Disliked{quote} Good question. I used historical data to test theories. If they looked appropriate, then I would use months and months and months of demo trading the live market across multiple currencies to fine tune it. My objective was consistency. It took a long time because I kept trying to improve the precision using the live market. I spent free time reviewing and creating better ideas, but always tested them live to be sure they were accurate. When I finally got to the point where the precision broke, I knew where to reverse it to. I found that...Ignored
From my perspective it is about using a lot of data, forward testing on demo or otherwise is not efficient, because by its nature it must run in real time.
I reason that using historical data is just as valid as using data going forward because both are unknown, providing care is taken not to fit the data to the curve, and the advantage is that using historical data provides the ability to test a lot of setups very quickly. My approach is to use one year of data to test a theory and adjust parameters, and then test the same algorithm on as much data as possible, in probability theory it is really necessary to test at least 10,000 events to determine whether a correlation is "reliable", it is also possible and desirable during this process to understand the maximum draw down from peak of a strategy. Having passed back testing then I would run a system forward with small money to check whether it performs as my back testing would suggest.
I've posted this before, but here is a test I have run for a Spot market system that will go into Forward test soon. I have tested the method all the way back to 2003 and the results look remarkably consistent, with the only minor losing years around the 2008 financial crisis.
Attached File(s)
Zen Testing.pdf
62 KB
|
158 downloads
I live entirely from the growth of my personal trading account