But before we can talk about how to implement this strategy, we need to understand what gives algorithms their power, and possibly their Achilles heel. First, algorithms are built on patterns, and the ability to recognize these patterns extremely quickly. But the patterns, complex as they may be, that are used to train the algorithms are necessarily extracted from observed history, whether at very high, or very low frequency.
Second, machines in general have perfect discipline, are perfectly rational within the bounds of their programs, do not have emotion, and do not get tired. In other words, machines are relentless and do not get attached to their positions; they keep doing what they are supposed to do.
Third, machines are fast, and this raw speed is both a strength and a weakness.
And fourth, machines and algorithms are designed and maintained by their human designers, and hence, for the time being, may have flaws, which will likely diminish when machines can design and build better machines (thankfully we are not there yet for trading algorithms).
Clearly, taking a machine head-on in any of the dimensions which is their strong suit is folly, e.g. trying to day trade against them. Humans are emotional, humans are irrational, humans get tired, and cannot stick to the plan under stress.
Right?
Almost.
There is no better example of how human weakness can be turned to strength than watching the ending of the first Terminator movie (I won’t remind you of the ending). The very fact that machines cannot think outside of the box and are relentlessly persistent can be used to counter their power. But this requires some knowledge of the rules that they follow, and pre-empting the next move they are likely to take. Since I won’t spoil the ending of the movie I will tell you the principle: the relentless pursuit of the objectives programmed by the designer is actually what makes the machine vulnerable, because it depends on assumptions or programmed responses.
Let us look at the financial markets to provide a couple of examples.