Replacing Analysts in commodity trading

In trading. And the problem is that it's very difficult to forecast because at the end, it doesn't matter the prediction if you make 80% of the predictions right. But the other 20% make you lose a ton of money, it doesn't matter. So at the end, what would you get that the trader wants? Profits? Yeah. The best risk adjusted profits. So it cannot be too risky, the the strategy. So is it possible to replace the analyst? I think so. Why? Because everything can be automated. So in my field, what we do right now is, like, there are 2 type of traders. There are the speculative ones and the physical the physical ones. The physical ones, they care about having like the asset delivered and they deal with a lot of complexity because they have to set up like a logistic infrastructure in order to receive the asset physically. However, the speculator 1 doesn't care, and he goes on quantitative trading, and sometimes he puts so much volume that it makes to move the market. So the problem is that the market moves in a way sometimes that it doesn't follow fundamentals. By fundamentals, I mean that normally a price of an asset should move by, by how much supply, how much demand. If it's cold weather, in this case, we're trading liquid liquefied natural gas and gas. So, the demand is driving by, for example, high industrial activity or like, as I said before, like, weather, the freight rate, whether it's cheap or expensive to move the liquefied natural gas. So I made a Müller and I'm going to show it to you. It's a convolutional neural networks. And the model, it's a bit like a complex infrastructure because I try everything. And I made all the possible mistakes that you can do. I did maximum overfitting. I did look ahead bias. I did, like, using the prices and not the log returns because the prices can have different stages. And then I decide for going, to a different architecture, which is having 2 towers. The first tower is like taking the the, fundamentals. For example, how much we are using the pipelines because the traders, they want to speculate by using more the pipelines if the price is expensive in order to sell. How much, are the storages in Europe, for example? The problem with this is that, as I said before, sometimes the markets disconnect and they enter in another kind of regime. So we need kind of have an awareness on this kind of status. And this status is determined by, for example, the volatility if the markets are nervous or, for example, the seasonality or, for example, the tweets of Donald Trump, the orange president. So after, building this model and having a prediction, and then I decide to try to put another layer, which is where the title of this presentation comes from, which is trying to replace the analyst. So when you have an outcome, you go to the trainer and you say, hey. I I think it's going to go up, but then there is a lot of opinions going on. So we want to try to have the least amount of emotions in those opinions. So we want to set up everything with numbers. What is the the council of critic sets? So I made, like, something called the triple critics, which is, kind of 3 models together, and they this is, like, represent 3 critics, and they discuss between each other, and they decide whether the signal of the model is correct or not and whether it's worth to invest or not. And then they are retrained on every iteration. So I don't want to go through much, into the details. This is the this is the code. Do you do you see it? Or is it too small? This is the code for the for the this is the code for the for the critics. And here is very, very simple. These are the 3 of them that they do they put a probability, and we divide by 3. And then whether the probability is high or low, we have an optimization with the threshold, and then the threshold is optimized to have the the the better outcome. So I'm I'm going to go a bit crazy now because we don't have I have only 1 minute. So besides that, I went a bit crazy with, OpenClaw. Sorry for I was using codecs at the beginning, and I made an app. Yeah. And I decided to automate how good was the how good was the strategy. And here is a way of automating the strategy. Then I wanted to check for specific fields. And here, you want you want to test a specific price and here you select your, your own, fundamental and then you test if that fundamental is good enough, if it's giving you enough risk adjusted results. Then as well, I did, 1 second. Yeah. I did an anomaly an Müller, check for, for example, the for example, for to check if an anomaly was popping up on real time, paying for for data. And then to to finish, so who is using, cloud code with CLI and voice? How many? Okay. Okay. Because I talked with the audience and I I I saw many of you. So I wanted to show you a trick that Peter Steinberger is doing to to do his work. So the trick is the following. You go, and with this, I finish my presentation. You go and you just, install, Hammerspoon. You just check it out with how to, code and and you have yeah. This is very important. You need to have, like, 4 repositories, and you need to have a hook that every time that you do an interaction with the with the AI, you have first a git pool, then the AI is ready to read the code. Everything is on the main. Right? It's the way he does it. And then you have 4 repos on the same time running parallelly. And then what you do is that every time you have an interaction, you have a git pull. It reads the code, and, of course, it writes. And on top of that with the hook, it commits any push. It doesn't matter. We go crazy. And then it's fine. You can go back. That's the thing. You can go back if something is wrong as it's ruined. It doesn't matter. And in order to go faster, you can do the thing that, so I did the the following. You activate the microphone and you just want to commit something you want to implement, and then you deactivate the microphone. And directly, what I'm saying is prompted on the on the cloud code. And this is going to boost Müller by 5 year productivity 100%. Yeah, that's it. Thank you very much. Thank you very much. Any questions from Jurgen? Are you rich yet? No. No. It's relative, right, in Zurich? 1 last question. Have you benchmarked your strategy against Nancy Pelosi? I have no clue about these, strategies to it. I'm more a fundamental guy. Alright. Thank you very much.