Richard Sverrisson, Editor-in-Chief, Montel: 0:07

Hello listeners, and welcome to the Monto weekly podcast, bringing you energy matters in an informal setting. This week, we will delve into the world of robots and algorithmic trading. AI has been making headlines this year, as it is set to make huge inroads into our daily lives, with new developments occurring almost on a weekly basis. Robots are well established in financial markets, but how prevalent are they in energy markets, and should energy traders be worried about their jobs? Joining me, Richard Svarasan, to unravel the world of high frequency algorithmic trading, is Imani Joss. A warm welcome to you, Amani.

Amani Joas, Managing Director, CF Flex Power: 0:59

Thank you very much, Richard.

Richard Sverrisson, Editor-in-Chief, Montel: 1:00

Can you just tell me a little bit about yourself and your role and the company you work for, Amani, just before we get started?

Amani Joas, Managing Director, CF Flex Power: 1:07

Of course. So my name is Amani Jawas. Thanks for having me again. I am working as managing director as one of four co founders for a company called CFlexpower. We're quite new in the market. We were established one year ago. And our focus is in three areas. We do route to market trading for renewable assets. Um, we do what we call slacks trading for, um, flexible assets, mostly batteries. And we also just launched a platform called PowerMatch for standardized, or as we call them liquid PPAs. And my background is in energy trading. We have been on the short term markets as a team, I think, with one of the longest track records with a focus on the continent, especially in Germany. And the four founders were like the core trading team at one of Germany's biggest trading companies. They got acquired by Shell and then we had a good chance to venture out on our own. And, and we're, we like what we do and we do what we like. So, so it's been going very well.

Richard Sverrisson, Editor-in-Chief, Montel: 2:12

So you're very much an expert in the field of short term trading and the slaves to the algorithm, if you like. Let's, let's start by talking about these robots in short term and intraday markets primarily. How prevalent are high frequency trading algorithms in these markets that you're dealing with?

Amani Joas, Managing Director, CF Flex Power: 2:29

So the markets that, that I would focus on today is really short term trading of, of power more, more so than gas. These are the markets usually on the big exchanges are Apex and North pool, and you have the day ahead market that's run as an auction. You have some intraday auctions, depending on the countries, and then you have a. Continent wide continuous intraday market. And I want to focus on, on that market because that's where we have seen the big rise of, of the algorithms and there they are extremely prevalent. So, um, over the last five years to growth in, uh, in positions being closed by algos has, has been quite rapid. So on EPEX, we see that 85% of trades at the moment are already closed by ALGOS and, and about 60% of volumes are closed by ALGOS. The difference comes a little bit because ALGOS tend to trade smaller positions. They're faster than humans trading with their fingers and the trend is continuing. However, it's important to distinguish, like when you say high frequency ALGOS. These are not the kind of autos that front run other people in the massive speed that we have seen on stock markets, but most of them are quite simple automated execution engines. But on this, they have been taken over, but it's different to say automated trading or autonomous decision making. So this is not the big AI revolution that you're seeing. It's, it's more a tool that, that traders are using.

Richard Sverrisson, Editor-in-Chief, Montel: 4:01

But we'll talk about more, more about that later, I think, and also about the, the future outlook for the, for the robots and these algos. But these are quite huge numbers, 85%, 60%, you know, could, could we reach a point where it would be 99% and sort of 80% of the volumes?

. Amani Joas, Managing Director, CF Flex Power: 4:18

Maybe, maybe a little less. So I think the trend is going, going towards that, that direction. But I do think that there is a role for, for humans and for manual position closing in the market, as I think we'll, we'll delve more into detail on later. But yes, I think, I think we'll get to around somewhere in our 95 positions in the, in the 85% of volumes being, being closed through APIs and, and with algos behind that, that's definitely the trend that we're seeing.

Richard Sverrisson, Editor-in-Chief, Montel: 4:45

Well, that's, that's, uh, that's quite a transformation. Then I think you're very familiar with what you do in, in your, in CFlex power, uh, and maybe some others, but, but our company using them differently, are these the same kind of algos or are they, are they, are they, you know, I'm, I'm a complete novice here, Armani. So, I mean, what kind of algos are we talking about here?

