A $160 billion global quantitative investment manager is supporting research that aims to determine whether cryptocurrency prices are predictable.
Using natural language processing, the Oxford-Man Institute research is attempting to identify tradeable signals in cryptocurrencies based on public communication from places such as online discussion boards and the news media among other sources.
A post-doctoral research appointment in the machine learning division at Oxford-Man Institute has been tasked with the crypto project. It is one of two recent post-doctoral research appointments at the institute - the second is focusing on the applications of Bayesian machine learning to volatility modelling.
Speaking to Financial Standard in London this week, Man AHL chief scientist Anthony Ledford said natural language processing (the interaction between computers and the written word) has become a key research focus for the quantitative manager.
|Sponsored by PIMCO|
Cyclical Outlook: Growing, But Slowing
"We've been exploring numerical data for decades and there's certain numerical data we can bring in to add to what we've done so far, but the direction where I think there's been real progress made within Man AHL is not numerical data at all," he said.
Ledford explains the language processing does far more than word counts and the extracting of sentiment from documents. To research cryptocurrency discussion, for example, the sector's dictionary must be understood.
"People use language in particular styles, so you do need to tune the [research] to that style," he said.
"There's different vocabularies used by journalists and analysts writing about companies, and for companies talking about themselves. That's still a fairly structured type of language. If you stepped away from that and go into the free-for-all that's the discussion boards on the internet, then you've almost got a new vocabulary and different language."
It is early days but Man AHL's natural language processing is also looking at the implications for ESG investing. Ledford said the language processing may identify ESG connectivity between companies that wouldn't ordinarily be considered.
So how does the chief scientist approach Brexit?
"[Generally] machine learning models require big data sets, and if you give me data on 500,000 Brexit-like events then I could tell you something about it," he said.
"All the modelling we do is about looking at large volumes of data about that kind of event. If you don't have the raw materials for this kind of modelling the question becomes 'how do you manage the risk?'"
Ledford says that more generally than Brexit, the manager has a base where it knows there's going to be some particular date in the future where there'll be some perceived level of risk.
However, that risk is often not reflected in the historical prices the manager observes.
"In fact, you quite often see the level of volatility in the run up to these types of events becoming quite subdued.
"This is an instance where looking at the data isn't really sufficient. It's not telling you about this heightened level of risk," he said.
"We've monitored this by looking at implied volatilities which are based on a forward view of risk in the market rather than a backward historical view of volatilities based on prices. We deal with it by monitoring the relative levels of the two sources. When they start to become out of kilter that is an indication there's a market risk not being reflected in the historical volatility and you want to account for it in your current risk scaling. Our system then scales the volatility in the portfolio."
Flagging the challenges of building his machine learning research team in 2017, Ledford said there is now one more spot to fill to make a final senior group of eight. He added that last week saw the arrival of 12 PhD students to the Oxford-Man Institute who will work on machine learning for quantitative finance.