A brainstorming session about forecasting time series in R.
First we discussed the topic data structures. Regular time series data can be represented using the ts class which is already included in the basic R installation (as part of the stats package). By contrast, irregular time series data can be represented by the popular zoo objects (or xts). Most (ARIMA-)forecasting methods require regular time series.
The following R packages have been identified to support time series forecasting:
- forecast: Excellent package by Rob Hyndman supporting all kinds of ARIMA models; even includes an automatic forecasting function (auto.arima).
- tseries: Package supporting basic ARIMA model but also includes GARCH for volatility forecasting.
- x12: Interface package to the X12-ARIMA program for sesonal adjustment
- seasonal: Interface package to the X-13-ARIMA-SEATS program for sesonal adjustment.
The following potential projects covering simple time series forecasting examples have been discussed/requested:
- Google Trends data (e.g. Influenza)
- Company data, e.g. Revenues
- Bike Traffic in Vienna
- Stock Prices, Volatility (GARCH)
- Birdlife data, see data for Bavaria e.g. 2016, 2015, 2014, 2013, 2012, 2011, 2010, 2009
The projects will be presented during the November R-Meetup (contributions welcome!).
The next meetup will cover the sparklyr package presented by Roland Boubela.