Amani Joas, Managing Director, CF Flex Power: 5:04

So yes, companies are using them very differently. And I think the, the most important point to, to keep in mind that there is. Very different levels of sophistication. So algorithms are essentially a tool and their sophistication ranges from that of a screwdriver to that of a full scale assembly bot that you would see in an automated factory. Um, and the use cases for algos are mostly, um, position closing of, of renewable positions on the intraday for asset optimization, market making, arbitrage trading. So this is where you see a lot of the algos. When it comes to prop trading, I think you still see a good mix of humans and algos that tend to have the best results.

Richard Sverrisson, Editor-in-Chief, Montel: 5:47

Is there a danger here that the machines, the algos are all following the same kind of signal? So if you say, okay, there's a sudden change in the weather or a big power plant goes offline and all the algos react in the same way?

Amani Joas, Managing Director, CF Flex Power: 6:02

Like the algos do run on similar signals, but also do humans. So, uh, so we all look at similar things. So when something, something happens, the market might overshoot a little bit and overcorrect and a classic thing that we see where, where the algos should react a lot is, is big changes on, on the balance of the system. Or if you have an NFRR call. You see outdoors going crazy and running into the same direction sometimes, but I don't think it's so much a danger for, for the system or for the market. It's more a danger for the companies using them and, and, you know, making trades at unfortunate prices.

Richard Sverrisson, Editor-in-Chief, Montel: 6:38

So then they have to put in the sort of stop signals there, or the, you know, Stop loss kind of elements. They have to be quite strict on those.

Amani Joas, Managing Director, CF Flex Power: 6:45

Yes, but there's a danger in that as well, because since you have price limits that are too restrictive, you might not close your position. And that's actually the danger that you're talking about in energy market if somebody not handling their positions and having, having large imbalances. So I think when you talk of danger, it's more, it's more managing your position rather than, you know, influencing the markets and letting it run wild in one direction or the other for, for a few minutes.

Richard Sverrisson, Editor-in-Chief, Montel: 7:10

So the managing of that position, that's when we come into, to the humans and the, the, the, the, you know, behind the machines here, isn't it Mani? So they're, they're the ones I'm just going to ask, you know, what is the role for, for creativity here on human intuition that you have on many trading floors?

Amani Joas, Managing Director, CF Flex Power: 7:27

So there, there's a quote that I like a lot by, um, Nassim Nicholas Taleb, who says finance run by technologists is just like food cooked by pharmacists, books written by printers, operas composed by sound engineers, and plain flown by mechanics. So in that sense, I do believe that, that humans still, human creativity still has a very large role to play. And I've heard the story so much over the last five years that, you know, the, the times of traders are over and bots are taking over. But when, uh, when I look, especially at prop trading desks and at us at FlexPower, I've not seen. An algo desk that, or a pure algo desk that can compete with the combinations of humans and algos. And, you know, algos are programmed to solve very specific problems, very specific situations. So they tend to struggle when, um, when something unforeseen happens. And, um, when it comes to position closing and asset trading, they're, they're quite good. But when, when something fails in the market. And, and things fail all the time that human creativity and problem solving is still tends to be very important. And I would personally struggle to have a desk that is only run by algos because I know how things fail and I wouldn't want to run the risk of having a large open position.

Richard Sverrisson, Editor-in-Chief, Montel: 8:50

I mean, certainly in the last couple of years, we've seen some incredible price movements in these markets, uh, for, for several reasons, of course, but mainly, uh, you know, due to, due to the war in Ukraine and, and the, the, the gas crisis or the energy crisis more generally. And how, uh, how can machines cope with that?

Amani Joas, Managing Director, CF Flex Power: 9:09

It's usually that, that comes back to, to the humans running them. So most algos, again, I think when we're talking algos right now, we're talking 95% automated execution. So the human will set certain parameters on which the algos trend and problems that we've seen in the Ukraine crisis. And I, and I've seen this on desks is that you would have price limits that were set too low. And let's say gas prices go up 20 euros. So power price levels go up 40 euros day on day. And you have a price limit that's set to lower and the algo doesn't, um, doesn't close the position then, um, then you're, then you're really running danger because again, they're programmed for it to do very specific things. Another example that, that we saw on April 10th, that was Easter Monday, where we had very low, um, very low demand, extreme, um, extreme sun. So very, very low or negative residual demand. And. On, on markets, we'd never seen really price levels below minus 80 euros because that's where renewables usually curtail. And on this day, prices went down to negative 2000 euros. So most algos would have lower price limits at like minus 80. So then they wouldn't close their positions and you could see some traders panicking and seeing that their positions aren't closed and just, just go wild and selling it at ridiculous prices. So these, these are the situations where, where you might want to have a human in place.

