Course book · Second edition · August 2026

Transformers for Market Prediction

A first-principles course connecting neural networks and transformer architecture to the mathematics of forecasting, evaluation and trading research.

Understand the machine before using it

The course starts with a neuron and builds towards modern time-series models. Small worked examples introduce each object before the general formula, with definitions, symbol tables and diagrams intended to make the machinery easier to follow.

From architecture to the research question

The second edition follows three stages: the model, the mathematics it relies on, and the market setting where it is evaluated. References connect early explanations to later derivations, so readers can follow a practical question or study the underlying theory.

What changed in the second edition

The first edition was expanded into a more self-contained course with a clearer prerequisite level, more explicit notation and additional visual explanations. Worked numerical examples are paired with verification scripts in the source material.

Where it fits

This is course material prepared for the trading research programme. It supports the platform’s experiments in forecasting and evaluation; it is not a peer-reviewed research paper or a claim of investment performance.