How Trading Bots Will Boost Global Innovation and Why I Invested in Almax Analytics
Consumer banking is having a watershed moment. On one hand, the experiment with negative real interest rates is putting pressure on traditional banking business models. On the other, nimble digital startups, such as Estonia’s Transferwise, have begun to offer better services cheaper. Even free.
But investment banks, the financial behemoths who arrange the sale of equities, help facilitate acquisitions and broker trades, have seen little effect from the ongoing technology onslaught. On the contrary, Goldman Sachs seems to thrive on technology disruption. But it too has to put up a fight to stay relevant.
As service fees remain the bread and butter of consumer banks, investment banks who traditionally do not take deposits nor give loans earn a bulk of their income from speculative trades on their own account. But what they both typically share is the so-called sell side, the middleman between the issuer of securities and the investors. This middleman, which houses the analysts who perform securities research and make buy and sell recommendations for clients, is increasingly facing automation and becoming a hotbed for new entrants.
While truly artificial intelligence is still early technology in the banking sector, automation has been there for decades. In 2007, together with two friends, we built a program that modeled five popular investment strategies using Bloomberg data. That amateur calculator, named Babelator after the Tower of Babel, beat its benchmark index by 5 percent and was probably more sophisticated than anything used then by the high street banks in Helsinki. It was proof to me that models can easily be automated so as to gain an advantage. Since then, I have been a believer in financial technology.
And now, as network and processor speeds soar, the processing power of the Internet is becoming powerful enough to accommodate functional artificial intelligence. While our Babelator had 76 704 variables and a data feed that updated in minutes, today’s systems can do more in a fraction of a second allowing analysis to expand from a pure quant realm to a more qualitative space, such as news.
This is where news technology gets truly interesting. Today, speeds are advancing so fast that the same basic technology that helped DeepMind beat Lee Sedol in Go is starting to allow programs to predict stock movements from newsflow. A task that has typically been performed by analysts.
While there have been news analytics players in the financial sector before, nobody has yet managed to automatically assess with confidence the stock impact of news. Almax Analytics, a London-based startup founded by Balázs Klemm, a former portfolio manager at AXA Rosenberg, and Peter Sarlin, a professor at Hanken School of Economics in Helsinki and an expert on neural networks, is devoted to fix this.
Using both structured and unstructured data from numeric and textual sources, such as press releases, Almax can already in today’s early beta version predict with an 85 percent interval the direction and magnitude of a security’s market reaction following news events. With the experience from Babelator, and with the ever accelerating pace of machine learning and network speeds, I found — together with a co-founder of RiskMetrics and the chief operating officer at MSCI— enough resolve to join Almax as first investor.
While trading technology has thus far helped create opaque instruments such as credit default swaps and intra-industry process advantages such as high-frequency trading, artificial intelligence is about to even out the playing field by helping eliminate arbitrage opportunities.
Soon, most financial opportunities bar original innovation may become automated, leaving only buy side risk-takers, such as Goldman Sachs, left with traditional trading opportunities. With money flowing from unproductive arbitrage trading to innovation investments, the pace of technology advancement and global innovation may accelerate even beyond what we have previously anticipated.