I launched 5 different AI-Generated Portfolios into the market. Here’s what happened over the past week.
Read the original story on Medium.
A Quick Recap on Last Week’s Article and Methodology
Last week, we initiated a real-world paper trading experiment to gauge the effectiveness of different trading strategies, including those generated by ChatGPT. The test involved three core approaches: the good old Buy and Hold, a ChatGPT-derived strategy, and optimized versions of both. Our goal was simple: to determine whether AI-generated strategies could be a disruptive force in trading.
Read: ChatGPT generated my algorithmic trading strategy. It beat the market.The experiment was configured to be uncomplicated yet robust. Using technical indicators, we established five portfolios:
- Control: Buy and Hold of SPY, an ETF tracking the S&P 500.
- ChatGPT-Generated Portfolio: Developed based on a strategy crafted by ChatGPT.
- Optimized Portfolio One and Done: A one-time optimization of the ChatGPT strategy.
- Sliding Window and Expanding Window: Two continually-optimized portfolios using different windows of past data.
Our initial hypotheses were formed based on backtests, which showed promising results, particularly for the optimized portfolios. For more details about the experiment, check out the previous article where we launched the experiment.
Caveats to Keep in Mind
As a reminder, the trading strategies we are testing are simplistic, involving one asset and basic technical indicators. This simplicity serves as both a limitation and a potential avenue for further improvement.
Unveiling the Re-Optimization Process
The week reported here used the original optimized rules, so all three optimized portfolios performed identically. The Saturday changes described next apply to the following week, rather than to these reported returns.
The optimized portfolios were re-optimized on Saturday, and as a result, we expect them to perform differently during the following week.
Read: A Real-World Experiment with ChatGPT-Generated StrategiesExpanding Window Optimized Portfolio
The Expanding Window Portfolio was re-optimized to have the same start date but a later end date. The result of this was a portfolio with two new rules that make it drastically different than the One-and-Done Optimized Portfolio.
Sliding Window Optimized Portfolio
In contrast, the Sliding Window Optimized Portfolio incremented the start date of the optimization and the end date. Interestingly, the sliding window portfolio generated similar strategies to the Expanding Window Portfolio
Both of these portfolios are incredibly bullish; a lot more so than the original One and Done Optimized portfolio.
It’ll be interesting to see what happens over the following week!
Performance Metrics: One Week In
SPY Buy and Hold
Our SPY Buy and Hold strategy, the benchmark for the experiment, recorded a loss of 1.05% over the week.
ChatGPT-Designed Portfolio
The ChatGPT-crafted portfolio stayed dormant, making no trades, and thus neither gaining nor losing any value.
Optimized Portfolios
All three optimized portfolios lost 1.21% during the reported week using their original rules. The Expanding and Sliding Window versions were re-optimized afterward on Saturday for the following week.
Aside: Email Verification and Referral Codes
In October 2023, we were working on an email-verification and referral rollout. The rewards described in this aside belong to that announcement, rather than current program terms.
The profile-page invitation below was part of that October 2023 rollout. Check current program terms before treating it as an available reward.
What’s Next? Stay Tuned for More!
The first week was far from conclusive: SPY lost 1.05%, the inactive ChatGPT portfolio stayed unchanged, and the three optimized portfolios lost 1.21%. Saturday’s Expanding and Sliding Window changes were completed for the following week; their new results are still ahead.
Thanks for reading!
Go to NexusTrade📚 Follow our thought pieces on Medium
🤝 Connect with me on LinkedIn
📸 Catch us on Instagram
🎵 Dive into our TikTok
No comments yet.