The recent stock market turmoil took the shine off February’s early glow and set the NASDAQ on course for its worst losing streak for twelve months. Much of this disturbance appears as a spill over of bond nerviness and the specter of inflation raising its ugly head. It is interesting, though, to see how much it is the ghost of a former era. The dip-buyers generation knows not inflation. Up until a year ago said same generation had not known a bear market. Now however, they have experienced a bear market, all be it the shortest in history and one that surely has cemented unwavering faith in dip-buying forever.
This raises a curious question: is ignorance bliss, or a profound vulnerability?
Did for instance the blind faith that the pandemic fall had a bungee cord attached, and that all dips should be bought, assist the rapid ascent of recovery, and consequently actually effectuate a new reality?
It is part of human make-up to rapidly identify and compound confirmatory biases. This tendency has its problems in that we tend not to see that we are involved in a self-fulfilling prophecy until it has blown up in our faces. On the other hand, without the optimism to see the positive signs of opportunity, there would be no growth only a fearful stagnation.
When building an AI trade decision-making process, we must be cognizant of how it receives its data. Trading is not a game with a set of clear, defined rules, it is elastic, plastic, smart and stupid, crowd and individual all at once.
Opinions as to whether the increase in long term treasury yields is a sign of a pandemic aberrated, economic overheat or as Janet Yellen recently suggested a sign of a stronger-than-anticipated recovery, will vary according to one’s personal perspective and with it the preparations (or lack of) for higher inflationary conditions.
What interests us however, as AI practitioners, is what this means in terms of the limits of knowledge and the degree to which knowledge or the absence of knowledge influences a decision-making process. Subtle inferential bias can emerge in that process depending on the amount of knowledge at hand. So, we are left with questions as to how much we should seek to know to assist, rather than detract from this process. The Big Data temptation is to accumulate all that there is to know or at least all there is to know up to some processing limitation. Know is the operative word here, does know mean knowledge of i.e. awareness of, or does it mean the accumulation of factual knowledge? This question itself, requires a further question, what do we mean by factual? As that the author considers that facts are always context dependent.
Thus, we are faced with having to analyze what we think we know, or in the case of an AI system what it is aware it is aware of. This changes concrete facts into floating opinions and the analysis brings one to recognize the manner in which we weight or rank those opinions (a structure that is often predetermined by the perspective that has been adopted from the start) hence the danger of falling into the trap of introducing latent self-confirmatory bias. In simple terms, if you see disquieting signs of inflation ahead you will find more, whereas if looking at the same signs you see a stronger-than-expected recovery then those are the signs you will find more of.
When building an AI trade decision-making process, we must be cognizant of how it receives its data. Trading is not a game with a set of clear, defined rules, it is elastic, plastic, smart and stupid, crowd and individual all at once. So, these varied and in many cases contradictory perspectives are both imbedded in the data but also imbeddable into the data. To avoid leaving itself in a position of static indecision the AI system must be able to discern the implicit meaning of that data. Engaging in this process though is not value free. By doing so it is therefore adopting an opinion. In designing a more robust system this can of course be accompanied by a skeptical alter-ego which can query the legitimacy of the system’s decisions (fortunately AI systems unlike their human equivalents don’t take criticism personally).
The key to obtaining useful actionable AI trade decisions is ultimately simplicity. Having the ability to cut through complexity and arrive at a clear decision, requires a recognition of the limits of knowledge, not only what it can know but what it is useful to know and more importantly its own ignorance. ■
© 2021 Plotinus Asset Management. All rights reserved.
Unauthorized use and/or duplication of any material on this site without written permission is prohibited.
Image Credit: Feelartphoto at Can Stock Photo.