← The HC Commodities Podcast
The HC Commodities Podcast
From Systematic Trading to Structural Edge with Hans Balgobin
01:14:34
What actually is systematic trading and the capabilities behind it? How can it generate returns? How can it inform the broader trading desks and organization's decision making? How is it delivering a competitive advantage and strategic edge to companies behind those returns? What have been the developments over the last couple of years? And how is such trading shaping the markets themselves? Our guest is Hans Balgobin, who has had a career as a systematic trader and building systematic trading platforms both in equities and energy, with the likes of HSBC, Shell and Millennium and is now joining Uniperin a systematic trading role. His views expressed herein represent his own and not those of any organization.
Loading summary
Transcript109 lines
- [00:00]
A
Foreign welcome to the HC Commodities Podcast, a podcast dedicated to the commodities sector and the people within it. I'm your host, Paul Chapman. This podcast is produced by HC Group, a global search firm dedicated to the commodities sector. Today we're talking systematic trading. What actually is it? What do we mean by it? How can it generate returns? How can it inform the broader trading desk and organization's decision making? What have been the developments over the last couple of years and how is it shaping the markets themselves? Our guest is Hans Balgobin, who's had a career as a systematic trader and building systematic trading capabilities both in equities the likes of HSBC and more recently at Shell and A Millennium, and is now joining Uniper in a systematic trading role. I should express that the views discussed in the episode are Han's personal views and do not represent those of any organization. We have a long and wide ranging discussion, not just on the technicalities and how to build these types of capabilities, but the strategic importance and impact both today and in the future. As always, you can really support the show by leaving us a positive review on the platform you're listening on. And as always, I hope you enjoyed the episode. Hans, welcome to the show.
- [01:26]
B
Thank you. Thank you for having me, Paul.
- [01:29]
A
So I am looking forward to this discussion. We are talking about systematic trading in energy, in commodities. This is a topic, I guess we. We last covered a couple of years ago. So in part it's to update where we are today, but I think also take the opportunity to lean into some of the definitions and really try and build our understanding of not only what it is, but now how crucial it is to a platform and the competitive advantages it brings. Can you just set the scene a little bit for us?
- [01:58]
B
So the way I am going to describe systematic is my own view on how it is in the space and it's a vast space. When we talk about quantitative and systematic, there is a big overlap in the middle in that Venn diagram, but it's not exactly the same thing according to me. But I'm pretty sure most of the industry would agree where the difference lies. So systematic is about being systematic, repeatable and rational in the sense that you would take inputs and get an output. Output would be your investment decision, your trading style. The quantitative side of things is about the use of the mathematical tool set. You know, having data using functions, using computation to take that information and then bring some kind of edge to it and then send it out to the market in some way. It could not just. It could not even be about investment. It could be about risk management, it could be about, you know, weeding out irrational decisions and potentially risk regulatory risks out of a picture. Now where it overlaps is the mathematical side of quantitative and the systematic. The reason for systematic, which is to have repeatable things where you can talk to the public, talk to your own bosses and say this is what I've done and you will get that in a repeated fashion in the future within what we can control. Obviously the market is a non stationary process, so it's very hard to get exactly the same outcome, but at least you can have the same methodology for taking inputs and getting outputs. And that's where the similarities between quantitative and systematic lies. The, the reason why I'm saying this is where it stops. You could actually have normal non mathematical persons being systematic. You could have a person having a set of rules that they adhere to all the time and that person would be a systematic person. And it could just be something in, in Excel where he or she says if this, this, this happens, I will do that. And that person may not have done any back testing, may not have done any, any, any mathematical thinking, processing on the back of that, but that person would be systematic.
- [04:14]
A
So that's very clear and very helpful. I think that distinction, this has evolved in largely in equities, right in the same process. So quantitative trading came first. You had a lot of rich data, a lot of publicly available data that could start helping in decision making. That obviously translated into energy trading, particularly in gas and power early on, just the provenance of these things. And then systematic is that final piece where you're saying, okay, I'm going to now create a set of rules that determine actions in a set of conditions. Just the philosophy there is the, is the benefit of adding that systematic piece at the end there. What is that? Is that because you're decisions get made faster? Is it that you're removing human bias like philosophically? Why do you want that last bit?
- [05:02]
B
So in my mind, if you want one word, it's about clarity. Clarity of purpose in the sense that what you would say to you know why you're doing what you're doing, you can say this is what I have set out to do and I will always do that. Now that clarity brings the symptoms of the, what you just described. So things around speed, we get faster because you're doing the same thing in a repeated fashion. You get taking out the human error, all these things get taken out, but it will eventually be about clarity. So I'm going to diverge a Little bit here. But you know how they talk about black boxes in AI? The reason why this is an issue is because you cannot explain it to people. So having some kind of clarity around why things are happening is going to not just be good for yourself because you can understand what you're doing. It's also going to be good in terms of if something bad happens, you can explain to a regulator why something bad has happened. You can say, oh yeah, exactly. I followed my steps and we all agreed when we first started that these steps are going to lead to good market outcomes and irrational decisions. But in this particular case, in this market, something happened that is not exactly what we wanted to have. But on average it's going to be fine. So this is about clarity in the end. And that clarity brings together with it many things like speed, reduction of errors. But also it's going to end up helping you in further development of ideas. You will be able to look back over your decision making and say, ah, something could be improved here in that fashion. If you were not keeping a systematic approach, you would have so many, so much messiness in your box in your decision making process, but it would be harder for yourself to analyze what went wrong. So it's all of that. It's clarity. Reward for me is clarity.
- [06:59]
A
And we're going to, we talk about and build up to how this complements existing platforms and a lot of that has to do with frequency and tenor and so forth. Yes, let's start with the. And this is the bit where I'm going to be so at sea that you know, I'm going to need your help. But obviously and I guess the, the lens here is that someone sat in the boardroom and they've got a head of systematic trading or, or something has gone wrong and they're getting explained what's going on. Just some of the definitions around what's in these models and how they're built. So can you just give us some sense of graphs and nodes and inputs?
- [07:34]
B
Yes. Okay. So if you think about how a person who has no obvious mathematical or systematic background will invest is usually they will have a physics in their mind of how the world works and they will look at the current state of the world that is inputs data and then they would say okay, this is where we are and this is what will happen next. Or given some very complex thing in their mind that may or may not be able, they may or may not be able to put on in programming on paper and do it themselves. But it's always state of the world model of a world that you hold private to yourself and you send out and you either make money or lose money based on how accurate your own internal model of a world is.
- [08:20]
A
That's, that's the brain, right? That's how our brains work.
