When it comes to the investing and finance realms & managing portfolios, one always seems to wonder whether to go the human intuition route or the algorithm route. If algorithms are the way to go, could it adapt? Are human fund managers better equipped? Or could the power of AI be harnessed to navigate investing?
According to Chetan Mehra, Head of Quantitative Investments at Bandhan AMC Limited, “When you’re using machine learning or AI, what you really need is good quality data, because, in machine learning, if you one doesn’t have data, there’s nothing to learn from. How one sets up their foundational data layer and how the systems are going to consume it make a huge difference, as it impacts the efficiency of saving and consuming large amounts of data. Before going into a machine, data has to be cleaned, so one has to ensure biases are not being introduced and that errors are not added to the data.
So, how it’s prepared is extremely important, because once the machine starts consuming the data, it won’t know there’s an error in the data. There’s an initial stage at which one has to have a separate set of models, which are not only getting the data out from different sources, but also looking for errors, data jumps and misprints. So, there are a lot of checks which one has to go through before the data actually becomes consumable. The stage of data analysis and machine learning for investments has to be done just after that. Everybody thinks data is just taken, chucked in and then, it’s all going to work. But, there’s a lot of grunt work before that”.
Mehra remarks, “Almost 80% of our work is ensuring that what gets consumed is reliable and of high-quality & that’s a lot of work… Machine-driven funds are cyclical sometimes, but most of them tend to outperform or reach their goal, because one has defined them well. It’s important to keep in mind how well a problem is defined, how prudently the tools to be chosen are applied and how well it’s set up. The whole idea of investing using machine learning is to remove biases and let the data do the talking. A lot of people may come out and say that they’re data-driven and have no biases.
That’s something that should be taken with a pinch of salt. Good science is not to do with just studying math or physics, it’s about being trained as a scientist to set up an experiment objectively. Before doing an experiment, one has to know what the hypothesis is, at what level it should be accepted or rejected, following the process systematically, rejecting or accepting based on evidence, making modifications for subsequent tests and more”.
And what are the steps involved in applying machine learning to portfolio management?
Mehra declares, “The first step is writing down what one is trying to solve. If one misspecifies a problem, they’ll have to revisit it, so it’s an iterative process. The first question one has to ask themselves is what they want to do and what they want to achieve & write it in different forms. The next step is identifying data sources and how much one has to pay for it. Once you know what the data sources are, then they have to look at the raw dump, whether there’s good-quality data and if it’s cleaned. And then, when you’re happy, then you begin machine learning”.
“A lot of investors don’t pay attention to risk and think as long as they make money, they don’t care. But, they should actually care, because one’s model may be capturing the upside at this moment in time, but there’s a possible downside which could make one lose a fair bit”, quips Mehra.
Watch the full interaction here:
Previous links:
How Relevant Is Crypto In 2024?
What Makes A Great Acquisition In 2024?
Could GIFT City Be India’s New AI Hub?
Can Blockchain Solve The US TikTok Crisis?
Will India Be The New Hard Rock Café With Lab-Grown Diamonds?