In this episode, I'm speaking with David Walsh, who is the Head of Investments of RQI Investors, a First Sentier fund manager. And we delve into the concept of a Quant Winter.
Some market participants argue a new Quant Winter is in the making, since growth and momentum factors are compounding with limited breadth, driven partly by the promise of AI. This can lead to distortions in the market and the collapse of quantitative models.
David has tackled this topic in a recent paper, called 'Lessons from the Quant Winter' and we discussed what it is, how likely it is another one is coming, the impact of monetary policy and innovations in quantitative strategies through machine learing and AI.
For the full paper, please see here: https://www.firstsentierinvestors.com.au/au/en/adviser/insights/latest-insights/lessons-from-the-quant-winter.html
__________
Follow the Investment Innovation Institute [i3] on Linkedin
Subscribe to our Newsletter
Explore our library of insights from leading institutional investors at [i3] Insights
__________
Overview of Podcast with David Walsh, RQI Investors
04:00 From engineering to investing
09:00 What is a Quant Winter?
11:00 Is a Quant Winter a form of mean reversion?
15:00 Do I see another Quant Winter emerging? Probably not. What we are seeing is an anti-value period
18:30 On average value should beat growth, because the market is behaviourally tilted towards growth and overpays for it
20:00 In a high volatility environment, a rotation towards quality is sensible
23:00 A good quant portfolio is not just about established factors, it should be much more about finding idiosyncratic sources of alpha
25:00 You don't want a 100 signals in your portfolio; you want them to be able to breath
26:00 Machine learning let's you build models in ways you couldn't in the past
28:30 How to deal with cost in implementing non-linear signals
31:00 Higher dimensional portfolio optimisation through quantum computing
33:00 Quant Winter versus a recovery
38:00 Is AI in a bubble? "My guess is that the air will come out of the balloon, rather than it popping"
41:30 The extent to which passive or passive-enhanced money has affected the market structure is definitely an issue that has been arising in the past five or 10 years. You can broadly read that the market is becoming less efficient
Full Transcript of Episode 130
Wouter Klijn 00:00I
Welcome to the [i3] Podcast. I'm here today with David Walsh, who is the Head of Investments for RQI Investors. David, welcome to the show.
David Walsh
Thank you very much. Great to be here.
Wouter Klijn
So I understand you studied electronic engineering before you got into the investment industry and specialised in reducing circuits and chips, is that correct? And optical communications? So how do you end up in the investment industry after that?
David Walsh 00:28
Yeah, well, that was the topic du jour when I was studying my electronic engineering, way back a long time ago, the idea of what was called, at the time, very large scale integration, which was the idea of taking a circuit board or a chip design, and shrinking it down as small as possible had a lot to do with the way in which the mapping of the transistors on the chip worked, the material science underneath which materials you're going to use. That was very interesting. Also worked and looked at optical communications quite an early part of that industry. A lot of the talk about the fibre technology and loss rates and speed rates and bandwidth and so on, are quite interesting topics that were quite topical at the time. Emerging industries, the material science and the optical technology side, were very interesting, and clearly emerging technologies that have gone a long way since then. I took that and worked in power and mining engineering for a few years after I graduated, not directly in those topics, but used a lot of the ideas and instrumentation design and the like when I when I was in the industry, it doesn't naturally segue into finance, but it's important to think that a lot of the problem solving techniques that you get from being an engineer or training as engineer, the way you approach problems, the technical issues you use, the things you realise you're missing, apply themselves pretty well to a finance study as well. So when I, some sense, moved careers from engineering to finance, a lot of the skills came with me.
Wouter Klijn 01:49
So when you talk about shrinking circuits and chips, is that related to like Moore's law, where, you know, trying to get more and more things on the chip and making them smaller, and then that would increase the computer power. Or is that totally off?
David Walsh 02:05
Yep, no. Same field, pretty much. The idea there is you can only shrink them down to a certain size, beyond a certain size. It's impossible to to get the chip widths, the better the actual tracks you use for transmitting electrons around the circuit. You can't get them any smaller than a certain size. So there's that sort of limit, physical limit. There's been a lot of work since then. That's a long time ago. A lot of work since then, that's evolved that technology. But the idea was, how far could you get it down before we started to impinge on on issues regarding impurities in the material, in terms of the track sizes, in terms of the transition times, in terms of getting things out of sequence. So the design was really important in that sense, that was kind of the the intuition behind it.
Wouter Klijn 02:44
Do you sort of with that background look at today's, you know, development around machine learning and AI and like this need for chips and the dominance of AMSL, with that background, does that surprise you? Sort of where that has gotten to?