- [08:23]
B
That's the brain. And now when we talk about systematic, we are saying we will try to extricate these pieces into their relevant components. So for example, the state of the world would be inputs would be data feeds, could be market data, it could be fundamental information, it could be news that you scrape from Twitter, from other places. So all of these are things that are available to you publicly or you have to buy it from vendors. And that is already somewhere you have to be careful about. Because if some data gets corrupted or some feed gets stale, you need to be monitoring that all the time. So that's first thing, monitoring the state of the data coming in, whether it's going stale, cleaning the data, sometimes all that is important. And that is one thing. You have to give a reassurance to your stakeholders that you understand what it means to consume data and how to deal with that. That's rule number one. Data comes in, then the next thing is about, I mean, there's two ways of thinking about that. There are systematic people or even quantitative people who are data driven, and then there are people like myself who are thesis driven. So the difference between the two, it gets blurry. But the difference between the two is a data driven person would actually say, let me run some kind of search algorithm on the data coming in through history and see what pops up, what pops up as repeatability. And then they will say, oh, this is repeatable. So either they can take the repeatability and make some kind of, you know, set of rules around that and then say if it's repeating in that way, it will keep repeating in the future.
- [10:06]
A
The danger here is you're like, you know, oh, when it gets cold, gas prices go up.
- [10:11]
B
That's right, that's right. And it's obviously, you know, there are other pieces to this puzzle where you're maybe by, by looking at a set of data and deciding that this is it. It's a bit like the allegory of play of a cave, the plateau's cave allegory where you think that this is what's happening, but there's always things that would make you overfit because you're, you're not seeing the whole picture.
- [10:34]
A
Yeah, yeah. Or it's not useful. Right. In other words, you know, like actually there's Other it hides a more interesting information and, or it's information that is just ubiquitous and you're not. There's no trading edge in it. Right?
- [10:46]
B
Yes, yes. So, so, and, and, and sometimes these things are the human bias of a researcher because that person can say, oh, I don't think there's something to be looked at here. So that betrays that there is a thesis even in this kind of data driven approach because we will have to filter what we think is important or not. So there is, there's that. Now let's talk about people who are a bit more humble like my, like myself, where hopefully I may be overstepping here. But, but when I, when I talk about humility, about I'm saying that there is a thesis to what I'm going to do.
- [11:25]
A
But the thesis, scientific method, right. This is like I've got a hypothesis, I'm now going to test it against that data.
- [11:31]
B
That's right. And the word humidity I'm using here is, I'm thinking carefully about the word humidity because what I'm saying here is a thesis. We know that we are coming with assumptions about what feeds we're consuming and what interpretation of these feeds of this data we are going to be doing. And we have to set all these assumptions in clear writing to everyone who can see that. And that bit is also we are in a way not being humble because we are saying the world works that way, but we are being humble enough to say, but we know that we have made assumptions and that is something that the persons who are being data driven sometimes miss. And this is what leads to what the industry knows as overfit. Overfit in my opinion is the biggest issue in all sorts of trading, human and quantitative system overfit. Overfitting. Overfitting is where you say this is how the world works. And I don't need any further input from outside the system. And it will work like that forever.
- [12:38]
A
Yeah, which it definitely won't in our world because obviously policies and market structure, you know, policies change, paradigms change pretty quickly, you know. Exactly. In the last five years the entire sort of world's oil flows have rerouted. Right. So that's right. That's just not available. So in that sense, I imagine most commodity traders only successful are have that humility.
- [12:58]
B
You have to have a bit of, you know, flexibility of mind to understand what is going on now. So we were talking obviously about how to talk back to your board member. So we talked about feeds and I was talking about how to do the interpretation side of things and different methodology. So in the middle here is how you actually process that information. Many kinds of processors happen, so the data driven ones tend to have things that would be looking in ultimately like a search driven approach to defining something like clustering algorithms or could be doing some very complex, maybe a neural network for example would be of that type of approach. Then you get a thesis driven approach would be I will create an equation in the middle and the equation would be set of rules that I could explain to people. And that is what I'm going to get signed off from my boss, from the risk and from the compliance side of things. And that is what after backtesting. Obviously backtesting is a parallel thing I'm going to talk about in a bit, but this is how you build that bit in the middle. Now there are many ways of doing that. I favor something called a directed acyclic graph, which is basically saying you have information coming in and then you don't rely on the one single process, you don't rely on only one phases. You build a tree of phases and each one of them would get assigned some kind of risk and then get mixed and matched in some ways. And ultimately that's going to provide the sizing and the risk taking that you need to do in terms of volumes, positions, etc. But the reason why I think about it as that as a directed acyclic graph is because I believe that information, when you process it is going to have to flow in one single time direction. You cannot get information from the future, but you need to be able to get information from the bus flowing in one direction so that you don't get something called a look ahead bias, which is another set of issues. I think it's a smaller issue, look ahead bias in the sense that most people know how to find, that most people in the industry know how to find and identify look ahead biases compared to the overfit side of things. But what I'm saying here is the middle side of things is you have a framework that allows you to take information and process it and that allows you to get signals that you can talk to your boss to compliance to risk and get sign off. And this says how it's going to work in the middle.
- [15:39]
A
I just have interest the guts, the mathematics of this. Is it sort of, is it regression analysis? Are you trying to look for sort of coefficients that are statistically significant, you know, of these, of a series of, of independent variables and then you're creating that tree?
- [15:53]
B
Yeah, I would Say look, in the end everything works on some kind of regression because you need to. Even clustering, right? Clustering sounds, oh, this is, you know, discrete and regression tends to be a continuous time series, etc. But everything goes back in the language of regression, you have to go back in the boss and then see what tends to move with what, what tends to correlate with what. Now a group of people who are talking about causal inference, I think that what we tend to do is, what we tend to see is correlations. Causal inference is where you say, you.
- [16:33]
A
Say it has, it's not just correlation doesn't necessarily imply causality. Right, but this.
- [16:37]
B
Exactly.
- [16:38]
A
You're actually saying. No, this is causal and it's in.
- [16:41]
B
In the English common law, when you're buying a house or something, you get, you, you get a big bundle of information to show the lineage of your house and how gets to you. So it's the same thing with gosit, right? Because you have to create a lineage of these things tend to move other things. And now I am on the nth node, on the nth generation of what is moving what. So you still have to create that whole tree of what is causal to.
- [17:10]
A
What this is the node then this is your.
- [17:15]
B
Well, I mean it's the same, it's a similar concept. But I'm saying more philosophically speaking, the causal discussion that is being had right now is more around. They are sure they have done their homework around what causes what from the ground up. That's why they come to the level where we are right now and say, I know that when you take temperature and you take time of the year and I take these, these, these are causal to demand in power. And then the demanding power is, then goes out to price in power based on things.
