Tata Mutual Fund launches AI and ML powered ‘Tata Quant Fund’
The AI based algorithms designed by the inhouse data scientists at Tata Mutual Fund has significantly beaten the markets sixty two percent of the times in terms of accuracy
This quant fund employs a proprietary quant framework that combine multiple rule engines and predictive models to create investment portfolios that are aimed at maximizing returns during up-trending markets while minimizing losses during down trending phases.
According to Utpal Sarma, Head – Business Analytics, Tata Asset Management, “Actively managed funds benefit from human intelligence that learn, comprehend, and respond to different market challenges in complex manners. Passively managed funds, on the other hand, are rule based and they avoid pitfalls of biases that accompany human judgement. Their strength lies in increased objectivity and elimination of human errors. These traditional styles have contrasting pros and cons. AI and ML adds a third dimension to fund management. They enable machines to mimic human judgement to certain extent while retaining the benefits of disciplined rule-based investing. Investment strategies employing such constructs tend to leverage market opportunities better while avoiding bias errors. Technologies supporting artificial intelligence today is quite evolved and robust. Their wide usage across businesses is testimony of their potential and capability. While quant strategies for wealth management features prominently in advanced markets, its only a matter of time that they gain popularity in India as well,”
AI outperforms market
Tata Asset Management has data of more than twenty years, which is employed by the algorithms. The AI based models have been tested for over 107 months. “We have erred for about twenty percent of the times i.e the market has outperformed us; we have been neck to neck for nineteen percent of the times. It means the market hasnt beaten us significantly however we have significantly beaten the markets in sixty two percent of the times,” informs Sarma, speaking exclusively to EC. The accuracy achieved by the algorithms has been far and wide in the back testing time frame for the 107 months.
The structured data from the last twenty years has brought about the best value creation. They have proved to be very helpful for designing algorithms, which gave the best results, “All the returns that we have been able to generate during the back testing period have come from structured data. The primary source of outperformance is not coming from exotic data structures. It is coming from the ability to develop a good algorithm to do the predictions,” says Sarma. The structured data includes the common macro economic parameters like numbers on GDP, inflation, short and long term interest spreads, credit spreads, exchange rates, international indexes and data on crude prices. The other data includes PE, RoE, EPS growth volatility, dividend yields, basic price momentum adjusted for volatility, etc.
When asked about the critical factors behind the high success rate of the algorithm, Sarma says it has been a judicious mix of the open source technology platform, statistical tools and the talented team of data scientists.
Interestingly, the data which spans across more than twenty years is hosted on the on premise IT infrastructure of Tata Mutual Fund.
Broad and persistent factors of stock returns are used for building rule engines for portfolio creation. Each rule engine uses scoring to create concentrated portfolios with attributes like ‘value’, ‘quality’, ‘momentum’, ‘size’ and couple of combination attributes. Thereafter the ML powered predictive algorithms decide basis prevailing market and macro-economic conditions, portfolio with which attribute is likely to outperform during the next month. The algorithms also predict absolute direction (positive or negative) of return for the next month. Long position in selected portfolio is taken only for months where predicted return is positive. During months where predicted returns are negative, the strategy uses derivatives to hedge the gross long equity position held previously. Tata Quant Fund portfolios are rebalanced monthly for optimal performance at low risks.
The predictive engines use more than past 20 years of market and macro-economic data to analyse hidden relationships and patterns. These correlations along with prevailing market and macro-economic data are then used by the engines for making monthly predictions. Thus, the investment decision making process of the fund is fully machine driven and free of human judgement.
The machine learning predictive models also recalibrate and re-adjust at a fixed periodicity by using new and incremental data. This enables the models to factor in emerging patterns and relationships. These algorithms are developed and managed in-house by a dedicated team of data science specialists.
We have come a long way from using a pocket calculator to analyse figures gathered from company reports to using machines which decide which stocks to be picked for investing.
Prathit Bhobe, MD & CEO, Tata Asset Management said, “Machines have massive computational power needed to process very large data sets, spot patterns and correlations, make decisions faster, objectively and without human biases. In the current world, computers are powerful enough to solve problems, a lot of data is available and we strive to use this data in combination with algorithms to its best”.
“Tata Mutual Fund has developed intelligent machine-driven strategies keeping in mind the appetite of long-term equity investors. This framework crunches massive amounts of data, recognises patterns and leverages the power of technology. The future of investing is in the use of quants and with us entering a new decade, we believe that the Indian market is now ready for tech-based investing”, says Sailesh Jain, Fund Manager, Tata Asset Management.
The machine learning predictive models also recalibrate and re-adjust at a fixed periodicity by using new and incremental data. This enables the models to factor in emerging patterns and relationships. These algorithms are developed and managed in-house by a dedicated team of data science specialists. We have come a long way from using a pocket calculator to analyse figures gathered from company reports to using machines which decide which stocks to be picked for investing.
The minimum application amount for this fund is Rs. 5,000/- and in multiples of Re.1/- thereafter and additional investment of Rs 1,000/- and in multiples of Re 1/- thereafter. The fund will be managed by Mr. Sailesh Jain.
If you have an interesting article / experience / case study to share, please get in touch with us at [email protected]