Can Autocorrelation Affect Exponential Smoothing Results?

[neznam]

Hi,
I have a conceptual question. Looking at exponential smoothing methods I came across relationship between the autocorrelation function and lambda. It says that if the time series doesn't apper to be autocorrelated then lambda is expected to have a low value .Any help will be appreciated.

1st order exponential smoothing
y(t)tilda=λ*y(t)+(1-λ)*y(t-1)tilda
where λ=1-θ
and θ represents the weights

 

[ych22]

Hello, in many statistical tests, inferences or applications, personal judgement is required. I would rephrase your statement to "If the time series appears to not be auto-correlated, we should set lambda to a low value." After all, choosing lambda is a personal judgement on the part of the statistician, there is no "expectation of lambda" here.

so:

forecast(t)= lambda*observation(t-1) + (1-lambda)*forecast(t-1)

So if the time series appears correlated, then lambda should be set to a high value. Thereafter, forecast(t) will depend highly on observation(t-1) and less on forecast(t-1).

 

[neznam]

Thanks a lot. That makes a lot of sense now :-)