- [17:52]
A
So matches and fire and my house burns down, but it's the matches that caused the fire and the fire that burns my house down.
- [17:59]
B
Yes. So it has to have an arrow. It has to have an arrow of direction.
- [18:03]
A
Just because I have matches in my home doesn't mean it burns down. Right? Yeah, yeah, yeah.
- [18:07]
B
But in a way this is what people learn as a grammar from, you know, subject and verb and all these things. So they are creating a, a language to understand reality. And that's what the causal people are thinking. It is still reliant, in my opinion on ultimately some correlations down, quite down. But if it assumes causalities end up being correlations, then the whole house of gods can fall down in that particular way of thinking. But let's say that in the end, everything is about regression because you're going to see what moves what. And then you have different kinds of techniques you use. You can use regression itself, logistic regressions, all these terms get branded around. You get more discrete approaches where you have to cluster things together. And then this works when the relationships are a bit non linear, where you get a regime shift and you need to be able to say, oh, we have shifted regime right now we are no longer in a peacetime situation, we're in a wartime situation. And then you have to go back and find where we had wartime situation. Can we infer what it means on the sensitivities of new variables or variables and then so on and so forth. This is the kind of stuff that you can get in that interpretation segment of a decision making process. So we've got data interpretation. I will quickly cover backtesting and then I'll go into the next bit, which is about risk management and sizing and then finally there will be execution. So backtesting is a technique that anyone can use. Right? You would do that yourself by looking back over the charts and saying, oh, I would have done this at that point. Would that have made me money or not, based on my way of thinking about things. And it's the same thing done mathematically, computationally. And then you can look at the statistics that come out of this. You know, things like Sharpe ratio sort, you know, max drawdown, all these numbers that come out allows you to quickly summarize what is the technique that you're employing, what it has done in the past, and then you can make an assumption that it will keep going like that in the future. And that is where it's a matter of faith at that point. But backtesting is very important. If your backtesting, for example, doesn't take into account things like slippage, when you execute, it's already getting a bit less, a bit flimsier than it could have been otherwise. So in the end, backtesting is something that has to be given a lot of attention to make sure that all the realities of the world are met in this bit. And then that is where you can get reasonable figures to show to your boss and say, this is how it would work in reality if we haven't forwarded any important bits and the next bit would be about risk management and sizing. So you get these signals that are coming from your individual alpha strategies, etc. We will be saying we need to buy once a sell. Maybe you created some, some strategies that can also give you some kind of confidence level. So not just buy, but buy. But we think that there's a 65% chance that, you know, we will get it right or not. So. So this is. You can get some of these more, you know, nuanced alphas in the interpretation layer. The next bit would be about sizing and risk management. So that's where you get various methodologies, you get something efficient. Frontier mean variance optimization, Markovitz mean variance optimization. All of this is around what kind of size, what kind of bet size should I use on all these signals that I'm getting from my system to create a nectar position that I would then trade in the market? By the way, this risk management and risk portfolio optimization layer should also be baked into your backtesting so that you can replicate that bit as.
- [22:08]
A
As well, I was going to say because this is where sort of your, the notional board member or you know, senior fundamental physical trader sort of looks askance at you and says well that's where all the problem is. Right. Which is in the sizing, you know. Yes, it's fine and it's fine in equities where you've kind of for the most part there's sort of unlimited liquidity in most of these markets.
- [22:27]
B
Yes.
- [22:27]
A
Or at least sufficient liquidity to where you're. There might only be, you know, three.
- [22:32]
B
Some of. Yeah, some, some of the markets in, in commodities or some of the very important markets can be pretty illiquid. However, let's say that if you're looking at something like Apple versus Envelophone in the UK which when I was in equities was quite liquid, there is not much risk for you to pick something less liquid. For example an AIM stock in FTSE aim and put a little bet on that because you know that you have got hundreds of stocks you can, you can bet and spread your risk around in commodities they tend to be pretty correlated things. So you would still have to, you know, put quite a big, a few eggs in one basket when it comes to communities. Well, in energy anyway, when you can get slightly more decorated stuff. If you go into Uggs, if you go into metals, there is some correlation in metals, but is not exactly, you know, diversified from energy, but maybe algs you can get some, some decorrelation. But yeah, so sorry, I, I cut you off there.
- [23:34]
A
No, no, I was gonna say. Well, let's return to. Because I think that's, that's also. Am I fair in saying that that's why we've probably seen the most systematic trading or the the tip of the spear in all this stuff has been focused on Nymex, on gas and more liquid markets. Right. Where there is also more data.
- [23:51]
B
Yeah. So data is important. Definitely. So okay, so we, we are on the, on that, on that fourth base component which we talked about, which is the sizing, risk management side of things. And all of everything that I'm saying here has to be baked into your backtesting. So I'm also going to talk about execution in a bit. But you have to assume that your backtest will have to have all sorts of levels of nuance when you're back testing so that you get stats as it would be when you start trading. I myself favor a style of backtesting called blind walk forward where you say your backtesting system doesn't, has no idea what's going to happen next. It's just being fed information and it's just running through things as if it would run in, in real life. It goes all the way to the, I don't know, yesterday night. Yesterday night or right now. And it tells you how things would have progressed if you traded in backtest all the way to now. And now you transition into real life and you start trading in real life and everything should be as it would have been in the past. So this is all being in backtest. But yes. So we talk about sizing and optimization layer which is going to be saying I will maximize on sharpe ratio given these signals over the last year or something like that. When you bake in information around slippage, cost, etc. All of that has to be in and finally you get execution. So execution can be in two ways. You can as a systematic trader if you think that your signals are for long running positions. You can even get a signal at the close in the afternoon and you get a CSV file and then you execute them manually to the levels that your system is asking you to do. Or if you think that you have a system that is going to potentially react intraday, that's where it gets a bit trickier. To react as a human and execute it yourself. You would want to use an algorithm and that could be things like twap, vwap, pov, percentage, rev, that kind of stuff. It goes over to the market and then it completes a feedback loop back into the system. So these are the parts that you need to give confidence to your stakeholders when you build a systematic trading process.
- [26:11]
C
The energy and resources sector is experiencing unprecedented changes. To help navigate this change and capture its opportunities, HC Group launched Enco Insights A global advisory network dedicated to the sector providing senior advisors and subject matter experts to investment and infrastructure funds, law firms and corporates. Enco Insights leverages HC Group's 20 years of connections in energy and commodities to give clients the expertise they need. Need when the stakes are high. And insight matters. Learn more@encoinsights.com the fact that I'm sort.