David Walsh 03:02
Not at all. No, I think it was a natural evolution. We would see of the of the the technology was growing back in the day when I was again while studying it a long time ago, the evolution of the technology was, was clearly progressing towards faster chips, greater memory, better software and the like. It's pretty early days, but you can see the trajectory it was moving on. A lot of discussions at the discussions at the time not about where the hardware or software would go, but how the usability of those technologies would move. And really, even then, artificial intelligence was a concept. At some point the industry expected to get to the point where it could replicate some kind of human behaviour, not to the scale we see it today. But certainly there was trajectory. I don't think it's been much deviation from that. The only real direction I think has been interesting, and I don't really understand this properly yet, is the idea of quantum computing, which gets away from the original idea of logic gates being zeros and ones to an issue where to a logic gate or a bit can have both zero and one at the same time based on some probabilistic distribution, which I can I've studied Quantum Physics A long time ago. I kind of understand basically the principles. I still don't get quantum computing the way I'd like to.
Wouter Klijn 04:04
Yeah, no, it's a fascinating topic, but very complex indeed. So that engineering background you now work as a quantitative investor. So you know that engineering background was that sort of a natural progression to become a quant, rather than, say, a fundamental approach?
David Walsh 04:21
Yes. So I transitioned across to finance academia first. So I was doing a higher degree in Finance as part of my studies, further studies. And I liked it, and transitioned across at an academic level, mainly because I liked the idea of research and pushing the ideas of knowledge and challenging myself. And I liked teaching. I like communication side. Did those skills necessarily move them across to quantitative investing? The answer is yes, I think fundamental investing has changed quite a bit since I've been in the industry. There are a lot more quantitative skills being used, but it's still not disciplined in a way that a quantitative investment process is. So having those tools and techniques that I'd learned as an engineer were not. Really directly applicable, but the discipline around problem solving, the idea of numerical optimization, the idea of constructing a problem in a certain way, those things lend themselves very naturally to quantitative investing,
Wouter Klijn 05:11
Yeah, but you're not getting to a stage where you built your own computers and systems for backing up the quant strategies?
David Walsh 05:18
No, no. There are many more, many people in the industry, including people that work with us, who are much better at that than me, my skills, are more around I think thinking about how the evolution of information works in markets, what things get priced in, what don't, what drives volatility, what events are happening that will perhaps override those that sort of quantitative discipline applied to the investment market. So it leads us on, naturally to where I've sort of come from in my come from in my industry, rather than going back to the technological side.
Wouter Klijn 05:46
Yeah. So part of the reason to do this podcast is because you wrote a paper that was talking about the quant winter lessons from the quant winter, which is perhaps not the topic you expect from a quant, because, you know, always works, doesn't it? What is a quant winter?
David Walsh 06:02
So quant winter was a term that was coined by a couple of industry people back in about 2020 also 2021 to reflect a certain period when a lot of quant factors didn't work as expected, both not in terms of factors not working, but in terms of the complementary factors that usually work not working either. So I went to almost a drought. So you could call almost a winter or a drought would be sort of either terminology. I don't like the winter term very much because it tends to mean that there might be seasons in quant behaviour. There's a spring and a summer. Yeah, that's not really how you want to think of the world, but a lack of of drivers of underlying fundamentals that have worked for a very long time, that certainly reflects a more wintry period or more drought period in terms of those factors. So it wasn't coined by me, but it's become a bit of a topic, particularly more recently, with some of the quant factors behaving in, again, peculiar ways, particularly for the last year or so. That suggests that maybe there's been some discussion, maybe there are reflections of that quant winter appearing now. Winter appearing now. Are we still saying the same things? Is it going to happen again? The Quantum winter itself probably can't be defined easily as being a single period, although it does tend to reflect times when certain style factors didn't work as expected, and the other factors, as I said, which were complementary to those didn't work as expected, either, and that period of nothing working was a winter or a drought.
Wouter Klijn 07:25
So what is, sort of the fundamental driver behind that? And as I was reading the paper, I got a sense that it points out that people were paying more for sort of the same earnings growth, that there seemed to be a bit of a mean reversion element to it, where you know that trend of paying more for the same earnings growth is not sustainable, so at some point that will break. Is that just mean reversion, or is there something else going on?
David Walsh 07:51
Yeah. So you don't want to try and bet against the markets when they're doing things which appear to be irrational. We all know that those sort of expressions, but what appeared to be happening was a narrow part of the market became more and more expensive after a long period when value was hadn't worked for about 2011 to 2017 or 18, value had been in a difficult period, low interest rates, low inflation and stimulated growth. Companies had driven a lot of the AI boom. A lot of the technology around that was developed in that period because it had a runway. A lot of investors were buying in to that period, what we saw from about 2017 or 18 onwards was a strong period when value underperformed, because that narrowness of growth continued very strongly and went upward. So in a sense, it's mean reverting. It kind of is a precursor to a mean reversion. I think that's why it's thinking of it. So in other words, markets became more and more expensive during that period, with the inevitable outcome that eventually it's going to be a reversal, which we saw in 2021 so that you get a reversal. And that was, in some sense, the main reversion that you speak of. But what was interesting as part of this was there was no way probably you could really have predicted or changed your position accordingly if you want it to be tilted towards value, you couldn't change that value characteristic. But normally, in an environment we know in quant investing, if value is not working, which is associated with mean reversion, trending signals like growth or price momentum tend to work, they didn't really work either. So it's because the market was so narrow, in the sense of a handful of stocks that were running, you couldn't really pick up a growth story unless it was very much a tech, US tech, expensive US tech story. And that's not usually, that's too narrow a bit for most investors to make. Certainly, for quant investors to make.
