Working Through Chatfield's Time Series Problems
Chris Chatfield's The Analysis of Time Series is the book most undergrad and graduate programs assign for intro to time series. It covers ARIMA modeling, Box-Jenkins methodology, spectral analysis, state space methods, and multivariate extensions in a fairly classical framework. The solution manual exists because the exercises aren't trivial. They're not trivia problems either. A single exercise on identifying an ARMA order from sample ACF and PACF plots can take you 45 minutes even when you know what you're doing. The solution manual walks through each chapter's end-of-chapter problems. I've used multiple editions across the years, and the coverage is generally thorough. Chapter 4 on ARMA processes, chapter 6 on model identification, and the spectral analysis chapters are where students hit walls most often. The manual shows the steps for computing sample autocorrelations, building likelihood functions for parameter estimation, and working through the Yule-Walker equations by hand. That last one is important because Chatfield makes you do enough manual computation that you actually understand what the algorithms are doing under the hood before they get abstracted away in R or Python. Most copies circulate as scanned PDFs on academic sharing sites. You'll find them under various file names, often with slightly different pagination depending on the edition. The third edition (1996) and the fourth edition (2004) are the most commonly assigned. The second edition (1989) has slightly different problem ordering. If your professor's problem numbers don't match, check which edition you're working from before assuming the manual is wrong.
I ran into a specific issue a couple years ago working through the exercises on seasonal ARIMA identification. The manual presents the seasonal differencing operator as (1-B^s)^D but in one of the worked examples for monthly data, it quietly treats the seasonal period as 12 without explicitly showing the intermediate step where you compute the 12th differences. That omission cost me about twenty minutes of confusion because my manually computed differences were off. Once I went back and expanded (1-B^12) term by term, the rest followed. It's a small gap but one that catches people who are following along step by step rather than just checking answers. There are a few things the manual doesn't make clear and that you need to figure out on your own. One is that Chatfield's approach to model selection relies heavily on the information criteria he introduces, but he doesn't spend much time comparing AIC against BIC or HQ in the exercises. In practice, these can give you different final models on the same data. Another thing is that the manual's handling of nonstationarity diagnostics assumes you've already confirmed stationarity through visual inspection and unit root tests. The book walks through Dickey-Fuller tests but the solution manual rarely shows the test statistic calculations explicitly, so you're often expected to run those in software and interpret them separately. When I work through these problems, I use the manual as a checkpoint, not a crutch. The most useful part is after you've attempted a problem yourself. If you look at the solution first, you lose the diagnostic thinking that the exercises are designed to build. The manual is good for checking your autocorrelation calculations and confirming your identification logic, but it won't teach you to think through a new dataset from scratch. That part comes from doing the problems without the answers in front of you.
One practical note: the manual's numerical answers sometimes differ slightly from what you get using modern software. Chatfield worked many of these by hand or with early computational tools, so rounding at intermediate steps produces results that don't always match what R's arima() or Python's statsmodels will return. If your software output is within a reasonable range of the manual's answer, you're likely fine. Big deviations usually mean you've specified the model differently, not that you've made a calculation error. The book itself is straightforward to use alongside the manual. Work the chapter summaries first, attempt the exercises, then check the manual. It takes more time than skimming solutions, but the material isn't shallow enough to benefit from shortcuts. Most students finish a chapter's problem set in a few hours if they're working carefully. Don't rush it.
Get the Full Details
