Power market education
Cart 0





AI for the Power Markets

The last bastion to hold out against automated portfolios, the financial power markets present difficulties to algorithmic trading that are unlike any other in the world. We believe that research conducted in the past few years has finally provided tools to automate trading these volatile markets. Years of experience trading manually, as well as profitable fully automated strategies in PJM, SPP, and MISO back these claims. We provide training and education services that allow technically savvy traders to adapt to the changes that have already begun.

Schedule A CALL

Call or email us for a chat to determine if what we offer can help your firm.

We are trying to prove ourselves wrong as quickly as possible, because only in that way can we find progress.
— Richard Feynman

Our Principles


Market Philosophy

We believe that the financial power markets, more than any other, require a thorough understanding of uncertainty in order to be successful. Point forecasts of prices, Gaussian distributed risk, and off-the-shelf machine learning products are worse than useless in understanding the subtleties of the market. We recognize that every model will have assumptions that are violated, and we spend long hours determining the implications; some times the consequences of these violations are negligible, more often they are not. We believe that models must be constructed with an emphasis on the peculiarities of the product they aim to mimic. This core tenet restricts the modeling choices to tools which can account for massive uncertainty, can deal with high-dimensional data, will work well with short-term history, and are amenable to fast optimization so that the trades can reflect the most recent market conditions.

Education Philosophy

Trading desks naturally separate those who have a knack from those who do not. Firms rely heavily on their traders, their risk managers, their quants to produce profitable trades. We hold firmly to the belief that continuing education makes for happier and more successful employees. We recognize too, that education is not a means unto itself, and that there must be some tangible benefit to justify spending the time and money. We strive to adapt our lessons to the needs of the company and of the individual so that we can provide the tools and understanding necessary to allow for the creation of meaningful products once we are gone. For this to be realistic, we recommend prior modelling experience and fluency in a programming language. Under most circumstances we will not spend much time during the lessons explicitly programming so this requirement is to allow for implementation after we have left. Beyond this requirement, we will tailor our instruction to the needs of the company: lectures, informal discussions, presentations, hands-on workshops are all options, the choice is up to you.

A Note About Neural Networks

In the past few years neural networks, in particular so-called "Deep Learning" has garnered much attention. While these algorithms are massively powerful, and we are just scraping the surface of their potential usefulness, we do not believe they are appropriate for modeling power prices. The reasons are mostly technical, but come down to the following rationale:

In general, the more powerful a model and the more flexibility it is given the more data is required to make it useful. This is because there are infinitely many potential instantiations of the model and each data point rules some out as being unreasonable. The more flexibility, the more data is required to hone in on the correct choice. Deep Learning leverages the massive amounts of data available in (for instance) images for facial recognition or voicemails for speech recognition to key in on the correct parameter settings of the model, and the performance is spectacular. In power we are limited to a few thousand days, many of which are irrelevant due to infrastructure changes. This is not enough data to adequately train a deep network. There is plenty of AI and Machine Learning research that works in this regime of "small" data, though it typically does not get as much attention. We believe that the technology developed for small data is more than sufficient to adequately model power for trading purposes.


copyright: David Kozak  // web design:  Squarespace // images: Unsplash


Why we exist

A former colleague once said "People who can't trade, teach". We respectfully disagree. The power markets present an unending dynamic playground in which to test hypotheses and learn the mathematical and computational limits of algorithms. Over time, we recognized that we preferred exploring this landscape to actually trading. Rather than relegate the knowledge to academia, we prefer to teach what we have learned to traders so that it stays practical and relevant.

About Us


I began my career trading virtuals for Endurance Energy in Boulder, Colorado in 2011. I have experience trading manually in MISO (Virtuals, FTRs, ICE) and SPP (Virtuals). In January of 2016 I started working for eXion Energy where I built fully automated trading algorithms for virtuals trades in SPP, MISO, and PJM. 

In 2015 I began pursuing a PhD in Statistics. The focus of my research is in optimization and uncertainty quantification for machine learning. My work at eXion provided substantial overlap to allow me to bring practical insight to my otherwise theoretical background. 

I have a dog, Caius, who keeps me from getting too lost in my research. Together we explore the mountains surrounding our home in Boulder, Colorado.