Wouter Klijn 09:34
So in this quant winter, a lot of factors are not working. Some factors are, I think, distorted a little bit in how they perform as well, when you look at that, is that mainly a problem from the perspective of a factor neutral investor, because usually, then you sort of don't want to take a bet on one particular factor. You try to have a broad spread that if one doesn't work, hopefully some of the other ones work. So is it mainly a problem from that perspective, or just did nothing work at all?
David Walsh 10:05
So if you were to bet on tech, as I said, I mentioned earlier, so if it was a tech bet, you're willing to buy expensive tech, expensive tech growth in the US, then you didn't have a problem. You were going to be overweight those names, and you'd do fine. But most managers weren't willing to pay up for the excessive earnings multiples, because the earnings had not yet been realised, even though the valuation is reflecting those earnings. So you're getting stocks trading on 30 and 30 plus times forward earnings. Those sort of numbers are pretty big for anybody to own. So that was a difficult thing to swallow, and that was what was driving the market lot of the time. That concentration was driving it. So in a sense that if you have a diversified factor model, like an alpha model, the quant model shop like us would have, you'll find a lot of things wouldn't work. You wouldn't necessarily be penalised by them, per se, on average. But in this case, we would have been, because we've been underweight a lot of those, those tech stocks. But generally, those sort of factor models would be have an alpha drag because of those. In fact, any model would have an alpha drag unless it had those characteristics that we talked about. I'd be, I'd be thinking that it's kind of a binary bet. A lot of ways, you either buy that story or you don't. And if you don't, no matter what your investment style is, you're going to be underperforming, whether diversified quant or a style risk premier capturer or a fundamental investor, the same thing will apply.
Wouter Klijn 11:23
Yeah. Now, you mentioned earlier that there are some people that saying maybe we're facing another quant winter. And to a degree, it does sound familiar, the concentration in the technology stocks, we do see that it impacts how momentum and growth work, because sometimes they seem to get a bit concluded in this narrow market. What is your take? Are we facing another quant winter?
David Walsh 11:47
Yeah, again, I'd push back a little bit on the term quant winter, although it's an easy thing to say, I get it, and it's catchy, isn't it? It's very catchy. And I get the point. I think probably the more the story that we are so uncertain of the direction, there's a lot of volatility in factors. There are periods and pockets when things are working. Things are working more quickly, perhaps, than we might have expected. Things which are in the past have worked aren't working. That sort of lack of product predictability probably isn't quite the same as a quant winter. It's more about an excessive period of volatility and uncertainty in the market, and so you're getting a lot of policy decisions, a lot of directions of the market, which are probably not what you would have taken as your bet being longer term, certainly. So do I think, to answer your question, do I think we're seeing a another winter emerging? Probably not. I think what we're seeing is an extensively value, anti value period in concentration and in expensive stocks and in tech unwinding. We are seeing that at the moment, and that unwind I would expect to continue. I don't think I would see a period where everything goes flat. More likely we'll see value continuing to do well, as we've seen over the last few months, after a period when it didn't do well, momentum will have difficulty. Growth stories, which are very uncertain in a higher inflation, higher interest rate environment, are difficult to pick. So some of those factors that have worked for a little while are not going to work. Would have called a winter No, I'd probably call it more just a high volatility. In fact, there's a lot of factor risk going on. So perhaps in an environment when someone like like us, who's investing a lot of these contractors, needs to be very cognizant of those. Probably would like to take some risk out of them, out of the process, or have it taken out simply because cross sectional deviation has gone up this uncertainty we're seeing. Yeah.
Wouter Klijn 13:34
So how do you prepare for that? I mean, the increase in volatility, does that also provide opportunities?
David Walsh 13:42
So yes, um, if it's cross so usually, quantitative models work better in environment when cross sectional dispersion opportunity set is bigger. Yeah, no question about that. So, so yes, you're exactly right. That's what you'd expect to see. So bigger dispersion, more likely opportunity set, provided it's the right sort of dispersion that you would see. So on average, that's true if it's macro factor risk or thematic risk that you're not normally betting on or is driven by factors you wouldn't normally see. What happens then is those factor opportunities don't appear. Contra sectional risk increases, and the model automatically adjusts itself by shrinking its positions and holding more names. That's part of the rebalancing process. When you use a risk model that we do, you tend to hold more names in period when cross sectional variance is higher. That's kind of what we're seeing at the moment. Volatility has been up a lot more uncertainty, and we tend to see a more of a risk averse process falling through. But we still have great conviction in our ideas. Longer term. You know, we were happy with those things. There's nothing in there we want to throw away. But it does seem to be that a lot of volatility in those factors and a lot of factor risk, which is not part of the normal Alpha factor risk that we need to take into account.
