An AI Agent created this trading strategy autonomously with one prompt. It’s up $3,430 in the past year
This technology will UPEND trading forever
When I first told my twin brother about my algorithmic trading agent, he looked at me sheepishly before being brutally honest.
“That sounds too good to be true”
My brother isn’t an AI expert, but he’s a software engineer at Meta. He also knows exactly how I feel about AI agents because I’ve went on tirades against them in the past.
Slow. Expensive. Brittle. What good is a Python script if it costs you $50 to run?
That’s why when I showed him my results, he was shocked. The methodology? Flawless? Lookahead bias? Non-existent. I successfully created a highly profitable algorithmic trading strategy with a single prompt.
And the crazy thing is… you can too without even needing to know how to code.
How does it work?
When I say that I built a fully autonomous AI trading agent, the first thing people imagine is some MIT dork with monocles covering his entire face using a NVIDIA H200 to dominate the market from his mother’s basement.
Press enter or click to view image in full size The average algorithmic trader 6 years ago (courtesy of ChatGPT’s image generator)You don’t have to be him.
To use my AI trading agent, you don’t even have to know how to code. If you understand the principles of financial research and backtesting, and you can understand trading ideas written in plain English, you can use this system to its fullest capacity.
It starts with a single prompt.
Press enter or click to view image in full size In the AI chat interface, we can start with a simple request and agent runner gets ready to workTo use this fully autonomous AI agent, you have to:
- Create an account on NexusTrade
- Navigate to the AI Chat page
- Say something like the following to Aurora
I want to create a portfolio with SPXL that:
* Outperforms Buy and Hold SPY in terms of returns, sortino ratio, and drawdown
* Matches out outperforms SPXL in terms of returns or sharpe ratio or sortino ratio
* Significantly reduces drawdown for SPXL
Press enter or click to view image in full size After the prompt, Aurora will ask follow-up questions to create the perfect strategy for youAurora will then ask a few follow-up questions, such as what your goals are and what your investment horizon is. Once you answer, she enters “agentic mode”.
Press enter or click to view image in full size The initial plan generated by Aurora, which involves research, portfolio creation, backtesting, and iteration/refinementIn agentic mode, we have the most powerful AI agent for algorithmic trading automatically perform financial research, creating algorithmic trading strategies, backtesting those strategies across a wide range of historical conditions, and iterating on her results to create the perfect set of trading rules.
The process is fully autonomous. It’s based on the ReAct (Reasoning + Action) Agent framework, a LLM framework that teaches AI models how to think about their environment and perform actions based on the current state. However, because this is not a technical deep dive, you’ll have to read all the juicy details in my previous article.
No, this article is about results. I’m going to show you a strategy that earned 34% in the past year with a single prompt.
It’s based on an extremely simple indicator – one that we talk about for risk management, but seldom used without our trading logic.
It’s called Maximum Drawdown.
Being Greedy When Others are Fearful
Press enter or click to view image in full size The final performance of the “winning” portfolio called SPXL Scaling into Drawdowns V5The AI started with creating 10 different algorithmic trading strategies. It backtested them, evaluated their performance, and continuously doubled down on promising strategies.
In the first loop, it eventually found one promising strategy called SPXL Scaling into Drawdowns. The strategy was simple – if SPY saw a large drawdown, start entering into SPXL, the leveraged SPY ETF. If SPY had a large dip afterwards (5% or more), sell all of the positions.
Press enter or click to view image in full size The exact trading rules for the initial version of this strategyThe rationale is simple. If SPY has already fallen substantially, it is unlikely for it to continue falling significantly. We can capitalize on the fact that the broader market tends to go up over time, particularly after a pullback. But does this logic actually hold up?
The answer is yes.
This simple strategy outperformed the market in many periods, including in raw returns, maximum drawdown, and risk-adjusted returns.
Press enter or click to view image in full size Early in the process, the AI discovered a potentially powerful trading strategyThe AI Agent doubled down on this idea, creating many iterations and versions. After 30 minutes, it arrived at its final answer.
Press enter or click to view image in full size The final answer generated by the AI agentIn its response, the AI notes that the strategy SPXL Scaling into Drawdowns V5 (Shorter Cooldown)’ consistently outperformed SPY across various market conditions from 2018 to 2022. It achieved significant drawdown reduction, often comparable or better than holding SPY alone.
