Machine learningProject record · Sep 2026

Machine Learning & Model Research

Leader-follower rotation and event ranking

Graph, GNN, and HMM experiments investigate which stocks recover after a market leader’s limit-up streak ends.

GNNStudent-t HMMTemporal encodersPlackett-LuceCalibrationEvent ranking
Why it sits here. Placed by the scope of implementation, the available contribution evidence, and the distinct technical capability it demonstrates.
03 / Somewhere between data and understanding.STUDY IN SPACE

01 / IMPLEMENTATION & CONTRIBUTION

What the work involves

Local research implementations compare a signed temporal graph, a dynamic latent GNN, and a Student-t HMM choice model, with separate data and evaluation contracts.

Technical depth

Within-event ranking, matured-label graph updates, temporal encoders, filtered HMM posteriors, weighted Plackett-Luce choice, out-of-fold calibration, lower-confidence-bound expected value and abstention.

The project family

A_Share/RotationA_Share/Rotation_hmm_v14_bg

02 / RESULTS

What came out of it

Alternative ranking paths use matured labels, filtered states, and explicit feature-availability rules. The GNN path excludes present-day industry classifications because they are not historical point-in-time labels.

03 / SUPPORTING EVIDENCE

Follow the source

Implementation notes, project records, and supporting artifacts.

Source context & project scope

No leaderboard or trading-performance claims. Production wording in local model docs does not establish deployment. Do not combine experiment variants into a single validated algorithm.

Defines an event-level ranking question rather than independent stock forecasts.

SOURCE · 2026-09-17

Filtered Student-t HMM, choice calibration, sizing and PIT data clock modules.

SOURCE · 2026-09-17

Dynamic latent rotation graph and exclusion of non-PIT industry classifications.

SOURCE · 2026-09-17

Non-ML signed graph updates only when future windows mature.

SOURCE · 2026-09-17

Same core question supports grouping as a related variant.

SOURCE · 2026-09-17
CONTINUE IN MACHINE LEARNING & MODEL RESEARCH

CIFAR-10 experiment harness