- [26:49]
A
Of marginally still alongside you and in some of this is, is I think a testament to you and your description of this. This is. So when we talk about execution, this is where we get to this world of high frequency versus mid frequency versus low frequency. So is. And. And then that has. Which I didn't really know, but now I know some work we've done that obviously that has significant compliance requirements, at least depending on which jurisdiction you're in. So high frequency be. I've done my systematic, I've set up my, my graph and its nodes and I've done all my back testing and my sort of sizing and all this good stuff. I've got my pathway to trade. I then hook it up to an algorithm that directly feeds into the exchange and it just runs. That's high frequency, right?
- [27:31]
B
It might be what you describe here can actually be both high and mid frequency. The mid high frequency side of things, I would say is where your, you know, the middle layer interpretation is actually. So it needs to be so fast usually that it gets baked in directly into the algorithm to the execution layer. So that's where you have to react in the milli microseconds. Sometimes maybe you're trading across various exchanges, making sure your orders land on each exchange at the same roughly millisecond. Because if you say you're sending an order to London versus New York and you send the same order across the wire and we are in Europe right now. So you would land in London first and then it would land in New York. If someone has a faster link than you between London and New York, maybe some kind of satellite or other kinds of technology, they could realize that you intend to also trading in New York because we have seen your London order trade before your order has landed in New York. And then we can do things like they pull their quote in New York and then all of a sudden you are legged out or you're out of your, out of your zone of optimality.
- [28:45]
A
So of course, I mean, that's interesting, right? So of course people are building systematic trading capabilities, not focused on the market but on what their competitors are doing.
- [28:53]
B
That's right. That's Right. And that's where, that's where the micro side of things is important. So the information around what people are doing themselves without any big view of a macro picture is so important these days. So you may have heard that in the last quarter or so Jane street has made more money than hedge funds, than banks, etc. So Jane street, when I was in KCG was a kind of high frequency competitor. I believe that they are managing to keep going that philosophy of keeping a microstructure information and squeezing as much as possible out of that alive. They are apparently also doing mid and longer term strategies. But I'm assuming that there is a bit of, you know, work that they are doing that brings in that high frequency back into the picture.
- [29:43]
A
What would low frequency be? And can you just give us, you know, when your average merchant, your average trader, your average, you know, in our world, in the physical energy commodity world, what prevalence is there of high frequency versus sort of this, what you're about to describe as low frequency?
- [30:00]
B
So you will be saying so high frequency or at least mid high frequency is coming, you know, quite, quite a bit now because you get batteries being optimized, you're getting information in some markets you can trade every 15 minute delivery in power markets, for example. And these kinds of markets I think are exactly what you need for algorithmic execution. And it could be even high frequency decision making on this information.
- [30:28]
A
This is where you have to upload your. In certain jurisdictions you have to share your algorithm with the exchange or with the, the monitor over, you know, with the, with the regulator.
- [30:39]
B
You have to. So if you different regions do different things. But for example, if you are a regulator by mifid, you need to explain, you need to, to send. With each order that your algorithm is going to send out, you need to attach a tag with your algo ID and that should then be able. So before that algo went live, there would have been some kind of compliance done where you say that your algorithm is not going to be a bad market participant, it's going to adhere to rate limit, size limit, price limits and it's not going to behave in a way that would be construed as, you know, destructive to the markets. And then every order that this algorithm will send will have a tag and then that tag is then going to be used in case there's an issue later on.
- [31:26]
A
And I hope no one's figured out what those tags are. And I mean there must be a fair amount of security behind you sending over your secret source to so.
- [31:35]
B
Well, I Think there's going to be some layers. There's always some secret source that is, that is kept in a trade sequence. Because if you think about how I've just described how an algorithmic system, systematic system is built, there are multiple layers to that. You could give algorithmic execution side of it for free to the regulator, but you would still be able to keep some stuff to yourself because of a previous processing you've done on various parts. So I would say it's, it's not as easy to just give a, to say a regulator would be able to just take my information I've given them and be able to recreate a profitable strategy because there's a lot of hard work. Even the data cleaning side of things is a bit of hard work. So it's not as simple as that.
- [32:28]
A
I'm just, I'm getting, I'm getting conspiratorial now. I've been sent off down this rabbit hole of thinking about just trying to figure out what your competitors are doing rather than. Anyway, you can tell I'm enjoying this. But. Okay, so. And I keep interrupting you, I'm sorry, what is low frequency?
- [32:41]
B
Right, so let's talk about holding periods. Right? So in the high frequency space, first we are reacting in millisecond and sometimes microseconds, but the holding period, so the time it takes for you to accumulate a position and then maybe divest, that would be in the minutes, tens of minutes, maybe one hour. So that's, in my opinion, that's how I think about high frequency. Mid frequency, you would react intraday, so you would still execute as fast as you can because you've got access to algorithmic execution. You will execute as fast as you want to execute, but you would tend to hold it for, you know, hours to days and very early weeks. And that's where you're using information that you believe would be around inventories that people have in their books. And then you would be using that together with some macro information, some fundamental information that tends to be in that kind of cyclicity. For example, I would say weather data, because weather forecast change, you know, every, you get one forecast from every six hours or so, I think from ecmwf, gfs, that kind of stuff. So you are using the kind of granularity of a change of frequency of these forecast to make your positioning align with that. So that would be in my mind what mid frequency would be in the hours to multiple days or even a week.
- [34:00]
A
So frequency is really about the holding period necessary than the number of Trades you do in any given moment.
- [34:05]
B
Moment frequency in terms of execution speed, that is execution frequency where you can go in and out very fast, milliseconds, that kind of thing. But then a holding period is what I think people have in mind in our industry as opposed to, if you talk about frequency to someone in equities, maybe they have in mind the execution side of things. But in commodities and energy, I would still think that people are thinking about holding periods.
- [34:32]
A
Okay, but the compliance piece comes in as a result of speed of execution.
- [34:36]
B
That's right, that's right. And there is always a chance that, you know, regulators can still think about. Oh, I'm interested in also your holding periods and sometimes you do see that. So companies, for example, for your, if you work for a hedge fund, sometimes your compliance team would say if you're buying for your own personal use this stock or that etf, you cannot divest it over a month. You have to wait longer than a month to divest it. That's where you know, compliance can look into frequency, into the holding period side of things. But typically compliance would be more interested in the execution speed. Interesting bit. I used to work for KCG. Night Capital became KCG after it lost about 400 something million in about 45 minutes in the US and that is what. And the reason for that is there was some kind of good that was checked in and that good had a switch that was in the wrong way and then that made orders being sent into the market at much higher rates than it should have been. And that made a 400 million loss in 45 minutes. On the back of that, a lot of regulation caught up. So in my mind there is three things that is called size of orders and price of orders. And also the rate orders are being sent out. These three things, you will see them being managed in many kinds of other rules. But in the end these three things, these three properties are going to be looked at very carefully by regulators. So low frequency in my mind would be things that are being held for weeks, months and sometimes even longer. So multi month, even a year. And low frequency is where the skill of the investor and skill in the sense that they have to build a very good model of the world in their mind. And that has to stay for as long as possible and it has to withstand shocks that could happen, you know, geopolitics or change in regulations, etc. That has to withstand the shocks and then it will still make you some money at the end of a period. So that is a different kind of systematic Trading required for this and it's a different kinds of thesis required because you don't have enough sample size for testing that with a good level of confidence.
