PDA

View Full Version : Best method of calculating trendlines



Susmita
06-26-2008, 05:00 AM
Hi

I was checking my business performance in terms of financial values. I was sitting with hell lot of data of last few years. And then suddenly an idea ping on my mind for which I want to start from trend analysis simply to understand my position in the market.

Please let me know which process should I follow.

Linear trend analysis,
Logarithmic,
Polynomial,
Exponential or
Moving Average.

Any help will be well appreciated.

Thanks
Susmita

kgun
06-27-2008, 11:31 AM
First, before you even think about functional from, you shoud make your data stationary. Then you should use cointegration analysis to identify comment trends.

As time ---> infinity, the t-statistic in a regression between two random walk series ---> infinity as well. That is called spurious regression.

There are a lot of software that you find in my link collection "Search" link (upper right corner) in second link in my signature that is well suited to this analysis.

Click that link + "Search" in the upper right corner.

Then on that page scroll down to or page search (CTRL + F) + Advanced topics

and scroll down to the sub heading

Software

The first link is one that starts with "Data mining ...

I will personally reccomend Gauss or PcGive.

There are a lot of clever statisticians in India, so may be you already have good enough software for cointegration analysis.

A classic book by K. Takeuchi (Japan), H Yanai (Japan) and B N Mukherjee (India):

"The foundation of Multivariate Analysis" A unified approach by means of projections onto linear subspaces

K.W search on Cointegration litterature:

Søren Johansen (http://www.math.ku.dk/~sjo/) cointegration

Susmita
06-28-2008, 03:40 AM
Thanks Kgun.

Thanks for each and every information provided by you.

Its really helpful.

Regards
Susmita