Amber Lab

Project AMBER

Can neural networks uncover useful patterns in financial market data when we test them on periods they have never seen?

Amber Collector

Amber Collector is the part of AMBER dedicated to gathering market data. Its purpose is to provide a consistent record that we can inspect and prepare for experiments.

We are developing two editions, one for Windows and one for Ubuntu. The Windows edition is further along in testing. The Ubuntu edition is being prepared for a dedicated collection server.

Read the latest progress update

Research workflow

  1. Collect market observations

    Record incoming observations and retain the timing and source information needed for review.

  2. Check data quality

    Inspect ordering, duplicates and gaps before preparing a dataset for a model.

  3. Evaluate on later periods

    Train and tune on earlier periods, then compare the model with a simple baseline on a held-out later period. Where an experiment simulates trading, document the cost assumptions.

What makes market data usable?