- [37:05]
A
Fantastic. So we're going to talk in a minute about does it work and kind of where there's correlation of where it works. But just before we get there, there might be someone listening to this thinking, you know, we probably should have one of these by this point.
- [37:18]
B
Yeah.
- [37:18]
A
How, how long? Just roughly speaking and obviously this depends on the size of the platform. You know, all of this stuff, how, how long does it take just to rough, you know, what's the investment piece to stand up one of these capabilities?
- [37:32]
B
So okay, I would say the way I build it and different people have different views on this. The, the most of the two most expensive pieces is data and execution access. The bit in the middle, which is processing that can be done very cost effectively if you're a coder, if you know what you're doing with maths and these days with LLMs, you can ask it anything and it can give you some code, it can give you a piece of logic. So the bit in the middle is not very costly once you know what you're doing. But data, you know, having access to historical data where you have to get a big dump from exchanges and then getting the live data that you keep subscribing to that bit can be costly. And we're not even adding vendors of alternative data, of fundamental data into the picture. Some of it you can get for free, which is publicly available, but then you have to invest time writing scrapers, scrape this data and then bring it in house. And that is quite a bit of an investment. And then the next bit would be about access to the markets. Now I would say access to the market is a decision that you make once you have done your test, once you've got your data and your interpretability layer, you've done your backtest and you're happy. So the cost that you can get out of is a cost of data, you need that. And there will be a bit of cost associated with, you know, running cost of your interpreter interpretability layer, which is going to be your servers, if you're in the cloud, aws, Azure, that kind of stuff. And obviously the people themselves that you hire, they are going to be very important. It's important that they have got the humility to understand where they are wrong. And these people can be a bit expensive as well. So.
- [39:26]
A
Yeah, yeah, yeah. As Rick Lee, who runs our technology practice practice with us, you know, he knows the answer to the question I'm about to ask, I should point out. And he's probably cringing as he hears it, but like just in terms of those, those titles. So data, data analysts and engineers, those are the people that are scraping and cleaning and collecting them and building relationships.
- [39:45]
B
With vendors and, and, and other places where you can get data as well. What's part of their job, in my opinion.
- [39:51]
A
Yeah. And then developers I met. Well the second part of the question said the coders and all that piece, those are the. What are the titles of the people plug in, do the APIs into the actual exchanges. That tough bit.
- [40:03]
B
Yes, there are two big classes. Typically you get quant devs and quant analysts. And the quant devs will be usually integrating data feeds into the platform, building the platform itself, platform which is going to be containing the business models, the mathematics, practical models that quantitative analysts tend to build. And then so basically they build a wrap around the work that quantitative analysts would do. In my opinion you can get quite a few people in the industry could be both devs and analysts and traders at the same time, but going and finding data sources, that is, they will find that very boring. But you will, you will have people who can do all both analytics and the platform building.
- [40:54]
A
And is it a benefit to have people that deeply understand the energy and commodities sector or in some roles, is that a hindrance? You know, if you're thinking about that kind of like I don't want my developer with buy, you know you're saying you want them with theses already, right?
- [41:09]
B
Yes, I say that you want to be able to have any domain knowledge. What you have is important. It's important to have people with domain knowledge in your team. But then you have one person who is more of a microstructure guy. So I would say now I do have domain knowledge, but when I left the equities world I was market making in a place called Winterflood and I joined Shell at that point. At that point I didn't have any domain knowledge in energy. I was more a macrostructure guy where I would say people behave that way. When you're seeing these kinds of patterns of buying and selling and accumulation, that is what's happening. So I would approach it from that side. And over time I have been able to integrate how the fundamental data that you get in commodities, how that should color the behavior of these counterparties that you're meddling in the micro world. It's a very blurry line, these two, micro and macro. But it's kind of a feedback loop. So what I'm trying to say here is even if you don't know domain knowledge, you will be able to get in and then, but very soon you will want to integrate as much domain knowledge as possible to try and, you know, get some, some edge. Because if you're just doing the micro side of things, you are one among many, many people.
- [42:32]
A
And it's commodities on hard mode as well. Right. Like, you know, we always say this is, there's so much the nuance to these markets from a regulatory standpoint, from a physical constraint standpoint, there are only two cargoes on the water. If one turns around, you know, that's 50 of, you know.
- [42:47]
B
Yes. Yeah. Getting information, maritime information from, from certain vendors, that is very important. There are some analysts in that. Just look at that. Yeah, so.
- [42:57]
A
Well, that's where a lot of the volatility is going to, as we've covered on the podcast. Right. We're definitely going to run over on this, but I'm enjoying it so much. We're going to do a long episode so our listeners will have to bear with us. But I think it's an important bit of work. Okay, so that's the team and I guess our answer there is there are these sort of discrete capabilities, skill sets, there's some overlap between them depending on what you're trying to do. What's fascinating there, you're saying, is actually, you know, with the right person, that middle bit is actually, you know, that's where you want to buy real talent and then you want leadership with domain knowledge on the other, on the two ends, the data in and the execution out. But obviously in a thesis driven world, that's where the trader earns their alpha, if you'd like.
- [43:39]
B
That's right. And the thesis. So a thesis is a is, is a skill. A creation of a thesis is a skill because anyone can have a view of a world. Right. You can be a conspiracy theorist and have a view of how world works, but that doesn't mean it's going to.
- [43:54]
A
Be as we, as we talk right now Congress is discussing UFOs. So just a. Yeah. Oh, wow. September 9th.
- [44:01]
B
Well, maybe that's going to be useful to some people.
- [44:04]
A
I imagine that would have a pretty significant impact on some algorithms.
- [44:07]
B
If, of course, of course, you know, foster online travel. Right. So yeah, so yes, it's important to have people who are not too dogmatic. But even then they need to keep in mind how the world works not just mechanically, but also from how people would React So the behavioral science side of things. So we need to understand when this price does this, it will cause a margin call. Even if fundamental picture is supportive, this situation would be a margin call because too many people have accumulated longs or something like that, and that would then push the price back down. So these kinds of marrying what happens across the micro or positioning side of things and the macro, which is the objective reality of what's happening in terms of demand and supply, that is quite important in a way. If you think about this, we talked about that before. But the trading decision making process is about information asymmetry. It's about what information is public and what is private. And the information that is private and impactful is very much underestimated in my opinion because it's about position.
