I get to work for this really cool company. It's a tech company. I'm not selling my soul. Look at social media. I'm like connecting families and friends and former colleagues and alumni.

But then it turns out what what I'm really doing is I'm selling people ads and that gets at some point maybe grading on your soul. And so you quit that job and you go work for a different soul crushing place that's just trying to make rich people richer. Cool. Nonetheless. Um that is one um source of increased supply. The other is the field itself has advanced. I mean, the problem of forecasting asset prices. is a really hard problem. Uh this these are incredibly noisy processes.

Uh, very close to random locks. Uh, you're out of sample R squared, for those of you who know what that is, or like low single digit. So like point oh three is a successful R square. And, you know, anyone in any other scientist most other scientific fields would look at that as a forecast out of sample R squared and go, you're basically noise. Like you don't have anything better than a coin toss here. So why are we doing this?

But that's a that's a successful that's a successful outcome in my world, right? So When you have that noisy of a process, the technique to uh and by the way, very limited data. Um You know, if you're doing self-driving cars, you can basically generate more data just by so like let's say that you have a self-driving car and you know you find that it's not performing well in ring. Well, let's go to Portland and Seattle for a while and train more cars there. Like you could just get more data.

For the capital markets, you can't get more data. The data are non-stationary, and there just isn't that much of it. You know, if you compare like the number of Facebook posts and number of Facebook users to like the total amount of data about US equity. And I mean like the daily data in the one versus the total cumulative data in the other. It one like they're just orders of magnitude different, you know? Uh we're talking about billions of data points per day, if not more.

and on in the kind of social media world. And we're talking about maybe billions of data points in the whole cumulative history of everything in US equity. And so You know, you can see that that's those are just different problems. Uh there are other differences too, but just for the sake of time, let's just acknowledge that these are very different problems. You may need different and better techniques and more powerful computing clusters to be able to tackle these differing challenges.

And so that's the other aspect of the supply side. But again, the demand side is there. So as conventional old school techniques have become more heavily competed, as the philosophies and concepts and ideas and theories that underpin those strategies have proliferated and been leaked out into

competitive firms by virtue of employees changing jobs or leaked out into the academic journals and textbooks and so on. You know, those old ideas are harder and harder to make money on. And so Just as some people are deciding, okay, well, the hell with it, I'm gonna go venture far afield in and trade China, even though it's a pain in the neck, uh, maybe uh some of them are saying, well, I'm gonna venture further afield in my technique.

I happen to be really good at this machine learning thing. And so I'm gonna do that where it's harder to compete. So there's that happening too. that there's a desire to take more fitting risk because the lower hanging fruit which doesn't have fitting risk, you know, it's not easy making money that way anymore.

I feel as though a lot of what we've been talking about here is very much applicable to, you know, the large quant funds uh and very much the professional space. I'd like to ask you What have you seen in the trend among DIY quants? So the at home retailer who has, you know, might have some machine learning skills.

you know, has learnt how to program, has wrote some sort of systematic strategies.'Cause I think when we last spoke in uh two thousand sixteen, you know, in the past four years, there's been a big rise uh in this space. Is there anything you've been saying here which is particularly interesting?

Yeah, uh, for sure. So there's a few dynamics there. It's a good question. In my perspective on retail is mostly US and then like random smatterings Asia. I don't actually weirdly have any impression at all of Australia or Europe as far as retail investing, but maybe you can tell me and and we can reflect on it a little bit. But as far as the US goes, um, retail investors were a steady kind of part of that market.

for a very long time. I can tell you, like when I was a child, I distinctly remember my dad going through the newspaper, like the physical print newspaper. And going through like which stocks he owned and how they did yesterday. And this is like the next morning, finding out how each stock he did. No clue what his portfolio was like because you know you'd have to like sit there and do that math.

You could certainly look at the Wall Street Journal or the local paper and it would have a printout of thousands and thousands of ticker in very fine print and what their last close was and what the change was yesterday and what the volume was and a few other columns. You know, folks like my dad have been doing that for a long time. When the internet kind of boom happened in the late 90s and commensurate with that, tech stock.

sort of IPOs boomed. And you know, pets dot com and and all that were out there. Um it's a now defunct company that sold like pet care products online. You know, there were a bunch of companies that went public. And a lot of interest on those, but also just in general on the markets, because the market was rallying really hard and it was pretty easy to kind of quote unquote look like a genius. You bought five internet companies and just were up like a hundred percent a week or whatever the hell.

And so that dynamic really did lead to a pretty significant increase in the participation of retail investors. When that crash. That definitely took a uh a big chunk of retail participation with it because those folks realized, well, no, it turns out this is hard and I'm kind of shitty at it. And so they stopped. That took another leg down just a few years later when the second that, you know, 2008 crisis happened. You know, the first crisis ended in 2002.

