I'm trying to find a topic for my undergrad dissertation and technical analysis came into mind. Only started reading up about it but to my horror, it has been proven that technical analysis is proven to be pretty useless. However I've not actually seen a paper that actually do a comprehensive of all the indicators, not to mention one that tests the performance of combinations of indicators. Anyone has any insight? It was said that TA is superseded by you quants (not that I know what you guys do anyway... ). Is it true? My sis is a dealer and she's pretty much still drawing lines on her little charts.Comments? Cheers!

*ahem*

Please see my blog.There is a post regarding the relaitionship between Technical Analysis and Quantitative Analysis.There is also a link to a wonderful post here in Wilmott forums on a debate for and against TA and QA.You might find some interesting topics to write about. MyBlog

Not all techinical indicators are useless. See the attachment.

QuoteOriginally posted by: JeriotI'm trying to find a topic for my undergrad dissertation and technical analysis came into mind. Only started reading up about it but to my horror, it has been proven that technical analysis is proven to be pretty useless. However I've not actually seen a paper that actually do a comprehensive of all the indicators, not to mention one that tests the performance of combinations of indicators. Well, you're not going to see "comprehensive of all the indicators" . It would be massive data snoopingas people invent indicators after the fact. But, -lots- of basic tests were done back in the 50s and 60s, untilthis type of research was pretty much played out. For surveys of it, seeRichard A. Brealey, "An Introduction to Risk and Return from Common Stocks", orP.H. Cootner (ed), "The Random Character of Stock Market Prices".regards,

QuoteOriginally posted by: JeriotI'm trying to find a topic for my undergrad dissertation and technical analysis came into mind. Only started reading up about it but to my horror, it has been proven that technical analysis is proven to be pretty useless. However I've not actually seen a paper that actually do a comprehensive of all the indicators, not to mention one that tests the performance of combinations of indicators. Anyone has any insight? It was said that TA is superseded by you quants (not that I know what you guys do anyway... ). Is it true? My sis is a dealer and she's pretty much still drawing lines on her little charts.Comments? Cheers!I'm not a quant ,but an interesting to do would be to do "extended" data analysis ,say you pick up 15 classic indicators and 5 exotic indicators ,you make some combinations(e.g. SMA of RSI crosses another SMA of RSI ,or STO of d(price)/dt) ,you take several timeframes ,and you display the results of the systems monthly(6-12 months of 5min data should be enough).After that ,try to answer to a few questions :1]Does TA give an edge? (answer is yes on the credit suisse report)2]Is it better to keep the optimized timeframe ,time value after the fact or better to trade several timeframes3]Should we have several trading systems in place or put all the eggs on the best horse(study the max dd and the volatility of the systems monthly)4]You dump half of the "systems"(the ones that did poorly) ,one "system" is constituted of a system + a timeframe ,so you could dump system stoch with timeframe x but keep system stoch with timeframe "y" ,after that you backtest the systems who survived on out-of sample data.Even a more interesting thing to do is to backtest the "systems" who did poorly in order to see if systems selection can be useful or if results are just "random".5]you combine the systems who survived into one indicator and backtest on out of sample data.6]you redo all of the above with more complicated stufff ,more indicators ,higher frequency data(ticks) ,Monte Carlo analysis ,etc...7]you stop wasting your time

One project that might be interesting is to see if you can sucessfully identify the "patterns" oftechnical analysis. For example, if you check out the encyclopedia on chart patterns by ThomasBulkowski, you find a lot of supposed chart patterns like "Broadening Bottoms" and "Bump andRun Reversal Tops," and the like. While the pictures seem to define clear patterns, is it in fact really possible to write code that canrecognize these patterns in a way that does not depend on a subjective human observer? Are theyreally meaningful patterns that can be unambigously picked out of messy tick data? Some thoughts...

I have been studying classical books on technical analysis, e.g. Murphy etc, and I dont think their stuff is 100% applicable anymore (not in interest rates at least). However, the main ideas of technical analysists, primitive as they may seem to a quant, are really interesting. It often reminds me of basic techniques in econometrics...just that they are applied with pencil and ruler. If you get the basic concepts, you can start constructing your own indicators and you will be noticing patterns yourself. At the very least, TA makes you think a bit more about what is driving the market in the near term (hint: it is NOT brownian motion, it is NOT random, and it is NOT rational), it gives you a feel for what constitutes high or low, rich or cheap, and it will give you a more intuitive sense for what is going on. I use TA in micromanaging the risks in my book and almost every other trader I have met is doing at least support and resistance lines (even on derivatives desks). I do not believe in any long term predictability (e.g. elliot wave theory), but I strongly believe the market is technical at very short horizons (especially intraday).If you need a start into computer driven TA, lookup a paper written by Andrew Lo of MIT. He investigates head-shoulder formations using pattern recognition. It is very interesting and it concludes that these patterns may actually work.

Very interesting topic and one I have been mulling over for a while. Not sure I'm ready to give up my disdain for some of the more voodooish aspects of TA but there are definitely some interesting links. For me the most interesting is the direct linkage between Hurst Exponent - its relationship through the Fractal Dimension to the Stable Levy distribution and further relationships to things like Fibonacci, Malahanobis distance and Frechet Hoeffding bounds for eliptical distributions. If the 'true' trend / de-noised signal is all that is important then technicals may have a lot to offer but I have yet to see a system that consistently calls the change before the fact. excessively high Hurst scores may be related to Sornette's log-periodic necessary but not sufficient conditions for a bubble / blow-off but still much work to be done here. The problem with signal extraction is that it seems to require high freqency data and that way lies madness as some others have already found after much effort. Will probably hire some people with a background in Sound engineering and DSP nonentheless.