66 Following

Arbie's Unoriginally Titled Book Blog

It's a blog! Mainly of book reviews.

Currently reading

Nonlinear Time Series Analysis
Thomas Schreiber, Holger Kantz
Progress: 29/320 pages
The Politics of Neurodiversity: Why Public Policy Matters
Dana Lee Baker
Progress: 9/239 pages
Ursula K. Le Guin: Hainish Novels and Stories, Vol. 1: Rocannon's World / Planet of Exile / City of Illusions / The Left Hand of Darkness / The Dispossessed / Stories (The Library of America)
Brian Attebery, Ursula K. Le Guin
Progress: 440/1100 pages
Life and Letters of Charles Darwin - Volume 1: By Charles Darwin - Illustrated
Charles Darwin
Progress: 332/346 pages
Basics of Plasma Astrophysics
Claudio Chiuderi, Marco Velli
Progress: 58/250 pages
Ursula K. Le Guin: The Complete Orsinia: Malafrena / Stories and Songs (The Library of America)
Brian Attebery, Ursula K. Le Guin
Progress: 359/700 pages
A Student's Guide to Lagrangians and Hamiltonians
Patrick Hamill
Progress: 7/180 pages
Complete Poems, 1904-1962
E.E. Cummings
Progress: 110/1102 pages
The Complete Plays and Poems
E.D. Pendry, J.C. Maxwell, Christopher Marlowe
She Stoops to Conquer and Other Comedies (Oxford World's Classics)
Henry Fielding, David Garrick, Oliver Goldsmith
Progress: 76/448 pages

Time Series Analysis and its Applications, Shumway and Stoffer

Time Series Analysis and Its Applications: With R Examples (Springer Texts in Statistics) - David S. Stoffer, Robert H. Shumway

I read the first ~100p of this book. I stopped because the subject matter had diverged too far from my area of immediate interest (which was covered in the first chapter) rather than because the book is bad. In fact I think it is a good introduction to the topic for those with an interest and a background covering "normal" statistics to a level most STEM undergrads would have. Perhaps one thing that became obvious to me by inference should have been made explicit at the outset, which is that the fundamental general approach is as follows:


1. Get time series and plot it.
2. Guess any trends and/or periodicities in the data (various methods)
3. Subtract them (various methods)
4. Examine what's left ("residuals") to see if it behaves like noise (i.e. has some known type of random distribution) (various methods)
5. If it does, YAY! You have a usable model of the time series
6. If it does not, either make further guesses about trends/periodicities in the residuals and repeat from step 2 OR
7. Go back to the original time series and start from step 2 with different guesses about the nature of trends/periodicities


A flow chart of this at the beginning of the book would make what the book is actually about crystal clear.


As mentioned in a status update, the book does not assume the reader is scientifically motivated and does not discuss the meaning or validity of any trends, correlations or periodicities discovered. There are applications where this is entirely legitimate, probably the biggest and most utilised being analysis of financial/economic data for purposes of investment or trading: One only needs a model that works and not an explanation of why it works in order to make practical decisions. I would advise budding scientists to approach with caution, however; this form of analysis can only generate empirical models and hypotheses about why they are true are a separate but essential part of the scientific process. So, for example, if one discovers a model of the form, seasonal oscillation + white noise, describing your time series, one can make predictions about the future but there is no explanation of why the seasonal variation occurs. You are only part way there, scientifically.