Bayesian Optimization with Gaussian Processes for Trading Parameters
Summary
The document explains Bayesian optimization as a way to search expensive black-box objectives, such as portfolio return or risk-adjusted performance. It describes using a Gaussian process as a probabilistic model: observations update the model’s posterior, while an acquisition rule selects the next parameter setting to evaluate. Expected improvement, probability of improvement, and upper confidence bound are named as common acquisition choices.
The workflow starts with evaluated points, fits or updates the surrogate model, selects another point, and repeats until a stopping condition is met. An example uses a synthetic trigonometric objective and a Matérn-kernel Gaussian process to plot predictions and an uncertainty band; it illustrates modeling rather than demonstrating an optimization result. The discussion says exploration and exploitation are balanced through the acquisition function, but provides no comparative performance evidence or practical trading backtest. It also notes that Gaussian processes make smoothness assumptions, which may limit their suitability for some objectives.
Key ideas
- Bayesian optimization models an expensive objective probabilistically to guide parameter search.
- A Gaussian process represents uncertainty about the objective and updates as evaluations arrive.
- Acquisition functions such as expected improvement, probability of improvement, and upper confidence bound choose candidate points.
- The iterative process balances exploration of uncertain regions with evaluation of promising regions.
- The example visualizes a surrogate fit but does not establish trading performance.
Tags
This summary was written by Stratmill's research agent from the original; it is not a copy of the source.