Sramana Mitra: In terms of the heuristics you use to make the correlations, what is involved in that?
John Plavan: We use machine learning, generic algorithms, network techniques, and basic statistical techniques. It depends on the type of data and how we have quantified it. We use a rotated empirical orthogonal function to define an actual weather pattern, which is a result of individual and variable weather inputs such as sea level pressure or jet stream winds or pressure. There is no set algorithm or function. It depends on the types of patterns, types of variables, and also the source data set we are working with. We are constantly trying to improve what we do. We are currently working on another revision of a neural network model that seems to have good promise. Our current product is built on a basic statistical relationship interface.
SM: Are there learning components in your model?
JP: There are. Every day that we bring in a new observational data set, we use it to update the training set. We then look for results in the verifications to determine at exactly what level we should waive certain variables to go into it. The product is learning over time as we continue.
SM: Please talk about how you price the product for energy traders.
JP: That is a good question. It is an annual subscription that we sell to clients. If you think about the application we have for energy traders, this could be incredibly valuable information to the degree that the rest of the market does not have it. In the early stages of our company, it is a very high-priced product for very valuable data. As the markets start to adopt our technologies faster and faster, the uniqueness of our information relative to their competitors diminishes, because an increasing number of their competitors have the same information. Then the pricing for the actual probability of an event will decrease.
We also supply a lot of analytics tools for meteorologists to use themselves. Our software outputs a specific probability for an event to occur. But that is built on the relationship of a variety of weather patterns. It is very hard for meteorologists to quantify those weather patterns in the manner we have done. So, we also sell a subscription of those analytics tools for trained meteorologists to use to come up with their own view of what the future of weather will hold in conjunction with what ours is. To summarize: Right now we have a fairly expensive annual subscription to the forecast model output data and a fairly low-priced subscription to the analytics tools that meteorologists can use to generate their own forecasts. That will shift over time as we get more model subscribers and the product becomes increasingly ubiquitous in the energy trading marketplace.