A great book on testing and optimisation - page 7
You are missing trading opportunities:
- Free trading apps
- Over 8,000 signals for copying
- Economic news for exploring financial markets
Registration
Log in
You agree to website policy and terms of use
If you do not have an account, please register
1. Forming a trading strategy in the form of a flowchart. ........
This is a very old and very effective technique to sift out everything left-handed. Somewhere on a 8" floppy disk, you can find a program that creates flowcharts on DEC computers.
Used in large teams.
Very good at identifying those who have not done their part.
One problem, if block diagrams are drawn "for yourself", they begin to reduce blocks and drawing itself loses its purpose,
i.e. ceases to be informative.
In search of grail drawing of flowcharts of course would help that 98% of ideas would be where they should be
- in an urn. However, who will draw it, the market is leaving!
Just for this kind of research, the 'Test and Optimization Management Software' was developed.
Based on the tasks described in the book Mathemat kindly provided
book (many thanks to Mathemat) the macro-programs allowing to make forward test were created...
Too bad, I'd love to hear the opinion of the "Philozoan sceptic" :-)
Continuing "about nothing", i.e. about optimisation and evaluation of results. Up to this point I have finished on the PROM parameter. I won't show the formula here, it can be found in the book on page 94. There are two other criteria based on this: "PROM minus maximum profitable trade" and "PROM minus maximum losing series". The author calls the latter the most reliable evaluation parameter, as it removes the influence of the most exceptional profitable series ("prepare for the worst"), and suggests ranking the results just by it.
Then there is the standard, well-known reasoning on how to find the best run we want out of all the optimisation results (the optimum must be surrounded by close values for it to be robust). After that, p. 98-101, the author outlines a technique for evaluating complete optimization results in general, and shows in which cases these results can be considered successful and in which they cannot. After that he concludes the preliminary chapter on optimization by reasoning about test space (for a single optimized parameter it is a curve of results with preferably flat extremum).
Chapter 6 is entirely devoted to a practical example, in which we again go through all the steps up to optimization (not including it). I will not repeat it but I can refer practitioners straight away to this chapter.
Chapter 7 is optimization together with forward analysis recommended by the author after multiperiod/multi-market test. This is where it gets serious.
1. parameter selection. It is clear that we have to select the parameters that have the maximum impact on the TS efficiency ("significant"). Those that have little effect on it, it's better to fix.
2. Choice of scanning range. Here everything is clear: If our system is a short-term one and is based on the moving averages, then in this range it is undesirable to include the moving averages with a period of 1000, because it is rather a long-term moving average for this TF. Step change - also everything is clear: the more steps, the more time is needed for the whole optimization. In addition, the author claims that too small a step can lead to a larger fit.
3. Data selection. We should, firstly, ensure a sufficient sample size (statistical validity) and, secondly, include a sufficiently wide range of market conditions in the data. The number of degrees of freedom (I mentioned it earlier) should also meet the appropriate criteria, so as not to run into a false fit. See picture for degrees of freedom:
4. Selection of a test criterion evaluating a single run. This was also discussed (e.g. PROM without maximum profitable series).
5. Choice of method for evaluating integral results of optimization. First - evaluation of statistical significance. Again a picture:
Second evaluation method - by test space. Picture:
If the system trades on a volatility breakout, a very good result curve is shown above. A flat maximum and profitability. This curve is quite promising.
Further the author speaks about multimarket multiperiod optimization. The author's justification is the higher statistical validity of the model. Frankly speaking, I have my doubts about multi-market part of such optimization if the model will initially work in one market. But the author doesn't insist on that either. It is imperative to test on different markets if a portfolio is to be traded. Picture:
How to combine all these tests is not clear to me yet. Probably do them independently and look for a common optimum.
Further there are considerations of integral estimation of optimization from the viewpoint of the system efficiency. I do not want to repeat myself as we have already briefly mentioned all this. If as a result of optimization the average TS effectiveness corresponds to the specified profitability criterion, has acceptable risks and wins compared to other investment opportunities (say, T-bonds), then the system is good and it can be subjected to the last test - forward analysis. Stop. Rest.
I downloaded this book in djvu format from ihtik.lib.ru under economy.
The link is: http://ihtik.lib.ru/economy_21dec2006/economy_21dec2006_495.rar
The link is broken.
I can share the unbroken one: http://bigfx.ru/load/8-1-0-4
It would be nice if at least 5% of forumers would not just download this book, but also read it... It could significantly improve the quality of testing/optimisation compared to what we see now. Otherwise, we are fed up with these endless cardboard grails and the bewilderment of their authors, resenting the results in real life ... .
When they download, and MAYBE read, there will be even more questions ;).... some will start counting coffee and gold......
Yura, thanks for the link. It's been a long time since I've looked at this topic.
2 rider: it's unlikely that 'when' will come.