I fine-tuned a foundation model trained on 45GB of personal data and then tested it against my own trades over a 15 year period and against the S&P.
Then I gave it all the tools of a quant fund, The system has the tools of a quant fund, factor models, HRP optimization, deflated Sharpe, probability-of-backtest-overfitting, macro regime overlays, etc..
But the intuitive leap baked by the personal data supersedes the financial data...
Then I put the model on its own vm and gave it API access and its own trading account with a small stake ($10k).
So far, I'm beating the S&P by about 5%. Thought it was interesting and wanted to share with you all. Let me know if you have any questions.
Also, do you have any ideas on what else I can do to make it more effective?
I've fed it all the Edgar market data I could scrape. Do you have any other suggestions?