In the previous posts, we introduced the quadratic optimizer with cost constraints for both N-alpha problems and N-stock problems.
We noted that the inputs were not robust to errors, and that the optimizer `maximized these errors’, resulting in unstable weights. Our first post of many on portfolio optimization will take a look at the seminal work of Oliver Ledoit and Michael Wolf on shrinkage of the covariance matrix towards constant correlations.
Moving on, we will look other forms of shrinkage, Bayesian methods and release proprietary benchmark datasets on active alpha returns for readers to experiment with using their own algorithms.
Paper with Code & Code Files: