Similarly we may see trade price ticket spreads stretch alarmingly at certain times of the day that give a totally different real world result from our testing.
Another issue is the sheer size of our sample. One year of data really only shows the current underlying trend. As we await the Trump/Clinton vote, for instance, we see a market responding with a slight skew to the national vote (this is typical), especially as we near the actual voting date. But our backtesting systems would not record this influence and will only record the hard data; they will also be oblivious to the market's political skew.
I have seen 30-year data sets that take trade theories to another extreme; again, we have to question their relevance as the world has changed over that period of time. Trading techniques, trading speed, and the sheer interconnectedness of world markets via the internet have all shifted things due to the mammoth change in market responsiveness and sensitivity to news events.
As such we need a real-time measure that isn't so slow that we amass irrelevant values but is still applicable to the ''here and now''.
In India, the substantial internet technology education that exists there is such that they can research theorems (however odd) easily because of the high number of skilled IT students in the country. One such project related to algorithms for statistical modelling. The objective was, bizarrely, to build a means for a computer to assess a material and consequently make a statue of an elephant via algorithms only.
Initially, the sheer amount of data required crashed the computer systems. Typically, every nuance of the elephant's shape, scale and proportion, as plotted in relation to a three-dimensional image, was beyond the computers memory ability.
It's more complicated than you might think. Photo: iStock
Back-testing a market can be much the same and just as frustrating... too many years of data produces only an average outcome and we need to be better than average with our trade structures.
So how did India overcome the back-testing algorithm for the construction of an elephant from stone, wood, or any other choice of material? In hindsight the answer is easy and uses far less information than we would think: they used the most basic Bayes theorem and reversed it. The code is irrelevant here, but basically they cut away what was clearly not an ''elephant'' and rapidly achieved their required result!
This is where we need to excel in trading. We need to use enough data to know what isn't relevant to the trade's opening and execution.
Axioms of choice A, B and the mote C are simplistic algorithms that are arranged along classic trader actions:
Action A = You open a trade and go long.
Action B = You open a trade and go short
Action C = You do nothing.
All three actions are independent trading decisions and all have equal value and weighting. But action C also tells us that you are not risking anything and will not earn/lose anything. In probability terms, the results will always be 0% returns whereas actions A+B will have a higher percentile.... even if it's only a 1% success rate, it has a higher chance than action C.
So should we ignore Action C?
Axioms of choice as described above are the core of the programming but have astonishing outcomes. We need to pay particular attention to what we programme, otherwise a series of unforeseen circumstances can arise.
Politicians, for instance, are forever resigning due to the axiom choice whereby they enact a policy only to see unexpected outcomes. Banks are no different, but perhaps the most shocking example can be found in an axiom that ran what was regarded as the perfect corporate trade model for a T-shirt company.
What happened was that the firm discovered its computer, when linked to a dictionary, could produce a printed product on demand whereby the company never stocked the printed shirt and could advertise the product for any size or colour T-shirt and printed to whatever text they advertised freely on Amazon. Thus they faced no stock control issues as they would only produce and sell an ordered item and their magic computer would simplyprint the shirt.
In the UK, there was a popular period when the phrase ''keep calm and....'' was printed on T shirts and the like. The T-shirt company in question, then, used its axiom choice program to merge the dictionary's many verbs to the ''Keep Calm and'' phrase believing they would earn a fortune. Ultimately, of course, the Action C (the mote) hit hard... some verbs were totally inapropriate, offensivem and unforeseen. No matter how you look at Axioms, there must be (for want of a better description) a damage control preset.
Without damage control, you will hardly be able to remain calm... Photo: iStock
Runaway trades can and will destroy accounts just as the T-shirt company (ironically called "Solid Gold Bomb
") destroyed its own viability and reputation.
Autonomy founder Mike Lynch utilises axioms in his Artificial Intelligence programming business which specialises in working through the three axioms (ironically) for due diligence. Corporate takeovers – whether hostile or mutual – require immense number of hours for manual due diligence and the computer can work through millions of data reference points to make an axiom showing the three potential outcomes and areas to research further.
It does not do this by cross-referencing everything but instead by considering, as in the elephant example, what isn't there... it searches for a completed audit trail and highlights areas that haven't been agreed upon etc.