11

I have a dataset where empirical intuition say I should expect a weekly seasonality (i.e., the behavior in saturday and sunday is different from the rest of the week). Should this premise be true, shouldn't an autocorrelation graph give me bursts at lag multiples of 7?

Here's a sample of the data:

data = TemporalData[{{{2012, 09, 28}, 19160768}, {{2012, 09, 19}, 
    19607936}, {{2012, 09, 08}, 7867456}, {{2012, 09, 15}, 
    11245024}, {{2012, 09, 04}, 0}, {{2012, 09, 21}, 
    24314496}, {{2012, 09, 12}, 11233632}, {{2012, 09, 03}, 
    9886496}, {{2012, 09, 09}, 9122272}, {{2012, 09, 24}, 
    23103456}, {{2012, 09, 20}, 25721472}, {{2012, 09, 11}, 
    12272160}, {{2012, 09, 25}, 21876960}, {{2012, 09, 05}, 
    7182528}, {{2012, 09, 16}, 11754752}, {{2012, 09, 23}, 
    23737248}, {{2012, 09, 26}, 20985984}, {{2012, 09, 10}, 
    12123584}, {{2012, 09, 06}, 9076736}, {{2012, 09, 17}, 
    20123328}, {{2012, 09, 18}, 20634720}, {{2012, 09, 22}, 
    23361024}, {{2012, 09, 14}, 11804928}, {{2012, 09, 07}, 
    9007200}, {{2012, 09, 02}, 9244192}, {{2012, 09, 13}, 
    11335328}, {{2012, 09, 27}, 20694720}, {{2012, 10, 26}, 
    12242112}, {{2012, 10, 15}, 10963776}, {{2012, 11, 09}, 
    9735424}, {{2012, 10, 08}, 10078240}, {{2012, 10, 31}, 
    10676736}, {{2012, 10, 20}, 11719840}, {{2012, 11, 05}, 
    10475168}, {{2012, 10, 01}, 9988416}, {{2012, 10, 24}, 
    11998688}, {{2012, 10, 12}, 10393120}, {{2012, 10, 23}, 
    11987936}, {{2012, 10, 19}, 11165536}, {{2012, 10, 04}, 
    9902720}, {{2012, 11, 16}, 10023648}, {{2012, 11, 21}, 
    10047936}, {{2012, 10, 10}, 10205568}, {{2012, 11, 08}, 
    9872832}, {{2012, 10, 21}, 12854112}, {{2012, 11, 04}, 
    10485856}, {{2012, 10, 07}, 9565248}, {{2012, 09, 30}, 
    9784864}, {{2012, 10, 29}, 12880064}, {{2012, 11, 10}, 
    8945824}, {{2012, 11, 15}, 9870880}, {{2012, 09, 29}, 
    9718080}, {{2012, 10, 18}, 10992896}, {{2012, 10, 06}, 
    9319584}, {{2012, 11, 03}, 9077024}, {{2012, 10, 03}, 
    10537408}, {{2012, 11, 22}, 9853216}, {{2012, 10, 11}, 
    10191936}, {{2012, 10, 22}, 12766816}, {{2012, 11, 07}, 
    9510624}, {{2012, 11, 14}, 9707264}, {{2012, 10, 28}, 
    12060736}, {{2012, 11, 19}, 10946880}, {{2012, 11, 11}, 
    9529568}, {{2012, 10, 09}, 9967680}, {{2012, 10, 17}, 
    12093344}, {{2012, 11, 20}, 10520800}, {{2012, 10, 05}, 
    9619136}, {{2012, 10, 25}, 11484288}, {{2012, 11, 17}, 
    9389312}, {{2012, 10, 30}, 12078944}, {{2012, 10, 14}, 
    9505984}, {{2012, 10, 02}, 9943648}, {{2012, 11, 24}, 
    9458144}, {{2012, 11, 02}, 10082944}, {{2012, 11, 01}, 
    11082912}, {{2012, 10, 13}, 9117632}, {{2012, 11, 23}, 
    10253280}, {{2012, 11, 12}, 10240672}, {{2012, 11, 06}, 
    9723456}, {{2012, 11, 13}, 9806880}, {{2012, 10, 16}, 
    12368896}, {{2012, 11, 18}, 9632800}, {{2012, 10, 27}, 10606656}}]

... and the ACF:

enter image description here

... and the PACF:

enter image description here

Hugo Sereno Ferreira
  • 619
  • 2
  • 6
  • 18
  • 4
    Perhaps your intuition is wrong? I personally like to look at boxplots by day of week. How do those look? Alternatively, you could look at seasonal plots, plotting your variable of interest against day of week for multiple weeks, like this (but with day of week instead of month on the horizontal axis): http://otexts.com/fppfigs/a10b.png – Stephan Kolassa Dec 05 '12 at 16:48
  • 1
    Have you looked at [this](http://onlinelibrary.wiley.com/doi/10.1111/j.1467-9892.1984.tb00392.x/abstract)? – tchakravarty Dec 11 '12 at 04:42

2 Answers2

16

First, here is your intuition illustrated in a simplified time series where the weekend is readily apparent in the ACF:

enter image description here However, this expected ACF pattern can be masked when the data have some trend: enter image description here enter image description here

A solution (if this is a problem) is to estimate and control for the trend when determining the seasonality.

R code that produced these plots follows:

# fourteen repeating 'weeks' of five zeroes and two ones
weekendeffect <- rep(c(rep(0,5),1,1),times=14)

plot(weekendeffect,
    main="Weekly pattern of five zeroes & two ones",
    xlab="Time", ylab="Value")  
acf(weekendeffect, main="ACF")

# add steady trend 
dailydrift <- 0.05
drift <- seq(from=dailydrift, to=length(weekendeffect)*dailydrift, 
   by=dailydrift)
driftingtimeseries <- drift + weekendeffect 

plot(driftingtimeseries,
    main=c("Weekly pattern with daily drift of",dailydrift),
    xlab="Time", ylab="Value")  
acf(driftingtimeseries, main=c("ACF with daily drift of",dailydrift))


# add larger trend 
dailydrift <- 0.1
drift <- seq(from=dailydrift, to=length(weekendeffect)*dailydrift, 
   by=dailydrift)
driftingtimeseries <- drift + weekendeffect 

plot(driftingtimeseries,
    main=c("Weekly pattern with daily drift of",dailydrift),
    xlab="Time", ylab="value")  
acf(driftingtimeseries, main=c("ACF with daily drift of",dailydrift))
MattBagg
  • 1,643
  • 14
  • 17
0

Have you used a differencing technique to make your data stationary? your ACF plot suggests that maybe you have not done this step. Once you have a stationary series then it will be easier to interpret the plots. I add two University sources that might assist you with differencing and interpreting.

The Pennsylvania State University

Duke University

Jordi Perepérez
  • 1
  • 1