10 July 2010


This concerns a paper by Jingtao Yao et al: "Foreign Exchange Rates Forecasting with Neural Networks". Neural Networks are computer programs which find underlying structure in data, and from this generalized structure, elaborate behavior for new data.

They are based on very simplified models of the way the brain may operate. The work I did was in optimization algorithms which find solutions then find better solutions. There is some evidence the brain may do something like this. My trading on the 1 minute chart which has yielded some interesting results, was based on this.

Chess programs based on these algorithms beat the best chess players in the world. I looked at distributed versions of these. It is basically the structure of problem solving (i.e. one does not worry about the structure of the brain, one looks at the structure of the problem solving act itself).

It does seem the market is a powerful network of problem solving activity, distributed and focused, as getting it wrong hits you where it hurts - your wallet. It may be this focus and distribution makes the collective expression of this precise, which gets mistaken for efficiency.

When Yao applied neural network programs to the forex market he got this result: the worst predictions using technical indiactors applied to the derived structure was for Yen. This is what interested me. Yen is a huge liquid market referenced to flows to and from Japan, another artificial conduit, though based on long term low interest rates.

Yen survives because it is a safe haven currency amongst other reasons. This is the worrying possible future for the dollar though. It is all based on deflation and the effects on asset values, pensions etc. From this, a political necessity driving central bank policy.

Yao thought the reason for the finding was the efficiency of this market, it moved from moment to moment digesting all information, there is simply no room for predictive inference. I might suggest though the distortion caused by the long term conduit may be a factor in this.

Note how the conduit set up between EUR/USD has caused valuation shocks, these look like chaotic surges, or Hurst bias changes, which make prediction impossible (the reasons behind all those inaccurate calls in forex recently). But because there is some evidence there are no long term memory processes in forex, one might think the conduit is the explanation in and of itself.

© 2010 Guy Barry - All Rights Reserved.