Machines Can’t Make Traders Obsolete. For Now

First, in the late 1940s, it was Alan Turing who came up with the idea that a computer can be taught to do things. This concept materialized in 1997 when IBM’s Deep Blue beat the world chess champion, Garry Kasparov. Then, in 2017, it was AlphaGo that defeated the reigning GO champion at a game with possible combinations exceeding the number of atoms in the universe.

The number of atoms in the observable universe is 10⁸⁰. The number of positions in a game of GO is 2 x 10¹⁷⁰ (moves played not included).

Ever since Ray Solomonoff’s first idea of an “inductive inference machine” and John McCarthy’s introduction of the term “AI”, we have always kept the fear of a more developed technology taking over our world. This fear proved wrong. Technology develops at an unseen pace and we are those who benefit the most. In medicine, for example, AI is actively helping predict diseases, while at manufacturing, it is the reason why entire factories are run by robots, with little-to-none human intervention. But when it comes to finance and trading, in particular, AI is still far from replacing humans entirely.

The current state of AI in trading

Stephen Roberts, one of the leading AI experts, admits that there is a certain amount of hype behind the application of AI in finance. The truth is that, currently, the technology is employed predominantly for daunting and time-consuming tasks, such as:

  • Fundamental & data analysis  —  by going through financial statements and large datasets, computers can calculate metrics, relevant to a specific investment decision;
  • Chart pattern recognition  —  when analyzing charts, AI solutions can spot important patterns which save technical analysts’ time, as well as minimizes the risk of missing a trading opportunity;
  • Social media signals and sentiment analysis  —  computers gather social signals and information from around the web. By employing Natural Language Processing, they form a so-called “sentiment score” that informs the trader about others’ thoughts and emotions towards a given stock;
  • Security selection and asset allocation  —  some AI solutions, also called robo-advisors, are capable of managing a portfolio, independently selecting securities and allocating capital;

Why traders’ jobs are safe? For now

As humans, we are often scared of the new and unexplored, which makes us exacerbate things at times. This is one of the reasons why researchers introduced the term “robotophobia” to describe our fear of being replaced or overrun by AI. Yet assumptions that trading computers will soon make their human colleagues obsolete is simply too futuristic at this stage. This, of course, may happen in the future, but for now, it is more fictional, rather than possible. Few reasons why:

  • Machine learning has its limitations

Machine learning, as the fundamental discipline behind AI, employs algorithms to learn from data and build models that generate outputs. A well-known fact is that an algorithm is dependable on the quality of the data datasetset that it is fed with. The truth is that historical data cannot reveal the entire picture behind an instrument’s price and is often biased. There are sentiments, irrational behavior, market corrections and unforeseen events that machine learning algorithms cannot take into account. The problem is that by analyzing historical data, we are basically assuming that the past will replicate itself in the future. But in a world, where financial markets are constantly evolving, such an assumption is becoming more and more irrelevant.

Artwork: SMBC-Comics
Artwork: SMBC-Comics

Currently, AI is limited to making accurate conclusions only when the objective is clear. All the generated predictions are a consequence of pre-defined rules and the characteristics of the dataset. A strategy like this works when everything goes as expected. What machines struggle to do is predicting market corrections and unforeseen outcomes. The same argument is valid for deep learning as well. Even now, it is unable to handle market noise effectively, which is an essential part of the trading process.
Our collective intelligence couldn’t predict the next US president or foresee the result of the Brexit vote, even though both of them were very well-researched scenarios. If we, as humans, cannot predict such outcomes, then expecting computers to do accurate conclusions in a much more complex matter, like financial markets, is simply too ambitious. At least for now.

  • It is intuition that can’t be programmed

Data analysis and output generation is just a part of the process. The more important thing is the ability to interpret results and make reasonable conclusions. Even if most AI applications are based on advanced unsupervised learning methodologies that make them capable of extracting noise from data, thus becoming better at making predictions, they will still lack intuition. Humans will remain in charge of making the final decisions. Computers’ raw processing power that allows faster, more efficient and complete analysis helps but doesn’t solve the problem.

At some point, markets tend to stop following stable patterns which prevents computers from being able to make accurate predictions. This is where human traders excel at. Our ability to make rational forecasts can help in situations which machines analyze straightforwardly. In a world where most of the market movement is believed to be a random walk, a computer will need a significantly better intellect in order to become a threat to humans.

Another fact is that we are prone to underestimating the complexity of financial markets. In reality, they are one of the most complicated structures that we know. And machines are yet to learn how to act in chaos. Currently, not even the most advanced AI algorithms know what to do in times of crisis, where humans’ emotional intelligence and ability to contextualize information thrives. This means the development of a flawless AI trading system is still far out of reach.

  • Arms race of equal-but-not-dominant machines won’t benefit anyone

Financial markets are a level playing field, but only in the long-term. Currently, companies like Aidyia and Sentinent, focused on evolutionary computation to develop the ultimate electronic trader, are about to bring the game to the next level. But even if they succeed and make such breakthroughs, in the long-term, the competitive edge may be lost. At some point, a market equilibrium will form, where the majority of the hedge funds are equipped with machines that are somehow similar. This will result in automated stock returns and no competitive advantage for the masses (except for those who can afford to use some advanced propriety AI-based solutions). Using one and the same datasets, as well as quite similar trading solutions, is a recipe for identical outcomes — no alpha for anyone. Basically, this means that when one wants to sell, no one will want to buy, which will erase common profit possibilities.

  • Machines need humans and vice-versa

In one of the Techemergence’s podcasts, Alexander Fleiss from Rebellion Research, one of the world’s most advanced and successful investment management companies, expressed an opinion that machine learning is taking our jobs. While such “survival of the fittest” scenario is valid for low-skilled labor, it is not the same when it comes to trading.

“If we’re honest, machine learning is synonymous with job loss.”

Alexander Fleiss, Rebellion Research

AI applications will complement human traders and not make them obsolete. A human-machine symbiosis like this will result in more streamlined trading processes. At Charles Schwab, for example, significant market changes are monitored and analyzed by human experts, so that any potential risks of automated-trading-gone-wrong can be avoided. Because leaving a trading machine without control may result in devastating consequences, like the exacerbating of the Flash Crash for example. The truth is that computers can handle the more time-consuming work faster, cheaper and error-free. This increases the value of specific human skills such as creativity, reasonable judgment, and intuitive decision-making process.

Our brains' cognitive abilities are what computers remain unable to replicate entirely.

Machines require clarity. Markets lack such

Markets are not static like a game of chess, where all moves adhere to strictly defined rules. They are dynamic, complex structures which make it harder for computers to trade flawlessly. Even if there is a super-algorithm that can consider all factors, relevant to an instrument’s price, the unpredictability of human behavior can still make it generate a couple of different outcomes from one and the same set of data. Advances in deep learning and evolutionary computation are taking us further by bringing better and more accurate predictions, but the point where the mystery of financial markets is solved still remains out of reach.

1 Comment

Leave a Reply

Your email address will not be published.