Experiments in Technical Analysis

Hi,
Trading Risk is basically calculating the standard deviation of historical returns or
its avg.return of my module(maybe MACD).This module or TA is a derivative of Price / data.
No quantitative or qualitative analysis of the scrip but Pure price.( MACD is only a Trend
following Momentum Indicator.)Here at this stage i am flooded with options,whether RSI or
Bollinger Band or MACD etc etc.All derivative of Price / Data.To minimize Filter Risk,
(the moment say i discard RSI for MACD i am running a filter risk)we recourse to backtesting.
Again another Risk factors Autoregressive risk (the assumption past performance may be
replicated in future)creeps in.This is now becoming difficult to Percieve the TOTALITY.
1st let us fix the ROOT.My EOD data ,repeats the occurence in fixed interval.(This realization
Hit me ,when Souyma_b,installed an utility in my computer to extract intraday data to Metastock
from PIB,and i found all my Good indicators are giving below par results barring few,i have
deliberately deleted that utility).Now my Time frame is fixed.
whenever i say datas occurence is in fixed time interval,there is a fallacy.An example :=
In Horse racing tract ,a term "Photo Finish" is used to determine the winner in a neck to neck
contest,a stationary camera at finishing line takes a picture,the horse which is most ahead
at that moment is declared winner.Visualize galloping horses in breck neck speed ,side by side.
A Photograph of that non-stationary objects is captured ,where the basic Physics of "Camera"
says "IT HAS TO BE STATIONARY",yes it was stationary in say 1/200 th of a Sec.
So my EOD data IS STATIONARY in that context.
Next quality of this Data.Any data service,giving me the
O,H,L,C,Vol,OI,accurately with Reference(NSE declared EOD figs ,in my case)is absolutely OK.
There is a Myth promoted by "Paid Data Service",until it is Paid ,it is bad quality.Tell me
other than the authenticity & pre adjustments( dividend / bonus) of data ,can there be any
Quality aspect of data.
1st Root is seeded here.
a) Data is Stationary.
b) Data is Quality Data.
Data Mining is used in all fields to identify a Pattern on a large data bank.Any pre filtering
of data before apllying a Price Derivative brews "Bias".We always do that,maybe for Liquidity
etc.or with other objective ,so this is the LOOP part after seeding.

After yesterday night's CV's post HIT me a serious review once again strenghtened my self belief.
Any tinkering of Data is Data Mining or TA which it self is the process for Signal generation.

Hello everybody,

Can anybody please tell me (espesially Karthik) what is metastock & what is amibroker? Kindly please tell me what is that programming kind of thing written under "Elliot Wave" & "Elliot Wave Signals"? as Iam very new to this forum & trading too...
Eagerly waiting for reply.

Abhijit.

Hi Anant!
Hope you are doing well.
I want to discuss some issues with you. I tried sending u a PM but it appears that your PM is disabled at the moment.
Hence, can you please send me an email on aca.ashish[at]yahoo.com or enable your PM?

Best Regards,
--Ashish

Is that CVs real name?

Yes, it'd help to know the break-ups of the steps outlined.

Regards,
Kalyan.

Please, please have a look at the subject of the thread before making a post.

There are entire threads on Amibroker & Metastock in the Software section.

Regards,
Kalyan.

In the zero lag MACD what happens to the divergence? Is there any variations? Your observations Please. Thanks
Regards
gacsekar

Hello friends,

The thread has now entered a new and exciting phase. The points CV has brought out are very important. This has forced me to think again about the questions I was asking myself since beginning. When I started learning about TA about two years back I had two basic questions which are unanswered even today:

1. Is price a random variable?
2. If it is not then is it a dependent variable?

If the price is a random variable then it should follow the gaussian distribution. In that case we can apply the various statistical parameters like mean, standard deviation etc. to a small sample of data and draw conclusions about the whole data set. The sample values also should be selected randomly. This means that sample data should be such that each value in the data set has equal chance of getting included. What actually we are doing is selecting a portion of the data where the values are not randomly included. The sample is a continuous series of values. Therefore the sample set is not made of randomly selected values. Hence the conclusions drawn do not reflect the properties of the whole data set.

In TA we also talk of linear regression, coefficient of variation, regression coefficient etc. This means we are treating the prices as dependent variable and not a random value. If price is dependent variable then on what does it depend. In other words what is the independent variable. In price charts we plot the price on Y-axis which by convention is dependent variable and the X-axis is time. Does it mean price depends on time? But it is not. Time axis is just a convenience to plot the prices. So, price is independent variable. Then why do we try to fit a regression line to independent variable? Why do we plot it on Y-axis?

I do not have answers. I am not sure whether price is dependent variable or independent variable. Unless this is clear there is no meaning in fitting some lines, indicators etc.

What we are supposed to do is fit some relation/equation to the data to explain its behavior. But what we are actually doing is trying to fit the data to a particular relation/equation. We fit a small set of data points to whatever we want and try to interpret the results as the properties of the whole data set. The results of TA on a small portion of data would apply to only that set but not the whole data.

There are many more questions. But let us answer the basic question. What type of variable is price.

More later.

Regards

-Anant

Hi anant,

Einstein said: " Make things simpler and not simple "

i think what ever approach you give to markets , what matters is that, are you a consistant performer and is the green buck under your table at the end of the day?

Regards,
Winston

Dear Friends

I think Winston has a moot point.

As I had pointed out before that most indicators are built on the assumption that the prices have gaussian distribution. Ananth has also brought out this fact in his post.

(What I admire about ananth is his clarity of thought and expression. He has a great why of explaining complex things in a simpler way.. His posts make a great reading…)

Ok, the question here is does approximation work? Assuming the prices to have a gaussian distribution is an approximation.. Is this approximation good enough for trading?

We can also approximate the prices to a gaussian distribution by applying Fisher transform? Will these Fisher transformed Indicators perform much better than the conventional Indicators which assume the prices to be gaussian distributed?

Many questions…good for a lot of discussions…

I am enclosing two charts here. One chart shows a fisher-transformed signal of a 10-day median price. Also shown in the chart is the conventional MACD.

Another chart shows a conventional stochastic D signal with a fisher transformed stochastic D.

I will leave it to you to draw the conclusions…

Regards

Karthik

Thanks Karthik, for the charts.

Just one clarification. Is the IFT for the Stoch or RSI? You have mentioned Stoch but the label on the graph shows RSI.

Assuming they are both for the Stoch %D, i see a definite advantage in gaging the turning points. (Less ambiguity)

And why have you excluded %K?

Regards,
Kalyan.