I saw a news article recently about how high frequency trading firms are turning to artificial intelligence (AI) to improve their trading performance and profit. Two large companies were mentioned with each having a slightly different approach.
AI in the front office
This May, Nomura Securities are releasing a new stock trading system, built around analyzing and assessing vast price and trading data resources. The system aims to simulate the insights of experienced (human) traders and make predictions based on historical
market conditions and the correlation to current asset prices.
Nothing particularly new about that, but the difference is they will be using AI tools to enable the system to enhance its price prediction ability as it gains more experience. Just like a mortal but at speeds and data quantities impossible for the human
brain to handle.
Goldman Sachs Asset Management has taken a slightly different approach, yet still harnessing computer-science based, machine learning capabilities. Their new system can, apparently, use AI to study up to a million different analyst reports and identify factors
affecting share prices.
If the system comes across negative news, such as a profit warning, it will automatically revise down investment opportunities – unless, one assumes, there is a shorting opportunity. Naturally it does the opposite with positive findings. Such machine learning
can then feed into high frequency algos and the cycle continues with yet more data.
Developments in this area are being driven by academic research, in particular through the AI department at the University of Tokyo that specialises in financial data mining and artificial market simulation.
Such progress seems to be the way forward for trading, which then becomes an arms race where firms compete to develop ever-faster technology to get ahead of computer-controlled markets.
In post trade, however, it’s all a bit different…
Human intervention in the back office
Where interactions at the front end tend to be mainly computer to computer, the post-trade operations world often includes many manual processes. The ‘blood and guts’ of any back office encompasses: clearing, settlement, payments, static data maintenance,
custody, treasury, trade reporting, compliance, risk and regulation functions to name just a few.
Likewise, the network of participants extends far further than market execution, and includes buy side, sell side, multiple exchanges, CCPs, custodians, administrators, trade repositories, banks, clearers, payment agencies, settlement agencies and regulators.
Post-trade participants handle this complex ecosystem with a vast array of people and disparate technologies, which inevitably leads to problems, some of them severe. Thousands of people around the world are currently carrying out repetitive post-trade tasks
across hundreds of firms. While certainly not denigrating the complexities of the front office technology environment in any way, in post trade there is no common standard protocol like FIX. The number of interfaces across post trade systems is huge and even
basic tasks such as keeping client and instrument static in line is still a daily challenge. The old adage of ‘the back office can lose more than the front can make’ remains true today. Much of this points to where AI in post trade can help.
AI is the future
For pre- and at-trade processes, AI is there to maximise trading/profit opportunities, whereas in post trade, machine-based learning can address the high cost of human capital, eradicate errors and provide signals to improve the whole trade cycle process,
even back to the execution functions.
Analysis of data and behaviour patterns can help point to the root cause of problems and thus enable firms to transform processes that typically rely on humans doing repetitive tasks. As players continue to seek new ways to keep their businesses alive and
healthy, the use of AI stands to become an area of huge competitive importance with firms looking to apply it in practical and pragmatic ways.
Wholesale adoption of the blockchain process is a few years away and, in the interim, addressing cost and accuracy with new machine-based tools has to be high on the agenda of many institutions.
What is clear is that open collaboration between the connected parties and academia is the way ahead to create a more efficient business model.