Rise of AI Summit 2020 · Berlin · Applied AI Day | Hybrid Area

How to Build a Stock Market Money Printing Machine with AI

At the Rise of AI Summit 2020, Raul Glavan, a quant trader speaking at his fifth Rise of AI event, explains how he builds and runs AI trading models with his own capital. He quit his job in asset management to trade full time and entered the World Championship of Forex Trading, reporting about 150% net return and a 2.3% maximum drawdown for the year at the time of the talk. He argues that building a profitable AI trading system is possible but very hard: easily accessible public data such as prices and technical indicators is highly competitive and has little predictive value, so an edge requires accurate alternative data, rigorous feature engineering, and avoiding errors like look-ahead bias. He walks through his six-step process, data sources ranging from sentiment analysis to insider trading ratios and unconventional signals like sudden LinkedIn profile changes that hint at upcoming layoffs, and his practice of building a separate model per asset. He closes with audience questions on market efficiency, data infrastructure, time allocation and advice for beginners.

Key takeaways

  • Glavan quit his asset management job to trade his own capital with AI, reporting about 150% net return in forex trading that year.
  • He argues easily accessible public price and technical-indicator data has little predictive value; alternative data like sentiment matters more.
  • He builds a separate model per asset, since one model cannot generalize across assets like stocks, copper and currency pairs.
  • Look-ahead bias, such as normalizing training and test data together, is a common and dangerous error he sees in many trading models.
  • He spends about 80% of his time on data acquisition and feature engineering, only 10% on the model itself.

Detailed notes

From asset management to trading his own capital

Glavan says he quit his job in the asset management industry the previous year to trade his own capital using AI and entered the public World Championship of Forex Trading to prove the approach works. At the time of the talk he reports being in first place for most weeks that year, with about 150% net return and a 2.3% maximum drawdown, verified daily by a third party.

Why a money printing machine is hard to build

He says building profitable trading models is possible, as quant funds, prop shops and hedge funds show, but genuinely difficult. Public data such as stock and asset prices is cheap, widely used and has almost no prognostic value on its own. Useful alternative or proprietary data is hard and expensive to obtain and often unstructured. Markets are highly competitive and efficient, transaction costs erode edges, and countless variables influence prices, making sustained advantage rare.

A practical process and common mistakes

Glavan outlines a six-step process, saying data acquisition and feature engineering take about 80% of his time. He warns against using technical indicators as core features, since they are just derivatives of price and often collinear, which slows models and reduces predictive value. He favors LSTMs for time series and decision trees for their easier feature-importance analysis. He repeatedly flags look-ahead bias as a common, dangerous error, for example normalizing training and test data together, or using economic data published later than the point a model is meant to predict from.

From easy data to hard and unstructured data

He distinguishes 'easy' public data, fundamentals, prices and technical indicators, which most traders already use and which has little value, from 'hard' public data such as sentiment analysis and insider trading. For insider trading he stresses using a person's track record and the ratio of a trade's size to their income, not the raw fact that an insider bought stock. He gives crawling LinkedIn for sudden employee profile changes as a signal of an impending layoff as one example of building an information edge from public but unstructured sources, and lists patent data, credit card transactions, IoT and weather data as further unstructured private data sources.

Model design and results

Glavan builds a separate model for each asset, saying a single model cannot meaningfully forecast an Amazon stock, the copper price and a currency pair alike. He shows feature-importance examples for an S&P model, dominated by corporate insider data and liquidity metrics, and for a copper model, dominated by supply and demand factors, along with recent buy and sell signals on a roughly four-week horizon. He contrasts an implausible back-tested claim of 90% accuracy and a profit factor over 30 with his own reported real-money results of 94% accuracy this year.

Q&A highlights

Asked what happens as more traders use AI models, Glavan says competition will rise, markets will become more efficient, spreads will narrow and profits will shrink, which he calls good for the market overall but harder for model builders. On infrastructure, he says he avoids cloud services and stores all data on his own servers out of concern for protecting his code and data. He estimates his time as 80% feature engineering, 10% modeling and 10% idea generation, says he relies on a network of domain experts such as weather specialists for commodities like orange juice, and advises beginners to learn Python and libraries like TensorFlow and Keras, then replicate and modify simple public models on GitHub.

Quotes

I would say feature engineering is 80%, model is 10%, and the other 10% is something like idea generation.

Raul Glavan

Markets are very efficient, and it's very hard to gain an edge against other market participants.