Wouter Klijn 14:51
Yep. So you mentioned the value factor there, and you said, well, that underperformance is sort of unwinding, but you don't necessarily. Expected to outperform going forward. What is your sense there in terms of, you know, the role of the value factor in portfolios? Because we've seen a long period where it pretty much did nothing or underperformed, then it started working a little bit again. But it's, it seems that you saying, well, we don't expect that to continue for a long time?
David Walsh 15:21
No, no, I think what I was saying was the value factor probably should continue to work for a little while, but the growth component of that probably, the growth factor probably would not so Okay, so I'm more of the view that historical evidence associated with value is going to continue to be true. We've gone through some periods when it didn't work. As you noted, 2011 2018 1020 period value was a struggle. It's rebounded strongly, given back some it's come back again. More recently, on average, value should beat growth, because the market is behaviorally tilted towards growth and overpays for it. Yeah. So you tend to get this bedding up of growth and then disappointment. The market sell off, you get glamour stocks, which are sold off, and you get under unloved stocks being sold off too far before that and bouncing again. That value premium has been shown many times through history. I would expect that to continue in the future, but longer term, but with volatility being as high as it is and uncertain drivers that we're seeing, macro drivers, thematic drivers, as well as these sort of factors which are highly volatile. I'm not going to say that we're going to have a period over the next 12 to 24 months when value is going to shoot the lights out, but I'd certainly expect it won't underperform by anywhere near as much as it has done during the period we saw prior to that.
Wouter Klijn 16:31
So it's not time to throw value out of the portfolio just yet, because it seems that when you look at sort of asset owner portfolios, that some of them have migrated from a pure sort of value exposure to more of a quality exposure, and basically, don't have a lot of exposure to it, to a pure value factor.
David Walsh 16:53
That's that's a good point. I think the the idea of a rotation towards quality measured however you can measure it, quality is a bit of a slippery concept, but, but measured however you measure it, that's sensible, because in a higher risk environment, you know, more likely better quality stocks, again, however you measure them, should outperform. But we would make the case pretty strongly that it makes sense to have a style balance for asset owners. They should be having quality in their portfolio, certainly, perhaps a little bit more of that, if that's their view, but you're going to be buying growth, because growth is where a lot of the returns are going to come from, and so you need to buy growth. You generally have to buy good quality, bottom up stock picking growth managers understand those companies, understand where growth is going to come from, not just narrow in tech, but more broadly than that. But that gives you a risk. You get an expensive growth e portfolio, particularly with the quality side that sits out there on its own without a balance now that stops paying off, you can significantly underperform. So we find, not only does value outperform over the long term, we like to think of value as being an insurance policy against those and then in client portfolios, we've had just a few clients have told us this. They like a low turnover, transparent, robust value process to balance out the styles they get from their other managers. So it's almost like a risk management tool in some for some clients, it certainly is. Others buy the value story and will own it longer time as a longer term as their core there's no question about that. But certainly for some larger asset owners we speak to, it's a balancing item. It's almost insurance policy in some ways. Yeah.
Wouter Klijn 18:25
So one debate in the quantity industry is focusing on, you know, how many factors are there, actually? And the answer, I find sometimes, ranges from three to five to like, 1000s. Yeah, sure. Where do you sit?
