Jumping

Jumping between two levels and many sublevels for many days now – starting at the very bottom where price of one security X is the only cause and effect. Prediction of X given the non linearity of prices is not an easy task and many models from time to time have been suggested:
-simple brownian motion
-more complex markov models
-garch/arch models
-kalman filters
-neural networks/genetic models
-vector quantization/information theory

A very simple approach will be to use high low regression divergence and any movement beyond say 95% confidence interval should provide a trading opportunity.

Now taking the 50,000 feet view, world can be defined as a vector space of n dimensions with the various dimensions being important economic variables such as interest rates, currency, inflation, commodity prices (oil/gas/gold), money circulation, broad market index, credit growth rate etc ( and of course the momentum/rate of change of many of these). The price of the security X is a function of the state S of this vector. Now as the state changes from S(t-1) to S(t), the predicted value Xp(t) and the observed value Xo(t) are compared and any difference in them provides a trading opportunity. This, while may seem like mean reversion, its now, because there is no mean here to which the price is expected to revert.

While combining the micro with the macro may sound like a great idea there are infinite challenges to implement this – even for the macro prediction, one can begin to ask if GARCH model is better then the Kalman filter. Then there may be divergences in predictions in the micro vs macro data points, which may be simply timeframe issues as some variable are sticky and move slowly, while momentum is an important factor in price movements.