Asia-based startups are beginning to make progress serving artificial-intelligence solutions to global asset managers.
To date the most activity has been in the U.S. or Europe, where global fund houses are based and where they have the most business. Kensho Technologies, from Cambridge, Massachusetts, is one leader in applying A.I. and machine learning to finance; it was acquired last year by S&P Global for $550 million, the largest deal for an A.I. startup to date.
Can an Asian startup generate that kind of enthusiasm? Here we look at three companies with their hats in the ring.
All three cited challenges in getting a mandate. Fund managers, already under commercial pressure, have limited budgets. These tend to be defined along old-school tech needs, i.e. trading connectivity or cyber-security. They don’t have space for newfangled things like A.I. startups.
Some appear at the cusp of breakthroughs, others aren’t there yet. Some have received generous funding, while others are still bootstrapping off founders’ money. But all of them are finding opportunities that are specific to Asia, a trend that is likely to accelerate.
Blue Fire AI
Anecdotally, the company that is generating the most unprompted buzz is Blue Fire AI, which DigFin profiled two years ago. In that period, while being funded by partners Luke Waddington and Samir Rath, the company had to find a path to commercialization.
Waddington, co-CEO and an ex banker, says they learned two things. One was that asset managers weren’t interested in claims of delivering alpha – delivering alpha is their job. What they wanted was a way to avoid nasty surprises.
The second deliverable was a way of piecing information together into a story for the purposes of portfolio construction.
“We built the equivalent of a FitBit for a listed company,” Waddington said. Instead of seeing a doctor annually (or filing a financial statement) the FitBit monitors heart, lungs and sleep continuously (or balance sheet, cash flows, shares, bonds). “It’s how you can see when the heart rate or the sleeping patterns aren’t right,” he said.
The output needs to be monitored by senior investment professionals, though, to ensure against false flags.
The company has won plaudits from asset-management industry groups in London and Singapore, and it’s now working with those member firms to develop solutions, which should be deployed within weeks. London managers are interested in applying risk alerts to Chinese companies (an area Blue Fire has invested in) while Singaporean managers want warnings of unknown risks.
The company is also opening a Toronto office where it is hiring data scientists and coders.
This startup is also storytelling, helping clients visualize things happening in the market. Based in Hong Kong by way of Beijing, MioTech is working for hedge funds, the research units of investment banks, family offices, and the investment advisors at private banks.
In 2017, the company received $7 million in financing from Hong Kong tycoon Li Ka-shing.
“We tell a visual story, not about the portfolio but about the investment thesis,” said Jason Tu, co-founder and CEO. For example, how does a relationship manager pitch an investment opportunity to a knowledgeable client who also has lots of information?
MioTech does this through topic modeling, a branch of A.I. that analyzes what companies are trending and what relevant people are talking about.
For example, one bank’s sell-side research team wants to know how the U.S.-China trade war will impact supply chains of, say, consumer electronics companies inside China.
Or a private bank’s R.M. wants to know how a Federal Reserve rate change will affect a client’s portfolio.
Tu says MioTech’s software can answer those questions a lot more quickly and accurately than a human analyst. And from there it can use data to build new investment ideas.
So far, early adopters of MioTech services are hedge funds and investment banks that can make quick business decisions. Private banks or those with clients are slower adopters. But for all of them, the challenge has become that the wide availability of information means clients may well know more about an investment idea than the firm.
This startup had early PoCs with the likes of Mirae Asset and the buy-side arm of Chinese broker Shenwan Hongyuan. But production has been put back.
In the meantime, Alibaba has made a seed investment into the company, and it is in a program sponsored by Alibaba and SenseTime that gives Squared-S access to some of their resources.
The founders, Soujit Ghosh and Seth Huang, are using the time to do more research into how to apply their A.I. models. The initial concept was portfolio optimization. However, they are currently working with two clients that want their software for something else.
In one case, a bank is using it to help predict very short-term movements in foreign exchange. In the other, it’s working with a hedge fund to identify moves in distressed debt. The startup seems to have found an immediate demand in these short-term market predictions but it is still working on a more long-term investible solutions.
Part of that process is now running proprietary money. “We’re not interested in becoming a hedge fund or getting a license,” Ghosh said. It’s more about proving the technology instead of relying on the usual backtesting. “A.I. is about doing things better, whether that’s improving my Sharpe ratio or reducing my portfolio’s volatility.”
People at banks and asset management firms don’t really care about the underlying technology. They just want something they trust will work.
Blue Fire relies on natural language processing (see here for our Glossary explainer), combining capital-market expertise with machine learning techniques in a quest of getting the machine to put text into context. “We’ve invested technology that can identify what matters” in a document, Waddington said.
From there, that understanding has to be related to a target (a company, a stock). “This is where the magic happens and you build relationships, you measure the strength or the distance of those relationships, and you measure when those relationships stop,” Waddington said. “Then you need more information to continue. It’s just like how your mind works.”
But today most of this exists only in theory.
MioTech is also trying to create these relationships. It uses knowledge graphs, a version of NLP, to create an ambitious, broader approach.
Knowledge graphs connect and expand data sets, and find sub-graphs or similar patterns within a graph to map that to a goal (what the client wants) and a larger training set.
Self-learning relies on better training data, which companies like Tencent and Netflix achieve through topic modeling. Lots of data from many sources is built around an entity, which could be a listed company, or a market event. So if two companies are in the news together (a lawsuit, an acquisition, a contract) then that forms a topic. A knowledge graph is a never-ending process of building relationships in data among these two entities.
MioTech has been working on knowledge graphs for listed companies and economies in the region since 2016. “We think it’s the biggest graph in Asia, or looking at Asian economies,” Tu said during a presentation earlier this year.
Because two companies have many other relationships (suppliers, competitors, former executives), then entity and event vectors for other connected players. And so the machine training keeps expanding.
Finally on top is a convolutional neural network, a class of deep neural networks used to visualize data. The goal is to create insights that are more meaningful than sell-side analyst reports, which use a lot of hedged language and don’t help investors make the binary, buy- or sell decisions they need to make. (Although it still takes a human to sell what might be a politically unpopular decision.)
Although businesspeople may not care much about the tech, startup execs are frustrated by how it is perceived.
Many in finance don’t understand how it works.
“There’s no backtesting in A.I.,” Tu said. “There’s only more training with new data. If the data isn’t available, we can’t produce what clients want. But ‘backtesting’ is what finance people keep asking for – it’s not what A.I. people ask for.”
And even if people do understand it, the internal makeup of a financial institution is difficult to work with.
“Corporate digital teams are not aligned with management and sales,” said Seth Huang, CTO at Squared-S. There is often suspicion that A.I. will take away jobs. “But I’m not trying to get rid of analysts,” he said. “Banks are trying to get rid of analysts.”
For machine learning to work in finance, it can’t just be about computer scientists. It requires experts in economics, capital markets and other fields to put it into context. But it’s becoming easier and faster to do so. “Machine learning is not mysterious,” Huang said. “It’s easy to learn, and there are plenty of libraries that are open sourced. It would probably take eight weeks to solve many of your company’s problems.”