As the number of energy assets grows through the energy transition, market watchers say more sophisticated trading tools leveraging AI are needed. |
Commodity traders have made big strides toward automation in recent years, but market watchers expect more sophisticated solutions driven by artificial intelligence to be rolled out more slowly than in other industries.
Intraday electricity trading in Europe, for instance, has become increasingly automated in recent years. Since 2021, more than half of the intraday trades on the European Power Exchange's (Epex Spot) have been logged through APIs.
The use of APIs is a strong indicator for automated trading or order routing, Epex Spot said in an email. In 2022, 56% of the volumes traded on Epex Spot's continuous markets came from programmed tools connecting directly through APIs.
Commodities trading house Danske Commodities A/S has invested heavily in algorithmic trading and has seen a shift from manual intraday trading to near-full automation in the past five years, said Tim Kummerfeld, director and head of intraday power trading.
"Today, traders are more like pilots steering the plane," Kummerfeld said in an email. "What our traders do is they shape the view on the market and then they steer the algorithms to execute in that market."
Algorithms are fed with large datasets such as historical price data, weather data and weather forecasts — data that explains fundamental changes in power markets, Kummerfeld said.
AI solutions where human capabilities end
Danske Commodities expects the sophistication of trading tools to grow beyond these current capabilities.
"We expect AI to play an increasingly important role in trading systems. The changes that you need to react on, they happen often so quickly that human traders cannot necessarily catch them in time," Kummerfeld said.
"That could be changes in cloud coverage in Germany that you don't appear to see early enough. But if you utilize AI and a system that captures these kinds of instances from satellite pictures, then you can react on them quickly and trade on them," Kummerfeld said.
What exactly "AI-powered trading" means is sometimes obscured in marketing lingo, said Jürgen Mayerhofer, CEO and co-founder of Austrian digital power trading company Enspired GmbH.
According to Mayerhofer, true AI converts forecasts, alongside a multitude of other data points such as historical price data or physical constraints, into real-time trading actions without any need for human decision-making.
Utilities, transmission system operators and virtual power plant operators in Europe task Enspired with trading their power plants and batteries profitably, Mayerhofer said.
Mayerhofer claimed that profitability of the assets typically improves by at least 30%. "[Customers] say 'my job is buying and building assets, your job is making money out of it,'" the CEO said.
The decentralization of assets, which was brought about by the energy transition, and the growing complexity of the energy system are boosting the business case for platforms such as Enspired that take a wider portfolio view, Mayerhofer said.
Traders in oil and even soft commodities such as coffee are looking to AI for efficiency gains, too.
"Previously you had almost two-dimensional parameters — if 'x' is greater than 'y', then do 'z,'" said Shahjahan Pramanik, director of commodity management and trading at consultancy NTT Data. "Now, you have a 3D view, ... you can have a much more fuzzy logic."
NTT Data works with a host of energy companies and traders across different commodities on the implementation of AI-powered trading solutions such as pre-deal analytic simulation, speedier trade entry and error prevention with tools that spot anomalies, among other services.
The need among traders and utilities for AI solutions is varied, Pramanik said.
"There are clients that have a bit of FOMO [fear of missing out], they keep hearing about AI and they say 'wow, we need to do this.' Then there are other clients who have real business problems," Pramanik said. Such problems could, for instance, be around manual data entry where there is a high volume of trades, increasing the risk for human error.
Secret sauce or slow boil?
S&P Global Commodity Insights contacted a raft of Europe-listed utilities about their activities in AI-powered trading. None agreed to provide details into their operations in this arena.
"It's their secret sauce. There's no standard offering, it's very bespoke. These trading companies, this is what gives them their edge," said Pramanik.
While silence on this issue is partly due to protecting intellectual property, Mayerhofer also said many utilities simply have not made meaningful headway.
Compared to trading in banks or at hedge funds, commodity trading houses are earning their reputation as slow-movers when it comes to the adoption of AI, market observers said.
That is partly because of the physical realities of commodities. "Financial markets don't have efficiency losses, container temperatures, physical constraints," Mayerhofer noted.
Pramanik also sees complacency in company cultures. "They talk about new tools and they end up using their old tools. They are still stuck in their ways."
Talent, data key constraints for AI growth
"To be honest with you, most of the AI work we do is not rocket science," Pramanik said.
Maximilian Kiessler, COO of short-term power trading service Powerbot, sees the power utility sector as being at the beginning of its AI rollout. "We see tremendous gains from very simple approaches, and marginal gain from higher levels of sophistication," Kiessler said.
A key component in this calculation is access to and cost of talent, both machine learning engineers and staff building the infrastructure used by such engineers.
In addition, systems need to be able to cope with growing volumes of data, because industry is increasingly using algorithms and forecasting tools for intraday trading, which requires crunching much larger volumes of data.
"A lot of large players are moving relatively slowly," Kiessler said, meaning existing simple AI models deliver a lot of gains. "As the market as a whole matures, the benefit of these models possibly will diminish. Traders will eventually need to get more sophisticated in order to get an edge."
Pramanik said many trading companies try to develop in-house solutions, but then run into challenges and high investment requirements.
Key for traders in successful leveraging of AI in forecasting is good data governance, Pramanik said. "If they can get that data cleaned then anything you put on top of that is much more powerful."
System-critical infrastructure
Part of some energy companies' hesitance in rolling out AI in different use-case settings comes from concerns over adverse outcomes, for instance, so-called hallucinations by language models, which generate false information and could cause adverse outcomes or poor decision-making.
That is why experts say large-scale deployment of AI in critical grid functions may still be years away.
According to Kummerfeld, caution is needed in energy trading, too. "Energy traders work on critical infrastructure when trading physical power, so it's important that we live up to our obligations that we have with transmission system operators and regulatory bodies," Kummerfeld said.
Energy traders and their trading solutions providers need to continue to grapple with technology and algorithms that are capable of integrating growing renewables fleets, for instance, by forecasting volatility far further in advance, Kummerfeld said.
Enspired's Mayerhofer said fears over adverse consequences of trading automation are overblown.
"If you know what can go wrong, why don't you fix it?" Mayerhofer said.
S&P Global Commodity Insights produces content for distribution on S&P Capital IQ Pro.