Turns out that all it takes to beat an army of financial analysts working around the clock is a robot. Yes, the first-ever fund managed by a robot is so far beating the market and its peers. San Francisco based EquBot is the first team to unite artificial intelligence (AI) with an active exchange-traded fund (ETF). Created with the sole purpose of freeing investments of human bias and democratising access to investment opportunities that AI can uncover, EquBot is the brain behind the AI-powered Equity ETF (AIEQ, launched in October 2017) and more recently the AI-powered International Equity ETF (AIIQ, launched in 2018). About a year ago, AI was focused only on the 6,000 publicly traded US companies, but today it is processing information on more than 15,000 global companies, says EquBot CEO and co-founder Chidananda Khatua.
In an exclusive interview, Khatua, the IISc Bangalore alumnus and former technology team leader at Intel, gives us a ringside view of how AI works, the checks and balances put in, and why incorrect investment calls are more important to EquBot's AI platform.
Why did you think of combining AI with an active ETF?
There has been an explosion of data and the financial sector is one of the data-rich sectors. Bloomberg recently reported that 90 per cent of the world's electronic data was created in the past two years. At EquBot, we believe we will be saying this same statement two years from now. AI is the most efficient tool and probably the only tool to use in today's and tomorrow's investment environment.
When it comes to global markets and investment management as a whole, we acknowledge that the ability to process more relevant data in a timely manner is a competitive advantage. The use of AI gives us a significant advantage in knowing what to trade and when relative to our set investment objectives, especially when we compare ourselves with the competing actively managed investment products.
With the enhanced ability to view the financial markets free of human bias with AI, EquBot made a conscious decision to create an ETF rather than a hedge fund so that investors of all types could access and benefit from the technology.
How has the AI ETF worked so far?
Both AIEQ and AIIQ have been performing at the top of their peer investment groups through Q3 2018 and both perpetually learning from every trade. We can say with confidence that our ETFs are better today than when they were launched and will be better tomorrow as our proprietary deep learning models develop.
Where can the AI ETF go wrong?
Financial markets are inherently risky and it should be understood that the AI system, like any other investment process, will never be completely risk-free. We actually view the incorrect calls as arguably more important to our AI platform from a machine-learning perspective.
There are a couple of key areas where we see significant learning in our model. Like an army of financial analysts working around the clock and legions of trading desks, our system creates multiple valuation models and analyses management teams, news and events, and trading signals. While our platform may observe characteristics of securities with high opportunities for market appreciation and decide to take a risk position, extraordinary events can lead to a non-optimal trading decision. We train our models using the market, and as more data is perceived, the system course-corrects similarly to re-optimise investment exposures.
How did your varied experience contribute to developing this product?
The EquBot leadership team consists of seasoned technology and investment professionals. We met each other at the Haas School of Business of UC Berkeley while doing MBA.
My experience involves decades of technology development, AI studies and graduate-level coursework at IISc Bangalore, Stanford and UC Berkeley. I have led technology teams at Intel.
Art Amador, CFP and EquBot COO, was a former VP at Fidelity Investments and was responsible for managing more than $1.3 billion of investments.
Chris Natividad, as EquBot CIO, has a diversified financial-services background, having worked for Franklin Templeton and Goldman Sachs. He has actively managed multi-billion-dollar portfolios for Apple and Gilead Sciences.
Please give us a nuts-and-bolts explanation of how AI works.
EquBot analyses both structured and unstructured information that can affect security prices. The EquBot decision-making engine makes decision on what to trade and when to trade to best meet the fund objective. We run multiple instances of AI solutions to process data, which include IBM Watson, Amazon ML Labs and Google cloud offerings. They operate underneath our proprietary EquBot algorithms. We look at over 15,000 global publicly traded companies and create multiple financial models using structured data, like 10k/10q regulatory filings, and unstructured data to analyse the management team, news and event data, and market trading signals impacting each company. Our platform generates numerous hypothetical portfolios, which combined with the daily market data, further train our models. At the highest level, we compare our platform with harnessing the power of an army of research analysts doing due diligence on every global company and legions of trading desks managing investment risk to better understand what to trade and when to trade it.
