Re: Need HELP!!! Trends and Statistics

dave_at_autobox.com
Date: 01/21/05 

Date: 21 Jan 2005 09:57:57 -0800

Jacob,

The concept of Trends is an integral part od time series analysis.

Models that include trends can look quite different depending on the
specific nature of the data that you are analyzing. For example

1. DETERMINSTIC TRENDS

Y(T)= B0 + B1*T + E(T) where T is the counting numbers E.G.
1,2,3,4,.....T

raises the question

a. Ho many trend variables do I need ? Are there two trends or one
trend breakpoint in the data ? Are there more than 2 trends ? each with
it's own slope or differential. How does one go about identifying
(automatically) the correct number of trends ?

b. is the distribution of E(T) Gaussian ? restated are the E(T)
residuals free of any autoprojective structure that would vitiate any
test of significance ? Is it necessary to

model E(T) = [PHI(B)/THETA(B)]*A(T) where A(T) IS A Gaussian
Process.

or

2. STOCHASTIC TRENDS

Y(T)= Y(T-1) + THETA0 + E(T) where THETAO is the one trend coefficient
or is there a need ( evidenced by the data) that multiple THETA0's are
needed reflecting change points or break points

and furthermore does E(T) need to be modeled as above.

Furthermore should one be using .5*Y(T-1) rather than 1.0*Y(T-1) or
more generally should we be employing ...

Y(T)= PHI1*Y(T-1)+PHI2*Y(T-2)+.....PHIP*Y(T-P) ....

OR

3. Some combination of 1 and 2

We at AFS have been in the business of developing computerized aids in
this regard for nearly 30 years. Please visit us at
http://wwwautobox.com

Dave Reilly
AUTOMATIC FORECASTING SYSTEMS

p.s. Your stat professors may not necessarily be able to guide you
properly . It all depends on where you went to school as most
statisticians are not properly trained in the dark art of time series
analysis.

If I can help ..please contact me at 215-675-0652

Relevant Pages

  • Re: goodness of fit test for data over time
    ... are evenly spaced times. ... The practice of fitting high order polynomials has long since been ... and time trends ... This is done for all other parallel time series. ...
    (sci.stat.consult)
  • Re: Getting an semi accurate forecast
    ... Chapter 20, "Time Series Seasonality," of my book "Data Analysis Using ... > decreases in sales trends year to year but also the month to month trend ... > decreases should carry more weight than say this April to last April. ...
    (microsoft.public.excel)
  • Re: point me in the right direction for time series analysis?
    ... If you do have enough repeated exposures to the stimulus you may have time series. ... In SPSS that would be TRENDS. ... amorphia wrote: ...
    (sci.stat.consult)