The Robot and Stationarity: The Best-Kept Secret in Trading Success?

I attach an old article by Dr.Brett N. Steenbarger on stationarity.
I believe all members interested by the Robot's design and concepts should study it carefully.
Indeed, I think the robots are using stationarity rules derived over one or more stationary windows of opportunity.
Neutral windows of opportunity like now are non-stationary (uncertain and whipsawing market).
Non-neutral signals detect rising stationarity early whether into mean-reversion or trend-following mode.
The trade timeframe is not fixed and is dependent on how long the new stationarity is active.
I haven't consulted Pascal about these assertions, so he's welcome to comment.
Billy

[ATTACH]11102[/ATTACH]

"The fundamental uncertainty of trading is highest in daytrading the stock
market—particularly index futures such as the SP/ES and ND/NQ. This is because
markets are nonstationary on an intraday basis—almost without fail. Markets are most
volatile early in the day’s trading, retreat to lowest volatility in early afternoon, and then
pick up volatility toward the close (only to plunge in volatility during Globex trading). It
is rare indeed that the distribution of price changes from 09:30 – 11:30 AM ET will
match those of 11:30 AM – 13:30 PM ET. Using the same indicators and indicator
values in morning trading as in early afternoon and Globex sessions is a sure road to the
poorhouse. Conversely, identifying regime change and valid relationships with each
intraday regime shift requires a nimbleness—and an ability to control losses—that most
traders lack."

So, true! Thanks, Billy.

I did some searching and was not able to find much about the methodology (without reading the Sherry book yet). If anyone knows of or has an idea how to implement a test for non-stationarity within a trading platform, please post here. The necessity of iteration will whittle the candidate platforms down (unless a dll is used), but TradeStation and eSignal are viable.

Honnestly I cannot say if the Robot's investment time frame corresponds or not to the "stationarity" definition.
Today, I can tell that the IWM Robot is "stuck" in neutral because the ATR is still higher than usual and therefore pushes short edges, while the MF signal is in overbought territory, which pushes the long probabilities down. However, since the MF directional model is in "buy" mode, the IWM Robot is just caught in a pattern that is not clear enough.

This type of environment is not built by chance: it corresponds to the combination of fear/uncertainty (High ATR) with strong available liquidity (pushing the MF into Overbought.) This is very unusual and shows - in my opinion - that there is a good dose of "control" in the market.

The spike in gold yesterday shows that it is difficult to control fear with liquidity. This means that if the markets needs to go down, they will, whatever the available liquidity. The only thing that needs to be avoided is a Lehman type of domino effect. It is the domino effect (the contagion of forced selling) that pushes markets down in an uncontrollable manner. From what I see, "control" is not lost.




Pascal

Yesterday, I mentioned that the proper attitude toward the robot was one of student to teacher. I learn from it; hence, I'm the student; yet I remain in a position to profit. When I wrote the sentence I wasn't entirely convinced of it because I hadn't entirely explained what I meant. The article posted by Billy provides a better description of my precise meaning:

[I]A losing streak with well-researched
trades is often a sign that the markets are changing. Standing aside, waiting for evidence
of the new regime, and remodeling the market over the more recent time frame
corresponding to the new regime may allow the trader to learn from losses—and recoup
them as well![/I]

I take the robot absolutely seriously as a well researched trading system. However, I could see from the first day that this latest GDX trade had a greater probability of failure (as did Pascal who hinted that position sizing on this trade would be very important.) Yesterday, in the first hour or so of trading my sense that the trade would go badly was confirmed by sharp price action. I was unable to free myself of my robotic trance until most of the move had vanished. But it was precisely the well-research failure of the robot that made this reversal in the market, this change of playing card decks, so utterly apparent to me. This is what I mean by "learning from the robot."

[QUOTE=Pascal;18027]Today, I can tell that the IWM Robot is "stuck" in neutral because the ATR is still higher than usual and therefore pushes short edges, while the MF signal is in overbought territory, which pushes the long probabilities down. However, since the MF directional model is in "buy" mode, the IWM Robot is just caught in a pattern that is not clear enough.

