I simulated a time series using expressions (3.10a), (3.10b) from (Hyndman et al., 2008). Next, I'd like to use a simple exponential smoothing method to forecast for the next period. For a given initial point $\hat{y}_1$ I obtained the smoothing parameter $0<\alpha<2$, etc.

My question is as follows: am I going to predict the next value of the local level $l_t$ or the next value of the observed time series $y_t$? Regretfully, I still do not understand how one can predict a quantity which includes a random component, and if I got it right, time series without trend consists of level and such a component.


1 Answer 1


For simple exponential smoothing, \begin{align*} y_t &= \ell_{t-1} + e_t \\ \ell_t &= \ell_{t-1} + \alpha e_t, \end{align*} where $e_t$ is a white noise error. So the forecast of $y_{t+1|t}$ is also a forecast of $\ell_{t+1|t}$. That is, both conditional distributions have the same mean, and the forecast is an estimate of that mean.

Everything you want to predict has a random component, otherwise you would know its value and there would be no need to predict it. Forecasts are just estimates of the means of conditional distributions.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.