Raul Glavan

Questions answered in this talk

Is it possible to build a profitable AI trading system?

According to Raul Glavan at Rise of AI Summit 2020, it is possible but very hard, because public price data is highly competitive with little predictive value, so success depends on accurate alternative data, careful feature engineering and avoiding errors like look-ahead bias.

What data sources give an edge in AI-driven trading?

According to Raul Glavan at Rise of AI Summit 2020, sentiment analysis, insider trading data combined with a trader's track record, and unconventional sources such as LinkedIn hiring changes, patents or credit card data can provide more predictive value than easily accessible price and technical-indicator data.

How much time should go into building an AI trading model?

According to Raul Glavan at Rise of AI Summit 2020, about 80% of the effort should go into data acquisition and feature engineering, 10% into building the model itself, and 10% into generating new ideas.

Mentioned

LSTM, decision trees, sentiment analysis, look-ahead bias, feature engineering, insider trading data, World Championship of Forex Trading

Keep exploring

More from Raul Glavan

More from Rise of AI Summit 2020

More on Enterprise AI & Adoption

Transcript

Auto-generated transcript, not edited. · 5,332 words

Show the full transcript

Next, we should learn if we don't want to set up a VC, maybe we can even earn money ourselves by using AI, applying AI. And I'm very very happy that we have a very special guest today who is in the room, I think. Ah, very good. Um Coming soon. Um Raul Glavan Galvan Glavan, sorry. He's a you could say quant trader. So, he earned a lot of money himself by just using different kind of algorithms, statistics, AI in several competitions. I read that he won against 2,500 other traders the Iron Trader 2011 award. Sounds a bit like the Iron Man, but he made a tremendous return on that. He also has a very very well performing WikiFolio in competition.

And he wants to share this experience and help us also maybe getting rich maybe just having a nice outcome of our own portfolio. So, I'm very very glad to learn more about that. Please welcome Raul. Thank you. So, hello. Hi everyone. My name is Raul Glavan. And yeah, I'm very happy to speak today about how to build up money printing machine with AI. And yeah, so we are going today to 20 minutes much of content. I'm a very practical guy. So, it's not too theoretically today. I'm just showing you how to do it. And to to think about why why I am and and why I rise of AI. This is my fifth time at rise of AI.

I'm I'm very happy and I think one of the one of the most uh loyal guests of Rise of AI, but today I'm here as a speaker and I'm very proud of it and I'm very happy to be here today and speak about this topic. Yes, so to be as I said before, I don't want to be too much on the theoretical side because you have a lot of dreamers in AI, you have a lot of strategies and concepts and all this stuff. And you could, should, perhaps, probably when in 10 years, if. So, this is not what I'm doing. I'm trading and I'm trading now. At this moment where I'm doing the speech, I'm trading AI.

My AI is trading for for me and I'm using my own capital and I'm going with my own risk. So, last year I quitted my job in the asset management industry to trade my own capital with AI. So, this is something from a very I would say practical perspective today. And we have also two polls, so please feel free to answer them and participate and I'm looking at the poll, so would be interested if you are using AI for trading or would you use AI for trading and also if if AI is if you have an AI what makes men money for you. So, yeah. So, let's start with the presentation. And like always, we have presentation finance, you have always a disclaimer risk disclosure statement.

So, just just to keep in mind when we are trading, this is always with risk and you have to do it on your own research and your own ideas and this is not an advice. Don't rely on any data, don't rely on anything you hear at this presentation, make your own research, make your own analysis and don't trust me, so. This is just the statement for this. Um, I'm not a financial advisor. So, usually when I make a keynote speech or anything, I'm just talking about what I'm doing and who is Raul Glavan, but we have 20 minutes and 20 minutes is not that much. So, let me Google that for you. Just type my name into it and I have a lot of interviews, I have a lot of articles.

Uh, you can go if you want in in different languages, in different lengths. Uh, 3 months ago I had a home story of 10 pages about what my favorite food is and what I have done as a teenager. So, if you're interested, go there. And, um, as I said before, uh, I quitted my job in the asset management industry. So, uh, I'm trading and something what is interesting for you, I think, um, when I'm when I started trading this year my own capital, um, when I traded my own capital, so then at this moment, uh, I entered the World Championship of, uh, trading. Uh, just to prove it, uh, on a public basis, um, if it's possible to trade with AI and if it's possible to trade successful with AI.

