Modelling market mode (trending vs about to reverse) I am interested in stochastically modeling whether the market is likely to go on in the same direction(trend), or reverse and head back. This is all for intraday purposes, next 1-2 ticks kind of strategy with 30sec - 3 mins holding times.
How can I attack this problem? Where do I start? 
 A: Here's an R model of "trending and reversing":
#Plot a random time series
x <- cumsum(rnorm(1000))
plot(x, type="l", main="Trend versus Reverse")


Notice that you'll always get trending and reversing, 100% of the time (run the code over and over).    And, those so-called "trends" and "reversals" were generated from a random sequence.   So, how can you tell the difference between "trending" and "reversing" that's just random, versus "trending" and "reversing" that is due to some trading situation?     If you think about it for a while, you'll come to the conclusion that you can't.   The whole concept is a waste of time (and it's not just trends/reversals that are a waste of time).
A: Finding turn points is a very difficult problem.  Having a covariate that is correlated at a leading time would help.  Covariates in conjunction with noticing a leveling off in the market index would be even better.
A: This is something that I'm looking into at the moment, although I'm using using daily OHLC data rather than tick/intraday data. The approach I've taken is to postulate 5 market types and then generate a large amount of synthetic data that conform to models underlying each respective market type. Using this data, I'm currently training a classification neural net to identify which model/market type "real" data most closely resembles over a specified look back period. Based on this classification one would then choose the most appropriate trading strategy: attempt to pick tops and bottoms in a mean reverting environment, and go with the flow in a trending environment. I'm blogging about my progress with this project on Dekalog Blog. 
A: I would look at 2nd and 3rd standard deviation Bollinger Bands and take a regression to the mean approach. Another approach would be candlestick analysis. Combine either with volume data obviously. If it's hitting new highs but volume doesn't confirm probably expect a reversal.
A: There have been studies that show that technical analysis "works" ie better than random results. http://en.wikipedia.org/wiki/Technical_analysis#Empirical_evidence. 
Most academic studies have been conducted by people that have never actually traded anything. Whatever method you're using has to be taken in the context of the overall market. A rising tide floats all boats. 
