Fractal confluence in smaller timeframe occurs sometimes due to stochastic streaks, not because of mass behaviour confluence, other than major intervention, of course. http://wizardofodds.com/image/ask-the-wizard/streaks.pdf
It is funny that you quote the paper from the last thread. Do you have any others? Recent? Moreover, did you read the paper carefully? It appears that you did not. Here some ponts from the conclusion you should pay attention to "the absence of strong dependence in stock returns should not be surprising" This is not the same as absence of memory. You and the authors probably confuse memory with strong dependence. "if some form of long-range dependence is indeed present in stock returns, it will not be easily detected by any of our current statistical tools" This is because statistical tools look for strong dependence only and memory is a much broader spectrum of possibilities. LTCM was a good firm too Your premise is irrelevant.
Chapter 6 proves that there is no long term memory in stock market prices. However, in this same chapter he does point out a significant shorter term autocorrelation that he says is on the order of a month or two. The chapter is concerned with perceived business cycles over the course of years and decades, which he is disproving. Chapter 2 linked earlier disproves that stock markets follow a pure random walk. He points out again that there is significant autocorrelation in the weeks to one month time scale. So I think the conclusion is there is no long term memory when long term means many months to years (i.e., the change of an asset price in a 6 month period does not depend on the change in that asset price in the previous 6 month period). But there is definitely short term memory (i.e., the change in price over the day's trading is dependent, albeit weakly, to yesterday's change in price). Here is another article that agrees with the many points of the Lo chapters: http://www.efficientfrontier.com/ef/199/momentum.htm
I like this discussion. I also like when people argue while maintaining the respect for each other's opinions. For the record, I do understand and respect your point of view. About 15 years ago I had very similar views on the market and was hoping to extract useful information out of the price series thinking that there might be a short-term memory that could be exploited. Unfortunately, my research has led me to believe that there are no exploitable patterns or traces of short-term memory in the price movements. Of, course, I could be wrong. If you have any kind of experimental data that might indicate the opposite I would definitely be interested to look at it. Cheers, MAESTRO
I am very familiar with the article. In my opinion there is a slight confusion between the RW types and their implementations as models for price behavior. This subject is extremely fruitful for anybody who wants to have an edge in trading. From my 20 years of experience I can tell you that there is hardly anything that is more productive than RWs for the algorithmic trading. Thank you for your valid and well thought through points. Cheers, MAESTRO
This thread is one of the very few threads on ET that is actually pleasant to read. I am glad that the most of participants exhibit different points of view but generally refrain from mutual assaults and childish remarks. I urge all of the participants to continue this very valid discussion and keep this thread going the way it has been for the past few months. Happy New Year!! I wish health and prosperity to everybody in 2012! Thank you for your time and contributions to this incredibly important topic. Cheers, MAESTRO
Some of you must try harder to understand mathematical procedures and assumptions. Any conclusion in statistics is only "in the sense of the specific statistical method used". A statistical method applies some metrics to data. The results obtained are good only in the sense of those metrics. Furthermore, nobody - not even the stupidest trader in the world has ever insisted that stock prices have a six month memory or in that order. Patterns are not directly related to memory. Patterns form and exist above and beyond memory in price series. They are results of trading behavior, not artifacts of series returns. Wake up people before you get too old one day and find out you have missed the train. Happy New Year.
I guess you are saying Prof Andrew Lo must try harder, LOL. I just re-stated what he wrote in his chapters. I agree that it is rather absurd to use autocorrelation as the sole measure to represent the concept of "memory". I believe one needs to carefully identify the cases in which certain patterns work, sort of a conditional probability or Bayesian type approach. Of course, the reason academicians focus on the pure fundamental and simple estimators is because one can apply it to a large sample size to try to prove or disprove statistical significance. The truth is, in the business world, it is rare or impossible to find a significant trend that hasn't been exploited. We are left with patterns that may or may not be significant, but there is no way to prove it's significance given the amount of examples we have. In other words, making money will always entail risk.
I would agree with that. Of course, the whole point here is to find those patterns. I have always found mine in the different types of price/time distributions. I am not saying it is the only right approach; markets are like chess - there are unlimited number of potentially successful strategies in them, however, everyone has his/her preferences. Mine are Random Walks! Happy New Year!