Adaptive indicators

[Uncle]

The good Doc has referred to Adaptive Indicators in another thread - to which I have no wish to make a further contribution -and I would be interested to know more about these indicators, either from a contribution from Doc or from anyone who has experience and can enlighten us.

John

 

[Dr Iraj]

Hi John,
Just noticed the post.

There are two methodologies in stock selection

a) Scaning for those stocks that fit YOUR CHOICE of indicators (rigid approach, YOU choose the indicator).
Example Buy if MACD crossover and RSI < 30.


b) Scaning for those indicators, which best fit stocks (Adaptive approach, survival of the fittest indicator). Indicators are not dictated by YOU, unlike the first strategy. (a pre built optimization algorithm selects the fittest indicator and signlas the exit/entry point)


I will elaborate more in weekend

 

[Uncle]

Thanks Doc..........look forward to reading the follow up post

Trust everything is OK with you.

John

 

[Hill Farmer]

Yes please Doc. I too believe that indicators need to reflect the individual characteristics of each stock. This may mean that some indicators will be more appropraite/inappropriate than others for given stocks, and, the sensitivity of the response is likely to be different for different stocks. Look forward to your thinking on this.

 

[Riz]

Thanks Dr Iraj,

Another enlightening post..I find choice of indicators of great importance..looking forward to hearing more from you in this subject..
regards

rizgar

------------------

 

[CHARTY]

If I can chip in on this discussion, one of ways you can use adaptive indicators, is by seeing which indicators, and which settings for these indicators, gave the ebst signals in the past.
I had gone long the Nasdaq ahead of Greenspan rate-cut gift, based on one of these indicators, for example (bandwidth on Bollinger bands). For academic reasons, then tested this strategy on other securities, and it did not prove to be so good.
It should be stressed, though, that, although statistically, what has given the best results in the past should give the more reliable indicator for the future, as they write in the disclaimers, "past performance is not a guarantee of future performance..bla, bla".

------------------
The above does not represent financial advice.

 

[Dr Iraj]

Dear all,
Lets see if we can clear few points regarding stock selection. First of all Never give up on TA. Technical Analysis is an extension of mathematics into human behavior . Since human can be unpredictable , you would expect TA to get it occasionally wrong . I have always found it to be easier to logic with my partner when she is in a good mood than when she is not. She can be rather Irrational when she is angry. Market is exactly the same . When market is uncertain the irrationality increases resulting into random movement of the share prices. Good news has no effect, upgrades means nothing and the rest of it. Please try to understand TA’a role.

The following subjects are essential for any technician .

a) Which stock responds best to technical Analysis? ( This is not an easy subject but I will have a go some other time to shed some light on the type of stocks which are easier for technical analysis. Mean while remember the more fundamentally sounder the stock the more correlation between the past and future hence the more reliable analysis. I leave it at that for the time being)
b) What indicator?
c) What settings.?

What Indicator and settings.

I need to explain stock’s personality. There is not much mention of this subject in academic literature, however this is the very most important issue in real life stock selection. EACH STOCK IS INDIVIDUAL. Stocks are individual because the kind of the investor/trader who buys them are different. Some stocks are darlings of big institutions some are in the hands of high risk punter (penny stocks). Stocks are like people and should be treated accordingly. The most success full traders get to know their stock’s very well and move in and out of their positions accordingly.

Now, since the stocks are individual then they should respond differently to different indicators. Some stock might show a support at Lower EMA with 10 days settings some other might prefer the 25 days settings and some might not like the Lower EMA and respond to Lower Bollinger Band. (Most technicians keep saying the key in TA is finding the correct indicator with correct settings).

OUR JOB AS A TECHNICIAN IS TO FIND THESE SETTINGS AND THE CORRECT INDICATOR FOR THE STOCK UNDER RESEARCH. (This procedure is called optimization and the optimized indicator for the given stock is called the ADAPTIVE INIDCATOR). An adaptive indicator is the optimized indicator. An adaptive indicator is the natural evolution of the rigid (no-optimized) indicator.

There are two ways of finding the most optimized indicator for stocks.

a) Buy a computer package which can do all that for you (The most modern and efficient software nearly all fall in the AI category and they are all neural network based. There are how ever others which use other optimization techniques. OMNITRADER is one of them ).
b) Do it manually. I will explain this as some of our members have no access to any TA software.

Suppose you receive a tip from a friend for stock XYZ to be a good candidate. You decide to technically analyze the stock. As a technician it is always best to buy at support. But what indicator do you use to find that support. There are namely LOWER ESA, LOWER BOLLINGER BAND, amongst other trading bands.

Lets try Lower Bollinger Band for support.
Step 1:- chart the stock XYZ
Step 2:- Plot the Bollinger bands for the stock.
Step 3:- Has Lower Bollinger band shown a good support (by a good support I mean the price graph bouncing off of the lower band as soon as it approaches the band) in the PAST few weeks (short term trade)? If not then reduce or increase the settings for the Bands in a manner that price graph touch the Lower BB. Try this manually by trial and error until you are happy with the FIT. If Lower BB did not give you the most optimized support level then try Lower ESA and carry on the above procedure with settings again.

What you are effectively doing is to curve FIT the stock price behavior to one of the Lower Bands for support . (It is very much like buying a correct pair of shoes for your daughter). If you don’t use Adaptation technique and use the standard settings you are basically treating all stocks in the same way. This is very similar to having a ONE SIZE
shoe and expect every body comfortably to wear it. Hmmmmm

Try the same procedure for RSI. Curve fit the best Value RSI for each individual stock. You will be shocked to see that some stocks are over sold at 30 where as some still are over bought at this value.

