Interview with Lars-Ivar Sellberg about using boundary conditions to monitor algorithmic trading
As the suggestions for how to come to terms with the potentially destabilizing effects of algorithmic trading on the financial markets are many, Lars-Ivar Sellberg, Executive Chairman of Scila AB, in a recent white paper presents a new innovative method for how algorithmic trading could be regulated and monitored using boundary conditions.
Today, various types of algorithmic trading make up a substantial part of the volume at many equity and derivatives exchanges. The question is how to best reduce the risk while not driving away liquidity. Suggested solutions include requiring a minimum order resting time, imposing market maker obligations for algorithmic traders and charging extra for excessive amounts of generated messages (that is, a high order to trade ratio).
The method presented by Mr. Sellberg is a more pragmatic alternative to the suggestion that trading algorithms should be disclosed to the supervising authority before they can be used. Variants of this suggestion have among others been brought forward in MiFID II and by the German government, and successful implementation could theoretically stop rogue or erroneous algorithms from causing any damage. But the question is, is this method realistic? Cinnober met with Mr. Sellberg to learn more about his proposed alternative.
Lars-Ivar Sellberg, Executive Chairman of Scila AB
Q: What weakness does a method requiring all trading algorithms to be disclosed to, and approved by a supervising authority have?
Sellberg: The intention is good, but in practice it would be very expensive and a strain on resources, what with the massive transfer of knowledge that would be required for the supervisor to be able to get involved in and assess each algorithm’s soundness. Not only is there a large number of algorithms to assess, but new algorithms are constantly created and updated as the market changes. This is obviously a time-critical process for those using the algorithms. It is just not realistic to have a supervising authority assessing these algorithms with such short lead-times.
Q: What is the basic idea behind your suggestion of using boundary conditions to monitor trading algorithms instead?
Sellberg: The basic idea is that a set of boundary conditions can be set up - for example, maximum number of messages generated by the algorithm and maximum rate of turnover - and these boundary conditions are then monitored at the marketplace, be it at an exchange, MTF or other type of trading venue.
One way of doing this would be for the supervising authority to set the minimum boundary conditions but allow the marketplaces themselves to add extra boundary conditions on top of these. One reason they might want to do this is because they might want to demonstrate a high level of market integrity in order to attract additional actors who today are critical to the effects of algorithmic trading and rather use dark pools and similar trading venues.
This method is feasible for all parties. It is reasonable to expect algorithm designers to be able to define the boundaries of their algorithms – otherwise, we should all start worrying right now. And technically, monitoring this at the marketplace would be relatively straightforward, too.
Q: What are the main benefits of your method?
Sellberg: Firstly, that it’s doable, it doesn’t require much resources but it offers many of the benefits of the original suggestion that the algorithms should be disclosed in detail.
Another benefit is that algorithmic traders wouldn’t have to expose the inner workings of their algorithms - something that they would understandably not be comfortable doing.
Hopefully, this isn’t something completely new. Algorithmic trading firms probably, and hopefully, already set these types of boundaries for their algos. But there is always a risk of making programming mistakes when developing your own safety measures and one of the benefits with this model is that you get an external second opinion.
Yet another plus is that when monitoring the boundary conditions, the marketplace will overview the activity of all algorithms, and can then detect if they start interacting with each other in a harmful way. This is something that can’t be detected by individual traders who cannot see the whole picture. An example of algorithms interacting in a harmful way is the infamous flash crash of May 6, 2010.
Q: What would it take to get this up and running?
Sellberg: The supervisory authorities need to decide on the conditions that are to be monitored. The marketplaces need to have surveillance technology in place to be able to monitor the algorithms’ boundary conditions. And the algorithmic traders need to map their algorithms with the boundary conditions and highlight the algorithm that generates the orders. That is, the orders need to include a field that states the identity of the algorithm that generated it. It really doesn’t take that much effort and it’s not very complicated.
Q: Who should be responsible for setting and monitoring the boundary conditions?
Sellberg: As I said earlier, the supervisory authority should be responsible for setting minimum boundary conditions. The question is, who should set the final values of the boundary conditions? For example, a type of boundary condition could be ‘maximum turnover’ where the value could be ‘EUR 100,000 per 10 seconds’. I think this could be left to the respective marketplaces, to calibrate the values suitable for each type of traded asset.
Q: Do you think that this method would have made any difference to the Knight Capital software glitch in August last year, in which Knight lost $440 m in half an hour?
Sellberg: Yes, definitely. The erroneous trading by Knight Capital went on for quite a long time, and most probably their internal controls were not functioning properly. With monitoring of boundary conditions, such as maximum turnover and price movements, an event like that would have been detected by the marketplace as the boundaries were breached.
Q: Could this method also be used to help increase market efficiency?
Sellberg: Hopefully, better confidence in market integrity leads to increased market efficiency as potential market participants who otherwise trade elsewhere, such as dark pools, might return and contribute to liquidity and hence improved price discovery.
There is also the aspect of facilitating more efficient resource planning for the marketplaces. Knowing the boundary conditions will help them compare added value versus added cost in terms of hardware and bandwidth of a new participant trading with algorithms.
Q: Can regulators use this type of market surveillance to help them better understand trading in their markets and thus, help them better regulate their markets by adapting rules and laws?
Sellberg: If this model was implemented, the data gathered by the marketplaces when they monitor the algorithms would make post-mortem analysis of flash crashes and similar events much less time-consuming than they are today. This would be a great tool for supervisory bodies for collecting the statistics related to algorithmic trading, and for properly analyzing and studying their actions in the markets. This would end the guessing games and myth-creation associated with today’s algorithmic trading.
Q: From your perspective, what is the most common misconception of algorithmic trading?
Sellberg: That it’s black and white – it’s either or. The truth is somewhere in between - there is a lot of good algorithmic trading and there is some bad, too.
Q: What are the biggest challenges in monitoring algorithmic trading?
Sellberg: Monitoring algo trading would be much easier if the orders showed which algorithm it was generated from – whether the boundary conditions method is used or not. This would of course be private information, only given to the marketplace and the supervising authority. Implementing this would not be a big deal. It’s actually rather remarkable that it’s not already a requirement today.
Q: What other priorities do you see in market surveillance for 2013?
Sellberg: The one thing I see as a priority is leveraging the large amount of information that is gathered in both surveillance and compliance systems. Many market participants and marketplaces have started to realize that this can be used for value-added services.
Written by: Anna Hallgren, Product Marketing Manager, Cinnober