Designing and Building a Profitable Automated Trading System

Palinuro, response in bold below:

Quote from palinuro:

ScottD,

Interesting work - thanks for the report. Are the test results improving as a consequence of these additions?

Yes, the results are improving per se, but not where they need to be yet. Our Dec 10 focus day, for example, is not a profitable day yet. There are certain areas where I am unclear, and I suspect those are the most promising areas for improvement.

Does 'awayness' represent MacD histogram > 0?

Nope. Think of the Rocket style thresholding. Also think of the absolute values so that you are accommodative of longs and shorts.

Your new pace table is in line with the NT 'autopace' results - did you ever find out why TN has lower levels?

As you know, calculating Pace over a non stationary window is actually very simple at a concept level -- grab the relevant RTH bar data, sort it from max to min, and then slice it into deciles.

I did my original Pace table by hand in Excel. Then I automated it. Then I found Vorzo's very nice version. Vorzo's matched mine within a very small variation due to rounding -- specifically the way that he chose to slice datasets where the number of elements is not evenly divided by 10 (decile slices).

If Trade Navigator shows significantly different pace levels than posted here as at Dec 10, 2008, then TN probably has a bug in the code or a problem in its historical time/volume data set. To do a precise diagnosis as to why TN values seem significantly too low, one would need to know if TN excludes any of the first or last bars of RTH (i.e. pre-sync and pre-end effect bars) and whether it excludes unusually slow periods like holidays. Also would need to verify that it uses 22 trading (not calendar) days of data. Once the algorithm is validated, one would need to compare the data sets.

& congrats on the recovery-
thanks!

- palinuro
More...
 
The attached dialog doc is the result of printing the entire thread and related diagrams, re-reading thoughtfully, marking up key concepts, and mulling over areas that require deeper thought and further dialog. (All while experiencing a nice level of physical pain, lol). The dialog doc is prepared in advance of doing the work on Release 0.5 of the Cashcow ATS. Certain Q&A also feeds creation of a standalone ChopHop indicator, lol, for wayward traders.
 

Attachments

Regarding the schematic that Jack posted, I had the following questions:

- Is Awayness based on the histogram or lines? How is it defined?

- Are Exits / Reversals subject to Suppression?

- Can Stoch14 generate a Reversal signal?

- There is a typo in the 'Long Exit / rev OR' logic box and the box right below it.
 
ScottD: For the absolute values of MACD used for suppression, why not run a step thru test on Tradestation of a range of values ? Some call it "optimization". Its a built in feature of Tradestation, as I'm sure you know.
 
Significant work has been done in the background, and we have reached a key milestone:

Jack has certified Cashcow ATS Release 0.7 as compliant with Beginner Logic, defined in the schematic.

It required 7 iterations to reach this point. Symbolically, we have renamed the latest to Release 1.0. Now we have a known, trusted code foundation on which to add further functionality and enhancements.

Latest chart attached with annotated trades. For design and testing purposes, the focus day remains Dec 10, 2008. On this day, the ATS generates profit of 22.75 points, or 0.9 x ATR. It’s a work-in-progress. Going forward, we will expand performance evaluation to multiple months of data.

Anyone with a track record of being helpful to others on EliteTrader is welcome to see the ATS code, just send me a PM. If you have a track record of mud wrestling, you can make amends by: doing the work to code up the schematic, sharing your results here, and asking clarifying questions. You may get helpful responses that way.

As a next step, we are integrating additional whipsaw suppression logic. In Release 1.0, there is a type of whipsaw condition that is able to get around the existing suppression logic. You can see examples of this in Bars 70, (73), and 75 on the chart. We will address that.
 

Attachments

Added whipsaw suppression logic to the Cashcow ATS.

ATS Annotated chart attached for Dec 10, 2008.

Profit : ATR ratio on this particular day improved to 1.4.

Next steps include finalizing the non-stationary window lookup tables and moving some actions from close-of-bar to intrabar.
 

Attachments

I intended to post yesterday's charts in near real time, shortly after each action. As it turns out, I was unable to do that, so I'll post the historical chart here.

In yesterday's forward testing, the Caschow Beginner Logic ATS generated a gross loss of 18.5 points.

There are a number of Logic modules teed up to improve performance as we move forward.

A really important thing to experience is that inductive curve fitting has no place in building an ATS. This system is based on deductive boolean logic.

As I'm able, I will try to post charts closer to the actual time of action.
 

Attachments

You must log in or register to reply here.
Back
Top