Perhaps the most significant buzzwords in investing today center around the big data revolution, machine learning, and artificial intelligence. But this nascent industry is much more complex and nuanced than the simple slogans suggest, a May report from JPMorgan’s Marko Kolanovic highlights. In fact, there is significant runway for expansion. For instance, while big data revolution has been the hyped term, actual adoption has lagged. Less than 1% of such information was analyzed as of 2013, with an enormous amount of data being created in a computer-dominated world.
In fact, 90% of current data was created in the last two years alone. This is information that hasn’t been analyzed — a rare unplowed investing field — and where new algorithms might be found. To understand the opportunity, Kolanovic along with Rajesh Krishnamachari and assistance from a team of nine researchers, put together a robust deep dive in the form of a 280-page report (take that Bill Ackman!) that details the structure of technological advancement in finance and the current and projected spending. But what was missing from the report — understanding how human thinking can be different and at times better than computer-based formulaic logic — is a key component of success to combining human and computer.
Despite the buzz, the trading runway for big data revolution, machine learning, and artificial intelligence is long
The financial services industry currently spends between $2 and $3 billion is being currently on “big data,” JPMorgan estimates. Still, the industry’s spending on is a fraction of the nearly $130 billion currently being lavished on understanding the increasingly omnipresent information that is being generated in a digitally connected world.
As the big data revolution and the intelligent computer applications analyzing information and executing trades grows, how will markets endure what are not pure economic forces tipping the scales of price discovery? What happens if a specific algorithmic approach becomes too popular?
These are just some of the impact questions that vex the future.
JPMorgan, for its part, thinks changes resulting from technology will be “profound,” impacting nearly every sector in financial services:
As more investors adopt alternative datasets, the market will start reacting faster and will increasingly anticipate traditional or ‘old’ data sources (e.g. quarterly corporate earnings, low frequency macroeconomic data, etc.). This gives an edge to quant managers and those willing to adopt and learn about new datasets and methods. Eventually, ‘old’ datasets will lose most predictive value and new datasets that capture ‘Big Data’ will increasingly become standardized. There will be an ongoing effort to uncover new higher frequency datasets and refine/supplement old ones. Machine Learning techniques will become a standard tool for quantitative investors and perhaps some fundamental investors too. Systematic strategies such as risk premia, trend followers, equity long-short quants, etc., will increasingly adopt Machine Learning tools and methods.
Kolanovic - There are significant pitfalls to big data revolution, machine learning, and artificial intelligence
While there may be significant opportunity in big data revolution and machine learning, there are also trap doors to consider.
Perhaps the most serious business mistake could come from overspending on the wrong technology or algorithmic approach.
With a new fad or even fundamental shift occurring at light speed – including the changing of core technical platforms such as Blockchain and the popularity of derivative building blocks such as Ethereum springing up virtually overnight – investing in the wrong platform might require a significant rewire of a firm’s core infrastructure.
There are also costly issues that are typically an inherent cost of research.
For his part, Kolanovic and company note the researcher’s lament, parsing seemingly endless mounds of data to only learn that a “blind alley” has been discovered. "Datasets that don’t contain alpha, signals that have too little investment capacity, decay quickly, or are simply too expensive to purchase.” Expensive mistakes can be very expensive. “Managers may invest too much into unnecessary infrastructure e.g. build complex models and architecture that don’t justify marginal performance improvements.”
Expensive mistakes can be very expensive when a shift makes an entire technology investment obsolete or just fails to generate significant returns. “Managers may invest too much into unnecessary infrastructure e.g. build complex models and architecture that don’t justify marginal performance improvements,” Kolanovic wrote, noting the constant pressure to deliver measurable investment results on a consistent basis.
But this isn't the only issue fund managers adopting technology face. Another involves human talent.
Contrary to the over-hyped promise of computers taking over even the most intelligent human thought processes, down on earth at this moment in history building a strong quant-based program requires an investment in human capital. Great trading and investing formulas are decidedly driven by humans.
This human talent Kolanovic and company categorize as a risk, then subtly pointing to a key line of demarcation between tech executives that succeed in finance and those that don't.
“Employing data scientists who lack specific financial expertise or financial intuition may not lead to the desired investment results or lead to culture clashes,” the report noted. In part this can be seen in the recent high-profile hiring of major tech talent – Bridgewater Associates comes to mind – only to see the relationship falter.
The brutally frank method in which many fund management firms critique investment ideas and actually seek divergent opinions is a mode of operation that thin-skinned technology savants might not appreciate. A completely different environment coupled with the need to constantly deliver tangible and measurable investor results can result in culture shock for some of the more pampered in the tech world, multiple industry sources inside quant research projects have observed.
This could be one of the biggest stumbling blocks in integrating Silicon Valley culture with that of the wider financial services industry and it boils down to an ego. The technology culture that pervades in some circles mandates that technology itself is the primary driver of success. In financial services, technology is not the key driver. Understanding and identifying market opportunity drives success. Technology is the tool that achieves a more important objective. Technology needs to have a market understanding, which is a knowledge that sometimes comes from humans.
Kolanovic puts this in more austere terms. “In implementing Big Data and Machine Learning in finance, it is more important to understand the economics behind data and signals, than to be able to develop complex technological solutions,” he says, noting that core economic and market structure knowledge is a key to success not technology alone.
Recognize limits in big data revolution, machine learning and artificial intelligence
It is meaningful to understand the limits of machine learning and artificial intelligence. Here the report briefly touches on some of the many issues without much acknowledging the value of human intelligence.
“Machine Learning algorithms cannot entirely replace human intuition,” Kolalonvic states. But is it really “intuition” that is the human element that goes into investment / trade decision making?
Perhaps the most well-known trade over the last decade – “The Big Short” – was made possible not by human "intuition" but rather by connecting noncorrelated dots that systematic machines might have missed. Consider that there was no accurate price information available to the market and marking the illiquid assets was done on a discretionary and opaque basis. But that very famous trade is but one example of major market moves that involved a never before seen a combination of elements.
In fact, many of the greatest trades in history likely might not have been picked up by a systematic program, including George Soros and his attack on England's central bank. Many great trades involve connecting dots in never before seen element combinations. Most recently Hugh Hendry's populism hedge was an example of connecting noncorrelated dots – information not in the public domain such as the derivatives exposure of large banks – and creating a tail hedge strategy. In the wake of the election of Donald Trump, the populism trend seems to have slowed, impacting the hedge's success. Just like the logic behind Kyle Bass and his Japanese yen trade, those are noncorrelated dots that eventually get connected. There are key factors in all these trade decisions that computer-based if-then logic might have missed.
Nowhere in the paper does Kolanovic address the linear limits of computer-based formulas to recognize how humans differ in their thinking systems. This is not to say the advance of technology won’t roll on – it will. It is just to say that those who recognize the limits of computer technology and focus on how it integrates with humans, augments human performance, are likely to be the ones that control computers rather than the other way around. This is the fascinating focus.