And, well, it worked quite well. So, I think most of the weeks in this year or a lot of weeks in this year I was on the first place in this, uh, World Championship of Forex Trading. So, I'm trading currencies. And, um, at moment I'm something like 150% net return this year and, um, we go later in the, uh, metrics. But, uh, just, um, just to about what I'm doing. And then, the next, uh, topic is, uh, how to build a money printing machine, do it yourself. So, is it possible to build a money printing machine with AI? Yes. Of course, it is possible. We have a lot of quant funds. We have a lot of people doing this and, uh, it is it is something what it what has been done in, um, in in different different endeavors in in different, uh, I would say, um, structures as a prop trading shop, as also as a hedge fund, also in the high frequency trading space.

People are doing this, but they are very private. The most people don't talk about this. They don't talk about this public because it's obviously uh it's very uh lucrative if you can uh do an own money printing machine. So, um in this this problem is very secretive everything. So, but this uh this uh data scientists, anonymous data scientists are very happy they have their own money printing machine. So, it is working. But, now the problem is is it easy? And is it easy is um no. The problem is very very hard to build machines like this. And the why? There are a lot of a lot of uh problems to why it is not working and why it is very very hard to do it.

Um first of all, um it is public data. Public data is very competitive. Everyone is looking on public data. Public data is for example uh prices of stocks, prices of other assets, and so on. This data is um everyone can access them very cheap, very easy. And this data has nearly no prognostic value. There is very hard to get any insights out of this data. If you want some insights out of this data, then you have to use um some uh alternative data. And this is proprietary data. This is private data. And this is very hard to get. There is also public data, a lot of good public data, but it is unstructured. It's hard to access.

You have a lot of put a lot of work into it. And based on this, it's very very um hard to to get to um this data. And this is why it's not that easy. Um the second thing is um if you want accurate data, and when you are a data scientist, you know accurate data is very hard because the most times you should get are just or often crap. And if you have crap data and you put it in your model, then you get all the crap uh as outcome of your model. So, uh you need very accurate data um and not a lot of missing values, doubling data, and and so on and so on.

So, and this is also expensive to buy this kind of data. And yeah, then the next thing is markets are very efficient. Of course, markets have a lot of competition. All around the world, people are in competition in this markets, and this is why markets are very very efficient, and it's very hard to gain an edge against other market participants. And you have also transaction costs. You have a lot of things. They they are very very hard to to beat at the end on a long-term basis. And yeah, the relationship of asset prices have a lot of possible unbelievable much of variables. They are influencing the market prices. And um yeah, this is extremely competitive.

Public data is nearly useless. So, it's not easy, but it's possible. Okay. This is my cookbook. So, we have six steps, but I'm trying to do it. And the first two I think are for me the most important. So, first of all, data acquisition, and also the pre-processing of the data and the feature engineering. I would say on a very practical base, 80% of my time I'm investing in these two steps. So, if you if you are into this field, please do a proper feature engineering, get a lot of good data, and build your own crawlers. And for example, I I've seen so many models in in in in my life because I have more than 1,000 models build it.

And I'm doing this for roughly 20 years. I started in very young age as a teenager. And I've seen so many problems in just feature engineering. Please don't do this. For example, missing values. Find something you how to interpolate them, or don't use it at all if you have too much missing values or doubling data. Um I have seen a lot of people don't measure their features about their importance. So, you have to do an let's say analysis of your features and how important they are or not. And you can do this for example decision trees. Decision trees is very easy to do it and then you get how important these features are. Other thing what I have seen a lot of times is for example collinearity.

You have and this is this just crazy. Don't do this. Don't use technical analysis. Technical analysis doesn't make sense. If you know what it is and this is the problem also. The most people don't have enough domain knowledge. They are just good in the number crunching. They are just good in the model side but they don't have idea about trading, about stock markets, about all the other things. And based on this they are using for example 20 technical indicators. But the problem is a technical indicator is just a derivative of the price. So, you are using 20 times different price series. It doesn't make sense. Please use not these technical analysis indicators. I'm showing you today a lot of cool ideas what you can do practical for your models but don't use them.

And also what I have seen this collinearity in technical indicators but in also a lot of other indicators you can use. Don't do this and it doesn't make sense and it makes your model slow and your predictive value will decrease. So, this is I think one of the most important things. And yeah, I'm I'm using most of time LSTMs. Sometimes decision trees. I like them because they are more basic. You can do this feature important stuff easier. With LSTMs you have more the the option of sensitivity analysis and this is very hard to do it. So, most of the time it's it's easier for you to start with decision trees. But when you are using LSTMs, they are very good for this for time series analysis, especially for financial data.