David Walsh 18:41
So there's a lot of discussion around what the academics called the factor zoo, lots of ideas and many, many different anomalies and factors in there. We tend to break it down, sort of at a high level, in terms of what we call composites or families of signals which follow certain themes. When it comes down to it, there's going to be themes which are, at one end, very risk based. So for example, volatility type signals, which will pay off over time. If you're tilted towards low volatility on average for the long term, it should outperform. Or there are periods, as we know, but it doesn't work. But that's quite a simplistic implementation. You just tend to buy low vol, low beta names and lab perform well, there's not much science in that, a little bit. And if you did build portfolio, build portfolios properly, you'll certainly see that that tends to act more like a risk premium. So you're picking up risk, and you're being rewarded for taking on that risk, risk associated with low volatility. You might do the same thing in taking on risk associated with value. For example, if you believe value is a risk, or if you believe that quality is a risk, you might find those as well. So at one end, you can think of these high level factors as being risk premia. I think any good quantitative portfolio should have a component of those within them, even if it's seeking idiosyncratic alpha, simply because we know that risk premia. Listing works on average, but your portfolio shouldn't be all about that. It should be much more about searching for more idiosyncratic sources of alpha and building those in. And that's when we get into this factor zoo issue that a lot of academic evidence, a lot of practitioner evidence, finds things that work, that perhaps worked thematically at the time, or perhaps based on a certain data set, or perhaps we're not able to be replicated, because this is replica replicability issue in finance academia, as is in every industry built you to do the same thing over again. That chasing of that next big idea probably is not really the way that you'd like to build your your alpha model, right? So I'd like to have, I'd like to think of our model as being a balance between risk premia, not too many of those, a small proportion of those that give you that long term risk premia harvesting issues associated with behavioural in markets, certainly behaviour in terms of the way in which investors work, the way that markets are pricing things in the data complexity issue is one that's really important. So issues around complex data models and how hard it is for individuals, or even invest many investors, to process efficiently and ingest that level of data. And I think those things, and I guess and limit, as well as some things associated with an inability to price in these things simply through mechanical issues, whether it's a short selling constraint or trading issues or institutional things that prevent you all of these things lead to various types of what I'd call idiosyncratic Alpha sources. Now, if you were to build a model that was a range of all of those things, you could end up with hundreds, as you said, of factors in there. The danger with that is you just squeeze out anything, you overfit. You end up with it with it, with a model which is very much data driven, can't be replicated in the future, no individual signal gets any oxygen. So as a balancing act, a little bit of art in building a model that says, I want enough alphas in now, have enough conviction over that are spread across a number of different insights that consistently work. So how many do you have? Well, 10 is too few and 100 is too many. It really depends on the alphas that you actually have, yeah. So really, it depends on the research you do. And that leads a little bit to the nature of, sort of the differentiation between quant managers that comes a lot of the time from the way in which you research and build those models, I think.
Wouter Klijn 22:15
And has that changed with the advent of machine learning in its current form, because you're basically looking with some of these risk premium for trends that are replicable, and machine learning gives that extra element where it can find nonlinear trends, so it potentially opens up a new source of factors. Is that how it works?
David Walsh 22:40
Or, yeah, very close to that. I think the there's a bit of a misnomer with AI, which I don't like the term artificial intelligence very much anyway, or machine learning, then it's going to find the answers for you. It's not really working that way. The insight, the understanding of economics behind it is really much more associated with what drives good returns machine learning tools, and has been a strong advance in many of these machine learning tools over the last three to five years, as we all know, let you explore those ideas in ways you couldn't in the past. It lets you pick up data sets you couldn't in the past. It lets you test things and build models you couldn't in the past. That's where a lot of the innovation comes from, not from finding new ideas, there's a little bit of that, but then taking old ideas and making them implemented, implementing them more efficiently. I think a lot of the ideas associated with machine learning allow you to extract those. I can give you an example. A lot of models that you might have in quantitative investing look at peer relative valuations. So you look at a peer group and you say that company is expensive compared to that company. So I'll go overweight, the cheaper one underweight, the more expensive one in that peer group. The question then is, how do you define the peer group? Do you talk about a sector? Do you talk about common geographic revenue? Do you talk about same customers? How do you do it? Well, in some sense, the machine learning tool in the past, you would have done that. You would have specified it that way, the machine learning tool has the ability to, in some sense, learn what the peer groups are over time within which your relative valuation signals can work. So while in the past, you might not necessarily have been able to pick up what those peer groups look like, now that relative valuation signal can be applied within peer groups that are determined by the machine learning tool. That's a really nice innovation, and much more powerful than a simple sector relative PE, for example, yeah.
Wouter Klijn 24:30
But do you find that that that group, that it then identifies, that that is a stable, permanent sort of group? Because sometimes you think that they find trends, but yeah, it might be a moment in time.
David Walsh 24:44
Yeah, really good point if you tune your model in us. This is part of the use of machine learning. If you tune your model in a certain way, you can find at that point in time the perfect peer group. And then tomorrow it doesn't work anymore, and yesterday didn't work either. Depends on how you tune it. If you. May tune the model too slowly, those peer groups don't evolve quickly enough. Which case you're lagging. So it's a question of how you build your machine learning model, how you understand it, and then the interpretation of what those peer groups might actually mean. So you want some kind of level of stability, but you want to be adaptive as well. And that comes down to the machine learning tool you use, the way in which interpret the evidence, the way you measure the importance of the input factors. All these sort of things matter.
Wouter Klijn 25:22
And the other thing that matters as well, and that is, I think, cost, because, from what I understand, that some of the non linear signals that are found by machine learning that they're real, they're sort of sustainable, but if you don't manage your cost properly, then get eroded away very quickly.
David Walsh 25:41
Yeah, two levels of cost for that. One is the trading of them. You can wear a lot more market impact by turning these things over much more quickly. So it's got to be an integral part of your process for testing and implementation of quant signals. No question about that. But the second thing is the cost involved in processing these things earlier on. When you do the analysis for these you're requiring fairly sophisticated tools that need to run for some time, and conventional computing hardware and systems are quite often not built for it. You need to be able to use much more scalable processes to be able to do a lot of the work that you need to do now. So cost in the sense of fixed and variable, cost of the hardware and software you use, as well as the trading costs associated with implementing them.
