Spot energy market forecasts: Fundamental, AI and Seasonality

Predicting the future is no longer the sole domain of crystal balls, fortune tellers and astrologists - at least when it comes to the energy sector. Soothsayers of market prices have been around since the dawn of trading itself - whether catering to grain merchants, fishermen or even nascent stock markets - but in recent years, the industry has blossomed like never before. It’s a result of a number of factors: increasingly accessible data; widespread availability of the technology and know-how to handle it; as well as faster-paced, more open, volatile markets.

Here at Montel we have a few services that offer forecasts and predictions. We see them as complementary rather than competitive, as each service offers predictions using different inputs. With Europe continuing to experience near-historic market volatility as autumn turns to winter, we spoke to Montel AI, Energy Quantified and Montel Price it to hear about the services they offer to those who want to look into the future.

Montel AI - Continuously improving, intelligent forecasts

Montel AI’s forecasts, both for individual power plant production and spot market forecasts (see graphs), are generated using machine learning techniques. These models are operated by artificial neural networks that are able to ‘learn’ - both from their own experience and from historical data - meaning they can improve their predictions over time.

The Montel AI laboratory was established four years ago, and now that the infrastructure (the artificial ‘brain’) is operational, setting up new projects is relatively simple. A new model can be “up and running in a matter of weeks,” says data scientist Tobias Foslid. “Where we have existing models, like for wind power, we can set up a new wind park in less than a day.”

The team’s current focus is on expanding their list of ‘generalised services’ for clients from traders to power plants operators and even suppliers. The team hopes to deploy a first intraday price forecast, initially for the Nordic region, late this autumn. It will be a directional model, indicating whether the market is likely to move up or down in the next hour. It’s a complicated problem to solve, admits Foslid, because the intraday spot market is less driven by fundamentals and more by market expectations. “Those aren’t objective, so it’s harder for the model to learn.”

Another project is a model that can predict the load profile for a complete area, helping suppliers who are finding themselves increasingly at risk due to volatile prices. “They want the load for their areas so they predict how much power they need to buy.”

Provide his team with 50 years of good quality data, and Foslid says that their models can perform as well as those developed by an expert with 50 years experience. “And with less work,” he adds. “If you have good data, you can model and predict anything,” he explains, but underlines that it’s not just about feeding more and more data in. “A lot of the work on how to do this properly, as in all machine learning and AI,” he adds, “is understanding the domain.”

Energy Quantified - Getting your analysts to the starting line

In terms of experts in their domain, it’s hard to find a more experienced team than Hugo Birkelund’s Energy Quantified, who have over 50 years of cumulative experience in energy market forecasting. With his colleagues Mikkel Sveen and Jon Moen Drange, Birkelund set out to create a service and a platform designed for, well, people like him.

As such Energy Quantified both is and isn’t a forecasting service. Yes, they provide forecasts for all manner of variables from weather to fundamentals and even river temperatures. But the real value, says Birkelund, is in the quality of the curated data that they offer, which includes historical, actuals and forecasts for every variable a market analyst might need.


It all started when Birkelund was working as an analyst, and realised the amount of time he would spend treating and formatting his data before inputting into models. “The observed weather doesn’t align with the weather forecasts,” on a technical level, he explains. “The map doesn’t match to the actual terrain!”

The market today, Birkelund adds, is very different to when he was “not that much younger, but still younger than now…”. “Most companies have the ability to create forecasts themselves, they have that kind of competence, they have the systems.” And most importantly, they might have people working for them who know their own business better than an external analyst would. EQ’s target clients are companies who have the skills to work with and refine data, but lack time and resources. The aim is to get them to what Birkelund calls the “starting line”, giving them access to data sets that are stored, curated and verified by the EQ team, so that they can fine-tune their own forecasts.

As gas prices across Europe spike, Energy Quantified is in the process of adding gas data to their existing power offer, making them into a true one-stop shop for European energy market data.

Montel Price it - Using the market to forecast the future

Montel Price it’s core products, Price Forward Curves (PFCs) aren’t forecasts either. As Marc Hasenbeck describes it, PFCs are a prognosis tool for companies, helping them to understand what their exposure on the futures market equates to in hourly, daily and weekly forecasts. “Market players have an individual prognosis for the future, and this informs their strategy in the futures markets, which creates market prices,” he says, “What we are trying to do is rediscover what made the market prices themselves.”

The central tenet, inherited from Hasenbeck’s experiences in the stock market, is that all market information is contained within the market prices and their fluctuations. Their own spot forecast model, which runs to 14 days ahead, is built around this simple idea.

“All fundamental information is already included in the market prices,” he says, “so in principle, in a liquid market, you don’t need a fundamental model to forecast prices accurately.” This is why Montel Price it have developed what is known as a time series based model, which they adjust slightly to take into account certain fundamentals. That maybe makes it sound simpler than it is, as Hasenbeck and his team still keep an eye on the model's performance, adjusting it to account for new information.