Time Series Data Library now on DataMarket

The Time Series Data Library is a col­lec­tion of about 800 time series that I have main­tained since about 1992, and hosted on my per­sonal web­site. It includes data from a lot of time series text­books, as well as many other series that I’ve either col­lected for stu­dent projects or help­ful peo­ple have sent to me.

I’ve now moved the col­lec­tion onto Data­Mar­ket which pro­vides much bet­ter facil­i­ties for main­tain­ing and using time series data. You can eas­ily search the col­lec­tion, graph any series, fil­ter by sea­sonal period, and so on. You can also export data in many for­mats. Each data set has its own short link; for exam­ple, the famous Cana­dian lynx data is at http://​data​.is/​K​y69xY.

One par­tic­u­larly use­ful fea­ture is the abil­ity to read directly into R using the rdata­mar­ket pack­age. All you need to know is the short link. For exam­ple, to down­load “Deaths from gun-​​related homi­cides in Aus­tralia, 1915–2004″, use the fol­low­ing R code

deaths <- dmseries("http://data.is/Ky6vVf")

The data is set to zoo class. To make it of ts class, use

deaths <- as.ts(deaths[,1])

In this case, deaths only con­tained one col­umn, but in gen­eral mul­ti­vari­ate time series can be down­loaded in this manner.

Data­Mar­ket con­tains thou­sands of other time series from orga­ni­za­tions includ­ing Euro­stat, the IMF, the United Nations, Gap­min­der, and many more. Some time series require a sub­scrip­tion, but many can be used freely. The time series in the TSDL will remain freely available.

I’m grate­ful to Data­Mar­ket for agree­ing to host my library with­out charge, and I encour­age every­one inter­ested in time series analy­sis to check them out.

If you use any data from the TSDL in a pub­li­ca­tion, please use the fol­low­ing citation:

Hyn­d­man, R.J. Time Series Data Library, http://​data​.is/​T​S​D​Ldemo. Accessed on <insert date here>.

The data files will remain on my web­site so that exist­ing links will not be bro­ken.

Related Posts:

  • Pingback: Hyndsight - Time series data in R()

  • Din Vad­hia

    Hi! Has the Time-​​Series Data Library been cat­e­gorised by lin­ear and non-​​linear?

    • http://robjhyndman.com/ Rob J Hyndman

      You need to define what you mean by “lin­ear”. I sus­pect you mean “is each time series bet­ter fore­cast using a non­lin­ear autore­gres­sive model or a lin­ear autore­gres­sive model?” I’ve never tested that. Pre­sum­ably there are some of each.

  • Nel­son Acuna

    I have a ques­tion. Are the datasets com­plete, should I ass­sume the datasets might have miss­ing val­ues or does it depend on the dataset?

    • http://robjhyndman.com/ Rob J Hyndman

      Some data sets have miss­ing values