Machine Factor Technologies

Storing, cleaning and enriching market data are extremely important in any financial machine learning research. Unfortunately, there are thousands of papers on various modifications of ML algorithms and so few describing the process of cleaning high-frequency tick data. In the latest post, our research team describes the process of cleaning tick data on the example of CBOT Corn futures dataset.
https://machinefactor.tech/cleaning_tick_and_quote_data

Cleaning Tick and Quote Data Financial machine learning done right

On November 20th, Alexandr Proskurin will present at 9th Data Science UA Conference (https://data-science-ua.com/conference/) . Alexandr will speak on how ML researchers can improve the predictions of ensemble models using Sequential Bootstrap. The lecture covers:
1) Details and motivation behind Sequential Bootstrap algorithm. Analyzing the algorithm computation complexity.
2) A toy example of SB algorithm.
3) How to solve financial ML problems using SB algorithm and mlfinlab open-source package.
4) At the end of the lecture, Alexandr will present our latest research result on predicting the performance of SB ensemble vs Random Forest without model fitting.

Feel free to use the promocode: DSUA_Proskurin

9th Data Science UA Conference Discover the latest results, algorithms & trends in the AI world

Last week our Founder and CIO, Александр Проскурин (Alexandr Proskurin) was a guest speaker at Kyiv-Mohyla Business School [kmbs] Master of Banking and Finance (MBF) program. The lecture covered applications of Vine Copulas in portfolio risk-management and algorithmic trading. At the end of the lecture, Alexandr presented the application of multi-dimensional copula in modelling portfolio returns on the example of AMZN, CLX, MSFT and MCD.

This year we've built a portfolio construction algorithm for one of our clients. The requirement was to build a passive investing algorithm meaning that rebalances occur on average once in a week with low drawdowns during market turbulence. Our team decided to use a mix of equity sector and factor ETFs with bonds and gold to build a well-diversified investment universe.
Firstly, the client felt sceptical about applying machine learning in portfolio construction, however, we've managed to present the interpretable side of financial data science so that the client can understand how the algorithm decides on each component weights.
Our research team was stress-testing the algorithm during 2008, 2015, 2018 market turmoils and each time the algorithm was showing lower drawdowns comparing to S&P 500 Index.
Yesterday, we received thank-you email with algorithm performance during market drawdown. Client's account value decreased only by 10% despite the fact that most of the portfolio components are US equity market ETFs.

Strategy backtest overfit detection is one of the services provided by Machine Factor Technologies. The implementation is complex in terms of both modelling and technology aspect and suits best for corporate clients. However, retail traders and small quant shops face the same kind of problem.
Recently, our research team found a great article with a simple and yet effective algorithm detecting strategy overfit. Right now we are thinking of creating simple and intuitive web-tool which detects strategy overfit based on backtest trials provided by a user. We would like to ask the community if there is a demand for such a tool.

What are your thoughts on that? How useful can be this tool in your research?

The variety of packages such as pandas, numpy, Tensorflow, matplotlib made python the most widespread language used to solve data science problems. However, open-source development needs to be supported and encouraged by its end users, we hope that mlfinlab package (https://github.com/hudson-and-thames/mlfinlab) developed by Hudson & Thames Quantitative Research will further increase python adoption in financial machine learning applications.