Hvass Laboratories

My finance research has been made into simulation models that everyone can easily run. Give it a try!

• SimSim.Run

How should you combine different assets into a portfolio when investing? In 1990 the Nobel prize in finance was awarded for a portfolio method invented in the 1950's that is still the standard used today. My work first shows why that method fails, and then I present a new portfolio method that is simple, extremely fast and very robust.

• Simple Portfolio Optimization That Works!

• Fast Portfolio Diversification 

• Portfolio Group Constraints

• Video

• Data and Computer Code

• Python Package

How does the valuation ratio of a stock affect its long-term returns? Is it possible for stocks and the entire stock-market to be over- or under-valued? In 2013 the Nobel prize in finance was partially awarded for empirical studies on this topic, although a formal explanation has never been given until now. My work formally explains how the valuation ratio matters along with growth and dividends in forecasting long-term stock-returns.

• Paper

• Video Talks

• Data and Computer Code

• Python Package

• Spreadsheet for MS Excel

Corporations worldwide spend around US$ 1 TRILLION on share buybacks every year, without knowing whether it is good or bad for shareholder value. My work provides the formulas for calculating how long-term shareholder value is affected by share buybacks in different situations.

• Video Talks

• Treatise / Book

• Introductory Paper

• Spreadsheet for MS Excel

• Short Blog-Posts

• New finance research on long-term investing, portfolio optimization, etc. written as interactive Python Notebooks with video talks.

• Old finance research written as traditional papers.

• I also designed and implemented the SimFin Python API and Tutorials for easily obtaining and processing financial data, and sharing your research with others. I am no longer involved in that project.

My research was on so-called Meta-Optimization where two levels of optimization are being used: The first level is optimizing a given problem as usual, while the second "meta-level" is for tuning the parameters of the optimizer in the first level. This can often lead to greatly improved results, because the meta-optimizer can find counter-intuitive parameters that cause the primary optimizer to perform much better.

• PhD Thesis and research papers

• Implementation in Python and other languages (SwarmOps)

• More recent and advanced methods in Python Notebooks (MetaOps)

• Video Talks

In 2016 there was a revolution in Artificial Intelligence. I wanted to learn the new methods, but the documentation and tutorials were of very low quality. So I made this series of tutorials while I was learning it myself. Hundreds of thousands of people throughout the world have learned Deep Learning and TensorFlow from these tutorials, which set a new standard for how teaching materials should be made.

• Videos

• Python Notebooks

STOP! Please do NOT write technical questions about TensorFlow. I am NOT interested in doing freelance projects. I am NOT interested in making new "e-mail friends". Too many people write me because they selfishly want something from me.

magnus (at) hvass-labs.org