Late two thousand two, I think, maybe even early two thousand three. Uh and then the second crisis started in November of 2007. So you didn't even have five, six years. between those two crises. And that was two fifty percent drawdowns in US equities in the space of ten in the space of ten years. It's pretty crazy. And unsurprisingly, that turned off a lot of people to investing in stocks at all. And if they were going to invest in stocks, it was just going to be passive.

So that kicked off that passive trend and also kicked off a deretailization trend in the US. Uh in some other markets, retail investors have been more present uh for longer and more consistently, like in places like China and Singapore and so on. Um Korea. Yeah. Folks just gamble. Like you go to the casino, that's a completely normal thing to do. In the US, that's not like a completely normal thing to do. Casinos are like illegal in most of the states in America, which is a funny thing.

There's just less of that kind of culture of just making bets and punting around. Um, which for better or worse is what it is. Now we've had this crazy bull market. And on top of that, a lot more tech is available than has ever been before. That is just the nature of things. That same still statement will be true like two years from now, unless we Truly devour ourselves. So the platforms out there, and I don't mean Robin Hood, which is to me just sort of a joke.

Uh I mean more uh like Quantopian and other similar platforms where anyone can get out there and just code up trading systems. And then there's brokers like interactive brokers where if you're like semi-pro, you have a few hundred thousand dollars, you could set up an account there and have some of the same tools and access as professionals have, although it's

Definitely a far cry, frankly, cost wise especially. Yeah, it's you know, it's kind of cool. So you can you can set up like a one person quant shop and I know lots of folks who do. And in fact, some of the folks we've allocated to

come exactly from that lineage where they didn't go they weren't like ex employees of two sigma. Maybe they've had some professional experience, but what they really did was they're just homegrown hobbyists who are really good at it and go, you know what, we're gonna we're gonna this is I'm good at this, I wanna do this professionally and scale up.

Uh and we've had some success with that model. I mean, but you know, we've also had some failure with that. It's not easy. But yeah, there's been this um kind of uh combination or confluence of things where we have this protracted bull market. that's encouraged retail participation and a lot of really specific brand excitement around names like Apple and Tesla. And then at the same time the availability of technology

to actually make it easy to code up strategies and backtest them and implement them and enter competitions. Um and those things kind of combined to to make it pretty It's been uh yeah, there's been like a big surge in that world.

Yeah. I remember last time when we spoke you you mentioned you had allocated to someone who was essentially just a one man band working from his uh apartment. Still have the money with that same guy. Yeah, I think you'd mentioned he was managing about fifteen million dollars. I'm not sure what that is today, but Seventy. Seventy million, one guy from his apartment. That's incredible. That that's not seventy million from you, that's you're just a portion of that. No, it's all from us.

All from you. Yeah. Wow. That's that's uh quite incredible. Let me ask you this question, and we'll wrap things up here. But let's say there is someone listening to this who kind of fits into that category of DIY quant. Uh, you know, retailer working at home.

They've developed a quantitative strategy. They've been having a little bit of success with it, trading it for some time, real money. They like the idea of running more capital through that strategy, scaling up, what would this trader need to do to be ready for investment? That's a great, great question, Aaron. Um so uh it depends is the answer.

Um They are willing to simply license their signal, meaning they would once a day or on whatever right the right frequency is for that strategy, send a list of trades that their model comes up with. to someone who has all the infrastructure and compliance and operational aspects to do that, um to to put on those trades and do it correctly, then the bar is a lot lower because they don't really need much of anything. If they want to actually like set up a fund.

uh or a firm that does this and you know you'll have to register with the relevant authorities as a as an investment advisor or whatever the equivalent is in your jurisdiction. Um you'll have to show to investors that you can actually process trades and all kinds of other things. Um, you know, that hurdles a lot higher. Like a lot higher.

And it's expensive and not likely to work. So just to be clear, um, you know, I think that the far more sensible path is to contract with someone who's already got a substantial operation and license your signals out. And then if it turns out you're really good at this in the long run, A, you'll just make money doing it. You may not have to quit your day job.

But B, if you decide you really want to do it, at least you'll have some track record of what your signals have done. And then when you try to go out and get unreal size, you know, and then at least when you try to go out and get capital. one of the big questions that investors ask, which is is this thing scalable? You know, if I run a five hundred billion dollar pension fund or a ten billion dollar fund of hedge funds or family office or whatever, um, I'm not looking to make

A$10 million investment in a in a hedge fund that might return 10% per year. Like making a million dollars of PL per year for someone at that size is like not worth the effort. not worth the marginal cost. And so you know you've got to really demonstrate that you are

uh able to scale. And if you could demonstrate that, oh no, my signal generated 20 million of PNL, admittedly it wasn't my firm that that put the trades on, then yeah, there's there's something there. But the more realistic thing is that if you're a hobbyist, stay a hobbyist, frankly, and license your signals to people who are doing this for real and um and you get paid.

What would be an example of someone who may be interested in in licensing signals though? Like would that be a fund manager or Yeah, so um and I I think they've stopped doing this now, but WorldQuant used to do it and globally, like they had farmers in Vietnam who were contributors of signals. Right. Um, but I think they've shut that practice down. I'm not a hundred percent sure.

