Welcome to the Launch of TheStockExperiment.com

I am excited to finally launch my project, TheStockExperiment.com. This has been a labor of love that has taken me over a year to reach the point where I can finally release an MVP. "Thankfully," I was laid off twice in the last year, and each time I had time to work on it.

I have always been interested in the stock market — how it works and how certain stocks achieve crazy success. Were there any early indicators that could tip someone off (or some machine off) to future success? When I started pursuing my Master's in Artificial Intelligence at the University of Texas, it felt like a good time to apply what I had learned to dive further into quant finance. I didn't even know the term "quant finance" existed until a few weeks ago.

If you don't know, quant finance is short for quantitative finance — the application of mathematical models, statistical methods, and computational techniques to financial markets and investment decisions. Instead of relying on traditional fundamental analysis or gut instinct, quant finance uses data and algorithms to identify patterns, price assets, manage risk, and execute trades.

There have been thousands of scholarly papers that have investigated this very topic. Sadly, there are more scholarly papers on using machine learning models to predict stock market returns than on curing cancer. I have leveraged several in my studies to find the right setup and models for success.

Over the last year, I have tried dozens of different models. I have tried to find the right set of features and hyperparameters. I have spent thousands of dollars on GPU cycles for tuning and backtesting. I have learned far more by experimenting than in the classroom. More importantly, it has been a lot of fun — and a small distraction as I concern myself with where I will find employment next.

The Beta

I ran a small beta for the last few weeks, testing my models while I refined another. I was mostly pleased with the results. No, I could not outperform the S&P 500 during the beta, but it was close. I had an error in one of my models that over-indexed on near-term quarterly results and ignored a company's longer track record. This is something I tweaked for the launch.

The Launch

I was hoping to go live in February, but refining machine learning models takes a lot of work, and I was only able to launch a two-week beta at the end of February. Since I am using 42-trading-day windows for each rebalance, I drew a line in the sand that March 2nd would be the day I would go live. Who knew it would be an unfortunate news day, with the war in the Middle East escalating. I do recognize that today I am seeing an uptick across all my models compared to the S&P 500, and I credit the upswing in the market for those gains. We will see in a month or two where things sit and whether I need to continue tweaking my models.

While yes, the end goal is to create a model — or ensemble of models — that can outperform the market, the main goal is to always be learning. Thanks for stopping by.