I like them very much. And also, this is this is also very practical. I have seen it in hundreds of models out there. They have a look-ahead bias. They are using data when training their models. At this time, when they are training their models, they don't have this data. For example, for LSTMs, you are normalization the data. Because you have an interval, and if the if the difference is too high, the deviation, the LSTM is not working. Okay. So, you are doing a normalization with your data. And I've seen so many models. They are normalization They are doing this normalization with your with their training set and their test set at the same time. So, at the moment you you are doing your your training set, you don't have the future data.

You don't know the normalization of the future in 10 years. So, it doesn't make sense, but there are a lot of models you see with this problem. And a lot of other problems. For example, if you have economic data, and they are published on let's say on the weekend on Friday evening, but you're you're calculating your model with data from Thursday, then it doesn't work. Because you are using data from Friday when you are on Thursday, you don't have it in reality. So, look-ahead bias is very very dangerous. And well, I can I can talk about this hours. Oh, no, weeks. But, we have 20 minutes. So, I have a lot more content around the model.

Yes, optimization. And please always do use humans and do do the steps always again. Challenge your code. Challenge your databases. This is very important. And really go and do the visualization work. I have I've worked with so many people, "No, I don't want to do the visual It takes too much time." Well, if you don't do the visualization, then you don't see the obviously problems. For example, outliers and so on. But, okay. We have to go on. So, this is something some people know me from the sentiment space. And they say, "Ah, Raul, I know what what data you mean. You are talking about Twitter data." And Twitter data is the hottest So, and I say, "No, guys, uh I'm using Twitter since 10 years.

Uh I was one of the inventors of Twitter data and um I I've done this speech. This is I've just what I want to show you. This is in proof from London 2012. I made a speech in London in 2012 to explain exactly how to do analysis with sentiment on Twitter data and also on other social media data in 2012. And my models are running in this time in 2011. I won a lot of competitions with sentiment analysis at the time. And no, this is not uh the fancy new This is just um very old and I'm using it. It has some predictive value and this data is very interesting, but it's not new. But, I'm showing you what is new.

So, this is not scientific, okay? This is just I'm I'm a coder and I'm doing thousands of models. So, please um it's just how I see the world, okay? Um this is the easy public data and this is what the most people are looking at. The most people are looking at fundamental data, price data, and technical indicators. So, this is price to book value in the fundamental space. Uh the price data, of course, is the stock price, for example, or the price of copper or the price of anything. And there is also technical indicators, like oscillators, like um simple moving averages, and so on and so on. But, to be honest to you, um this is the easy data.

And the easy data, millions of funds, of people, of traders are looking at. This data has little value. You should use instead this one. It's hard public data. It's just what I call this, okay? And hard public data is for example sentiment analysis. Sentiment analysis has a value because it's not so easy to get the sentiment of the market, of the traders, of other participants, and this is very interesting data. We go later into detail of sentiment analysis. I'm showing you some of my sources. And you see that what I'm doing is a lot of data. I'm I'm very much I think it's it's just I'm I'm crawling data and I'm preprocessing data and I'm assemble data and I'm so I'm very much a data guy at the end.

And service for example, LinkedIn layoffs is just it's just an idea for me, but get you have a definite idea how to get an edge on this market. For example, um you are crawling LinkedIn. And in LinkedIn, uh you are crawling one company and this one company for example is um is um in you can see that 1,000 people employees are changing their profile at the same time. Now you know, wow, there is something happening. Probably and this is what most of the times what happen is there will be a layoff. There will be a job cut. So but this information, you have to crawl in the media. If this information is not public at this time, you have an information advantage because no one knows at this time in the market that for example this company is layoff 1/3 of the employees.

But if you know this information, then you know usually markets are reacting in the same way once when it's a stock, so it's rising. Stock stocks rise when people in this kind of of layoffs are happening. So if you have this information before the market, you can participate. You can make make your trades of it. And this is just one very simple example how you can use for example alternative data and how you build your own crawlers with public data. This is just public data. They are accessible and you can do this on your own. You just need to build the crawlers and the databases. And then you have some very insightful data. Um yeah, there are a lot of flows between asset classes.

The CO2 report, I think it's very useful. Insider trading, of course. But also insider trading don't use it as the most people use it. It doesn't make sense. It's not one insider buys the stock and then it it has an information well. No. Um for example, if you use the insider trading data, you need a track record of this guy who is trading this company. Because if he's a CEO and it was CEO before and previous he was also a CEO. So you have a track record of 20 or 50 transactions of this guy. And if this guy is doing well as an insider in his company, then you have an information advantage. Number one.

