Investment choices have a long-term impact on growth and earning capacity – making the wrong choice can be extremely costly. Today, top funds use big data and deploy technology to invest money in the stock market globally. Then, why not adopt tech-based investing in India?
To that end, the Mumbai-based startup Upside AI uses fundamental analysis and machine learning to invest in the Indian stock market. Founded in 2017, the startup is led by Kanika Agarrwal, Atanuu Agarrwal, and Nikhil Hooda. The story begins at a trijunction – three co-founders with complementary skill sets. Both Kanika and Atanuu come from investing backgrounds, having spent their careers in private equity and venture capital, while Nikhil has a B. Tech and PhD in Computer Science from IIT Bombay.
“Given our areas of interest and experience, we knew we would build something at the intersection of technology and investing. That’s how Upside AI came to be,” said Kanika Agarrwal, Co-Founder, Upside AI.
Product
The company manages money for high-net-worth individuals with a minimum ticket size of Rs 50 lakh. There it manages a few strategies, including Upside Navigator, which is an asset allocator that reads macro signals to decide how much to invest in equity, debt, and gold. The idea here is to reduce volatility while trying to beat the NIFTY. And as the company says, it has done exactly that by switching out of the equity in December 2021 and avoiding the massive fall in the markets in 2022.
It also runs a couple of equity strategies – Upside 250 for buying large/ mid-cap stocks and Upside Flexicap, which is primarily a mid/ small cap product. The research division also publishes small cases for retail customers where one can invest as little as Rs 50,000. There the company has three buckets of products across equities and asset allocation. Moreover, the products of the startup are built on three pillars:
- rules-based investing
- keeping the rules dynamic to change with markets
- focusing on fundamentals (company and macro) and not on technicals
The startup doesn’t do trading, derivatives, etc., but simple, easy-to-understand products that can be accessible to a broader audience. “The only difference is we use machines instead of humans,” said Kanika.
Clearing the model’s explainability concerns
The startup clears step one for them is data clean-up. This means for back tests to ensure the firm is eliminating survivorship bias, considering data as of reported date, standardising/ flagging missing or non-standard information (like change in financial years, missing numbers, etc.).
Kanika further said, “The universe we track is the stocks listed on the NSE. Therefore, financials for approximately 1,500 odd companies and macro data. After that, we parametrise each company to be represented by financial ratios that are typically tracked by financial analysts like growth, return, valuation, and more.”
The ML model kicks in after this when it is trying to solve a supply-demand problem. There is a supply of companies where it is training themselves on the meaning of a fundamentally good company. On the demand side, its training is answering “ what does the market think is a good company.” “Essentially, we are trying to buy companies at this intersection, which is what the ML training is focused on,” Kanika said. The algorithm goes over multiple time periods generating portfolios to understand their market reaction. Over 50 million iterations make a “decision” and generate a portfolio to buy and hold.
Talking about the accuracy of the model, the startup says that it has spent a year building the system and another year testing ln live markets. They run multiple diagnostics, out-of-sample testing, checking for overfitting/ underfitting and spurious correlation. Further, the team runs blind backtests, i.e., if the system made a decision in 2007, it only has information until 2007 and not after. Finally, Kanika clarifies that any product the company builds takes a year of building and testing before we even take it live.
The road ahead
With a long-term vision to become India’s leading tech-first asset management company, the startup aims to build products using rules and systems. They believe a systematic approach to investing is the best way to create long-term wealth.
“Over the next five years, we want to be able to reach 50,000 customers who can benefit from diversification into technology in their asset allocation. The idea is to be able to attract best-in-class talent in machine learning that can keep us ahead of the curve as the country increasingly adopts technology in investment,” concluded Kanika.
Source: indiaai.gov.in
