FX Alumni Offer Ai-Generated Trading Signals
Posted by Colin Lambert. Last updated: June 28, 2022
Coverage of The Full FX Trade of the Year is about to get a little spicier, thanks to the three founders of AiTonomy, who have issued a challenge to the publisher of this newsletter to use their Ai models to take on his AUD/CZK trade. This is how they are going to do it…
Mention automated trading to people in the FX business and they nod, often knowingly, because, as has been well-documented, more and more trading has been systematised. Mention Ai trading, however, and a more sceptic response is likely, especially if one throws in social media as a source of information, because most of us have seen what happened to hedge funds that tried to run a strategy based upon Twitter for example.
In spite of the inevitable scepticism, something they confess to sharing at the outset, three FX industry alumni have formed AiTonomy, a company that seeks to provide a structured and data-driven approach to trading, but with a real twist. “This is not a high frequency model, it generates one, maybe two, trading signals a day,” explains co-founder and chief revenue officer, Ashley Daffin, who has previously held roles at HSBC, Thomson Reuters and NYSE Technologies and acted as an adviser to several fintech firms. “Hence why we thought it would be interesting and fun to see how it coped with your Trade of the Year.”
Daffin’s co-founders are Andrew Brown, former global head of FX at HSBC, BBVA and Santander; and Amos Keshet, who has previously run FICC IT at BBVA, Credit Agricole and Barclays amongst others. Together the three are behind KAiT – Knowledge at AiTonomy – which incorporates a number of models seeking to use a mixture of public knowledge from social and mainstream media, alongside financial market data, to generate trading signals.
“At a high level, we harvest chatter across a large number of public sites, combine it with financial information, and use the Ai to transform this into various sentiment indicators, using Natural Language Processing (NLP),” explains Keshet, who is CTO of the firm. “This generates a vector of sentiment to the next part of the system, which is a proprietary deep learning neural network designed by us, and this predicts movements for the next day.
“Another bespoke algorithm generates a predicted high and low and a closing range and then the signals are provided on the prediction vectors to create the orders,” he adds. “The models are also able to generate high volatility flags, which signifies the possibility of a trigger event, which may generate a ‘flip’ signal or just the need to cut the position.”
AiTonomy has been applying the models to various markets for over nine months, initially it started with Bitcoin, but Daffin says it then moved onto Ether, US Treasury Bonds, the Dollar Index, the S&P, Nasdaq and various individual stocks. “We wanted to see how it worked across different market structures and products,” he explains. “Markets behave differently so we want to understand how easy it was to pick up the sentiment and generate sensible metrics for the day.”
Two methodologies have been selected to take on the AUD/CZK, both use the same signal, however the first operates with an intraday stop-loss (0.3%) and take-profit (0.6%) order. The second trades at the signal entry (any time after 4am GMT) and is closed out at 21.00 GMT regardless. As the model operates intraday there is no carry involved.
The models have been back-tested rigorously, Daffin says the firm wanted to ensure that all trades are timed correctly and that the model can actually trade in the market at the stated level before calculating results, and, to date, they have been good. “The challenge with all great ideas is that at some point you have to prove it works in a real live environment in front of real people, in a manner in which it can be scrutinised,” observes Brown, who is CEO of the firm. “Having come from a point of being somewhat cynical when Amos first approached me with the idea, the results have been unusually good. It’s been a low cost, low capital, and, so far, high Alpha model.”
Although KAiT started in the crypto world with Bitcoin, the best performing assets thus far have been the 10-year Treasury Bond and Dollar Index, even though, until relatively recently, neither was particularly volatile. “This highlights the benefits of a model that generates Alpha incrementally,” says Brown. “If you are able to access the right data sources, there is a tremendous amount of information out there about pretty much everything.”
The first cut of KAiT looks at the next day’s close, however Keshet says in time the plan is to extend the time horizons for the models, however one thing will remain the same – it will not become a trading model generating a large number of signals for a shorter time horizon. Rather, KAiT is seen by its creators as a complement to those taking a medium/long term macro view. “There are intraday opportunities within a cycle that can really improve returns,” observes Brown. “People can have their medium/long term macro position, but also, independently, exploit opportunities for intraday alpha.”
Ahead of the public launch of the “challenge”, AiTonomy have put in place the framework to trade AUD/CZK and the results have been (inevitably, but sadly, for this writer) very positive. The early observation ahead of official launch from Daffin is that AUD/CZK is a “crazy cross” (he’s not wrong of course), however, “The harder it is, the better it is. For us this is a good test because this is not a cross people look at, we don’t have any preconceptions about daily volatility or how people trade it, but the model doesn’t care – it just needs data.”
AiTonomy is currently in early stage talks to roll the product out on the street – “It is important to us we work with the right partner,” says Brown – but the results thus far suggest the firm has found a good niche in the market. While there are thoughts of transiting the model into a live trading environment, the founder say they prefer to keep it as a prediction service, one that can help traders of all types.
Either way, it is early days for the firm, which is putting its product to the test in a very public environment…and in a very unusual currency pair. The greater automation of markets and indeed the more public discourse around events therein, no doubt provides great ammunition for the models, the proof will now be in the pudding, or rather the returns.
For his part, in spite of his early scepticism, Brown is optimistic. “There is no doubt the world is changing, and technology is far more involved in every asset class. This provides a fascinating framework for people to trade just about any asset class. They can have a crypto package, a currency package, a stock package et cetera, or combinations of those and more, because one thing is certain, there is chatter about everything that is going on in the world that can be mined.
“Obviously there are a lot of people looking at big data out there, but we believe we have something a little innovative and different at this point in time,” he adds.