In April of 1968, a computer beat a human being for the first time, albeit on movie screens.  In 2001: A Space Odyssey, HAL 9000, the first in a series of increasingly unstable artificial intelligence systems, was not content to beat the astronaut Frank Poole at chess. A few minutes later, HAL killed him, along with a group of other people who were hibernating in the spaceship. However, the sole survivor managed to deactivate the computer by making it regress to childhood, thus temporarily winning the war between man and machine. HAL was the first example of a psychotic computer appearing in the movies. It was then followed by increasingly humanoid machines such as the replicants in Blade Runner, or Ava in Ex Machina, tending toward both self-determination and schizophrenia as well as an irresistible and sometimes understandable  aversion to the human species. 

Over the last two decades, fiction has transferred from the screen to become reality. IBM's Deep Blue broke the ice with chess. Like HAL which had replicated the Roesch-Schlage match of 1910 but had cheated, calling checkmate early when it could have been avoided, in 1997 Deep Blue also took advantage of our fragile psychology. At the end of the first match – lost to Garry Kasparov – Deep Blue made a move that was assessed by most as useless and illogical. The Georgian champion's post-match analysis showed that move 44 pointed to a capacity for calculation impossible even for a grandmaster. The move was also extraordinary for a new-generation computer, resulting in a potential checkmate only 20 moves later. But that move, first misunderstood then celebrated, had been the result of a random choice that the PC made when the program joined the match. At that point, poor Kasparov gave up the challenge before even starting to play. In other words, he had appreciated his own finite nature. He conceded the second game, which could have resulted in a draw just like with Poole, and he made other simple mistakes before losing in the sixth match in only 19 moves, his shortest defeat.

Something similar is happening today on other chessboards, this time in financial trading. Humans are conceding the match ahead of time here, too. In the last 20 years, in fact, we have not limited our use of computers to play chess, an application too trivial considering that your smartphone has a chess skill rating of 2900 Elo points, higher than that of the irascible and ingenious Bobby Fischer or Kasparov. We have transferred this potential to financial trading, from foreign exchange to commodities. Computers do not have the same emotional fragility as humans; their trading choices are made with unlimited memory and strict logic, thus enabling them to make the most efficient move from the rational options available in their memory. These skills, once set up in advance by the programmer (Deep Blue or HAL were two heavyweights in instantaneous computation but had a rigid memory), are now self-learned by the computer itself. We enter here into the magic of knowledge and self-determination. Humans only determine how the machine will develop its skills, while it is up to the computer to use algorithms (Algos)  to define key correlations between variables and determine the most efficient actions to take. Today half of trades in the most advanced futures markets are made computer-to-computer, with the remaining 40% of transactions made passively by replicating indexes or key variables. In fact, only 1 in 10 trades is set up by a biological neural network with all the strengths and fragilities of the human mind. The other 9 trades entail the use of an artificial decision-making system by at least one of the two parties.

In commodities in 2016, algorithms controlled over 60% of oil trades, 45% for grain, 54% for precious metals and 90% for foreign currency exchanges.

Automatic trading, while instantaneous and unaffected by mood, has hidden pitfalls that we have not yet fully understood. In fact, algorithms that connect different variables, opening or closing thousands of financial positions in fractions of a second, generate a process that maximizes the weighting of short- or very short-term variables, the most frequent newsflows and details coming from the most transparent economic regions. The weighting of fundamentals is lost, while the role of short-term correlations and temporary arbitrage increases. In addition, whether the price signal is adequate to build a sustainable business is not considered in any way. The objective of quantitative trading is margin trades, not finding a position to be maintained for 12-18 months like even the most hawkish traders, traditional hedge funds.

Moreover, the specificities of every single market, whether oil, copper or coffee, are being mitigated, with an increase in the weighting given to macroeconomic information to guide choices in individual sectors.

In short, such a quantitative process leads to an exaggeration of the value of input data, increasingly maximizing the weighting of correlations of those published variables which are becoming increasingly more relevant in the deep learning process and limiting consideration of the very short-term, where correlation between data, like weather forecasting for the next few hours, is more immediate and direct.

The value of the most frequent data also increases the weighting of statistics that have historically been more marginal. For example, in the oil market, electronic data preference has for some years focused on the number of rigs active in U.S. onshore, information published by Baker Hughes since 1944, and although this data has been almost completely irrelevant to trading for decades, it is now considered a proxy to measure the growth of U.S. and therefore global offerings.