Our charty correctly says that past is not a GURANTEE of future behavior . Of course it is not and we don’t expect it to be either, as long as we can perform 55/45 odds to our side then we will be in profit in next 12 month. (after all Casinos have only the ZERO to shift the odds to their side. Have you ever seen a poor casino owner?).

WE DON’T NEED MUCH ODDS IN OUR SIDE TO ACHIEVE A LOT.




[This message has been edited by Dr Iraj (edited 07-01-2001).]

 

[Hill Farmer]

Dr Iraj, a nice and very clear post from you. This should provide motivation and a good basis for some of us to start thinking about how to tailor the individual indicators. I have an observation. In embarking on such data-fit exercises, we may just be data mining/over-fitting with the result that, although the past can be accounted for the 'forecast' may not be good. Iy is the old wel known issue that past data fit does not guarantee good forecasts. What is needed is a statistical/econometric analysis regarding the robustness of the fit and the attendant forecast. This will typically involve a hold-out sample and formal statistical tests regarding the forecast. Perhaps it would be very useful if the formulation/structure of this type of analysis could be discussed on the BB.

 

[Dr Iraj]

Hi Hill,
Your comments are noted. They are most relevant to my post. Thanks.
Study of Chaos theory (weather prediction) and Random theory (lottery prediction) are the most relevant subjects in my research work. Finding that HIDDEN ORDER and milking the underlying pattern inherent in what seems to be a Chaotic set of stock’s historical data is a must for those who are interested in predicting the future trends.

I am currently looking into PACKARD’S work on Local Predictability.

 

[CHARTY]

It should also be said that Bollinger bands, by the way they have constructed, are probably one of the most self-adaptive indicators, this is why I use them so much (and you may be interested to know that some of my colleagues they hate them and find them completely unuseful!).

------------------
The above does not represent financial advice.

 

[Dr Iraj]

Charty:-

The curve fitting adaptation is not exactly the same as band narrowing due to change in SD as always happens once the volitality increase in the stock. These are two different things. Adaptive inidcators use Optimization techniques (such as Genetic algorithms, Hill climbing,...) to find the best solution. Bollinger bands narrowing is another story.

 

[ZigZag]

DR Iraj: Just found your excellent thread which I am surprised it has not been updated. Perhaps it has moved? Adding to your adaptive approach have you considered regression analysis for your tool box. For short term trading & change of trend alert for medium term trading I use a 5 day period. You mentioned you were looking at Packards local predictability . Can you tell me of any web sites which covers the topic. I too favour statistical trading plans. Good Luck

 

[Shadower]

You might want to look at http://www.google.com/search?hl=en&ie=UTF-8&oe=UTF-8&q=statistical+arbitrage+packard+farmer

for a quick round up of what Dr Packard and his mate Dr Farmer have/are up to.

Besta luck

Dave

 

[bonsai]

Dr Iraj
good stuff !
not sure that optimising software is all that easily available
to many traders here.
but 2 good points I would emphasise are
1- Know who are the dominant players in the stock and understand their objectives. (And track the options movement as well)

2- Retune your favourite indicator when you see a top or bottom.

 

[Glenn]

Not wishing to offend or disprove anyone, just stating my point of view :-

All indicators, no matter how they are selected or applied are simply an analysis of the past; quite often with a lag compared to the current price action. They often get you in late and out late - hardly efficient.
If you study the OHLC bars themselves, you can improve on this considerably once you get the hang of it.
And if you watch the price quotes, last price and size, you get an even better view.
I seem to be one of few who only use naked charts, having long ago discarded all indicators. Glad I did.

I think that people who believe that they are successfully using indicators are in fact subconsciously or even consciously looking beyond them. Indicators start out as a prop to your view of the chart, smoothing out the wrinkles, and providing a clearer picture - of the past.
Over time they become less useful as you watch the real chart and price action going on.

I also think that the relentless search for some kind of mathematical pattern which fits 'human' behaviour is both a waste of time and completely unnecessary. I have been through all this myself having majored in Maths during my education.
Imho it is easy to assume that there is something slightly beyond your current thinking which could be what you are seeking, and to then spend inordinate amounts of time looking at ever more exotic possibilities.
In fact doing this can become obsessive because it takes place within people's comfort zone. Easy to make the excuse that you aren't trading because you are doing more 'research'. Easy to spend a lot of time messng about with fancy sounding theories and endlessly arguing the pro's and con's.

After years of experimenting with indicators and such (I'm nearly 57 now) I discovered that I didn't need them, in fact they were a hindrance or worse, a misdirection. The same applies to backtesting.
Just thought a different experience might be useful to some, and in consequence save them a lot of time.

Keep it simple imo.
Each to his own.
G

 

[Shadower]

I'm still sat down and there's no spit on me screen, but I disagree with a couple of your points ( which probably means I've mis-read them :cheesy: )

I think a largely mechanical trading system can be a useful tool, as it is mostly not a case of interpretation but a ( head down ) 'statement of statistical fact'. Now obviously that last bit is treading very close to an oxymoron and it is dependant on the interpretation of the results of the statistics of the guy writing the system rules. But, if those rules give you a 5% lead over the market, I think it is worth doing, even if it only means you 'win' 55% of your actions against a 50/50 hit rate.

Having seen some of the results from the stuff that Packard, Farmer and the rest of the Prediction people have done I'd have to say it works, and works jolly well too.

I think with other indicator systems - MACD crossovers, RSI conditions - that I've looked at, they've *all* been additions to basic chart reading, decision re-enforcement I guess.

Besta luck all

Dave