Looking for an Time Series Analysis Text (Apologies, I could not think of a better title).
Two days ago, I was in the Library at Köln (Those who know it, knows that the books are not ordered by subject - but by date of buying, and one has to do catalogue research). and I was casually browsing the racks for things that might interest me. Now I did find a book.
All what I remember is, it was a book of Non-linear analysis of time series. The author, at the beginning of some chapter (either 3,4 or 5 - or may be I am wrong),  argues that in contrast to the previous chapters where he presented interpolation (e.g. B-Splines) and nearest neighbor techniques, in the present chapter he will present techniques that also make assumptions on the underlying dynamic process that generate the time series. 
The actual sentences were like, not that i remeber the actual words for which I could possibly search insides of amazon or google books, but it went along these lines: "it thus seems to be a good idea [...] to account for the underlying processes that generate the time series". 
However, afterwards, I found this interesting. So I wanted to read more, but I had to come back to Bonn - thus I wanted to check the book out, but I had some fines and all - thus I just kept the book there and thought I will be back later.
Problem is, as I was rushing back to catch the train to Bonn, I forgot to write down the name and author of the book. Going through the search result returned by the catalogue does not seem to lead me to the book I was looking for - the book feels like as if vanished.
Can anyone help be by telling me the name of a book where the author argues as above? 
 A: 
it thus seems to be a good idea [...] to account for the underlying processes that generate the time series

This is a very general statement, which in a field like econometrics is in the foundation of time series analysis. The matter is that unlike in Physics and other natural sciences, the series are often not repeatable. Not all field have a luxury of an experiment, the researchers in social sciences are often confined in the observational realm. Also the data series are very short, consider GDP time series: they're at quarterly frequency and go back to a few decades at most.
In these circumstances, in social sciences and some other fields a typical time series tools such as spectral analysis with FFT are difficult to apply. So people came up with other means, such as autoregressive processes AR(P), where P is the order of the autocorrelation: $x_t=\phi_0+\sum_{i=1}^P\phi_ix_{t-i}+\varepsilon_t$. So when you use a certain P or stating that $\varepsilon_t\sim\mathcal{N}(0,\sigma)$, you're making a statement about the underlying process. This is called data generation process or DGP. This particular DGP leads to certain shape of the spectrum, as identified by Yule-Walker method.
So, your recollection of the statement from the book in my opinion boils down to using parametric methods vs. nonparametric methods of time series analysis. The former, such as FFT, do not take into account the underlying process. And parametric methods such as AR(P) explicitly account for your theory around the underlying DGP.
