Financial Machine Learning

It is extraordinarily difficult to succeed in financial machine learning. That’s why we started, a financial machine learning SaaS for professional traders and asset managers, whether they are programmers or not.

We highlight some of the difficulties of financial ML below, and how we can help you overcome them.

Financial data scrubbing

This may seem like a task for plumbers, but is actually the crucial first step in successful financial machine learning, and it takes surprisingly deep domain knowledge to accomplish. For example, do you know that many fundamental stocks databases have look-ahead bias? How would you find out if your news sentiment data is sensible? With years of financial and data science experience behind us, we can help you find out.

See Part 1 of Dr. Chan and Dr. Hunter’s Lifecycle of Trading Strategy Development course to learn about detecting and fixing myriad problems of financial data. has a team of financial data scientists that can do that for you.

Interpretable Models

Machine learning models don’t have to be black boxes. We can show you clusters of the most important input variables (“features”) for your predictions, and automatically include only those in a model.

See our blog post “The Amazing Efficacy of Cluster-based Feature Selection“.

Don’t have enough good features? Our team of quants can help identify and create them. For more on feature engineering, see Dr. Chan’s talk.

Predicting “non-reflexive”* targets

If returns can be predicted, returns will change in response to the prediction. On the other hand, if weather can be predicted, weather will not change in response. Yet accurate weather prediction can benefit agricultural futures traders. can help identify and predict non-reflexive targets (e.g. earnings, same-store sales, etc.) that cannot be arbitraged away. E.g. We already successfully predicted the Non-Farm Payroll Surprises using alternative data.

* Reflexivity is a term used by George Soros to describe the effects of arbitrage activities on the financial markets and the general economy.

Capital allocation and risk management via “meta-labeling”*

Determine Probability of Profit for a trade through machine learning models, and allocate capital and manage risks accordingly. These are part of “quantamental strategies“: applying machine learning to help discretionary or fundamental investment managers quantify and systematize their ideas, factors, and knowledge, done without the necessity of disclosing the fundamental strategy to the consultant.

See Dr. Chan’s Quant World Canada talk, Toronto, November, 2018.

Predicting the Probability of Profit is exactly what SaaS does!

* Meta-labeling is a term coined by Dr. López de Prado in his book “Advances in Financial Machine Learning”.

Market simulations for performance evaluation

Using neural networks to capture essential market patterns, and generate realistic simulations for trading strategies evaluation and risk assessment.

See Dr. Chan’s QuantCon keynote speech, New York, 2018.

In Summary

  • None of above applications are “black-box” trading strategies: they improve instead of supplant existing trading strategies.
  • No need to disclose proprietary strategy specifications.
  • Upload your past trading history plus some external data to, click a button, and find out the Probability of Profit for your next trade!
  • Dr. Chan has conducted machine learning research since 1994 starting at IBM T.J. Watson’s Research Center, and financial machine learning research since 1997 starting at Morgan Stanley’s Data Mining and AI group. Dr. Chan also manages QTS Capital Management, LLC. a hedge fund that uses financial machine learning technology to great effect.
  • His team of highly skilled researchers and financial data scientists not only produces extraordinary returns to investors at his hedge fund, but also helps external clients achieve the same and without conflicts of interest due to the nature of meta-labeling.

Our Research Articles and Presentations


Explains in details how you can use our SaaS to improve your existing investment or trading strategy.

Go To Article


Blog post on on our News Sentiment research.

Go To Article


Quantopian webinar, October, 2018. (Video.)

Go To Article


At Friedberg Mercantile Group event in Toronto, May 2018.

Go To Article


At QuantCon, New York, April 2017.

Go To Article


York University Schulich School of Business Professional Seminar Series. January, 2013.

Go To Article


At Society of Technical Analysts, London, Oct 2014.

Go To Article


At Market Microstructure, Liquidity and Automated Trading conference, London, July 2015.

Go To Article


A talk about Machine Learning applied to finance and investment.

Go To Article


Webinar at, October 2019.

Go To Article


An open source kernel in Kaggle Two-Sigma competition.

Go To Article


At QuantCon, New York, March 2015.

Go To Article


Keynote speech at QuantCon, New York, April 2018.

Go To Article


Preprint by Xin Man and Ernest Chan comparing MDA, LIME, and SHAP feature selection methods in machine learning. Accepted for publication at The Journal of Financial Data Science, Vol 3, Issue 1, 2021.

Go To Article


Working paper by Radu Ciobanu and Ernest Chan on using alternative data to predict NFP numbers.

Go To Article


UNICOM conference keynote speech, London, June 2019.

Go To Article


At Grupo L&S conference, Brazil, Nov-2017.

Go To Article


Market Technician Association meeting, London, Apr 2013.

Go To Article


At QuantCon, New York, April 2016.

Go To Article


At Quant Invest Canada conference in Toronto, Canada, October 2012.

Go To Article