Number two, for example, you have to build ratios because the um payments salary of CEOs and other insiders are often are often public. And because when they are public, uh you can make a ratio. For example, if an um CEO earns 10 million a year and he makes an insider trade with, let's say, 100,000 US dollars, this is not significant. But if this guy who earns 10 million invests 10 million in his own company, then it gets significant. So you have to do this number crunching really really on detail. It's not not that easy. Oh, he's buying a stock. No, you have to analyze um has he a positive track record? What is the ratio of of his income and so on and so on and so on.

You have to do a lot of work for every single of these sources. And yeah, social media. But now, this is uh where the real magic comes into. There's also unexplored unstructured private data. And if you have access to this data, it's it's very interesting and you should use it. Um agricultural data, for example. Um ESG data, um, patent analysis, for example, how innovative stocks are. You can compare them, and so on and so on. Um, credit card transactions, product reviews, IoT data, your your whole house will be uh connected. You get a lot of interesting data out of this, construction data, uh what weather patterns. It's it's just crazy because I've seen so many models and it doesn't make sense.

Um, some of uh of my friends are doing this at a fund and and they say, "Well, uh I'm using this model for all assets." And I say, "You really use the same model for forecasting an Amazon stock like for forecasting the copper price?" And you're using the copper price uh model for analyzing um EUR / USD currency pair? It doesn't make sense. So, you have you to do uh one model for one asset. This is what I am doing. I have for every asset I'm trading an own model. I know it's a lot of effort, it's not fun, 100% and but uh it's only way. It's only way it works. Sorry, it's very competitive what we are here doing here.

And now I'm just showing you for one of these points, it's just sentiment analysis. What my data sources are, and just for equity, and just one uh excerpt, just a small sample. So, just for equity, sentiment in the long term, I'm using this kind of data, in the medium term, and in the short term, I'm using this type type of data. So, that you can just As I mentioned, I'm using from the space of sentiment analysis, equity, something like several hundreds of se- several time series data. And all of them are compiled, are built, are I'm testing them since, as I said, nearly 20 years. So, um as I said, yeah, it's possible to build these money printing machines, but I have not said it's easy.

So. And I want to show you for example um feature importance for my uh in this type it's S&P uh model. So, it's also just an excerpt because I have a lot of features, but you can see for for example I'm using a lot of corporate insider data because corporate insider have an prognostic value and um they are very interested uh interesting source of um uh forecasts. And also S&P liquidity premium, mutual fund flows, and so on and so on. These are the top 10. And then I can show you This is a bit borderline because now I'm showing you signals, okay? Don't do this this is not an advice, okay? Never trade uh a net Don't rely on this.

But you see the the green dots are entry points. They are buy points from the last um 60 trading days 60 trading days are something like 90 days in reality, so the last 3 months. And uh you can see that there are two signals in S&P. And yes, of course they are they are perfect in this uh case, yeah. And also for example, this is my uh feature importance of copper. And you can see it's whole different than my features I'm using for S&P. Um for example, you can see here of course copper very very important is the production. Supply demand uh the the all the economic stuff then is more important when you want to trade for example commodities.

And also you can see my uh signals for example for the copper price also the last um 90 days and yes, also hitting very good. And um yesterday, this is why I updated the presentation. Yesterday uh I got a new uh sell signal. It's the red dot, so but don't do it. And um I'm doing it because uh I'm I'm doing a good of money with with my models, but uh please you don't uh don't do it. And um for for you as you can see in this in this signals of signals I would say it's something like 4 weeks. This is the time horizon of every signal I'm doing. Okay. So, you've seen how the signals are how the feature engineering and now you need some results.

I think because now it's just a guy who's talking with but what what are your results? And this is I want to show you also results of my trading. So, what if someone comes to you with a back test and in this back test he says to you I'm 90% of the time over 90% of time I'm correct with my with my trades. And with what what about when this guy says he has a profit factor over 30? One unit of loss has 30 units of winnings. So, what would you say if if someone is telling this to you? And you would usually say and the maximum drawdown is just under three under 5% you would say it's stupid.

It doesn't make sense. And because look ahead bias, markets are efficient, sample size is too small, over fitting and so on. And then let's go into real results. And the real results are even better. So, this year I'm trading with a 94% accuracy. This are real trades with my own money. This is an account in the US and this is the world championship account. So, it's all reviewed every day from a different party on a third party basis and they are scanning all my trades. Yes, sample size is small, of course, and I'm running far above my expected value. I'm running better than expected, but it's not bad for me, but I think I will get a drawdown in the future.

