There was an interesting development from the Royal Bank of Canada last week when it announced that it had launched a new AI-based trading platform for its equities business, and I, for one, am very interested to see how – if at all – the trading style offers something different.
I do believe that when it comes to certain aspects of markets, the world has become enraptured with short term performance – investment professionals no longer have the luxury of one bad quarter, they are judged monthly, if not weekly. This is unsurprising given how the speed of markets has changed, I suppose, but does it lead to better decision making over longer term? Those of us with pension plans are repeatedly told to look at the performance over a five year (or more) time horizon, something that tells me there is significant beta in the returns we pay handsome fees for, but when it comes to picking markets the time horizon seems persistently under pressure to the downside.
My interest in the AI product is whether this further diminishes average hold times (on positions not trades!) Human traders take positions often with the intention of holding them for some time but can easily be knocked off kilter by news or other inputs. That is the nature of markets, but the really great traders know when to wear the short-term pain, even take it as an opportunity to add to a position they really believe in. Of course, these traders also cut positions when they realise they are wrong (all the great traders also know how to take a loss), but it is whether that unwillingness to react to every nuance in markets is in the purview of the RBS strategy that interest me most.
The inherent challenge in AI-based trading remains its reactive nature – generally speaking it is reacting to data, therefore it is hard to be proactive in nature. I am not sure that an AI-based strategy at the current stage of development can choose to ignore data? The RBC initiative is, according to the inevitably laudatory nature of a press release, an advance on the use of AI in developing trading strategies, and it is important to note that it has been back-tested thoroughly, including the current COVID-19 period, which is highlighted by the bank
That does not answer all the questions, however, because in many ways the early weeks of the COVID markets were a trader’s dream – there were lots of reversals but they generally speaking lasted for some time. This means there were lots of mini-trends to jump on and make money, but in the shorter term (minutes) markets were not that choppy. Reversals were sharp, of that there is no doubt, but overall most moves gave anyone reasonably quick to react, time to get on the move.
So the big test of this strategy is still to come, when we get choppy, reversal-prone markets with little or no direction. So far this century similar market conditions have caused no end of pain for trend followers, so the question is, will they, upon return, do the same for AI? Or will the strategy evolve into yet another short-term programme.
I keep coming back to the experiment run by FinLabs some 15 years ago using the Harvard Supercomputer, which ran all sorts of strategies through and kept coming back to the same answer – exploiting latency in the system (or being quicker than anyone else) is the surest way to make money. AI has moved on since those days, but has it moved on far enough to come up with a different strategy? It will be a fascinating experience to find out, because great traders often shine in choppy conditions, often by just surviving them through a mixture of inaction and patience. The big question will be, can an algo?