Yeah, we know there's there's there's a few out there. I mean, some of them are looking for more like professional CVs, like people who used to work. at at firms that are doing this professionally and maybe have have gone off on their own. Yeah, there there are a few out there. I I'm not comfortable naming names to be honest, just because No, I wouldn't expect you to. Yeah I don't have but yeah, these are professional quant hedge funds.

Uh okay. Would you have any luck, you know, maybe with some smaller family offices? Do you think that's something they may be interested in? Sure. If you have a relationship, uh, but you know, as with so many things, like these days there's so many people unemployed, especially in the US. that their resumes going into the inboxes of every person that's every employer that's got a job opening. It's not quite hopeless.

if you randomly send your resume into a vacuum, but you definitely know it's a lot better if you have a connection. Well, how much more so for something like this where you're asking for millions of dollars or tens of millions of dollars of risk capital and a budget. Um you know, that's not just a normal job, right? That's a job with some real strings attached. So if you have like a good relationship with someone, then sure it's I think worth barking up that tree.

I I honestly think it's a good idea. like lower probability than a lottery ticket if you're just like casting your net out into the ether. Trust is a big part of this, you know?

I know you've got a run. I was gonna ask you really quickly, uh, about your involvement with DARPA. Is that something you'd be interested in talking about? Yes, but I'm not like under NDA or anything and I can talk about it really briefly. Um so it was really cool and fun and interesting. So basically uh some years ago I was approached by um a professor who's a friend of mine. Who is friends with some folks who are at DARPA?

or were at the time and said, Hey, would you be interested in connecting with them? They're looking at uh financial warfare kind of things. And so this is part of the Department of Defense, the US Department of Defense. And this was not an exercise in looking at, you know, aggressive tactics, rather in defensive tactics. And strategies for protecting American markets and financial systems from various parties that, for either political or economic reasons, may want to

cause problems, right? So causing outages at at exchanges or causing um outages with, you know, payroll processing systems or any number of other vectors of threat. And there are of course continuously both governmental and non governmental parties that are you know very aggressive in in that stuff. And so it's, you know, it was this super fun intellectual exercise of like, what are the vectors of threats and what could we do to detect them early, prevent them, mitigate the spread of the damage.

Um so you know, we do things like war games and um and table exercises to to sort of flesh out some of these things a little bit. And the part of the goal was to see is there something measurable and quantifiable and within DARPA's remit to actually focus on. to work on this problem. So the way DARPA works is it starts as a seedling, which is all this thing has been so far, and then it only becomes a program if the seedling shows that there is

not only something to do here, but something for DARPA specifically to do. Because DARPA isn't, you know, doesn't have a completely carte blanche remit. It has a specific remit. And so if it if there's stuff to do, but it's better for like the Treasury Department to do, well, then that's fine. Like hand that off to the Treasury Department and they can decide what to do. You know, if it's fits for DARPA, then it has to fit some criteria and

So yeah, they're still in the evaluation phase. It it does appear that there's enough for some stuff to be done as a program, but that has not been finalized. It's super fun. Really interesting.

It does sound like an interesting project to be involved with. So you're just sort of consulting on it at the for for the time being? Yeah, and I mean I'm an I'm like a volunteer. Uh so I'm I'm consulting informally, no security clearances, no NDAs. I literally had one condition was that I get to put it on my LinkedIn because how cool is that? Yeah. Yeah. Exactly. That's why I asked you about it.

And I w I wasn't really sure the link between DARPA'cause I just know them as creating the the robot dog and other crazy robots. I wasn't really sure the link between that and financial markets but yeah. You know, economic defense and military defence are sort of more interlinked than ever, you know? Yeah, yeah. Wars have been fought over economic interests, I think, far more than people generally realize. Right.

For sure. Well, Rishi, I know you need to shoot off. Uh I just wanna say, uh it was really nice to catch up with you and uh thank you very much for coming on the podcast and um sharing a bit of an update on some of the things you've been seeing and what's been uh developing in the markets. So if someone wants to follow you, uh would you like to share your Twitter handle and maybe your website and also mention your book too.

Sure, Aaron. Thanks a lot for having me again. Again, good to catch up. Uh I'm on Twitter, not super actively, but I am on Twitter at Rishi Knarang. And I did write a book, the second edition of which came out in two thousand thirteen, called Inside the Black Box, that should be readily available in places like Amazon. Uh what was the other thing I was supposed to share? Uh your website. I don't really have one. Okay. Ha ha. The T T website, no?

Uh yeah, there's nothing on there. But yeah, my company is www.t2am.com. Okay. And Inside the Black Box is a fantastic book. I've read it a several times. Uh have a copy myself. Uh would definitely recommend it if you're interested in this type of thing. So yeah, until next time, Rishi. Uh we'll we'll chat soon. Thanks, Aaron. Stay safe, man. Be good. You've reached