Of course, prices rarely revert to their mean because price series is nonstationary. Cointegrated series (by definition) must revert to the price of an index or another security. But that's pairs trading (different topic)...
This is starting to turn into the showdown of the mathematicians 2007....tickets are free...come on in ... but seriously... who wants to answer my original question so me and the majority of elite traders can understand the answer?
Nothing from my individual work that I can share, but if you google it you can find examples. Here's one, China's grain market in the late 90's: http://scid.stanford.edu/pdf/scid212.pdf Granted, that's pretty obsucre, but it was the first one from searching I came across. Have seen some more recent papers dealing with market microstructure that touch on it, but I can't remember where. Stationarity is highly dependent on the time-frame and window of time you're looking at. It's not uncommon for price series to go in and out of periods of stationarity and back. If you're looking at the complete historical series, yeah you'd be hard pressed to find it, but in shorter windows they do exist... be pretty hard to make money if they didn't. Again, though, knowing any of this doesn't neccessarily help and definitely isn't any kind of a requisite for making money with mean reversion or trading in general.
I'm not a mathematician, not even close. Seriously, any of the math concepts mentioned is maybe first year math/statistics. Regardless, seems to me you answered your own question on the top of page 2. Don't trade stocks that meet that criteria and see how it pans out. Longer answer is that only you can know the attributes of your system that you'd want to filter for/against and how to define them. There are all sorts of trends, each of which may or may not do well for mean reversion strats depending on time frame, volatility, etc. Learn more about the nuances and properties of why your strat works and why it doesn't and I think you'll be able to find the answers you're looking for. Also, answers that the majority of people can understand are usually not the most profitable.
Look... Cointegration is an econometric relationship. I tried a similar data mining exercise using indicator/log return relations without success. Too many significant "effects" simply disappear as I moved the observation window closer to real-time. Price series are a nonstationary unit root process without an existing average. There isn't much meaningful statistical analysis that can be done directly on price (moving averages are very strange in that regard). Returns are stationary and you can track mean (drift) and variance (short term volatility) and sometimes higher (central) moments like skewness and kurtosis... So, the formula I posted earlier provides a good approximation of a detrended series within a given time horizon. Ehlers' book "Rocket Science for Traders" has a chapter (with TS code) on detrending by removing the DC component. He basically decomposes the price series into a DC component (trend) and a cycle (smooth series). I think it's mostly bullshit... focus on derivatives instead.
============= NasdaQQQ;I was thinking something like that; but dont think he wants or is thinking/writing that.Laugh out loud For up trenders, you want 52 week new highs[like SPY/ES/DIA,] plenty of them; higher highs, higher lows, higher closes. Downtrenders want lower lows ,lower highs, lower closes If filtering out those is your cup of tea, filter on QQQQ has been in a sideways trend for many years but started uptrending; but its still in long term downtrend.
Skimmed the paper, didn't notice they were using co-integration, just glanced at their conclusion using the unit root tests. Look... why don't you just do a google search for "stationary price series"? I know there are papers studying price shock events finding stationary price series out there because I've read them in the past. Also, I know there's some microstructure papers looking at bid-ask series that find the same. I just disagree here. From my own tests and trading, I've found price series move in and out of being stationary/non stationary. I've found negative autocorrelation in stationary price series that do indeed exist and make great mean reversion series without needing to force an MA onto it. If you choose to believe otherwise based on your own analysis then that's fine, to each his own. If you don't know what you're looking for and why, then I would agree 100%.
Writing your sentence in the present tense might help you think more clearly. No one can consistently correctly and reliably predict the future.