Hi, I have an nxp data set of time series. For my final year project, I would like to partition this data set into a smaller number of units ( < n). Where each of the units contains time series that move closely together (i.e. in unison), and the "simmilarity" (in terms of occurence & amplitudes of peaks and troughs) between the segregated units is "low". I know these are slightly "fuzzy" terms, but I want to get to the essence of what I'm trying to achieve (time series partitioning), without throwing in words like correlation or cointegration which are "trigger" words. I would like to know if anyone has done this kind of partitioning before ? Any tips or suggestions as to how to carry out such a partioning in R will be MOST welcome. Ideally, I would like to automate this partitioning through an R "script" since n will be approx 500, and it will not be feasible to do this manually. Any pointers/help will be very much appreciated Regards _________________________________________________________________ -.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.- r-help mailing list -- Read http://www.ci.tuwien.ac.at/~hornik/R/R-FAQ.html Send "info", "help", or "[un]subscribe" (in the "body", not the subject !) To: r-help-request at stat.math.ethz.ch _._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._
Neil Osborne wrote:> > Hi, > > I have an nxp data set of time series. For my final year project, I would > like to partition this data set into a smaller number of units ( < n). Where > each of the units contains time series that move closely together (i.e. in > unison), and the "simmilarity" (in terms of occurence & amplitudes of peaks > and troughs) between the segregated units is "low". > > I know these are slightly "fuzzy" terms, but I want to get to the essence of > what I'm trying to achieve (time series partitioning), without throwing in > words like correlation or cointegration which are "trigger" words. I would > like to know if anyone has done this kind of partitioning before ?I haven't, only with univariate time series in a regression framework. The next version of the package strucchange will contain functions to estimate breakpoints in regression relationships and thus for partitioning univariate time series. This is mostly based on work of Bai & Perron, (1998, 2001). But maybe this functionality could be used in something like a cointegration regression as well, although I haven't seen such an application and I'm not entirely sure whether it would be sensible. Hope that helps Z> Any tips or suggestions as to how to carry out such a partioning in R will > be MOST welcome. Ideally, I would like to automate this partitioning through > an R "script" since n will be approx 500, and it will not be feasible to do > this manually. > > Any pointers/help will be very much appreciated > > Regards > > _________________________________________________________________ > > -.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.- > r-help mailing list -- Read http://www.ci.tuwien.ac.at/~hornik/R/R-FAQ.html > Send "info", "help", or "[un]subscribe" > (in the "body", not the subject !) To: r-help-request at stat.math.ethz.ch > _._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.- r-help mailing list -- Read http://www.ci.tuwien.ac.at/~hornik/R/R-FAQ.html Send "info", "help", or "[un]subscribe" (in the "body", not the subject !) To: r-help-request at stat.math.ethz.ch _._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._
Neil There is some related work by Stock and Watson, Forni and Reichlin, and Quah and Sargent (see below) which tries to extact "dynamic factors" from large cross sections of time series. As far as I know, no one has gone the second step of clustering the data, but their work might be a good starting point. (Sounds like a pretty ambitious term project paper though.) As far as I know, no one has implemented any of this in R (but I would be happy to hear otherwise). I am working on something that is (marginally) related. Paul Gilbert Forni, M., and L. Reichlin 1996. "Dynamic Common Factors in Large Cross-Sections" Empirical Economics 21: 27-42. Quah, D., and T. J. Sargent 1994. "A dynamic index model for large cross sections." in Business Cycles, Indicators, and Forecasting, J. Stock and M. Watson (eds). NBER and Univ. Press Chicago. Stock, J. H. and Watson, M. W. 1999 "Forecasting Inflation," Journal of Monetary Economics 44: 293-335. Neil Osborne wrote:> > Hi, > > I have an nxp data set of time series. For my final year project, I would > like to partition this data set into a smaller number of units ( < n). Where > each of the units contains time series that move closely together (i.e. in > unison), and the "simmilarity" (in terms of occurence & amplitudes of peaks > and troughs) between the segregated units is "low". > > I know these are slightly "fuzzy" terms, but I want to get to the essence of > what I'm trying to achieve (time series partitioning), without throwing in > words like correlation or cointegration which are "trigger" words. I would > like to know if anyone has done this kind of partitioning before ? > > Any tips or suggestions as to how to carry out such a partioning in R will > be MOST welcome. Ideally, I would like to automate this partitioning through > an R "script" since n will be approx 500, and it will not be feasible to do > this manually. > > Any pointers/help will be very much appreciated > > Regards-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.- r-help mailing list -- Read http://www.ci.tuwien.ac.at/~hornik/R/R-FAQ.html Send "info", "help", or "[un]subscribe" (in the "body", not the subject !) To: r-help-request at stat.math.ethz.ch _._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._