The Irrational Exuberance of Artificial Intelligence
Friday, July 13, 2018
The current technological revolution born from the immergence of social media, long-form digital media and cheap, unbridled bandwidth is shaping the way humans communicate with each other. Regardless of medium, YouTube, Instagram, Snapchat, Facebook etc. sitting between the chaos of commentary, conversation and debate on one side and the orderly presentation of feeds, content and feedback on the other is a robot, charged with the insanely important task of feeding us what we want to see and hear, engaging us to the point of distraction and pleasing to the point of addiction. But we are also beginning to see the societal and political impact of this person-pleasing robot. If you pander to the exact tastes of an individual, they become shrouded in a cotton wool that they themselves have spun, fed a continuous diet of things that agree completely with and thereby further embolden their viewpoint and opinions. This leads to social division at the macro level. Pockets of individuals battle-hardened with the espoused cause of an activist tribe after falling into an artificially intelligent rabbit-hole simply because they liked some innocuous comment 24 or 36 months prior. What can this message tell us about the role of artificial intelligence in trading?
First, behind every AI program is a human programmer. And in the cases of both social media and stock market activity, the algorithm is dealing with the complexities of human sentiment. With social media, the gauge of this sentiment is usually through language – liking one Trump article will create a steady stream of such articles and based on the language in that article, the likelihood is that you will not get many articles relating to those championing the threat of climate change for example. Over time, it has been shown that your news feed will carry a given sentiment and that news feeds will therefore play a part in creating sentiment. In markets, an algorithm tries to gauge sentiment in numbers – if a 1,000-lot order joins the bid, hit market with buy order for example. Why? Because the increase in buy orders, according to this AI example, implies a bullish sentiment. But now the program has just hit market with a buy order and in doing so has had its own bullish impact on market sentiment. Just like the social media robot, his market-making cousin creates sentiment as well as trying to decipher the over-arching sentiment.
It has been said that AI will assist in furthering the arguments of the Efficient Market Hypothesis. However, computer programs, like the traders that program them, are caught in cyclical overshooting. What was termed irrational exuberance by Robert Shiller refers to the human tendency to be unreasonably optimistic in strong market run-ups and illogically pessimistic in bearish market falls. Programs can be so fast and furious that they often exacerbate rather than improve this irrationality in certain instances creating devastating flash crash scenarios rather than unperturbed efficiency.
But with volatility comes opportunity and it can easily be argued that another program will step in with speed in that scenario to bring the market back to equilibrium. From experience, humans can often react with great rationality in flash crash scenarios than computer programs that are typical built to work within the confines of regular market volatility rather than exceptional, “black swan”-type moves. This is largely because the development of sophisticated algorithms was an expensive endeavour both in terms of time and financial resources. Developing a program to deal with incredibly uncommon events that are very unlikely theoretically, will not have the same Return on Investment projections as those built to deal with routine, daily market movements and trends.
This may all be changing as innovations in algo-creation software makes the endeavour much cheaper and more accessible. For example, the industry benchmark in trade execution software, Trading Technologies (TT), has added as part of its core offering a module called ADL which allows non-programmers to structure an algorithm graphically, without the use of computer language. Let the games begin!