Automation has been stealing jobs since the industrial revolution and yet we seem to be ever fascinated about the the practical applications of its most potent form (artificial intelligence) to our day jobs.
Those of you who either confident in your abilities to adapt and/or jaded about the idea of work in general may be disappointed to hear that we've still got a while before AI and ML have meaningful roles to play in the commodity sector.
Aviv Handler has quite a diverse background -- a technologist at heart with a deep understanding and interest in regulation -- and this is why he's going to be panelling at ETOT once again this year.
Ahead of this, we got the chance to ask him a few questions on trade surveillance and, of course, "those darn robots" (AI).
What's often the trickiest aspect of trade surveillance?
There’s two. One is getting the data and the second is calibrating the system.
To make trade surveillance work several types of data are required. The key ones are trade data and order data for which you require own data and market data, so; four in total. Other than your own trade data, which is generally available, the other three types of data are difficult to obtain.
"Calibration is a difficult and ongoing process which never ends."
The data is usually held in different formats and in different places. In addition, the systems themselves ingest the data in different formats. While there is standardisation to some degree in gas and power, there is not in other commodities such as oil, metals and agriculture. There are also various commercial services available. In general, the reality of data collection is tough and takes time, despite it possibly appearing easier on paper. On the calibration side, once you have your data flowing in to the surveillance system; there is the question of when an alert should be generated. The process of setting system parameters in linked to the level of risk of unwanted behaviour. It is important to avoid the temptation to calibrate the system to the level of resources they have than to risk. Calibration is a difficult and ongoing process which never ends.
Is it possible to truly track all communications? Does this inhibit the trading process in any way?
Tracking activity tends not to inhibit the trading process. It is not always possible to track everything although technology tracks more and more. The issue with gathering more data is calibration. When you have more data more alerts are generated. The question is again how to tune the system or systems.
There are two types of surveillance system.
1. Trade surveillance
2. Communication surveillance system.
Communication surveillance is used increasingly in energy and commodities as well which is
a very different ball game.
"Although there is interest the application of AI and also machine learning in trade surveillance in energy and commodities is in its infancy."
While the trading process tends not to be inhibited by systems, the increased scrutiny from regulators, and definition of prohibited behaviours does have an impact. This includes behaviours previously considered to be “acceptable”. Dealing with such behaviour will hopefully provide confidence in the market.
How exactly is AI currently implemented in this field?
Although there is interest the application of AI and also machine learning in trade surveillance in energy and commodities is in its infancy. While the market is learning, AI and ML is not something yet practically used much in energy and commodities. It is being used increasingly in banking.
In your own opinion, what is the most difficult process to implement within the AI?
Finding the most suitable technology and model is currently a challenge. In terms of machine learning, training the system can also be difficult.
Keen on AI and trade surveillance in energy trading?
See more at this year's ETOT - view programme here