Hey everyone, just wanted to say thank you.

I never thought my first ever post – anywhere other my blog – would get this much deserved criticism and support. I just wanted to let you guys know that I’m committed and dedicated to making this project as great as I can.

This past week, I’ve made around 80 commits. Some of the key updates:

  • Added Sharpe and Sortino Ratios to analysis
  • Fixed the baseline comparison for proper benchmarking
  • Improved documentation and added helpful GitHub prompts
  • Started publishing logs regularly
  • Began building a version for non-coders (early, but in progress)
  • Tons of smaller fixes, refactors, and cleanups

Overall Ratios:
Sharpe: ~0.8803
Sortino: ~1.8735

New Substack post: https://nathanbsmith729.substack.com/ (link actually works, I promise)
GitHub: https://github.com/LuckyOne7777/ChatGPT-Micro-Cap-Experiment

Yes, luck has obviously played a role in returns. I only have a sample size of one currently, but I really am looking to scale for better overall accuracy. I start school soon (I'm a junior in high school), but I'll do my best to stay consistent with updates and continue growing and learning. I started coding in Python around four months ago, so I’m far from perfect, please reach out if you have any advice. And sorry to the people I haven’t responded to yet; I’ve been swamped, but I check Gmail way more than anything else.

If you’ve got any suggestions for places this project might fit (forums, newsletters, communities, wherever), I’d seriously appreciate it. I just want to share what I’m building and learn from people way smarter than me.

I'll wrap it up here, seriously, thank you to everyone who even cared enough to comment.

Love all you guys, and I’m only getting started.

– Nate

Posted by OpenArcher7341

4 comments

  1. S&P 500 data: Downloaded using Yahoo Finance API for the ticker “^SPX” and “Close” price, adjusted to a $100 baseline.

    ChatGPT data: portfolio data was gathered from ChatGPT, and calculations were made using Pandas and Numpy in Trading_Script.py. Data was exported to CSV file (chatgpt_portfolio_update.csv) and plotted using MatPlotLib (Generate_Graph.py)

  2. If chatgpt was capable of outperforming the market everyone would be using it and the gains would asymptotically approach zero.

    It’s called the efficient market hypothesis.

    Compare the performance against the s&p 500 across a longer time horizon, even then it won’t be evidence of anything. You’ll outperform 50% of the time and underperform it the other 50%.

    What would be more telling is testing it against different share markets and combining results.

    Fundamentally though chatgpt is only really good at finding speculative trends in online news stories and influencers from the first 10 ish relevant websites, so it’s not going to be amazing. Investment banks target and scrape tens of thousands of news articles and influencers to build their speculative fomo focused models, chatgpt isn’t going to outperform that, it might be very useful as a plugin to better interpret language for these models, but on its own it isn’t designed to be predictive of stock markets only context around what it’s been trained on.

  3. Luck.

    If there were easy gains to be made trading stock using chat gpt, lots of people would do it and the gains would zero out.

  4. Just wanted to say that you’re doing incredible things for a high school junior. Whether this is ultimately successful or not, you’ve got a bright future ahead of you.

Comments are closed.