Your forecast page as of July 16th says that you are "working on" a new
experimental model that does well usually but fails in times of high volatility
. . .
I wish to see if I understand this correctly. You have several dozen models
working at any one time, and for a model to do inputs, it must maintain a certain
accuracy % rate. If models are more accurate, they get a higher weighting. This
new experimental model is one that would be added to the list of several dozen,
if you can make it work with some reasonable accuracy?
How would this model have done in the 3 major recent times we have had
drawdown while the ANO was > .2, speaking of mid-January into feb and then early
May, and then, again in mid-June?
This model is completely different from our standard forecast model. It
is a numerical experiment to test an idea of forecasting the length of market
cycles rather than the market's direction. It doesn't use neural networks and
it doesn't employ evolutionary learning models. It is also not a regression analysis.
As I said on the forecast page, it does very well for long periods of time but
fails during times of high volatility. I think now that it may be possible to remove
this weakness but more coding and testing is necessary. When I am further along,
I will do some serious backtesting and will report results.