For a leveraged ETF, that’s incredible.
Wanting to see how this strategy performs on more recent data, I saved it to my NexusTrade account and started my recent backtests. You can see the exact trading rules and copy the portfolio here.
Did AI create the ‘Holy Grail’ of trading?
To test to see if this strategy was truly revolutionary, I decided to test it across 3 different recent periods in the past:
- The past year (08/28/2024 to 08/28/2025)
- The past two years (08/28/2023 to 08/28/2025)
- Year-to-date (01/01/2025 to 08/28/2025)
Here’s how it performed
Performance of this strategy in the past year
Press enter or click to view image in full size The backtest performance of this strategy across the past yearAcross the past year, the strategy did well. It earned a percent gain of 34.30% (versus SPY’s 10.10%) with a higher risk-adjusted returns (1.16 vs 0.52). It also achieved comparable drawdowns (24.38% vs 19.95%), just like the AI suggested. This is awesome, but the graph tells a different story.
The strategy doesn’t trade unless the market falls.
Press enter or click to view image in full size A closer view of the strategy generated by the AITo confirm this, we can look at the past two years, where the market was incredibly bullish for 2024. It confirmed my suspicions.
Performance of this strategy in the past two years
Press enter or click to view image in full size The backtest performance of this strategy across the past yearThe backtest across the past two years paints this picture flawlessly. In this period, SPY outperformed the strategy, earning 49.74% in the past two years, with higher sharpe ratios, and comparable drawdowns. In contrast, the strategy performed the exact same as it did across the past year, not making a single trade in 2024.
This indicates that the entry conditions might be too restrictive, or that we should have more strategies (like holding some amount of our portfolio in SPY) so that we don’t miss out on the market when there is a bullish rally.
On the flipside, the strategy continues to do well when there is a downturn. If the market has a flash crash or a bearish period, the strategy shines, as demonstrated by the performance year-to-date.
Performance of this strategy year-to-date
Press enter or click to view image in full size The backtest performance of this strategy year-to-dateAs I speculated, this strategy does phenomenally when there is a flash crash.
The strategy outperforms SPY and even SPXL (not pictured) while maintaining a higher sharpe and sortino ratio. This suggests that if you have suspicions that the market is overbought right now, this strategy is perfect to capitalize on the downturn.
From these 3 periods, we see that the strategy outperforms in some markets, but not in others. So what does this mean for you?
What do these results mean?
These backtests suggest that this strategy is particularly suited if you have a suspicion that the market is overbought. The rules are capable of capitalizing on drawdowns, particularly short-to-medium term flash crashes.
But the strategy isn’t perfect.
Like the backtests suggest, the strategy underperforms in highly bullish markets, oftentimes failing to make a single trade. This means that the entry rules might be too restrictive, and we might need to loosen them or add additional rules to capitalize on the uptrend in bull markets.
So if the AI didn’t create the ‘Holy Grail’, what does this actually mean for AI agents in algorithmic trading?
What are the broader implications of AI Agents in algorithmic trading?
Using Aurora, I created a market-beating algorithmic trading strategy with a single prompt. This strategy outperformed in many periods, but failed to do well in highly bullish markets.
I don’t know about you, but these results have me salivating.
This is the first public implementation of a “Cursor for Algorithmic Trading”. While you likely can’t COMPLETELY vibe-code yourself a working trading strategy, an experienced trader can easily use these results as a perfect starting point of a trading strategy.
It’s like using Claude Code for developing investment strategies.
The crazy part is – this is the worst it’ll ever be. As models get cheaper, faster, and better at following instructions, the agent will inherently get smarter. I can add successful ideas to Aurora’s knowledge base, and it’ll increase the likelihood of creating better strategies.
All with a single prompt. The implications are insane.
If you’re curious on how to apply AI for creating your own algorithmic trading strategy, check out NexusTrade. You can save up to $1,000 your first year by using the code “AGENT50” at checkout.
Or get left behind, as the world learns that AI isn’t just useful for writing emails and generating spaghetti code. The choice is up to you. Will you make the right one?