What is the role of IBM Watson in this?
EquBot was part of the 'IBM Global Entrepreneur' and 'With Watson' programs, so the relationship has both technical and marketing benefits. These programs allowed EquBot to both utilise IBM AI service at an advantageous rate and post 'With Watson' marketing collateral where appropriate.
As previously mentioned, we utilise IBM Watson for several of its AI product offerings to analyse unstructured data. We feel EquBot proprietary models adequately harness Watson services and provide significant value-add to EquBot insight and decision engine.
What are the checks and balances you have put in the system because a lot of information online is not accurate?
We acknowledge that a lot of fake news and irrelevant information gets published every day. EquBot has patents pending surrounding our IP, which focus on finding the credibility of financial and investment-related news around the globe.
Given the explosion of data, we can acknowledge the importance of determining what information is credible. Our current investment ETFs are designed for long-term investors, so with our platform instances, we can also train our models to prioritise information that drives long-term profitability and growth.
EquBot's technology should not be seen as a 'black-box' approach. We have designed our investment platform for observability so that we can more easily identify if there are data-integrity issues.
Finally, operationally, we have established a system of data checks, whereby we can see if there are any issues with the data processing as the system works around the clock.
Is there any bias in the way AI operates? What is it?
Right now, our ETFs are operating autonomously without human bias. We feel human bias from a historical perspective has detracted from investment gains. Human involvement occurred during the initial platform training and set-up but now it primarily revolves around data-integrity checks and sourcing new data sources for the system to ingest.
There is some talk that AIEQ is out-performing by buying little-known companies that paid off big. Is this correct?
Not entirely. Relative to the S&P 500, AIEQ does have a higher exposure to mid- and small-cap companies. We have noticed our platform's ability to continually analyse data and shift the risk in the portfolio towards positions with the highest opportunities for long-term capital appreciation throughout the year. While the higher exposure to smaller names did drive some outperformance during 2018, it is the data that drives our investment platform decisions. Going forward, it may not be the smaller-company exposure that drives alpha or outperformance.
So, is machine learning's power much better in the small-cap and the mid-cap space than it is in the large-cap space?
No, machine learning can be influenced by the amount and quality of information available. We have noticed that there are instances of large-cap names producing significantly more news and electronic data, so this can aid in our machine-learning modules. We typically avoid such high-level generalisations as the data for every investment opportunity is unique with respect to the current market environment it is operating.
Do you think AI can do the same job for ETFs in emerging markets like India?
Absolutely, we are already collecting data on emerging-market countries as a result of their proximity and interactions with the developed markets we currently cover in our existing investment products. We will continue to monitor investor interest and demands as we introduce new products into the global markets.
Can a competitor work with IBM Watson and do the same thing as EquBot?
EquBot uses AI tools across different platforms, including IBM Watson. At the core of it, EquBot's proprietary models use different outputs from various platforms to make it work. Yes, it is theoretically possible for a competitor to build a similar platform like ours, However, we believe our platform will continue to advance in technology implementation and grow the learning data set to stay ahead in the technology curve.
Can you tell us how your system has grown in the past year in terms of the amount of data processed and the number of investment insights uncovered?
About a year ago, we were focused only on the 6,000 publicly traded companies in the US. Today, we are processing information on more than 15,000 global companies, so by virtue of this expansion, we are now capturing investment insights from over 20 countries and continue to add new geographies. On the processing side, we are continually adding to our growing data set. For example, we are processing additional data like news articles, trading signals and financial indicators each day.
How confident are you about the AI working in bear markets?
We can see that during more volatile markets, there is more information. During bear markets, we also observe increased investor irrationality - massive sell-offs, illiquidity from lack of trading, etc. Our ability to process more data in an unbiased manner is a competitive advantage. So, during the last downturn in February, we noticed our system further diversifying into new positions to reduce risk and taking on new exposures at bargain prices in an unbiased manner while the rest of the market was selling or doing nothing. The performance and learning outcomes were very well received, so we hope to come out of this recent bout of volatility on top. During bear markets, we believe investors prefer portfolio managers with more information and not less.