And yeah, so I have a maximum drawdown this year of 2.3%. And net result net return of 150% this year in forex trading on the World Championship. So, yeah. Hopefully, it was interesting for you. You get some insights. Uh, it was a pleasure for me. And, uh, yeah, we are at the beginning of the greatest technological transformation in the history of mankind. I'm very happy to speak today here at Rise of AI. And, um, thank you for your attention. We have to try to keep a distance here on the stage. Thanks so much for this, um, and we have a few questions for the audience, so I would try to rush through, Okay. um, them.

So, Daniel Forster has a very good question, I think, uh, that you can apply generally to models. So, what happens if everybody or more people are using AI models, and especially if the larger funds are doing that as well? I guess they are already doing that. So, They are doing this, and, uh, the competition will rise. So, the the markets are getting more efficient. And, when markets are getting more efficient, it's it's better for everyone. So, uh, bid ask spreads are declining, the liquidity is higher. It's more fair for everyone if if the markets are getting more efficient. For example, when you are going to retirement, and you have a portfolio of your stocks, and you want to sell them, then you are getting the better price because you are more in in in the reality, what what is the intrinsic value of your stock or your commodities or your currencies or what you have.

So, it's a good thing, um, but at the end, it will, um, kill the the profit profits and the margins, and the competition will rise, and it will getting harder to build this models. Yes, it's very hard then, yes. Right, Oliver Rath has a question more around the amount of data, how you store that, uh, how you handle it. Well, um, I'm I'm doing this, uh, completely on my own because, uh, I've seen a lot of things in this financial world, and to be really honest, I'm I'm a bit paranoid, but I'm don't using cloud services. I have my own servers, and I storage everything on my own with my own infrastructure. And, I think this is the only way you have to do it because um, when you are have having something that really works, everyone want to have it.

Want your code, want your data, and uh data is is the most important thing and your best source for your models. So, for me I'm I'm a bit paranoid and I have everything on my own. Following question from myself, how do you spend your day? How much time is it on training the models, looking for new sources? Well, well to be honest, the the most time is on the data side and then there's also a lot of training, but training is paralyzed because I I don't have to sit there and just watching the epochs running. But at the end I'm I'm always running some optimization on my models and I'm always watching if there's something interesting in the data.

But most of the time I would say it's feature engineering is 80%, model is 10% and the other 10% is something like idea generation and I think this is how I spend most of the time, yeah. Maybe about idea generation, Tina Kantsler asked, what about special knowledge? Maybe domain specific knowledge? Do you have experts, people that you talk to? How do you gain that knowledge? Well, it's it's it's a good question at the end. Um you have a lot of of a lot of network you are building in this time and this is very important because when you are with these guys and someone is very strong in weather. And weather is very important for example if you want to trade orange juice because orange juice is mostly built in the US in Florida and then it's very dependent on on this weather.

But I have no idea about weather. So, if you have someone who is an expert in this kind of area and has a very specific domain knowledge about how to forecast weather and so on, then you are going to contact them. And this is what I'm doing a lot of lot of times. I'm talking with the experts in their field about how to get data, what is a reliable data and how to how to use it. And this is something I I have built a network over the last years, yeah. I think you inspired Oliver by the way he has another question. He wants to start he wants to start trading. So what trading basics do people need to know for AI trade well where to start I guess.

Okay. Sorry we have 20 minutes and 10 minutes well um We can post one URL where you can start. Well at at the end I hope you can code in Python. This is very very important for you. Also you need to learn the libraries Tensorflow Keras and all this stuff. I think this is also very important and then if if you if you want to learn go on GitHub and just just or take the first codes you can see and try them. Yeah try them in your own notebook is this working is this not working because the the most models I've seen they are public in in the internet very very simple. They are not complex and you can replicate them and just change the numbers change the data and so on and this is I think how you can start practically.

You have no idea then about neural networks okay but you can start with some models yes. Cool and we post in the chat and I think I also already posted your website so we can read a little bit about you because you share your knowledge. Thank you for that. Thanks for coming here and I hope you will enjoy the conference. Thank you. It was my pleasure.

Rise of AI 2027 · Berlin · May 2027

Be in the room for the next one.

Ten years of talks started as conversations in one curated room in Berlin. The next edition brings 300 selected leaders together in May 2027.