Re: [R] R time series analysis
For each arima model, can you output an associated confidence interval for the predicted value at each time point? -- View this message in context: http://r.789695.n4.nabble.com/R-time-series-analysis-tp2527513p2530595.html Sent from the R help mailing list archive at Nabble.com. __ R-help@r-project.org mailing list https://stat.ethz.ch/mailman/listinfo/r-help PLEASE do read the posting guide http://www.R-project.org/posting-guide.html and provide commented, minimal, self-contained, reproducible code.
Re: [R] R time series analysis
On Sep 7, 2010, at 7:51 PM, lord12 wrote: For each arima model, can you output an associated confidence interval for the predicted value at each time point? ?arima0 arima0 will return ... a list with components pred, the predictions, and se, the estimated standard errors as time series when se.fit = TRUE. -- View this message in context: http://r.789695.n4.nabble.com/R-time-series-analysis-tp2527513p2530595.html -- David Winsemius, MD West Hartford, CT __ R-help@r-project.org mailing list https://stat.ethz.ch/mailman/listinfo/r-help PLEASE do read the posting guide http://www.R-project.org/posting-guide.html and provide commented, minimal, self-contained, reproducible code.
Re: [R] R time series analysis
On Sep 7, 2010, at 9:33 PM, David Winsemius wrote: On Sep 7, 2010, at 7:51 PM, lord12 wrote: For each arima model, can you output an associated confidence interval for the predicted value at each time point? ?arima0 arima0 will return ... a list with components pred, the predictions, and se, the estimated standard errors as time series when se.fit = TRUE. See also: predict.Arima {stats} -- View this message in context: http://r.789695.n4.nabble.com/R-time-series-analysis-tp2527513p2530595.html -- David Winsemius, MD West Hartford, CT __ R-help@r-project.org mailing list https://stat.ethz.ch/mailman/listinfo/r-help PLEASE do read the posting guide http://www.R-project.org/posting-guide.html and provide commented, minimal, self-contained, reproducible code. David Winsemius, MD West Hartford, CT __ R-help@r-project.org mailing list https://stat.ethz.ch/mailman/listinfo/r-help PLEASE do read the posting guide http://www.R-project.org/posting-guide.html and provide commented, minimal, self-contained, reproducible code.
Re: [R] R time series analysis
lord12 wrote: I have a data file with a given time series of price data and I would like to split the time series into a test set and training set. I would then like to build an ARIMA model on the training set and apply this model on test set. I had recently the same problem and, after checking documentation and mailing list archives, I discovered that it is not possible to apply the same model on a different data set. Of course you can create the model on a part of the dataset and then check the prediction with the remaining part, as a testing set. But, if you have new data you and you want to apply the same model on them...nothing! I checked the source code of ARIMA functions but it was too complex and I hadn't enough time to learn all that stuff. However I found a little workaround: 1. I calibrate the model on the training part 2. I create a new model with the same parameters, using fixed (check arima documentation) on the new data 3. go to step 2. every time you have new data It worked for me. -- View this message in context: http://r.789695.n4.nabble.com/R-time-series-analysis-tp2527513p2528200.html Sent from the R help mailing list archive at Nabble.com. __ R-help@r-project.org mailing list https://stat.ethz.ch/mailman/listinfo/r-help PLEASE do read the posting guide http://www.R-project.org/posting-guide.html and provide commented, minimal, self-contained, reproducible code.
Re: [R] R time series analysis
How do I get the predicted values and the errors for each arima model? -- View this message in context: http://r.789695.n4.nabble.com/R-time-series-analysis-tp2527513p2527533.html Sent from the R help mailing list archive at Nabble.com. __ R-help@r-project.org mailing list https://stat.ethz.ch/mailman/listinfo/r-help PLEASE do read the posting guide http://www.R-project.org/posting-guide.html and provide commented, minimal, self-contained, reproducible code.
Re: [R] R time series analysis
How do you evaluate the predictive models? For example if I have: arima1 = arima(training, order = c(1,1,1)) arima2 = arima(training, order = c(0,0,0)) x.fore = predict(arima1, n.ahead=5) x.fore1 = predict(arima2, n.ahead = 5) How do I know which arima model is better for prediction? -- View this message in context: http://r.789695.n4.nabble.com/R-time-series-analysis-tp2527513p2527672.html Sent from the R help mailing list archive at Nabble.com. __ R-help@r-project.org mailing list https://stat.ethz.ch/mailman/listinfo/r-help PLEASE do read the posting guide http://www.R-project.org/posting-guide.html and provide commented, minimal, self-contained, reproducible code.
Re: [R] R time series analysis
Hi, basically, you know 5 periods later. If you use a good error measure, that is. I am a big believer in AIC for model selection. I believe that arima() also gives you the AIC of a fitted model, or try AIC(arima1). Other ideas include keeping a holdout sample or some such. I'd recommend looking at a time series textbook. HTH, Stephan Am 05.09.2010 22:37, schrieb lord12: How do you evaluate the predictive models? For example if I have: arima1 = arima(training, order = c(1,1,1)) arima2 = arima(training, order = c(0,0,0)) x.fore = predict(arima1, n.ahead=5) x.fore1 = predict(arima2, n.ahead = 5) How do I know which arima model is better for prediction? __ R-help@r-project.org mailing list https://stat.ethz.ch/mailman/listinfo/r-help PLEASE do read the posting guide http://www.R-project.org/posting-guide.html and provide commented, minimal, self-contained, reproducible code.
Re: [R] R time series analysis
You also may want to look at auto.arima in the 'forecast' package, which will return the best ARIMA model based on AIC/AICc/BIC values ?auto.arima hth c On 09/05/2010 06:02 PM, Stephan Kolassa wrote: Hi, basically, you know 5 periods later. If you use a good error measure, that is. I am a big believer in AIC for model selection. I believe that arima() also gives you the AIC of a fitted model, or try AIC(arima1). Other ideas include keeping a holdout sample or some such. I'd recommend looking at a time series textbook. HTH, Stephan Am 05.09.2010 22:37, schrieb lord12: How do you evaluate the predictive models? For example if I have: arima1 = arima(training, order = c(1,1,1)) arima2 = arima(training, order = c(0,0,0)) x.fore = predict(arima1, n.ahead=5) x.fore1 = predict(arima2, n.ahead = 5) How do I know which arima model is better for prediction? __ R-help@r-project.org mailing list https://stat.ethz.ch/mailman/listinfo/r-help PLEASE do read the posting guide http://www.R-project.org/posting-guide.html and provide commented, minimal, self-contained, reproducible code. __ R-help@r-project.org mailing list https://stat.ethz.ch/mailman/listinfo/r-help PLEASE do read the posting guide http://www.R-project.org/posting-guide.html and provide commented, minimal, self-contained, reproducible code.