Algorithm-based trading technology, already established in financial markets, is now making its way into the world of energy trading. But uniting the requirements of trading automation with the special demands of the energy industry is no easy task. In a four-part series Wolfgang Eichberger, Co-Founder of VisoTech, shares the challenges of developing a fully automated solution for intraday power and gas trading. If you’re considering developing your own algo-trading solution, read on so you know what’s in store for you. In part one, Wolfgang gave a brief history of trading technology . Part two covered one of the hardest aspects of developing an automated energy trading solution. Now in part three, find out more about the other challenges.
Algorithms: the auto-trader’s brain
Trading algorithms put the intelligence into an auto-trader, enabling it to analyse huge volumes of data, carry out a specific trading strategy and make highly complicated decisions. Failure to consider a worst-case scenario or to provide adequate limits and protections in a trading algorithm could have drastic consequences, so the bar is extremely high. Position closing, for example, is one of the simpler algorithms. Yet just in the last two years we have changed it about 30 times, due to new technical restrictions introduced by the exchanges or because market behaviour had changed.
Position closing is one of the simpler algorithms. Yet in just the last two years we have changed it about 30 times.
Other algorithms, such as flexibility marketing, are an order of magnitude more complicated; in addition to placing orders, it also needs to account for factors such as price ranges, ramp-up of power plants and more. The algorithm requires matrices that have to be correctly calculated in order to determine at what price what volumes are available to be marketed. To give an idea of the complexity, the last version of flexibility marketing took us eight months of iterative development and testing, in cooperation with customers. A particular challenge was the mechanism that profitably reopened an already closed position in order to generate additional profit.
To test on live exchange data without executing real trades, we had to develop a special toolset.
However, before an algorithm can go live, it has to be extensively tested. No algorithm can simply be developed and taken live, risking the company’s entire cash limit on the exchange. Testing occurs on live exchange data without actually executing real trades. To do this we first had to develop a special toolset – also a significant task that requires ongoing maintenance. Since in this case orders can’t be carried out on the exchange, we had to create our own matching system – in reality our own exchange software.
Performance: the auto-trader’s heart
Naturally performance is a critical factor for an automated trading system, but the performance requirements are not the same as those in financial trading, so they require different solutions. Financial markets have hardware directly connected to the exchange with fiber optics, able to place orders for discrete products in the nanosecond range. Energy trading, on the other hand, offers no direct connection to the exchange. Trading takes place across the internet via API and is relatively slow. This places a rational limit on performance; it is sufficient if an order is placed within milliseconds.
The performance requirements are not the same as those in financial trading, so they require different solutions.
However, energy trading involves numerous products that interact because they cover overlapping delivery times, and the trends is towards an ever-increasing number of products. Meanwhile, trading volumes continue to increase, resulting in growing volumes of messages to be processed. And trading on multiple exchanges or using multiple simultaneous strategies further multiplies the computational complexity. Also, compared to financial markets with fixed hours of operation, energy trading takes place around the clock. All of these factors lead to different, but significant, performance requirements requiring clever solutions.
Next week in part four, the final instalment of this series, Wolfgang reveals one of the secrets of his development success, and why it truly does take a village to make an auto-trader.
Register for our newsletter and receive monthly updates about our company, products and events.