By Gaurav Chakravorty, Co-Founder, Qplum
The only way the investment industry can deliver high quality returns at low recurring cost is by using artificial intelligence.
Adoption of artificial intelligence (AI) is happening faster in the institutional investment community than many asset managers want to believe. Proprietary trading firms and family offices have been the innovators and early adopters.
Hedge funds also fall into this early-adopter group; and lately we have seen a strong shift away from discretionary and a resistance to “pedestrian quant” strategies among hedge funds. Even individual investors (typically “laggards” in the asset management industry) are embracing AI-based methods, as speakers observed at the Equities Leaders Summit held in Miami on 4-6 December 2017.
However institutional investors, such as pension funds, endowments and insurance companies are stubbornly sticking to old data-less methods. In this article I will show the trajectory of evidence-based or model-based investing and why chief investment officers should reinvent their processes using AI.
It follows a recent podcast with Nvidia, a leading maker of GPU cards, who invited me to explain why deep learning promises to package the best algorithmic investing knowledge into a system that works and is universally accessible to all of us.
Evolution of data-driven trading
The earliest instances of data-driven systematic trading are value investing and trend-following. However, among these I am focusing on trend-following since it was a method where the outcome of the model was not filtered by human intuition. Trend-following was the original quant success in the Crash of 1987. Ten years later statistical arbitrage (StatArb) became the quant trade, and trend-following had a crash of its own in 1997. Another decade later, on 6 August 2007, StatArb crashed and high frequency trading (HFT) emerged as the top quant trade.
Now, the overwhelming consensus is that HFT isn’ profitable any more apart from the for biggest firms. There is a ten year cycle of changing quant trades due to changes in cheaply available technology. The dominant quant trade has changed over time due to the availability of new technology. New technology is why trades crest and fall.
This brings us to the exponential growth of quant we see today.
No mandate for guesswork anymore: allocators are overwhelmingly choosing systematic/quant
A partner in an institutional allocator summarized the current view: “There is no mandate to invest in anything discretionary unless it is activist investing. We feel the need to invest in strategies that are substantiated by research, by a scientific method.”
Allocations to quant funds are set to exceed $1 trillion. That would be about a third of the net allocation to hedge funds.
“Pedestrian quant” strategies don’t work: allocators need to stop chasing the past
Neal Berger and I described in a recent Opalesque CIO roundtable why “pedestrian quant” strategies aren’t working any more. However, the machinery of relationship-based consultants and outsourced CIOs set up for institutional investors makes them very slow to react to this change. The dominant institutional investing approach is still to make an asset allocation pie chart and shove everything remotely data-driven into a small allocation to “alternatives”. A flat CIO structure would be an improvement, and help provide investors with what they really need: solutions not products.
In a panel I led with Michael Weinberg and Michael Dziegielewski, Weinberg tried to differentiate between quant and autonomous learning.
“Where we are now is far more revolutionary for asset management than simple AI-driven quant. What we have now in investment management is the equivalent of AlphaGo Zero, but for investing. Just like AlphaGo Zero generates its own new ideas, so do Autonomous Learning Investment Strategies (ALIS). We think that this disruption to the asset management industry will come from outside discretionary and even quantitative managers. These so-called ‘third wave, ALIS managers’ exploit the confluence of data, data science, machine learning and cheap computing: their brains are wired differently.
They are often hackers and computer gamers with a healthy disrespect for convention”, he said. Weinberg also mentioned the fact that allocators need to spend a lot of time understanding the approach of the manager and their competency in deep learning and AI. It is not just about chasing returns.
This brings us to the multidisciplinary nature of AI today.
AI is learning from many fields not just finance.
The progress we see in AI today is very multidisciplinary. Every time we make a breakthrough in speech recognition technology, the same can now be directly applied to portfolio management. Every time we go up a notch in computing power and big data technology to support AI algorithms, we can use the same in portfolio management. In a recent article, ITG’s Ben Polidore mentions how cheaply available computing power is growing to a supercomputer level.