jwat

I checked out this very cool site. Thanks for the recommendation. I really like what they are trying to do. The fundamental problem writing very specific (coded) algorithms is that it is difficult to code up every situation that could develop. You are often forced to set arbitrary thresholds. Humans can be a bit more flexible in their interpretation of price data. However, there is the emotional baggage which can be detrimental to wealth...

There is a great site where you can test out technical analysis by comparing it to randomness -nonrandomwalk. You have to love the name. At the site, you practice trading blindly (you don't know what stock you are trading or what year you are trading it) on historical data and then compare how you did to a set of traders randomly trading the same stock. The results are often humbling. With regard to Malkiel, his experiment has a fundamental problem. Just because coin-flipping can produce classical patterns, that doesn't mean that the patterns in the market are random. One would have to look at the frequency with which patterns appear in coin flipping versus actual data. To my knowledge he did not do that. People who have (Andrew Lo) have found that chart patterns occur with demonstrably more frequency than would be expected if markets were just coin-flipping. Consider another example from a standard probability class. Imagine a thousand apes randomly typing on a thousand type-writers for an infinite amount of time. What is the probability that they will eventually, in combination, type the complete works of Shakespere? The probability is one. It is an interesting mathematical proof. Does this mean that the works of Shakespere were random?

I've done a lot of work with trying to use trend-following systems as applied to stocks (backtesting). They simply do not perform very well. I've read about the success of some trades applying trend-following strategies to the commodity and forex markets (like the Turtles for example), but those same strategies do not seem to work on stocks on the whole. I suspect this is because stocks (on the whole) are not as 'trendy' and tend to mean revert more often (versus commodities). Questions for the forum: Has anyone ever developed a technical indicator or other quantitative measure to determine if some stocks are consistently more likely to trend than others? Has anyone ever applied a comparitive study between the forex/commodity and stock markets with respect to how likely they are to trend?

I think you've got an interesting model/framework here, can you adapt it to do some simple forecasting? Try defining states, as you've done, develop some timeframes, and see if that maps to any future tendencies in price. It seems like something that would be fairly easy to backtest.

I've seen a bunch of applications that do this that you have to pay for, but I discovered a free web app that lets you do this at www.nonrandomwalk.com I like how it forces you to trade blindly and evaluates your performance. There is a lot of bad advice out there in technical analysis world and you definitely need to test out what works and what doesn't.