Can linear regression analysis really predict the future?

This isn't so much to generate debates or flames, but for those whom kindly PM'd me in the past asking for the 75% rule in simple terms, I did a small writeup and shared some R code. Hopefully, it changes the way you see random walks a bit.

One of many insights I've had (in this case public, but often overlooked) on the issue of gaussian based random time series.
Please PM me with any egregious errors, but proofs of the basic concept are contained in the book I mentioned on the writeup.

Cue... yeah, but markets aren't gaussian... 10...9...8...:D

http://intelligenttradingtech.blogspot.com/2011/03/can-one-beat-random-walk-impossible-you.html
 
Quote from dtrader98:

This isn't so much to generate debates or flames, but for those whom kindly PM'd me in the past asking for the 75% rule in simple terms, I did a small writeup and shared some R code. Hopefully, it changes the way you see random walks a bit.

One of many insights I've had (in this case public, but often overlooked) on the issue of gaussian based random time series.
Please PM me with any egregious errors, but proofs of the basic concept are contained in the book I mentioned on the writeup.

Cue... yeah, but markets aren't gaussian... 10...9...8...:D

http://intelligenttradingtech.blogspot.com/2011/03/can-one-beat-random-walk-impossible-you.html
More...

Increments of Uncorrelated Time Series Can Be Predicted With a Universal 75% Probability of Success, D. Sornette:

http://arxiv.org/abs/cond-mat/0001324
 
Quote from tradrejoe:

For those of you who went through the exercise of using historical data and linear regression analysis to predict the future prices of trading instruments, have you ran into situations where the best beta coefficients that generates the best curve fitting *does not* really predict the future? In fact, often times if you go back in history and pretend you were operating the prediction system in the past, the more testing the more your accuracy converge to just 50%?

What is the correlation between the ability of a set of time series data to fit a price curve and its ability to truely forecast the future with greater than 50% accuracy? Do we just pile up everything closely related to what we try to forecast (even sun spot movements) and go as far back on the time lag as we can without crashing the supercomputer? Does anyone have any experience to share? Thanks for your insights ahead of time.
More...

Nope, it won't be useful. Any possible correlation would be spurious.

Tom
 
Quote from jem:

The fed study spoke about bounces on round numbers. Round numbers are technical points.

How is that different?
--

By the way I expected the response because I knew that chart would be too challenging to the worldview of those who believe t/a to be garbage.

I asked a simple question - look at the chart and explain to me whether you still think that market is random.
More...

I think the point here is that the signals are self fulfilled... If large number of persons/systems rely on the same support level to start a trade at the same time (with the Spontaneous Synchronization rule), then that support level works.

The same for EMA crosses, if lots of persons rely on the same EMA periods, then maybe the system will work.

So, the idea is to re-evaluate continuously the parameters of the EMAs to match the current value of most of the traders. That way a system can be "tuned" with the "frequency" of the crow.
 
Linear regression analysis is like any other type of analysis. The opinions derived from the model are usually more a function of the analyst's interpretation than the quality of the model. One of the daily tests I conduct is whether there is the presence of a unit root or not and there are days when it is present and I discard it and vice versa.

Quote from tradrejoe:

For those of you who went through the exercise of using historical data and linear regression analysis to predict the future prices of trading instruments, have you ran into situations where the best beta coefficients that generates the best curve fitting *does not* really predict the future? In fact, often times if you go back in history and pretend you were operating the prediction system in the past, the more testing the more your accuracy converge to just 50%?

What is the correlation between the ability of a set of time series data to fit a price curve and its ability to truely forecast the future with greater than 50% accuracy? Do we just pile up everything closely related to what we try to forecast (even sun spot movements) and go as far back on the time lag as we can without crashing the supercomputer? Does anyone have any experience to share? Thanks for your insights ahead of time.
More...
 
I wonder about the following thought experiment: if you create a price trajectory based on a random walk with parameters of your choice, with suitable memory length (i.e., dependence of current price on N last prices + random factor) then what would happen if you tried to "predict" future prices from your random walk trajectory? Naturally your prediction accuracy would be better and better the more memory your random walk trajectory has. I suppose you could measure the memory length or persistence length based on the autocorrelation function of your price trajectory (even for real price trajectories). Just some thoughts.
 
Quote from tradrejoe:

For those of you who went through the exercise of using historical data and linear regression analysis to predict the future prices of trading instruments,.. *does not* really predict the future?

Thanks for your insights ahead of time.
More...
=============
Well actually any trend line maybe helpful;
not to predict the future. Trends are not predictions;
nor are trend lines predictions.

200 dma is more useful[again , not as a prediction];
perhaps that is because much more investors, much more traders look there...............................................

:cool:
However you may want to consider what General [Top trader]Haggerty said;
many institutions will not admit to using a 200 dma...

:D
 
You must log in or register to reply here.
Back
Top