Chetan Pandya, who set up Kotak Securities’ institutional business, is now a fintech expert wearing multiple hats. Increasingly he sees machines, algorithms and quantitative models taking over the investing space.
AI is a different ball game from quant-based models, however, because it has to be self-learning and not formulaic.
Since it’s a black box, pretenders can claim something is AI-based, but genuine adoption is hard, requiring a disruption of existing processes.
That is still at a nascent stage in investing even if there’s a lot of hot air on the topic, says Pandya. “This space is full of people who are using quant-based strategies and they’re making good money. But to say they’re AI-enabled would be a stretch," he says.
He does see wealth managers with AI credentials appearing on the horizon, however, so the shift to AI is starting to happen.
One example is IDFC’s Neo Equity Portfolio which uses AI for investment. Chetan Mehra, director of portfolio management services and alternative investments for IDFC, has a PhD in computer science from the University of Southampton, where he used multiple interacting intelligent agents to build financial market portfolios.
Mehra started his career in the UK in 1997. He worked on option pricing models for Bankers Trust, which was acquired by Deutsche Bank, and equity portfolios for wealth management firm Brewin, before co-managing a global hedge fund. “I then did my PhD, a bit of consulting for AI-based firms, and then joined IDFC AMC last year," says Mehra in his characteristic understated manner.
DISCOVERY FROM DATA
He starts by explaining the essential difference between the usual quantitative models and what he’s doing for IDFC’s portfolio management scheme. “In a quant model, you have a prescribed formula from a book for which you find the right parameters and it becomes your model. In AI, you feed in the data and the algorithm discovers the model from the data; this is the key difference between the two approaches. The good part about an AI-based model is that it will adapt itself as new data comes in."
One of the biggest challenges is finding the right data to feed the machine, because data sources are less reliable in India than in the West. “We’re paranoid about the data having errors and gaps. We spend 75% of our time cleaning it. If you get it wrong, it’s like putting diesel in a petrol car," says Mehra.
Finding good data sources and eliminating bad ones involves taking into account many factors: inflation, unemployment, interest rates, oil prices, earnings of companies, tenure of board members, sustainability, education levels of employees, and more.
“We are constantly on the hunt for newer data points that improve risk-adjusted return. The data is fed into the machine and it throws a list of optimal stocks we should hold," he says. The objective is to figure out which variable gives an edge to the portfolio. And it takes a deep understanding of both AI and finance to make it work.
“Financial data is 95% noise. Not every algorithm from machine learning can generalize into a good model for finance, owing to noisy data," says Mehra. “You’ve got to specify the problem so that you can solve the ‘right problem’. The way we create a factor that goes into the model has a deep impact on its performance."
The portfolio manager supervises machine learning because leaving it all to AI could be dangerous. “We don’t let AI run amok. There’ll be limits like the upper limit for a holding. We control what data is going in, so it’s unsupervised and supervised machine learning. It’s like saying, ‘this is the football ground; if you go beyond the rectangle, you’re out’."
Globally, AI-based investing has gone through a hype cycle with inflated expectations followed by a realization that it has to mature to fulfil its promise. But hedge funds have been adopting it.
Financial data tracker Preqin says AI hedge funds outperformed all hedge funds by three percentage points in three-year annualized returns. A pioneer in this area is Renaissance Technologies founded by AI scientists.
It’s a trend gaining momentum because there’s no arguing against how much faster an AI machine can process data and adjust models. But it’s still a phase where the pretenders have to get sorted out from genuine AI.
Sumit Chakraberty is a contributing editor with Mint.
Write to him at firstname.lastname@example.org
Subscribe to Mint Newsletters
* Enter a valid email
* Thank you for subscribing to our newsletter.
Never miss a story! Stay connected and informed with Mint.
our App Now!!