Wouter Klijn 26:19
Yeah. So we started off talking about a quantum winter, where you have certain factors not working. You got some distortion with factors. Do you think that machine learning can also help in sort of managing the risk, in the sense that it would pick up these signals earlier from distortions taking place?
David Walsh 26:36
Yes, indeed. So there's two strands. So that's a really good question. There's two strands to that. One would be trying to assess the way in which these signals are being implemented, sorry, being priced in. Now, what we've seen particularly at corners in the market. In Australia, we saw this in October and November. We saw sharp sell off and then rebound. A lot of factors behave very differently around those corners. Now, could you have a machine learning tool that allowed you to adaptively pick up what the factors are being priced in and change your model accordingly. Yes, of course you could whether that generate you get your original question, which is another good one, what about the cost involved in this? If you're turning over your portfolio too quickly, is to chase these things. Does the net reward? Is net reward positive? After costs, you might generate turnover. You burn it all up in transaction costs you earn you nothing. So you need to be able to assess those in history to work out whether it cost you when the time comes. But the answer is definitely yes. Machine learning tools could do that. They can also adaptively weight signals through time. So you need a machine learning tool that tells you what the best trade off between returns and risk and cost might be in terms of the mix of signals you use. That also is something that tells you in the normal run of things, rather than actually in terms of more in more volatile times too. So I'd say definitely, machine learning can help you in terms of portfolio construction, in terms of model selection and in terms of adaptability in the market. One thing I'll get back to a moment. I read an article this morning on quantum computing, again, trying to understand it, which I don't one of the tools that's being used there is for much more higher dimensional or granular portfolio optimization. So it's actually applications of that that apparently are toy examples that are being used at the moment that demonstrate how a much more efficient portfolio potentially can be built, not by changing the machine learning tool, by changing the particular application of a technology, the hardware you might do give much more space to search that space more efficiently than you're otherwise able to. So not just a machine learning tool, it's a hardware solution.
Wouter Klijn 28:25
Yeah, so with your background, then you look at quantum computing, and it's sort of a high level. Do you think that the gain there is mainly in faster computers, or could it also be that the fact that I work differently allows for a different type of software or program to be implemented?
David Walsh 28:44
I think the first one, again, I put the condition on that I'm still learning this stuff slowly, and I couldn't profess to be anything like an expert in what they do or how they might implement it. However, I think it's much more associated with the the selective problems it can address that current technology can't address. It has ability to compute many more computational cycles in a more complicated way than current computer. Computer literature, sorry, computer architecture can solve that particular problem. Lends itself very nicely to quantum computing, and I think that's where a lot of it will happen. I don't think quantum computing will come up with new ideas, as in the first pass it might, because of the machine learning tools you can evaluate on that lead to more things, but as a solution to a problem. At the moment, it's addressing problems which current hardware cannot really address because they're too complex or too big. That's where it'll probably come through, in my opinion. Anyway, my unlearned opinion.
Wouter Klijn 29:41
Yeah, interesting. So I think what part of the reason that you think it is not a quantum winter is because you talk about a sense of recovery, that we're in a stage where actually quite a lot of factors are working. There's some exceptions to that, maybe a low volatility. Not as robust at the moment, but it's basically a period of recovery, rather than a period leading up to our problem. And then I thought about that, because in the context, as well as the broader sort of monetary policy, we had a period where there was very loose monetary policy. It basically prevented the recession. That was the most widely predicted recession we were going to have. Is there an element there where perhaps there is the system had didn't have the time to flush that out, and there's a bigger crisis building because of that, the monetary policy has basically prevented the crisis from happening.
