Autocorrelation in time series data

[Philipp78]

I am running risk and return analyses for a number of time series (stocks and real estate). My first question is how can I compute Durbin-Watson statistic and 1st order autocorrelation for the return series (monthly log returns) ? Secondly I am also assessing the autocorrelation using the corelogram. However, I am not quite sure whether I should plot the the correlogram of the raw series (level), the first difference or the second difference, given my input data comprises monthly log returns? Would very much appreciate if someone could provide me with a hint for these question and could point to relevant literature in respect of the second question.

Many thanks & cheers,

Philipp

[startz]

The Durbin-Watson is automatically calculated when you run a regression. The first-order autocorrelation appears when you do a correlogram on levels, although you could also do

Code: Select all

ls r c ar(1)

[Philipp78]

Thank you very much for your prompt answer. Just one follow-up: In relation to the Durbin Watson Test - Can this only be calculated using a regression? I simply would like make a comment in relation to the autocorrelation of the return series (monthly log returns) I am focusing on and as such I do not actually need to run a regression. Furthermore I have a found in a numer of papers a reference to autocorrelation statistics or serial correlation statistics. However, it is typically not really explained what this autocorrelation coefficient really is and how it can be calculated. As an example I attach a paper examining real estate returns. Would be great if someone could share his insights in relation to what static measure is typically meant and how this can computed with Eviews.

Would be great if someone could give me a hint. Many thanks!

Philipp

[jalle7]

Hi,

Can you carry out a Breusch-Godfrey serial correlation LM autocorrelation test on pooled cross-sectional data in eviews?
If not arw there any other tests for autocorrelation that can be carried out on this type of data, other than looking at the Durbin Watson statistic?

Many thanks