Weekly data on the U.S. oil inventory are hyper-analyzed in two publications, one is issued by the American Petroleum Institute (API) and the other by the U.S.’s Energy Information Administration (EIA), the two reports appear only a couple of days apart. The data do not match, and the trends can even be opposed, with an accumulation in inventory according to one agency and with a decline from the other. Even so, comparison with commonplace expectations is immediate and affects prices. In this case, the inventory delta is an estimate of the demand and supply balance in the U.S. market and even a global market proxy.

As with rig data, the delta adds global value to local statistics covering the U.S. market that only accounts for 20 percent of world consumption and 10 percent of the offering and only a few percentage points in terms of exports. Local U.S. dynamics can in turn be conditioned by the refining cycle, by exports, by weather factors, and by local pipelines. The Algos provide no in-depth study, only an immediate comparison between expected value and published statistics.

Automatic trading is not limited to the oil market. Computers’ fingers are also covered in chocolate. In January 2016, the outlook in the cocoa market was bullish in anticipation of a significant harmattan, a sandstorm that periodically affects the countries on the Gulf of Guinea, where 70% of the world’s cocoa is produced. But, contrary to expectations, the market has seen a fall in the price of cocoa, one so violent and sudden that it cannot be explained with the fundamentals. The reasons for the fall are related to the decline in the Chinese stock market and fear of a hard landing for the China’s domestic economy. However, the Chinese only consume one percent of the world's chocolate. Only a correlation between the collapse of the Chinese economy and a similar economic crisis in Western countries could justify such a trend in the price of cocoa.

A further effect of this digital cacophony is the explosion of volatility in moments of uncertainty with faster machines simultaneously making similar decisions with the same data set, and when this volatility explodes, the system becomes radically out of control. These events are known as flash crashes, inexplicable meltdowns within a few minutes. To give an idea, in the oil market in the last two months of 2018, as part of a marked price drop, fluctuations of over 4% occurred in one-fifth of sessions, practically once a week. These were not true crashes, only frequent skidding. It is hard to justify such volatility with the fundamentals or with new data. It was in fact due to trading mechanisms acting simultaneously, driven by macroeconomic news or the oil industry itself and in the wake of a trend to sell off all global financial assets. For example, on December 24, oil dropped by 6%, then recovered by 10% on December 26. The only major news in that period was the Christmas ham or more probably, a more balanced position in macroeconomic outlook.

The most sinister aspect of Algos trading is, however, the exit of traditional operators and their daily contribution. In fact, the difficulty of operating in excessively complex and volatile markets leads humans, like the chess grandmasters Poole and Kasparov, to throw in the towel. Paradoxically, the same people who were accused of feeding market volatility, hedge funds, have to foot the bill. The idea of being able to identify a weakness in the market based on knowledge of the fundamentals is overwhelmed by the speed of analysis by quantum funds, which ensure higher yields. The large, historical hedge funds are forced to shut down, just like the neighborhood bookstores driven out of business by Amazon. In 2018, only 130 of the 368 hedge funds working in commodities six years earlier remained.

Even Andy Hall, nicknamed “God” for his abilities to predict trends of crude oil prices, shut down his Astenbeck Capital Management Commodity Fund in 2017. The same has also happened to the specialized funds at Clive Capital and Centaurus Capital. Brevan Howard closed in November 2018, too. Also in cocoa, the infamous “Chocfinger,” Anthony Ward, melted down his fund in 2017. According to Ward, automatic trading in the past created distortions of between 10-15% in the values of the fundamentals, an “irritating but often manageable” level. Today that value would be between 25 and 30%. The rise of the machines is injecting great volatility into the markets. In the words of Andy Hall as he closed his fund, “investing in oil under current market conditions using an approach based primarily on fundamentals has therefore become increasingly challenging.”  We are conceding the match and enjoying the thrill of digital trading. We refer to the need to focus on the long-term, for a fairer and less speculative market, but at the same time we are applying an increasingly obscure, volatile and short-vision model. “Greed is good, greed is right, greed works. Greed clarifies, cuts through and captures the essence of the evolutionary spirit.” No longer is this Gordon Gekko describing the true strength of Wall Street. It is now a modern supercomputer that has taken the baton, with even more disruptive firepower and cynicism. It learned to do so all by itself, in only a few years, too.