Let’s Think Outside the Box For Trading Strategy Optimization

I had 3 epiphanies about algorithmic trading…

The first one came yesterday morning: every tutorial on trading strategy optimization is exactly the same. The author introduces the idea of “optimization”, shows some Python code, and ends the article with simulated backtest results. For somebody interested in algorithmic trading, these articles tend to leave a lot to be desired.

The second thought came last night: could trading-rule search benefit from the larger parameter spaces and computational budgets used to train neural networks? A sufficiently expressive neural network represents a function; the training procedure searches its parameters. That distinction matters for the analogy.

The third thought is to use a genetic algorithm to search a model’s parameters. The model represents the function; the GA scores candidate parameter vectors, selects parents and changes the vectors. For a neural network, the fitness objective could be negative prediction loss. Reaching similar predictions to gradient descent is a hypothesis to compare under the same architecture, data and compute budget, not a guarantee from raising the mutation rate. The lottery-ticket paper concerns sparse trainable subnetworks; it does not establish this GA-versus-gradient-descent claim.