Richard Sverrisson, Editor-in-Chief, Montel: 10:34

Absolutely. Absolutely. But. You know, it's very interesting, this interaction between human and machine. And a few years ago, we were talking about our machines taking over. We're all going to be, as I mentioned, slaves to the algorithm, but it's more the other way around, isn't it? That the algorithms were slaves to, to the humans behind them.

Amani Joas, Managing Director, CF Flex Power: 10:51

I, in the status quo. Yes. So, um, so I mentioned a bunch of times now, um, that, that we're talking automated execution. So. In that sense, algos really much are your little helper tools that you send out to the market to, to do very specific jobs. But when we talk about algos really taking over, we're talking about autonomous decision making and self learning AI. And um, if, if that were happening, like you, you would have heard about it because this wouldn't be a gradual change. This would be a complete game changer. And a metaphor that, that I like on this, because people are looking at, um, very impressive technologies such as chat GPT and chat GPT is a large language model. The trained on a specific and non moving set of data with immense calculating power over a long period of time. Then we, when you ask it questions, it can use this information to answer flexible questions. But when you think of, um, autonomous decision making on markets, I think the better comparison is self driving vehicles. And you know, as well as I do, that this is a problem that has not been solved, even though the biggest tech companies in the world have put immense resources on solving that. And the difficulty there is that you're in a dynamic environment. Now, when you look at the real world, um, when the, the, the rules that we set for humans, like an 18 year old can get a driving license after, you know, six months of training, if they're physically and mentally quite capable. Um, if you think about the sophistication of working at a trading desk, I can guarantee you that getting a job at Slack's power and being a good trader is way more difficult than getting a driving license. And, and think about, you know, you're driving a home and your AI is learning and your self driving vehicle, and you have a curve bending. I know into your driveway, that curve is going to be the same every day. And the machine can learn this. And sometimes you will have people there. So you have this complexity, but now when you think about a market, let's say, you'd say your curve is our, or a quarter hour, 13 to four. And you learn in that quote hour yesterday that there is a certain curve and today or, or on the next day there, there's a certain curve in 13 Q4 as well. But in one day, it might slope down on the next day. There might be a mountain on this curve because things change so constantly. And then the really difficult thing is that the speed at which you enter the curve and the size of your vehicle. We'll bend the shape of the curve like you quite by making a trade, you bending the space around you because you're influencing your environment. So figuring that out for for an autonomous intelligent engine is extremely complex and we just haven't reached that level of sophistication yet. And we also don't have. The expertise in the markets to, to solve these problems at the moment. So do I say that this part will never be reached? No, definitely not. But I think it's quite a while away.

Richard Sverrisson, Editor-in-Chief, Montel: 13:52

Yeah. I mean, as you said, these markets are very, very, you know, dynamic. They change constantly, you know, on some days there can be a perfect storm of factors, um, you know, a whole cocktail of events that, that, that maybe are completely unforeseen, you know, and, and machines that. It must be, it must be very difficult for machines to be able to, to, to understand and react to that.

Amani Joas, Managing Director, CF Flex Power: 14:15

Yeah, definitely. And, um, so, so the big problems that I see on, on the autonomous decision making part, um, one example is just, you know, handling technical failures. So that's often an underestimated issue for autonomous decision making. Because streaming live data and making decisions based on them is complex enough if everything goes well, but very often it does not, you know, like markets have technical issues, um, and break down websites for essential input data crash, scrapers fail, data pipelines break, trade collection malfunctions, and all these processes are annoying for humans, but we're quite good and flexible and finding workarounds and solving these problems, but any of these issues would be catastrophic for an AI That handles the large unsupervised positions as the AI is just not able to pick up the phone and find a workaround.

Richard Sverrisson, Editor-in-Chief, Montel: 15:08

But how, for example, they're quite well established. I said in the intro that they're quite well established in financial markets, a lot of these algos. But how does that compare to what's happening in energy, specifically short term power markets?

