# time series

• March 30th 2010, 06:50 AM
eulerian
time series
What is the difference between a time series without autocorrelation, and a stationary time series?

If they are the same, why do we need to do both the Ljung-Box test and unit root test?
• March 30th 2010, 10:06 PM
CaptainBlack
Quote:

Originally Posted by eulerian
What is the difference between a time series without autocorrelation, and a stationary time series?

If they are the same, why do we need to do both the Ljung-Box test and unit root test?

Consider a time series where the value at epoc $i$ is normally distributed with zero mean and variance $\sigma_i^2=|i|$ and independedent of the values at all other epocs. The autocorrelation of this process is clearly zero for all non-zero lags and it is also clearly non-stationary.

CB
• March 31st 2010, 02:28 AM
eulerian
Quote:

Originally Posted by CaptainBlack
Consider a time series where the value at epoc $i$ is normally distributed with zero mean and variance $\sigma_i^2=|i|$ and independedent of the values at all other epocs. The autocorrelation of this process is clearly zero for all non-zero lags and it is also clearly non-stationary.

CB

One definition of a weakly stationary process is $E(yt-\mu)^2={\sigma}^2$or with constant variance.

So, why do the above process with non-zero value of autocorrelation function for zero lags is stationary?
• March 31st 2010, 02:52 AM
CaptainBlack
Quote:

Originally Posted by eulerian
One definition of a weakly stationary process is $E(yt-\mu)^2={\sigma}^2$or with constant variance.

So, why do the above process with non-zero value of autocorrelation function for zero lags is stationary?

Try that again what you have written is incomprehensible.

The process I have described is non-stationary

CB