Posted Apr 6, 2018 by Martin Armstrong
QUESTION: Mr. Armstrong – Neural nets and deep-learning algos are not all that new. However, with the advent of quantum computing power and huge cloud data stores recording every flinch people make, the business world is abuzz that this is the portent for AI to now transform business, business management, and our lives. Of course, trading systems are now being ‘trained’ on datasets. However, regardless the algo, it’s learning capacity is still limited by the extent of the data that it is presented. Since you have spent so much exhaustive work amassing such a long-term financial (and governmental, etc.) database, using even such things as coinage records for your forensics, I have to wonder if these other systems will still have a long run to go before their forecasting power can match that of your models. Do you expect this to provide Socrates an advantage over the new wave of AI market/trading forecasters that will last for some time yet?
ANSWER: The long-term database is essential. That cost more than $100 million to assemble and quite frankly, nobody seems to be willing to spend that much. This is why all prior models have collapsed creating economic catastrophes such as Long-Term Capital Management debacle. They collapsed again in 2007. Nevertheless, then you have the problem of Neutral Nets are just incapable of handling the vast array of variables. The attempts to create trading models are all flat-model based. Our system has made so many accurate forecasts for so long on so many markets around the globe that I do not even comment on. It is far too much for me to even write about. That is the whole purpose of Socrates.
I had to design a completely different programming technique to work out the complexity. Just image calculating every market in the world in 35 different currencies. The number of variables is beyond comprehension. If we are talking about a limited number of variables for normal business operations, Neural Nets are fine. When it comes to market forecasting, they can develop a Deep Learning system on a single market, but without correlating this with all other markets, you will NEVER see the contagion coming. When everything crashed in 1998 because of the collapse in the Russian debt markets, the illiquidity caused funds to sell other assets to raise cash to cover losses elsewhere. The Russian bond collapse caused a massive sell-off in Japanese yen/dollar rate that had absolutely NOTHING to do with the fundamentals in Japan. Contagions always emerge externally so you can create a model and it will work for a while and then you lose everything.