1

Possible Duplicate:
PACF manual calculation

I am trying to find a formula for how to calculate partial autocorrelation between variables. We know that aucorrealtion between variables at different lags are given by: $$ \hat\rho_h=\frac{\sum^T_{t=h+1}(y_t-\bar y)(y_{t-h}-\bar y)}{\sum^T_{t=1}(y_t-\bar y)^2} $$ I know also that partial autocorrelation is the autocorrelation between y[t] and y[t–h] after removing any linear dependence on y[1], y[2], ..., y[t–h+1]. But how do you remove any linear dependence on y[1], y[2], ..., y[t–h+1]? Does there exist some formula for this?

Community
  • 1
dato datuashvili
  • 723
  • 2
  • 7
  • 21

1 Answers1

1

Googling for "partial autocorrelation" will lead you to Wikipedia:

The partial correlation between $X$ and $Y$ given a set of $n$ controlling variables $Z = \{Z_1, Z_2, \dots, Z_n\}$, written $\rho_{XY\cdot Z}$, is the correlation between the residuals $R_X$ and $R_Y$ resulting from the linear regression of $X$ with $Z$ and of $Y$ with $Z$, respectively.

Stephan Kolassa
  • 95,027
  • 13
  • 197
  • 357
  • but there is written method very unclear,i can't undertsand how to use it for practically,let's take some example of time series,otherwise i can't guess it – dato datuashvili Jan 17 '13 at 15:18
  • any help?please – dato datuashvili Jan 17 '13 at 15:25
  • Are you unclear on the concept of residuals from a regression? If so, I suggest you read the first few chapters from any textbook on regression. This will help you much more than any cookbook recipe we could possibly give you here. – Stephan Kolassa Jan 17 '13 at 15:28
  • no i am not unclear,just i need to know practically how to remove linear dependence between variables – dato datuashvili Jan 17 '13 at 15:34
  • generally more help is when you give them direct hint,not something different unclear method,let us suppose that i have data from 1981 year to 1995 like this.23 45 12 10 9 8 6 4 5 33 7 66 54 89 76 how to calculate partial auto correlation between time 1982 and 1989? – dato datuashvili Jan 17 '13 at 15:44
  • ok i will ask supervisor about it – dato datuashvili Jan 17 '13 at 15:46
  • You have a very... optimistic... opinion of how quickly we can help you. I actually found this here which may be almost as helpful as your supervisor: http://stats.stackexchange.com/questions/18599/pacf-manual-calculation – Stephan Kolassa Jan 17 '13 at 15:54
  • no it is not helpfull much,because it did not describe full process,it measn that PACF is ACF including last lag? – dato datuashvili Jan 17 '13 at 16:02
  • it is not duplicated,but ok thanks for helping – dato datuashvili Jan 17 '13 at 16:07
  • @user466441 I'm sorry to see that you did not receive a satisfactory answer to your question. The "possible duplicate" should have helped you to some extent, but I think what you want is a specific numerical example. I won't go through an example using regression to estimate partial autocorrelation coefficients, but I have written a short document to show you how they can be estimated using the Yule-Walker equations. You can access the .pdf file at my dropbox: https://www.dropbox.com/s/phbdr1k0twxcs47/yulewalker.pdf Maybe this will help you out. – Graeme Walsh May 16 '13 at 01:32