I guess quite a number of readers will have visited Bletchley Park in Buckinghamshire, the site of World War Two code-breaking; compared with the laptop on which I’m writing this, the machines that were developed there to crack German Enigma codes worked at a pace that now seems anti-diluvian, but they still made their calculations or tested their countless millions of possible solutions far, far quicker than could be achieved by any human mind. They were the beginning of the application of digital technology, which was then given an enormous boost by the development of silicon chip technology a generation later. What they did was follow a pre-programmed path, but at such a speed as to be able to sort the correct answer from the wrong ones in a time-frame that made them useful. They were using static programmes to analyse massive volumes of data.
If we fast-forward to today, that is effectively still the model for the majority of algorithmic trading. Static programmes, reacting to data input secured from market sources. Now, algorithmic trading is becoming a bigger and bigger part of financial markets. It is estimated that 40% of new funds launched in 2015 are based on automated, algorithmic trading - and that’s added to those already in existence. Like all traders, from the beginning of time, they too are trying to be the first to recognise the developing trend so that they can act quicker than their peers. Last week, I wrote about how it seems to me that the ever-increasing rate of the dissemination of information has robbed the ‘insiders’ of their previous edge; one way of re-establishing that edge is to be able to learn and act on information quicker than your competitors. So we hear the stories of high-frequency traders - for whom speed is the very essence of their strategy - installing ever-quicker data transmission systems, like the straight-line New York-Chicago fibre line, or the proposed mast on the Kent coast to look over the horizon to facilitate faster information-gathering from between London and Frankfurt (good luck with that one when it comes to the local planners, by the way; my crude calculation suggests it will have to be a very tall mast erected on a gale-swept coast - what could possibly go wrong?).
"Can Machines Think?"
But even those of us who are not physicists can kind of understand that there will come a point where the incremental gain from ever-higher speed becomes questionable versus cost. So while right now looking for speed is a viable strategy, it’s likely that something else will also be needed. So, back to Bletchley Park, or at least one of its greatest denizens, Alan Turing. In 1950, he began a discussion paper with the question “Can machines think?”, and that question is what is driving the latest cutting-edge technology of the financial investment world.
The concept of machines thinking, or perhaps we should strictly say, learning, is part scary science fiction and part picture of a world where we all indulge ourselves in leisure, leaving the machines to handle the work. Although it’s arguable that this was not quite what he meant, I’m going to quote Richard Brautigan: “All watched over by machines of loving grace”. I’m not here going to debate the situation where they may not be motivated by ‘loving grace’ - that’s the scary bit…
If we accept that ultimately speed of transmission will not any longer confer its benefits, then we are left looking to see how else the ‘insiders’ can secure the advantage they seek. (For the avoidance of doubt, my use of the word ‘insiders’ does not imply any wrong-doing; I’m simply referring to those who are professionally involved in markets, who would legitimately wish to preserve their professional status.)
Well, given that the starting point here is the explosion in availability of information, then it would seem logical that the ability to analyse that information quickly and effectively is what a trader would be seeking. Now, as far as pure market figures are concerned - prices, volumes, open interest and so on - the static algorithms can make a pretty good fist of it, provided the original programme points them in the right direction. But since we have already broadly accepted that the computer can do all this stuff faster than the human brain, it would give a significant edge if it could be left to develop its own analysis of patterns and ‘learn’ what those patterns may be indicating. In other words, if, instead of being programmed to find specifically pre-defined patterns and act on them, it were prepared to find patterns for itself and analyse what were their implications. That may seem fanciful, but it is already happening. There are hedge funds who are already developing Artificial Intelligence for use in financial markets (interesting that hedge funds are now significant employers of physicists and computer science experts) and indeed the next step is also underway.
That next stage is undoubtedly far more difficult. That’s where we set the AI to look out into the world, beyond market numbers, to teach itself how events may interact and impact upon the markets in which it is operative. That means trawling the internet, social media, news reports, weather patterns - an open-ended amount of information. Is it fanciful to suggest this is possible? No, it’s not; undoubtedly some of the funds using AI in their trading are already pioneering this kind of stuff. We can’t know yet with what degree of success, but it’s highly unlikely to be a blind alley. It’s a daunting prospect, but then, so were most new technologies at some time.
Is this development good for the markets? Actually by now that’s probably a sterile question. Trying to attribute ‘good’ or ‘bad’ to scientific advancement is a bit pointless. It happens, and we apply it. Was nuclear fission ‘good’ or ‘bad’? ‘Bad’ for those in Hiroshima and Nagasaki, ‘good’ for power generation - there’s no moral dimension to scientific fact. In this particular case, I accept we are walking on the edge, because we are talking about creating machines which can do what until now has been the preserve of sentient beings - think, and learn from experience. May be walking on the edge, but it’s very difficult to foresee it not happening. As we all know, you can’t put the toothpaste back in the tube. So it’s best to try and understand, rather than follow the luddite road………..
(Incidentally, to see how a machine learns, it’s well worth looking on online for Deep Mind, and watching how the computer learns to play those old 1980s computer games. In that case, it does it by increasing its knowledge of the pixel-format of the screen, so that’s an example of AI working in a closed system. Nevertheless, it’s fascinating.)