Dynamic Model Selection for Stock Return Prediction Via Fused Penalization

Clustering-Based Sector Investing

“…Bryzgalova et al (2020) suggest to use Asset Pricing restrictions to guide the pruning procedure while using random forest. Goodarzi et al (2022) use fused LASSO to perform dynamic model selection. In our work we apply a classical unsupervised learning algorithm like bisecting K-means to find those groups of firms whose returns comove as tightly as possible, thereby engineering a pseudo-supervised classification technique.…”