Amani Joas, Managing Director, CF Flex Power: 15:22

I would say there are two differences. the level of sophistication. So energy markets, especially short term markets are not the markets where the biggest hedge funds in the world have been investing a lot of money, getting the best technical talent to, to tackle these problems. So the amount of resources that has been thrown off at these problems is just, it's just lower. And the second problem that you have that is more systemic, if you wish, is that short term markets you have. Every day in Germany, at least, or in the continent, 96 quarter hourly products and 24 hourly products. That's 120 products that are traded simultaneously or at least until, until the time they run out. So you're dealing with streaming data and input factors. So decision making is, is very, very quick and the volumes are not very large. Whereas in financial markets, you actually. Might have more time to decide whether you want to buy a certain stock or not. So you can throw more calculating power at a single problem. Whereas on short term markets, you're throwing calculating power at a lot of small problems. And again, you're streaming live data. So the start, this problem just becomes a lot more complex than, than in other financial markets that we see.

Richard Sverrisson, Editor-in-Chief, Montel: 16:37

Could you see a role for these types of robots now goes in, in energy also in forward energy markets?

Amani Joas, Managing Director, CF Flex Power: 16:47

Um, yes. And they, they do exist in Ford energy markets. And I, I think they're, they're more based on, on fundamental models where they have more time calculating what, what they want to do, but on those markets, I'm, I'm less of an expert. So I would be, I would be speculating a little bit. And then in those markets, what you do see a lot, but you also see that in short term markets is where algos have a lot of values in market making and arbitrage. So on that level, there are tech companies that only do this and are very successful.

Richard Sverrisson, Editor-in-Chief, Montel: 17:21

These machines, these algorithms, automated execution, and they're designed. To react very quickly, of course, and they're well intentioned. They're there to make money for the companies who, to program them. Obviously, are there any malicious spots out there is algos that are not so well intentioned, Damani?

Amani Joas, Managing Director, CF Flex Power: 17:41

Well, non programmed by Flextar, I can say.

Richard Sverrisson, Editor-in-Chief, Montel: 17:45

Okay, that's good to hear.

Amani Joas, Managing Director, CF Flex Power: 17:47

Um, so yes, a little bit. I don't, I don't think it's a big deal. So at the beginning, um, of, of the algo trading, we, we saw some bots. That would send out the, the order book in various, like just setting very, very small orders so that you couldn't see the order book properly as a human anymore, which was annoying. I don't even know exactly what their intention was. The people regulating or the boards regulating the exchanges got, got rid of them quite quickly. There's some algorithms that do a bit of spoofing that can also be annoying, but I don't think it's a large issue because at the end of the day, the way I see it as a trader, if you, if you show a price, you're willing to trade at that price. So you always have the risk that somebody lifts your offer. Um, and if you didn't intend at that price, you, you also have a risk, then that's your problem. So like algo is threatening the market in any, in any way. I think, I think it's well under control.

Richard Sverrisson, Editor-in-Chief, Montel: 18:46

Okay. And can you just explain spoofing in this context?

Amani Joas, Managing Director, CF Flex Power: 18:50

So spoofing would be, uh, if you have a spread between the bid and the ask, and you've put in a higher and higher orders on the bids to have other people jump in front of you to push up the price in that sense, so you don't actually have a buying intention, but you try to push the market into a direction.

Richard Sverrisson, Editor-in-Chief, Montel: 19:08

That that suits you, or you hinted towards the regulatory oversight that the exchanges are aware of some of these issues is, is it fit for purpose? What the way that these, this is what the regulatory environments and the supervision and oversight.

Amani Joas, Managing Director, CF Flex Power: 19:22

So if you think about it on a, on a deeper level, like nothing, nothing fundamentally less changed through throughout those, because from a regulatory point of view, you're still making a trading contract. Um, two parties come together or in that sense, you, you make, you make a contract, um, uh, that's, that's funneled through the exchange. Um, And I'm, I'm reluctant, so it's been going well, the process has, has been moving forward. I'm always reluctant to call for regulation because often times more harm is done than good. So I don't see, uh, I don't see a big regulatory challenge there. The big regulatory challenge that I see for governments is more. Um, that they underestimate the value of, of data and, and of life data and, and of transparency in, in helping the market or in helping actors making, making the right decisions. And that's even more true for algos because algos may make even wronger decisions if they get faulty input data. So, uh, that's where I would want to see the regulator move forward.

Richard Sverrisson, Editor-in-Chief, Montel: 20:26

And is, is cyber security a big issue? Obviously you're, you're dealing with very powerful machines, so you have to be completely secure, I would imagine.

Amani Joas, Managing Director, CF Flex Power: 20:35

Yes, that is, that is definitely a big issue. So, uh, if, if an alga operate, any trading operation, but especially if your alga operation Got hacked that that would be catastrophic. So you have to have very secure systems in place. I haven't heard of very big failures there in power markets, but this just might be due to the fact that we're still too, too small to, to be relevant at the moment.