This type of environment is not built by chance: it corresponds to the combination of fear/uncertainty (High ATR) with strong available liquidity (pushing the MF into Overbought.) This is very unusual and shows - in my opinion - that there is a good dose of "control" in the market.
Pascal[/QUOTE]

Hi Pascal,

One interesting question about the robot, and referring to the figures you published on the main robot page (which are fantastic, BTW), is whether you are constantly adding statistics to the tables as time progresses, and whether you use a sliding window or go back as far as data allows. From your comment above, inferences from the data are also mixed with (or guided by) your own observations and judgment about key factors of market regimes. In any case, it seems the methods you use to feed the robots may have something to do with answering the question of whether it detects and/or adjusts to regime changes. The multi-timeframe pivots and their clustering probably also help navigate regimes.

Excellent references from Billy and others here. I really enjoy the discussions here.

-Mike

[QUOTE=mklein9;18031]Hi Pascal,

One interesting question about the robot, and referring to the figures you published on the main robot page (which are fantastic, BTW), is whether you are constantly adding statistics to the tables as time progresses, and whether you use a sliding window or go back as far as data allows. From your comment above, inferences from the data are also mixed with (or guided by) your own observations and judgment about key factors of market regimes. In any case, it seems the methods you use to feed the robots may have something to do with answering the question of whether it detects and/or adjusts to regime changes. The multi-timeframe pivots and their clustering probably also help navigate regimes.

Excellent references from Billy and others here. I really enjoy the discussions here.

-Mike[/QUOTE]

I am planning to add the trades since March 2011 to the main Robot decision tables (not using a sliding window.) This is a rather long work that has not been automated. SO this is something I'd plan to do about twice a year.


Pascal

In doing some more research on regimes and their discovery, I happened across a really excellent blog called Quantivity. The author (who remains anonymous AFAIK) started looking at regime discovery in mid/late 2009 and has published a series of posts over time on the topic as he/she surveys the area. This one: [url]http://quantivity.wordpress.com/2010/02/15/regime-discovery/[/url] has this bit:

[QUOTE]An alternative approach is trying to build regime discovery models which quantitatively identify regimes early in their formation, and probabilistically inform how to trade ahead of the herd (whether by seconds or days).[/QUOTE]

Other posts in this series touch on identifying regimes through changes in the relationships between volatility and price ranges. From what I've been reading, regimes are usually considered either mean-reverting or trend-following; some people also use an "unknown" or "indeterminate" third regime type.

To me this recalls a number of things discussed here:

1) The 20DMF attempts to find accumulation/distribution that causes price movements off of an equilibrium, especially when prices have been stable for a while (i.e. early hints of a regime change). Pascal has said that this signal is most useful when occurring after some period of consolidation (i.e. a "mean reversion" regime).
2) ATR (which Pascal has said is a good measure of short-term volatility) is a key input to the robots, in choosing the statistical tables to use (thanks Pascal again for the diagrams) and probably other ways
3) Extremes of OB/OS also appear to help foretell regime changes, at least when rebounding from extreme negative
4) ATR is also a key input to the exit strategy

I also manually marked up a chart of IWM with the robot calls and found that in most cases, the IWM robot was frequently accurate at anticipating or being fairly early in joining regime changes, either in terms of entry or exit (in the case of a signal change causing an exit).

What is further more amazing is how well the robot navigated the extreme volatility in Aug-Oct 2011 which seems clearly a very different type of regime from anything before July. The one question in my mind is whether its success during this recent volatile period was maybe somewhat accidental (I don't remember whether most of these signals were based on short term edges or not).

Of course, my reading of whether a regime change occurred or not could well be off the mark, as there is not really any definite agreement on when such a thing occurs.

-Mike

Mike,

This is very interesting to me. Can you paste your marked-up chart?

Many thanks,

[QUOTE] Mike,

This is very interesting to me. Can you paste your marked-up chart?[/QUOTE]

Nickola, this is very embarassing but it's a hand-drawn set of signals on a Yahoo HLOC chart shot with my phone. I've attached it for now and really should do this more professionally :-).

Where there is a "Sig" notation it means the position was exited due to a signal change. Otherwise it is a stop.

-Mike

[ATTACH]11118[/ATTACH]

It looks good to me. It is always interesting to read the charts of others. How would you improve it? Cheers,