- [45:27]
A
That's exactly where I want to go. Right, okay. So. So the final segment of this of course is does it work and not just obviously what returns this unit, this desk produce itself, but also how it changes and facilitates decision making across the entire organization. Yes, the first question there is, and I've been guilty of making this assumption, but that most of these models certainly if you're in a hedge fund, so if there's. We cascade it down from very transparent markets or organizations that have very little sort of physical footprint in the sector. So let's say a hedge fund, they are very good at, they've got lots of money so they can buy lots of data and they're very good at synthesizing and capturing all that public data and, and, and subscription data. The argument sort of. We've. The thesis that we've laid out over this podcast over a while now and we're seeing it is actually it's the large strategics, the oil companies, the refiners, the AG houses, right. That own, there's four of them and they own the entire value chain, you know, of assets, they presumably are producing significant data that is proprietary, that is private.
- [46:36]
B
Yeah.
- [46:37]
A
Were that to be cleaned system, you know, put into the system would make them the absolute winners for a period of time versus those hedge funds that are relying on.
- [46:49]
B
I know where you're going with that, but it's not as simple. It's not as simple because even within company big companies that are, that have got infrastructure and hence data, there is a regulatory, you know, Chinese wall between proprietary restaking businesses and the data that can flow from the other sides of a business into that. And that is itself going to be a bit of a leveling of a field with the normal only financial market participants Where I think it is still useful however, to be in a big infrastructure community firm is the idea exchange that you get. Because talking about things in the ideation world abstractly is not something that you can regulate because it's just about intellectual exchange. And that is where you get faster integration of ideas. But you can just keep using the normal kind of public data. But your interpretation of that public data because of your interaction with people on the ground is more interesting, more intricate and more likely to be closer to the ground Truth that is where I think systematic works with big firms is.
- [48:07]
A
That, is that just to understand that. Sorry if I'm. I'm a large like making it up oil major, right? And yes, I guess this is the, you know, I, okay, I, I guess this comes down to a little bit of that. I, I decide am I regulated from a regulatory standpoint? I can't run my own systematic trading, use my own proprietary data, hook into you know, ICE or whatever it might be Nymex and trade away algorithmically.
- [48:34]
B
So you know the remits, remits which are reports that you send out if for example, you own a refinery event and there's an issue, there's going to be some kind of breakdown of the refinery when you have to maintain. So that information has to be made public to the world via the remit system and you have to make that information public to the world and you cannot send it to your prop desk before you send that to the world.
- [49:10]
D
Hello, I'm David Hunt, founder and managing director at Hyperion Search. Founded over a decade ago, Hyperion Search has helped organizations from major utilities to startups recruit their leadership teams and key individual contributors to accelerate both their growth and the energy transition. Our three main verticals are renewable power, energy storage and the mobility. The energy transition and the talent that delivers it has been our passion since day one. To find out more, visit hyperionsearch.com or listen to my Leaders in Clean Tech podcast, available on all platforms.
- [49:43]
A
But I can, you know, I'm still absolutely able of course to make all of my trading decisions. You know, I, I still can't trade on that information ahead of time. I get it right, but in terms of my sort of trading enabled capability, the decision making we're making around whether it's a longer term structure or whether it's, whether we're going to buy whatever it might be, that trade, that physical trade, presumably that can be incredibly enhanced. The toolkit that surrounds a trader to be able to make those decisions is, can be significantly enhanced by capturing this Proprietary data, mixing it with some public stuff and then having decision making supported by systematic decision making.
- [50:26]
B
Yes. So what I would say is let's assume that the proprietary data is something that is going to be made public so that everyone gets it at the same time. I know it's no longer proprietary at that point, but the understanding of what it really means when that data goes out, because people in the maintenance who maintain these oil rigs, these refineries, etc. They know. And you can talk to them because they are your colleagues. If you're working for these big firms, they know what it means. They know that everyone is going to be seeing the information that you're going to be sending out. They are going to be sending out at the same time, including the proprietary desk. But you, as a proprietary desk, you can go and talk to these guys and say in the past, when these things have happened, what does it mean? How does it impact your ability to bring it back in terms of maintenance? And you build a middle around what we will end up doing as opposed to we are just sending out mechanically some kind of information for everyone to see. So does that make sense? I think that's where the edge lies.
- [51:35]
A
Yes. I think we need to define what we mean by proprietary data. So I'm thinking of it in the sense of I'm a refiner, I have, if I've done a good job with my paperwork, it might be quite hard to find. But you know, I can look back over all these market events and see how my fleet of refineries responded. Right. Decisions that were, you know, there's, there's also that, that's. In other words, your, your historical set of data should be more enriched if you've done a good job. Or at least people should be thinking today about how they're going to capture that information more.
- [52:06]
B
So in that sense, I think we are. I'm talking about the same thing as you in that sense, because what we are talking about here is, you know, how things would react on the back of information that goes out. Exactly. So a hedge fund, a purely financial player, would not have access to the nuance around this. And you have access to that because in a way it's proprietary data because, you know, you talk to people, but it's not market moving data.
- [52:37]
A
But you're able to say we also know though that like for example, our traders are out there talking and they know that. So, and so, you know, actually there's only three ships on the water at the moment. Or you know, in other words, there's, there's, presumably, yes, richer data available, there's.
- [52:53]
B
Richer historical data, but maybe, maybe we should, the language around this is important. So I would say what, what we have described here is not data that is going to be added to the system on the fly. It's more information that you take into account when modeling, doing the middle part, the nodes. Yes. You take that into account when you build that part of it. And once you build that, then it's here and there's no issues, there is no regulatory issues around this. But we have built that with a richer understanding of a world.
- [53:27]
A
Well, I think that's, that's, that's useful. Okay, so secondarily, in the category of does it work?
- [53:31]
B
Yes.
- [53:32]
A
Does this work well in where does this work best? Does it work best in, in a system like, let's just say lng, where all these sort of variables can kind of be quite, you know, there's a finite number of ships, there's a finite number of terminals, you know, you know, like the port depth and all this, you know, where does this work?
