[UrmaBlume]

What is present in many commercial automated systems and missing from almost all "retail" developed and operated automated systems is that the retail systems are not adaptive.

So many retail traders have done the tremendous amount of work to develop, backtest and optimzie a profitable trading system only to see if work for a while and then fail. This failure is oftentimes not a failure of concept but a failure to make the parameters of the system adaptive/responsive to current/changing market conditions.

For us this is basically a 2 step process - first we take our measure of the maket from the most appropriate of many inputs and then the software makes the appropriate adjustments to system parameters. Different inputs from different time frames adjust different paramters.

Our read of current market conditions is just about the only place where we use time charts. The chart below is our measure of the percentage of average commercial commitment to the market in 6 different time frames ranging from 2 minutes to 405 minutes.

All of our inputs are normalized for time of day. We divide the day into 405 (the length of the day session in minutes) segments for EACH time frame. For example we take a reading of commercial commitment for 60 minutes at 1130 that reading is compared to the average of the 60 minute value at 1130 for the previous 62 trading days. At ll31 we do it again and we do it for each of 12 time frames in at least six different indicators.

We display some of these measure of market condition in the HUD.

We consider the percentage of commercial commitment shown below as well as total volume, price ranges and certain internals from different exchanges. All of these inputs/measures of current market conditions are normalized for time of day and scaled to percentages of average to facilitate their inputs to certain predictive technologies such as genetically optimized neural networks and MARS (Multivariate Adaptive Regression Splines) from Salford Systems in San Diego. Of note is that after training and development an easily constructed script can auto convert the MARS function to Easy Language for quick and easy deployment and real-time predictions.

Different levels of these indicators make adjustments to parameters such as speed and phase adjustments to different data smoothers such as the JMA as well as stop losses, price and volume targets and take profits.

If you want it to last - that is, if you want it to Live Long and Prosper then make it adaptive.

[Tams]

Originally Posted by UrmaBlume »

...
If you want it to last - that is, if you want it to Live Long and Prosper then make it adaptive.

didn't Darwin say something like that?

[UrmaBlume]

Originally Posted by Tams »

didn't Darwin say something like that?

You are right again Tams and so was Darwin.

We have seen a lot of systems and this lack of adaptive parameter settings is a VERY common weakness. Adaptive - not optimized.

[zdo]

UrmaBlume,

well, DUH...

zdo

[zdo]

UB, we struggled with adaptivity via time of day for quite a while… surprised you find it useful… finally went back exclusively with MarketTyping algo’s that are good around the clock and applicable to any instrument… use it to move the weighting of systems around much more than to push parameters though...

[UrmaBlume]

Originally Posted by zdo »

UB, we struggled with adaptivity via time of day for quite a while… surprised you find it useful… finally went back exclusively with MarketTyping algo’s that are good around the clock and applicable to any instrument… use it to move the weighting of systems around much more than to push parameters though...

The efficacy of time of day normalization depends on the input - for some it adds value and others not. As to the weighting of systems within a family of systems all trading the same market we consider that to be just another parameter along with those mentioned.

What you said, well DUH ZDO

Originally Posted by zdo »

UrmaBlume, well, DUH... zdo