True AI and Fake Neural Net Forecasting Programs


QUESTION: Hello Marty,just came across an interesting project called Numerai

with the ambitious goal of creating a hedge fund to manage ALL the money in world(!) in an efficient manner using AI (machine learning, neural networks and whatever the data scientists can come up with).

At the core of this project is the historical data made accessible for free to the data scientists in encrypted form, but still usable since the structure is still there. The data scientists then compete with their AI methods against each other in auctions when evaluating their methods on a to them at auction time unknown validation set (latest data set).

The hedge fund will be the combination of the best AI methods driven in a purely evolutionary fashion with the purpose to allocate the money to where it is needed best. A nice feature is that the AI methods developed and put to use by a data scientist can be kept secret from the other data scientists and the hedge fund itself, The data scientist only needs to make the output of the AI algorithm available, pretty much like you do with Socrates :-).

The data scientist can always take his/her algorithm elsewhere if not happy with the hedge fund.

I think this sounds like a really innovative idea, but there might be some pitfalls that you would be able to highlight. For example I would expect your Socrates database to bigger than the Numerais at the moment.

Best Regards

/M

ANSWER: The problem with such a project is that the best of genius will fail. The complexity of the markets globally is far beyond description. Even the Global Market Watch is best in the financial main markets and less in agriculture and individual stocks. Why? It is all complexity. The GMW has now exceeded 150,000 possible patterns each day and counting. I just uploaded the latest 6 months of creation of patterns. It has blown my mind looking at what it has done. Moves greater than 40% on the upside it reclassified as phase transitions.
I have experimented with Neural Nets and other interesting attempts to recreate the structure of the human mind. Neural Nets are notoriously subject to complete failures. There is no such program that has been able to demonstrate consistency. This is all due to complexity.
The effort to create option models known as BlackScholes completely broken down and resulted in the Long-Term Capital Management debacle. That was covered in the book – When Genius Failed. The failure took place because they did not have the database. They only tested the algorithm back to 1971.
There is no such computer model that can possible accomplish that and quite frankly anyone claiming to have some Neural Net Artificial Intelligence you should beware because it is just a sales pitch. Ask who wrote the code. You will find no real system other than a simple expert system based upon IF this THEN that ELSE expect this.
I was recently approached with a plea to buy into our technology by a very major company offering unlimited resources to help expand it to all sectors. I must admit, that is an interesting concept.
Anyone who claims to have such a system is pulling a sales promotion. You cannot rely upon a program to forecast the future on the cheap without a profound and in-depth database. The British Pound for example, all the data even in major data banks go back to 1971 and many only on a monthly closing basis. You cannot forecast trends without the full picture. A computer will not be able to learn all scenarios if the data does not cover all scenarios. Plain and simple. Just ask – SHOW ME THE DATA and WHERE DID YOU GET IT? 

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s