David Walsh 30:41
Yes. So there's probably two strands, if I can answer that in sort of two different ways. There's always been talk post any kind of major sell off or major economic disruption, and the support that's provided by by government institutions like quantitative easing, post GFC that prevent the natural run of capitalism, the creative destruction of capitalism for taking place. Yeah, really, companies that should have should go broke. I mean, should have been allowed to go broke. And we didn't get much of that. We had Lehman Brothers, right, but that's pretty much it. And Bear Stearns, obviously. So there's, there's that issue. I think there should have been. I mean, nobody wants to see this sort of thing happening. But in practice, if you believe in a capitalist system, that's a natural result. And things go wrong. Things are not working should be stopped, and things are working should replace them. Now, the creative destruction idea that wasn't really allowed to happen through implementation of lower interest rates, deliberately low inflation, productivity was very low. That stimulated a lot of and simulated a lot of growth in some areas, which turned out to be good, but prevented others from recycling the capital being recycled into more efficient places. So definitely, that's, that's an issue we've seen. I think, I think that probably explains a lot of the, or some of the growth we've seen in tech, in particular, a long runway of low interest rates of the market getting used to low interest rates and willing to buy longer dated cash flows and earnings that's certainly been built into the prices for a lot of tech companies in environment which is much more strict and higher interest rates and short term, it's unlikely that we've got the same runway that they've had. So you can sort of see that they would have performed the way they've done, and it's been great for a lot of reasons. Do I think a sort of thing is going to continue? That's really a different, much more, much more interesting and complex question, because I don't. I think in an environment we're seeing higher interest rates. Inflation is sticky. Despite posturing, interest rates are probably going to stay higher longer term. Certainly in Australia, we've seen that that's an inflation story, as much as anything else that will probably tend to play out in in a slowdown of growth plays nicely to value style. Getting back to what we talked about earlier doesn't mean it's building up something else to happen. To get into your question, I'd probably think that one of the the main things we've seen associated with the results of those long, long growth those companies didn't go broke. They probably should have. We've seen that in the past, but also the exuberance we've seen in Tech has played itself out in excessive spending, both in the past and into the future, on CapEx for data centres and for growth of hardware, extrapolating into the future, things which probably shouldn't be extrapolated in the way that they have been, and I'd be quite suspicious of that can't be continued, particularly the capex side, which just seems outrageous and unlikely to be able sustainable both, not just from environmental perspective, something from terms of finding the cash to invest in these for the return on invested capital you expect you probably won't see. So my expectation is a lot of those things will be scaled back, that those those dreams will not become reality. We won't have data centres in space, none of that going on, which I found also hilarious when I heard Elon talking about it. I think it's much more that those things will be scaled back. We can't have a sort of CapEx, we currently can't live with the depreciation that comes from that capex, in accounting sense, into the future. So I think those things be scaled back to be more reasonable. A lot of those businesses we're talking about do generate cash. They are next generation. They will be there in the future in some sense, perhaps not those companies, but the ones that come after them will be around. It's no questions technological shift, but it won't play out the way it's playing out at the moment.
Wouter Klijn 34:22
It doesn't seem at the moment, though, that they're scaling back, because I think they recently announced a combined US$600 billion in more capex that they're planning to spend, yeah, and at the same time, people have sort of said, Oh, it's not so much of a problem, because these companies, you know, the cash flow positive and have a lot of equity. So it wasn't, you know, a debt fueled spending, as some of the crisis have been in the past, but some of these companies are now starting to raise debt.
David Walsh 34:48
Yes, yes, that's right. There's 100-year bond or something that was Google listed the other talked about the other day. There's certainly been some debt fueling, I think a free cash flow for these companies is not as high as they thought. Open AI is a good example. Microsoft has plenty of cash, but the free cash flow, since companies is falling at the same time as the capex is increasing, the only way continuing to capex is to raise capital through debt or equity, and debt being the way they're talking about I'm confident or respectful of the market mechanism here. I think the market knows this, and it's already starting to price in some of the issues associated with this, this exuberance we've seen that's been going on for some time has probably rationalised itself a little bit, and I think it's more like that. These plans will simply be scaled back. There'll be some remnants of it. There are large data centres in the Midwest of the US and so on, which are being built now, and probably will continue to be built, some sense. But that long term growth of CapEx to solve a problem which doesn't yet exist simply to serve these things seems unlikely. So my view would be, it's more likely the air would come out of the balloon rather than a popping.
Wouter Klijn 35:47
So this type of big, you know, more macro orientated developments, how does that feed in with sort of your quantitative process? Does do you build signals on the back of that, or is it more it informs the broader portfolio.
David Walsh 36:01
Yeah, so it's more. It will not really find its way into portfolio construction wouldn't really find its way into alpha generation unless we can find something which we believe will consistently work through the cycle. It should really be kind of all weather. If it's thematic change that's happening then, then it's less likely to find its way in, it'll be more noticing what the risks are, understanding the processes and the outcomes that we see. And then when clients are curious, we can inform them. Team members are curious, we can inform them and understand how it works, because we know we can't predict the market well enough to understand a lot of these sort of underlying moves that take place, and the impact they can have on on returns, on portfolios. So just as being informed about the market, rather than making changes to the process, what we do in the process should be longer term. I don't think there's any real need for us to try and implement macro factors directly in, on thematic grounds like that. If we find something that's useful, very happy to do it. But I don't think there's going to be macro factors. There's too few macro drivers to make enough breadth in the portfolio decision making to allow you to put them in as an alpha factor. One thing I would talk about which has spurred some research that I've been thinking about more recently has been this capex idea. So there are certain regulations which are used around the world, accounting standards regarding revenue recognition, when you recognise revenue, and how you measure that. And it seems to me that revenue recognition is an area which is not well understood in the in the technology side. And if you recognise revenue at the right time, then you probably are understanding much more about the underlying development of the company and the quality of the company that goes with it. I don't think revenue recognition is very well understood in the IT sector at all at the moment, and how it's actually working. It's just a exuberance pay for the future growth story. So that leads me to think, How can we better incorporate revenue recognition more broadly across the market, across the portfolio, not just in tech.