Richard Sverrisson, Editor-in-Chief, Montel: 21:00

You mentioned the availability of data on websites, et cetera, being bots that scrape the data, et cetera. Energy market data is fundamental. It's of key importance for companies like yourselves and the availability of that. What's your view of the current sort of state of data transparency, uh, in, in, in the markets in which you operate the money?

Amani Joas, Managing Director, CF Flex Power: 21:23

From the private sector side, I think it has improved a lot and companies like Montel or, or Volu are doing a great job in, in cleaning data and organizing it better. And. Again, especially for, for AI and, and any sorts of models, my experience when, when I was working back as, as an analyst was that 80% time of the time is usually spent cleaning and organizing data. And then 20% of the time is spent doing actual analysis. So if you have companies organizing that. In, in a better way, that's, that's extremely helpful and gives us a large push when it comes to do to the regulatory side. I think the situation on, especially on the continent, this is quite bad. So if, if you let me take a step back, so for us as traders, prices are just pieces of information, right? And they used to signal abundance or scarcity. And thereby they influence production and consumption decisions. So if we as traders or also algorithms act on information that is, that is incomplete, incorrect, or untimely decisions are made that don't help the efficiency of, of the system as well as they can. So, um, the availability of data is, is very important. And when I look, especially at Germany, I think it's, it's a third world country compared to the UK when it comes to, to data transparency. Where in the UK, a lot of live production data is published and the perfect example of this is, um, the state of the, of the grid balance, which is published in real time in, in the UK and it's published in real time in the Netherlands. And obviously the state are. Influences short term prices immensely, maybe more than, more than anything else in the, in the last half hour before trading. And in Germany, you have a 20 minute delay of this data being published in 20 minutes in short term trading. That's light years. That's, that's a very long time. And you have some actors that, um, provide balancing services. So when they get called on these balancing services. They have this information 20 minutes before the rest of the market, and they act on that. So, when, when I think back to, to my, my compliance courses, I learned that if you have information that is material, private, and, um, influences prices in a predictable way, and, and you act on that, um, that's called insider trading. And, and this is happening in the market at all times, and especially the German TSOs, by not publishing this data in real time. Facilitate this practice. And I think that's very problematic. Yeah, no, absolutely.

Richard Sverrisson, Editor-in-Chief, Montel: 24:02

Thank you for highlighting that. I think, uh, so you would urge them to, to provide that information to all market participants at the same time.

Amani Joas, Managing Director, CF Flex Power: 24:10

Very much so. And, and even beyond that, I would urge the regulator to publish all real time production data by assets, say bigger than 500 kilowatts and production schedules by all power plants and in Germany, again. Look, we, we have this thing that data privacy is a big thing and sharing data is a bit scary also. Due to our past and in many, many ways, but in terms of markets, if that data was published, then we could do so much better job at running the system efficiently. And the fact that the state has often delayed by, you know, a half an hour or, or an hour, it is, is, is really, is really holding us back from pushing the energy transition to, to its next level.

Richard Sverrisson, Editor-in-Chief, Montel: 24:59

Obviously insider trading is what no one wants and I think that that's a serious issue. Um, but the other one is obviously market inefficiency.

Amani Joas, Managing Director, CF Flex Power: 25:07

Yes, I think, I mean, you know, on the insider trading side that some companies are making a bit more money than they should. That's, that's an issue and I think that's unfair. But the bigger issue is exactly what you said, that if the market has timely and correct data. Then it has better information and then we can steer the grid better. We can steer production assets better. We can steer consumption better and, um, we'll, we'll save a lot of, a lot of resources and money. And at the end of the day, often also also emissions, because if you ramp up an inefficient plan for the wrong reasons, then, uh, then this is not good for, for anyone, including the climate.

Richard Sverrisson, Editor-in-Chief, Montel: 25:47

Yeah. Well, absolutely. And I think, uh, there's, let's, let's hope that the TSO see sense and put this at this data out. And certainly, uh, we at Montel will be, uh, joining your call for, for greatest transparency as we have done over, over, over previous decades. So, uh, money, uh, thank you very much for joining the Montel weekly podcast.

Amani Joas, Managing Director, CF Flex Power: 26:05

Thank you so much for having me. It's been a pleasure.