Wold's representation theorem states that every covariance-stationary time series $\{Y_t\}$ can be written as the sum of two time series, one deterministic and one stochastic: $$ Y_t=\sum_{j=0}^\infty b_j\varepsilon_{t-j}+\eta_t $$ where, among other things, $\{\eta_t\}$ is a deterministic time series, such as one represented by a sine wave.

I am puzzled by the inclusion of a time-varying deterministic part. Does it not violate the definition of stationarity? Since the expected values of $\varepsilon$s are all zero, $\eta_t$ gives the expected value of $Y_t$. If $\eta_t$ varies over time (such as a sine wave does), the expected value of $Y_t$ varies over time, thus violating stationarity. What am I missing?

Update: A follow-up question.

  • $\begingroup$ Richard, I am leaving a comment for I have not touched the topic for quite some time. But doesn't deterministic mean perfectly predictable? Mainly the term is not a time trend and that means it shouldn't violate the stationarity assumption. $\endgroup$ Commented Sep 22, 2023 at 8:42
  • 1
    $\begingroup$ @User1865345, if $\mathbb{E}(X_t)\neq\mathbb{E}(X_s)$ for some $s\neq t$, that looks like nonstationarity to me. By definition, $\mathbb{E}(X_t)=\mathbb{E}(X_s)$ for all $s,t$ if the process $\{X_t\}$ is stationary. $\endgroup$ Commented Sep 22, 2023 at 8:49
  • $\begingroup$ Check this note. Note how the concerned term is constructed. $\endgroup$ Commented Sep 22, 2023 at 9:06
  • $\begingroup$ @User1865345, thank you, this looks very relevant! And a bit challenging – no bedtime reading... $\endgroup$ Commented Sep 22, 2023 at 10:37

1 Answer 1


The component $\eta_t$ is not "a" "deterministic" time series. It will be a time series such that it will be covariance-stationary (at least) itself.

Here the terminology "deterministic" (a terminology issue, once more), should better be "conditionally deterministic", because what it aims to convey is that $\eta_t$ is perfectly predictable given the past values of the whole process. It does not mean "deterministic" as in "a deterministic rule/function to generate it".

While a covariance stationary process could conceivably be broken into two non-stationary ones, where together the one would "cancel out" the non-stationarity of the other, it could not be decomposed into one stationary and one non-stationary component.

  • $\begingroup$ Thank you. So $\mathbb{E}(\eta_t)$ is constant over time, right? I am still trying to get a better feeling of this "deterministic" term. I wish it were as simple as in my follow up question; is it perhaps? $\endgroup$ Commented Jan 13 at 18:18
  • $\begingroup$ @RichardHardy Yes. As is the variance. The use of the term deterministic is unfortunate here. $\endgroup$ Commented Jan 13 at 20:27
  • $\begingroup$ @RichardHardy I posted also an answer to the follow-up question. $\endgroup$ Commented Jan 13 at 21:59

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.