Wouter Klijn 37:53
Yeah, yeah. Interesting. We looked at this quantum winter concept, and that's more about, you know, how different factors behave at the same time. Do you also keep an eye on sort of market structure at sort of a higher level? And the reason why I'm asking is because we've seen, especially in Australia, because of changes in regulation, because of mergers and peer awareness, that funds have been using more systematic strategies passive, but a lot of the time it's more passive and enhanced. So there's more money flowing in that side. Does that create distortions that sort of affect your process?
David Walsh 38:33
That's an open question, a good one, and an open question, the extent to which the growth of passive money, or very low risk, active risk, funds, ETF flow, retail flow, the impact that has on overall market structure and the pricing in of of ideas and what's driving them. That's definitely an issue that's been emerging, particularly in the last 45 or 10 years or so, and more recently, with the growth of index portfolios. The flow into index funds and the flow into benchmark hugging funds has a number of different effects. You'd imagine. One of them would be that conventional things, which we priced in, are now being priced in by fewer investors. Which case it should create an opportunity. There's less efficient markets. You'd think that's possible. One answer. The other reason these things never get priced in that no one buys them because you think it's a good idea, but people stop buying that because just go to the to the index, which case you're betting on the wrong horse. And that's possible, too. The emergence of these sort of things causes distortions, a lot of inflow, a lot of capital chasing fewer names. Market goes up and up on the back of these pricing issues that we see, this, this sort of flow issues we see, I can't see that changing, and it's definitely going to have an impact. It's just very cloudy as to what the impact might actually be. I think you can broadly read that the market becomes, as I said, less efficient. And there's an argument, also an interesting argument is. Well that it actually becomes potentially better for investors in an environment like this. And the reason for that is goes like this. If your index funds are growing, it means that the ability for stock to be borrowed is increasing. So if you're going to borrow stock and short it, you can now do more of it. So availability of shorting is higher, which means bad news should be priced in more easily. See the logic. So that would suggest that the short sale constraint and the ability to borrow is being reduced by the growth of index funds. Now, whether that's be true or not, I haven't seen any evidence on that, but you can kind of get the get the logic behind it. So it's too many things at play at once, on average. Definitely, it's been a change to the industry. It's not going to go away anytime soon. I would say that it's going to make everyone's investment processes more difficult, because less is being priced in by the things we used to get priced in by and more as being simply just buying at the close to meet a benchmark.
Wouter Klijn 40:51
Yeah, yeah, yeah. Fascinating. So if I might end up with a slightly tricky question, because we talked about the quantum Intuit. You don't like that term because it implies cyclicality, yep. So we get it, it's not cyclical. But we did recently a poll at our equity conference, and quite a few quants in there as well, and said, if it was to be divided in four seasons, if it's not winter, where are we at? And we got the majority that actually said, summer. Oh, right. Where do you sit?
David Walsh 41:21
Yeah, well, I think we've definitely come out of a period when quantitative investing was not front and centre with a lot of asset owners allocations for a variety of reasons. It could have been the performance from way back during the GFC. It could have been through some more volatile periods. It could have been quant winter that inverted commas quant winter period. But we certainly seen a lot of factors working at once, and a lot of emergence of quant businesses over the last few years that have really driven good performance have gained great reputation and have shone in the light of or perhaps in the lack of light from other managers who've struggled to perform. Deep failure, managers haven't done so well. Strong growth, managers haven't done so well. So that emergence, along with the idea that machine learning and data becoming more available, lends itself more towards popularity for quant investing. Are we in summer yet to use your analogy? No, I don't think so. I think we're in the upswing. I think there's still a lot of scepticism about the long term performance of quant factors have with quant models, have? We've certainly seen strong demand in our quant processes, consultants, clients have been very interested. I think what that means a lot of people is from one person different to another. You know what that quant process means? So some think of it as being very idiosyncratic factors that are all quite unique and different, and all driven by AI and black box and who knows? Others think of a simply risk Premier. And that variation probably won't resolve itself, but it also tends to lead to a lot of uncertainty, what, whether you're where you are, in that, in that, in that, in that seasonal period. Yeah, we're talking about, I think we're not there yet in terms of summer, but we're certainly growing towards it. Quant performance has been good. The risk controls are important. Technology is driving a lot of it. I can definitely see it being a, being a strong period for quaint investing for the next, certainly, next few years.
Wouter Klijn 43:17
Yeah. So spring, perhaps?
David Walsh
Let's say spring.
Wouter Klijn
Let's say spring. Well, David, thank you very much for coming to the office, and thanks for this conversation.
David Walsh 43:24
Oh no, that was fantastic. Thanks.
