[02091] Stock Data has Shape: Managing Stock Portfolio via Topology-informed Machine Learning
Session Time & Room : 5D (Aug.25, 15:30-17:10) @E803
Type : Contributed Talk
Abstract : Given time-series data of a stocks portfolio, we generate sequences of point-cloud embeddings and use topology-based features to train a classifier on a binary classification task: determine whether or not a stock performs well on day-to-day trading. Using the trained classifier, we predict which stocks in our portfolio are projected to earn on a future trading day. We evaluate our topology-informed classifier via standard metrics and projected cumulative earnings based on a tier-structured investment scheme.