- [53:53]
B
So I'm going to say there's two types of utility that you get out of Systematic. The first utility is the value of repeatable processing where it helps management talk to their stakeholders, to shareholders, to other people. To say that we have got a component of our, you know, our revenue which is going to be always reacting in the same way. So that is in neutral. It doesn't have an edge in itself, but it helps you talk to the firm helps. It helps a firm talk about how it's going to work and hence build relationships and reassurance to the markets. The other bit is, I guess, around the alpha. The edge itself, I believe that, that anyone, even the person, what we call discretionary trader right now, is also in a way systematic because we have built some. Everything is a spectrum of systematism because they would be doing certain things similarly over time. So what we are saying here is the systematic bit allows you to be more disciplined, but it's really up to the person using the systematic toolkit to have alpha in the end. So maybe it becomes easier and it helps you create more systematically capable people over time. But initially, the person using that systematic toolkit is more likely to be the important factor in determining the profitability of that system, rather than just purely saying, oh, systematic is going to make money or not. So yeah, now that's not to say that there are certain things that are just copied across in the industry. So, for example, CTAs, it's a bit of a cliche, but I would say they are trend following strategies. So if you're creating a systematic desk and they tell you they would be using CTA style strategies that will have correlations across many systematic desks in different firms using CTA strategies because trend is your friend is a very sticky and very correlated way of making money.
- [56:15]
A
Yeah, yeah.
- [56:17]
B
But then if you tell someone that we are doing statob now, startup is monuments because startup is different kinds of relative value valuations of things that someone can be doing. Also, if something is overvalued or undervalued, the person who is deciding, who is creating a model can decide that it can be overvalued. But the threshold of a trigger for actually taking risk on the back of that over or undervaluation will be their own methodology which would be different from a different stat orb person. So stat hub in my mind would be more decorrelated among themselves versus a CTA style systematic definition desk. So yeah, it depends on your mix. So for example, myself, I run many, many, many nodes because I don't want to decide and impose what strategy should work at a given point in time. I have a group of trend following strats, I have a group of reversion strats. There would be things like looking at relative value and carry in different kinds of ways. So all of us that would be competing for attention from the portfolio optimization layer, which is just after the interpretability layer. And then that is what decides to allocate weights. It's because I don't want to say that I knew exactly what's going to happen. What is going to be the next thing that takes over the regime in.
- [57:45]
A
Terms of that does it work third leg, is this fair to say at the moment? And maybe if I you get, you know, that probabilistically shorter tenor. Right. High frequency, very short tenor holding is more successful on balance than as you go toward, you know, inversely proportional to the longer term piece. Right.
- [58:05]
B
So, so, so yes, in, in the. There are many reasons for, for that being true. I would say first of all the, the liquidity means that you have more data in the shorter tenors. And that means. And more data is good for people who are quantitative systematic because it helps them with their backtesting. As I mentioned, the data side of things is not enough because you need to have that flexibility of mind so that you don't overfit because you could have decades of data and that would still not tell you how the regime would change how the market behavior will change in the future and then hence these decades worth of backtesting could be worth not much for a given piece of time. You are always making a bet. You're having faith that some repeatability would happen. And typically yes, you have to have that faith. But data is important and in short term tenors you have more data and it helps you out. So you get more samples and more lives to do it in short term, data, data. And that helps you systematically.
- [59:11]
A
Yeah. And then okay, so then it's been fascinating. But so as you look forward, you know, over the, I mean the developments in this have been quite rapid over the last five years from it being kind of, you know, relatively esoteric, the purview of the hedge funds and, and the CTAs to now obviously being a strategic pillar for many of the, of the merchants and the strategics as well. Where is this going over the next five, 10 years?
- [59:40]
B
I see two things. I see that there is going to be internal pressure to do more systematic anyway because data availability is getting higher. So what market will drive itself in terms of a skill set required, will get more quantitative and systematic over time. People will have to upskill, etc. But it doesn't need to be a very, you know, rough, you know, realignment in terms of skill set. It would be, it can be a gradual alignment of skill sets and you would teach people internally to start using a system. You would make your system what are being built will have to have some kind of democratic process so that people can get involved and participate in that. So that's one thing. But then the other things, that is what's happening in the actual physical world would be batteries for example. Batteries have got the Internet of things connected to the Internet. The sensors and the data are being sent out there. So you could actually get information on very intricate, very granular pieces of infrastructure and decide that to balance a grid right now we need to do this and that. So that's just one example. But I'm assuming there would be other things. Sensors would be going into grain silos to try and see what levels of things are happening. You've heard about AI coming into the meteorological forecasting side of things to get more faster forecast, that is more data being produced and maybe you can have an in house forecaster that you can run much faster than waiting six hours for a next forecast to come out. So all of that is getting closer. And when I say data, data is the main thing, as I said, it is the most expensive part of the puzzle. So given that it's getting more ubiquitous. I think it means that it generally generates more people in the systematic space focusing on that. Because to interpret all that deluge of data you would just need people who can actually take that and make decisions. That's not to say that you will not have people who can, you know, just having a very cursory look over the headlines. And these people have got some kind of model of a world that is, you know, touching on just the right things. And just from looking at the headlines we would still be able to make a judgment of what's going to happen next. But the chance of that continuing to be, you know, as, as good as we have had it in the past, I think gradually decreases as a market gets more efficient and more good at interpreting data over time. I think that is where it's going.
- [62:25]
A
Yeah. So to theory that out because that is fascinating. Right, so okay, in, in the day ahead market or in the very short term, all markets become ever more efficient and more, you know, the percentage of trading becomes ever more systematic systematized. What does that do to where does that push value? Right, the adjusted risk returns. Because presumably that means suddenly there's people are building ever more expensive massive machines to participate in that. That drives consolidation amongst market participants. You have to be of a certain scale and size to be able to do the, to participate effectively. The number of dollars left on the table are fewer, it becomes less profitable. Systematic trade, you know, and suddenly the whole world is thinking, well it doesn't really, there's no money in short term trading. Everything's about taking mega long term bets on, on an industry, a sector, a country or whatever it might be.
- [63:22]
B
So the way it happens is the game, the game that is being played across different kinds of tenors. The game gets very crowded in the parts of the curve where there's enough data or liquidity being thrown around. And that definitely pushes the more human style players down further down the curve because that is where there are fewer samples of what happens in terms of is there going to be some kind of geopolitical situation down the line. And you can systematize, you can backtest that as much as you want. Every war is its own thing. Every political lineage or trajectory is its own thing. And I think the human, the branching factor that you get in this game means that your human still has a closer to. Everything goes closer to 50, 50 chances for the machine and for the human. And at that point the human can decide that they will still have an edge there because the human are still the the apex predators on Earth, they have a view of the minds of humans and other societies. So what we think is better than what the AI would be interpreting for these. So they should still have an edge on the AIs on the longer term, fewer sample side of things. So that is an obvious thing. So shorter term becomes almost like a repeat, like a completely solved problem and systematic systems will fight among themselves to optimize that. And then maybe that is what's happening by the way around why Jane street is making so much money. Because where it's focusing is just now mastering that purely through automation. There are other players like XTX as well doing very well but that's on the equity side of things and FX maybe, but this will happen in commodities as well. Now maybe some, I've been talking about tenors, but let's talk about asset classes. So AGS probably is a bit harder to fully systematize compared to things that are a bit, you know, flowy, more flowy like power and I would say metals, some, some of the precious metals are still harder to fully systematize. So that's where you can, can look into that. But, but the rule of thumb is around data flow I think the flux of data being higher means it's going to be systematized and it's going to push human players into other less data centric places.
- [66:00]
A
But the lessons, what I find fascinating about this, and this is why perhaps we're danger of the longest pod yet but the same application, the same thought process, the same scientific method essentially can be applied to decisions of any tenor and scope. Right? So yes, this is sort of the idea that okay, my originator or my M A activity rather than it now being a bit of, kind of analysis by my sort of, you know, MPVs and a few assumptions and so forth. And I know obviously organizations do this already, Monte Carlos and all this good stuff but actually you can use, you can go to your systematic desk and say can you test these thesis, can you. I want you to help me understand the valuation of this asset I'm going to buy and how we're going to hedge it out and what it might do over the next 10 years and that becomes a very powerful brain in the organization is what I'm saying. So yes, you can plug it straight into the market and you can, you know, do what you do but actually having that capability built itself becomes kind of, you know, whatever sci fi movie you want. Right? That's HAL you're asking on the TV screen, hey, you know, x And all of that's going to get supercharged by AI as well. So having that clean data.
- [67:18]
B
So you mentioned Monte Carlo actually a good thing that you, you mentioned that. So traditionally Monte Carlo is about volatility over time and how things can change. They are going to be. I'm being unfair to people building Monte Carlos here because some people could have very specific trajectory dependent Monte Carlos and that could be saying when this kind of thing happened, it actually is a convergence point in the timeline. That actually means that all the many of Monte Carlo realities would converge towards that point and then go back out again. So you can get these kinds of convergence points if you want to build Monte Carlo. But the reason why it's very interesting because there is, I think it was either the AlphaZero or AlphaGo, one of these deep mind AI models when they maybe the chess or the go game, they were using something called a Monte Carlo tree search which basically says you, the AI, the neural network is creating situations in, in its embedding space. And then they are going to be running, you know, simulations in multiple simulations. And then they will say okay, I'm running the simulation, this trajectory here, how likely? With a counterfactuals being that I've generated over time through my training, how likely is that going to be sustained? And then you get not just a line in the Monte Carlo that is going in one direction, the line can get fainter because it's with probability of that line keeping going. You can actually assess that and start saying yeah, that is not likely to happen. So you get these kinds of opacities in space, in temporal space around what is going to likely happen. That is maybe the way you deal with lack of data. Because you can say that in these very long term, very few sample size, a very small sample size problems that you're trying to solve 10 years down, you can still create what you would call temporal convergence points that you could say yes, I don't know exactly how it would happen, but there's a high likelihood that in three years given that these, these other things will tend to happen in various parts of the world that would create a convergence point in Middle east there would, that there would be something there that would touch on a product that I'm trading back then. So this is a kind of stuff that you could do. Again this is very speculative and maybe that is harder than what I'm thinking about right now. And it's very sci fi but you know, that's how you could actually make it happen.
- [69:52]
A
Is there anything that you can look at in the equities worlds where they've got this sort of 5, 10 year head start in reality that we can extrapolate over.
- [70:02]
B
Well, so one thing that I would say is so in equities world where you get, you have even more, what I would call analysis reports, analysis reports than you have in the communities world. You have it for every company that is listed. You have it for ETFs which are like aggregations of various kinds of weights of companies and other vehicles. The reason why I'm saying that the analyst reports themselves are going to be interesting because you generate data just by having analyst reports come out. So even if there's something 10 years down the line, just by having many, many analyst reports being sent out, it creates a reality of its own and then that impacts information very far down the line around this. I'm not saying that this is going to exactly happen as these analysts would say, but at least you can trade a consensus further down the line. Maybe that is how again, it overlaps a lot with the chat we just had about Monte Carlos where if you have more and more analysts predicting further down the lines, maybe especially in the net zero space, for example, a lot of that is happening, then you can use that information to make relatively large bets that are going to run for a long time down the line. So in that sense it's a bit of a. Some kind of techniques that have been taken that can be taken from the equities world. Obviously you have high frequency space that has done, has taken over equity space. That will happen. I think that will happen in the short tenors a lot. I'm sure the market makers, the high frequency market makers are discussing power markets across the world right now and we will try to solve that. And these are very, very intelligent people and we will see what happens.
- [71:59]
A
These are the sort of, you know, rather sort of opaque groups that sort of buy office space as close as possible to the LME or to ice.
- [72:09]
B
Yes, yes. So colocation, you know, that will be part of it. But I'm assuming there is also micro alpha that we would be looking at where they are predicting, you know, minutes in the future on how things would be happening and there would be market making around that view. There could be, you know, some firms I know, they are not even trying to get humans to code the thesis. They are actually data driven. I think at that level you can afford to be data driven because there's so much information that data driven and without a thesis you can get away with it because Your market making and holding for many minutes and tens of minutes. So. So, yeah, but these tenors would be electronified over time and it would push people out in that direction.
- [72:53]
A
Yeah, it's interesting. It's also still in the, in a way, it still comes back to, you know, how the markets have ever operated. Right. You know, rather than analysts reports in equities in, in a world of highly consolidated oil majors, you know, it's just, it's a few looks at what activity they do in the window or whatever that can determine everything.
- [73:12]
B
But yes, yeah, obviously regulators tend to look very closely at that. Right. So and, and try to prevent bad behavior.
- [73:20]
A
But my point being they, they still are sort of, you know, they. Yeah. Like understanding what your. Your counterpart is doing is.
- [73:27]
B
Yes. Can be.
- [73:28]
A
Yeah. Anyway, well, Hans, it's been absolutely fascinating. Obviously wish you the best in Uniper and you know, hopefully we can have you back on in a, a year or two and see whether any of these predictions have come true. But where the market is and I think it's, you know, it's a bit of a. Well, I found it fascinating to actually understand a bit further into what we talk about when we sort of throw around these terms and, you know, how it might shape the market and what the future might hold.
- [73:57]
B
It's a pleasure for me to be on your podcast, but in my mind, you know, having these long chats and conversations is actually a way for, for me to diarize my own thoughts and I'll probably, you know, listen back to what I've said in, in one year or two years and say, oh, I've got that wrong or I've got that right. So, yeah, same for me.
- [74:18]
A
Thank you for listening. To find out more about HC Group, our global offices and our expertise in search within the commodities sector, please visit www.hcgroup.global.