The AI boom is undeniable. It has touched almost every part of our lives, from how we write emails to how we search for information. Yet, leveraging Artificial Intelligence in the financial markets isn’t as clear-cut as it is in other industries.
In this article, we will explore where AI shines for the aspiring profitable trader and, perhaps more importantly, where it falls short.
AI as a Trader: The Hard Truth
Every time a new model drops (GPT-4, Claude, Gemini), the retail crowd asks the same question: "Can it trade?"
Luckily, others have answered this question for us. All of the strongest current AI models were given a real $10,000 account to trade live markets. They’ve since been trading for about three weeks and have returned the following results:
Trading Performance of various AI models in the Alpha Arena competition
Despite having access to the internet’s vast information, not a single public model was able to reliably generate profit. Most were down significantly. Even those that managed to stay neutral or slightly green posted extremely low Sharpe ratios (risk-adjusted returns) and win rates hovering around 30%.
It is safe to conclude that public AI models, in isolation, don't really know what they are doing in the live market. The results suggest these models are essentially engaging in a "random walk," meaning that they are guessing.
Why Can't It Trade?
Even as AI improves, it is unlikely that asking a chatbot to trade your account will ever be a winning strategy. This is primarily for two reasons:
- The models available to the public are always generations behind what institutional speculators use. Hedge funds hire hundreds of PhDs to develop proprietary models. Since trading is a zero-sum game, relying on a public model to beat a proprietary institutional model is an extremely difficult battle to win.
- LLMs (Large Language Models) are great at theory. They know what Risk:Reward is. They can explain Value at Risk. But markets are messy, emotional, and noisy. There is no single "correct" answer in trading. An LLM looks for logic in a place that is often driven by irrationality.
The Pivot: Statistical Edges
We have established that LLMs aren't great traders. However, they are incredible data analysts. It makes sense to lean into their true strength: handling objective, quantifiable tasks.
In trading, we often hear phrases like "this setup is high probability" or "price is likely to reject here." Especially on social media platforms like X (formerly Twitter), opinions are often thrown around as if they were facts.
In the past, manually testing these assumptions took weeks of data entry. Now, with AI, we can cut that time down to minutes. We can move from guessing to knowing.
A Practical Case Study: The Weekly Open
Let’s look at a realistic scenario to see how this works.
Imagine you are scrolling X (Twitter) and see a guru claim: "I'm short EURUSD. Price always returns to the Weekly Open before Friday." Their setup might look something like this:
Instead of taking their word for it, we can use AI to verify the edge. Here is the workflow:
- Download minute-by-minute OHLC (Open, High, Low, Close) data for EUR/USD for the past few years.
- Upload this file to your LLM of choice. Ask it to calculate the probability of price revisiting the Monday 00:00 UTC open price before Friday's close.
- Ask the AI to generate a distribution chart showing how quickly that retest happens.
Link + description to Axiory historical data:
Get MT4 And MT5 Historical Data | EAs Trading | Axiory
When we take all of these steps, we get the following results:
Executive Summary: The Edge
- Odds of Revisiting: 100%.
- In every single week observed (43 out of 43 weeks), the price revisited the Monday 00:00 Open level before the end of the week.
- Timeframe: The retest typically happens very quickly.
- Max Time to Retest: 10 hours.
- Median Time to Retest: 5 minutes.
- Actionable Bias: Since 100% of retests occurred within the first 10 hours of trading (by Monday ~10:00 UTC), if the price trades away from the open on Monday morning, there is a historically definitive statistical bias for it to return to the open level within the same day.
Now we know that price is indeed very likely to retest the Weekly Open. In fact, because we specified a minute-by-minute timeframe, the odds over our sample group were 100%. Of course, part of that is price simply oscillating around the open, and we only tested 43 weeks of data, which isn’t completely statistically significant.
However, you can play around with this quite easily. Just mention to the LLM that you want it to look at the data on a 30-minute timeframe rather than a minute-by-minute one, and give it enough data so it can test this over 200 weeks (4 years).
Why Bother Tracking Statistics?
Financial markets reward specialists, not generalists.
New traders often assume a pattern works everywhere. They might think, "Every market respects the 0.50 Fibonacci level" or "Every asset fills its gaps." This is rarely the case. What works on GBP/USD might be useless on Gold. Each market has unique participants, liquidity constraints, and drivers.
By using AI to collect statistics, we stop relying on influencer logic and start building a Personal Playbook. Your edge in the market compounds with every piece of verified information you acquire.
The Laundry List: What You Should Test
Above all, it is essential that you turn to your own insights. What patterns have you noticed in your markets? Is there a point in time where price tends to move aggressively? Is there a certain type of level you notice that serves as a magnet for price?
Often, we cannot remember these kinds of things on the spot, but while you’re trading, you might suddenly see something. That’s when you need to write this down. When you have time to sit down and test it, you go through the steps outlined above.
However, we want to ensure you get started in the best way possible, so we’ve prepared a few ideas for you to try out.
Market Microstructure
Using AI to figure out the following:
- Average Daily Range (ADR): What is the average difference between the daily High and Low? This helps you understand baseline volatility and how to size your positions.
- Volume Profiles: What time of day does the volume spike? Don't trade during lunch hour if nobody else is.
- Price Insensitive Flows: Which market participants drive prices? Are they price-insensitive? What are their goals? In Oil, airlines are hedging their risks, creating price incentives, and predictable flows. In indices, index funds are legally required to rebalance their exposure to make sure they still maintain the advertised exposure. You can position yourself at month-end and quarter-end before them and wait for the flows to kick in.
- News Reaction: How big are the swings after major announcements? On average, how does price react to FOMC announcements, NFP releases, PCE, crop reports, EIA reports, …?
Specific Setups
- Weekly Open Price: How quickly does the price revisit it? Can we frame a trade around this?
- Friday Mean-Reversion: If Friday was a trend day, prices often mean-revert into the evening as other market participants are reducing their risk into the weekend. Is this applicable to your market?
- Bullish/Bearish Engulfing: What are the odds that the next candle is in line with the direction of the Engulfing?
- Gap Fills: How often do weekly opening gaps get filled, and how long does it usually take?
- Mean Reversion: If the price is 2 standard deviations from the VWAP, what is the probability it returns to the mean within 4 hours?
Bonus: The Quant Prompt
Want to try this yourself? Below is the prompt structure we used to calculate the Weekly Open statistics in this article. It uses a Role-Context-Task format to ensure you get the best possible results.
Role: You are a quantitative developer at a leading proprietary trading firm tasked with assisting me, a trader at the firm, to improve my edge.
Context: I trade EUR/USD, and I want to use statistical analysis to improve my edge and create a directional bias. I’m particularly focused on the concept of a Weekly Open retest, and how quickly price tends to revisit the level at which prices open on Monday 00:00 UTC. In this message, I am attaching minute-by-minute data formatted in an OHLCV format.
Task: Calculate the odds of price revisiting the 00:00 UTC Weekly Open on Monday by the end of the week (Friday’s close). Additionally, make a distribution chart that shows how quickly all of the Weekly Opens in this timeframe get tested.
The Bottom Line
It’s clear that AI will influence trading, but the answer isn’t as simple as handing full control to a bot.
A more effective approach is to use AI as a support tool, helping you gather statistics on the markets you follow so you can better understand their microstructure and identify consistent patterns. Let AI handle the heavy analytical work, while you focus on the precision and judgment required in execution.