Do you see patterns in Random Walks?

Quote from ssrrkk:
Statistics are used to estimate population parameters from samples. Probabilities are assigned for inference, either for predicting future values or assigning confidence / significance for parameter estimates.

Bayesian inference is one way to equate past statistics with future probabilities.

In the context of statistical tests, I am referring to the case where probabilities are used to assign confidence or significance to the parameter estimates. So if your past statistics indicate that 80% of the time, a certain pattern resulted in a profitable upswing, then I would like to know what is the 95% confidence interval on that 80% statistic. If it is 80% +/- 40%, then I am less confident about betting on it. Alternatively I could come up with a hypothesis test and ask: is this 80% (profitable) significantly different from 50% (not profitable, money-losing with commission+slippage). The null hypothesis says it is indistinguishable from the 50% case. If the p-value from this test is 0.3, then that means that 30% of the time, you could have gotten the 80% by chance, i.e., that it is just a fluke, there is no mechanism behind it, or it is a curve-fit.
More...

+1, good post:) , Bayesian Inference sounds useful...have you applied it?
 
Quote from CoolTraderDude:

Forgot to add... While markets may have a fractal nature I don't actually believe that you can predict anything with fractals.
More...

thanks for the links. From my limited understanding I would say, if anything, the market is multifractal, but then again this would be a more general discription and there are better and more complete statistical models that do a better job at explaining the data generating process.

I would also say that as far as predicting goes, as long as you can better predict the distribution (though unobervable) in the future, then you're able to better monetize market features.
 
Here is an interesting pattern in SPY as of last close from the price action blog with one showing only in the history and zero showings in some other markets. The author tries to provide an explanation for that. Anyone else has one?

http://t.co/mEsKnNQw

Do you think this a a random pattern or it has some meaning?
 
Quote from phattails:

thanks for the links. From my limited understanding I would say, if anything, the market is multifractal, but then again this would be a more general discription and there are better and more complete statistical models that do a better job at explaining the data generating process.

I would also say that as far as predicting goes, as long as you can better predict the distribution (though unobervable) in the future, then you're able to better monetize market features.
More...

I agree...!:)

Quote from intradaybill:

Here is an interesting pattern in SPY as of last close from the price action blog with one showing only in the history and zero showings in some other markets. The author tries to provide an explanation for that. Anyone else has one?

http://t.co/mEsKnNQw

Do you think this a a random pattern or it has some meaning?
More...

"Double outside of inside day..." Rare indeed... I think his explanation is pretty good. Traders bought too much then froze up... Went down some and then they frantically re-bought the dip...

I don't know if it has any real meaning though... If you change time frames you won't see the same thing.
 
Quote from Antisyzygy:

This thread is sort of dead but here is an interesting link.

http://www.bearcave.com/misl/misl_tech/wavelets/hurst/
More...

Thanks for thie excellent reference but I think there is a fundamental flaw that makes all this work a waste. Trends can show up in random data too. Just toss a coin and you will see that. Go to "price action lab blog" and search for the post "Fooled by Randomness Through Selection Bias". There is a link to an applet where you can generate all kinds of nice looking trends by tossing a coin that can fool you in that the market is not random.

This is what I got after six trials for 1000 tosses:

attachment.php


Again, the fundamental flaw is that trends deny the random-walk theory. There are random trends and deterministic trends. The Hurst exponent cannot tell you which is which. This is the flaw. It is simple. I wonder why those people do not think of that before doing all that hard work.
 

Attachments

Quote from intradaybill:

There are random trends and deterministic trends. The Hurst exponent cannot tell you which is which. This is the flaw. It is simple. I wonder why those people do not think of that before doing all that hard work.
More...

just by saying 'there are random trends and deterministic trends' it seems like you admit that you CAN TELL which is which and 'which is an illusion' .. how do you do that? if there wasn't a method one wouldn't be able to tell the difference between the two..In other words how do we know that 'There are random trends and deterministic trends' ? In the market that is..
 
Quote from vinc:

just by saying 'there are random trends and deterministic trends' it seems like you admit that you CAN TELL which is which and 'which is an illusion' .. how do you do that? if there wasn't a method one wouldn't be able to tell the difference between the two..In other words how do we know that 'There are random trends and deterministic trends' ? In the market that is..
More...

Please not not distort what I said. I said there are deterministic trends and random trends. What does that relate to the market? I don't know. I just argued that finding a trend in the market does not imply the market is not random. What you do not understand from the above?

Just to make it clearer for you: I know there are trends in random data. Then, finding trends in market data cannot imply the data is not random. All that math check the presence of trends. To claim the trends imply no randomness it is equivalent to claiming that the trends in market data are deterministic. That is their claim, not mine.. Get it?
 
Quote from vinc:

just by saying 'there are random trends and deterministic trends' it seems like you admit that you CAN TELL which is which and 'which is an illusion' .. how do you do that? if there wasn't a method one wouldn't be able to tell the difference between the two..In other words how do we know that 'There are random trends and deterministic trends' ? In the market that is..
More...

Stats 101: statistical significance. You can set up a hypothesis test, where the null hypothesis H0 is that your proposed trend is random, and H1 is that it is not. You must then somehow construct the expected background probability distribution of the random case (there are two ways to go about this: one is to go back to the fundamental distributions of your process and build it based on that; the other is to empirically construct it using MC sampling). You then compare your proposed trend with the background (random) probability distribution. If your trend falls within the extreme tails of your random distribution at level p=0.05, then you can reject H0 at that p level (i.e., there is a 95% chance it is not random).

Of course, there are many ways to screw this analysis up and trick yourself into thinking that you have something when in fact there is no statistical significance.
 
You must log in or register to reply here.
Back
Top