Why would it fail during a bear market, it's market neutral. Not to mention that it will benefit from increased volatility (wider/faster swings). The reason why your backtest is failing could be as simple as you're not calculating the minimum swing correctly.
Gary, We had some precipitous drops in 2002, and the radical downward swings, and resultant cashing in the chips at market, seem to have made this a loser. Perhaps my statement of it not working in a bear market is a bad comment. What I meant to say is that in the bear market of 2002, based on my interpretation of the algorithm, the system did not fare well.
abogdan, As was previously pointed out, you can't just open up a B/A file in WL and display a chart. As was also pointed out you can import the data and format it in such a way as to make it available for testing. As an aside I quickly wrote a small VB program that formats my B/A data for WL use. I am more familiar with VBs file abscess routines than in pascal so I went that way. In either case (since you are VB capable) testing B/A strategies with WL is easy to do. On to Hayman. I have no capability to state this as fact but I think it is a fair conclusion from my knowledge of the system and having programmed it a bunch of differing ways and studied the results: The system is fine on bear or bull trend days. Its the oscillating days that have killed it so far. My guess is that in your testing for whatever reason there were more flat days during the bear market time. I would be surprised if there is anything inherent to a bear market that would effect this. I have not proof of this...just opinion. JB
ig0r: >The reason why your backtest is failing >could be as simple as you're not calculating >the minimum swing correctly. Hey Gary, forgetting the above for a moment (which may or may not be true), do you have any results that disagrees with the overall system unprofitability that Hayman was proposing? It would sure be great to see it if you did. JB
No, I certainly don't. If I have time next weekend I'll convert the TS code into WL and do some tests myself. I am basing my view of profitability on abogdan's success with the system combined with the TS backtest
hayman: I just reapplied TS code to 900 days. I did not see anything unusual. It was still positive. I also did not reoptimized anything, just applied to historical data started Aug, 2000. I'm trying to use WLD now to present more or less complete system to you guys. As you know, I have never done it before. All my systems were written in VB6 with the back end (B/A data conversion) in C++. I used eSignal API to download real time and historical B/A data with sizes. All the optimization programs that we are running are written in visual Prolog. For all my digital filtering we used DASP tools. In real time we actually use AMD boards. It might take a little while, but I'm sure I'll come up with something. Also, as soon as we proved this algorithm I'm planning to move on to the next step were we used correlation filters to further reduce flipping. But let us finish this step first. Regards,
Related thread...testing results presented for discussion. http://www.elitetrader.com/vb/showthread.php?threadid=27978 JB