In an interview with ETMarkets, Agarrwal who has over a decade of experience in public and private investing in both India and the US said: “Over time our alpha is made in small bumps, while generally trading in-line with the market.” Edited excerpts:
Artificial Intelligence or Machine Learning became popular a few years ago, but it seems to have taken retail investors with a storm – at least for those who understand the technology. How is it helping investors in a large way?
AI / ML are not new concepts, they have been around for decades. What is different is that (i) computing power has grown a trillion times since then, and (ii) the wide availability of structured data.
The top 5 hedge funds in the world are quant funds and a majority of the trading activity in the US is system driven. In India, however, we are still in the early days.
Most technology-based funds are focused on technicals / HFTs and less on the fundamental side, where now investor wealth is concentrated.
Fundamentals-based investing remains the domain of humans in India. That will change over the next 5-10 years and hopefully, we at Upside AI will be at the forefront of that trend.
ETMarkets Smart Talk: Atanuu Agarrwal explains the role of artificial intelligence in investment
Atanuu Agarrwal, Co-founder, Upside AI who has over a decade of experience in public and private investing both in India and the US said that fundamentals-based investing remains in the domain of humans in India, but that will change over the next 5- 10years.
It will be difficult for investors probably here in India to adjust to an AI-type environment and not a human interface. Psychological speaking how does it work here?
I think we need to think about it in two parts – HNI and retail.
HNI investors in the US and other developed markets are very comfortable investing in quant funds or hedge funds that will never fully reveal their approach or portfolios.
At last count, 1 in 3 hedge fund managers in the US uses quantitative strategies. In China, quant AUM has doubled to nearly $ 100bn in 2020.
However, even there the interface is human – all these funds have sales forces or are distributed through wealth management firms that have RMs, etc. I think this is what will happen in India as well over time.
Retail investors on the other hand have become increasingly comfortable buying things without any human interface.
10 years ago there would be no way that my mother would even consider buying groceries or clothes online. But now digital adoption is extremely high, even at the high end of the market.
In fact, even when retail investors buy a mutual fund, now of them have no idea (or interest in knowing) who the fund manager is or what is the underlying portfolio.
How many numbers of parameters are analyzed to make a buy or sell decision? And can the decision go sideways? What does the history suggest?
Upfront I would like to clarify that I am generally talking about Upside AI’s approach here – there is no way for me to know exactly what other players in the space are doing.
At Upside AI, we don’t use any leverage – no F&O, margin, etc. Neither do we engage in any shorting of any kind. So, we are not going to have flash crashes, etc.
We are simply buying a portfolio of stocks / MFs / ETFs and rebalancing it periodically based on signals from the algorithm. We also don’t trade very often at now, once a quarter.
As a result, in our equity-only products, what we have seen in our live portfolio as well back tests is that we more or less mimic the index in our drawdowns and volatility metrics.
During March 2020, when the market was down ~ 30%, we were also down a similar amount. It was on the upswing that are returns were higher and hence, we are able to deliver alphas over the long term.
Over time our alpha is made in small bumps, while generally trading in line with the market.
In terms of parameters, we feed in fundamental data (P&L, balance sheet, cash flow) for all companies listed on the NSE and macro data such as GDP, inflation, liquidity, etc. since when (mostly early 2000s) these data points are available.
We then run millions of iterations based on proprietary frameworks – we don’t use any off-the-shelf products; all our algorithms have been developed from scratch in-house.
This is really the perfect use case for using ML techniques – it’s simply not possible for a human being to crunch this amount of data and draw inferences from it.
(Disclaimer: Recommendations, suggestions, views, and opinions given by the experts are their own. These do not represent the views of the Economic Times)