The following function estimates MAPA and produces out-of-sample forecasts. It is not suggested, unless only simple single origin forecasts are required.
mapasimple(y, ppy=NULL, fh=ppy, minimumAL=1, maximumAL=ppy,
comb=c("w.mean","w.median","mean","median","wght"),
paral=c(0,1,2), display=c(0,1),
outplot=c(0,1), hybrid=c(TRUE,FALSE), model="ZZZ",
type=c("ets","es"), xreg=NULL, pr.comp=0, ...)Vector with forecasts.
Array with MAPA components, if output="all".
In sample observations of a time series (vector). If y == "paper" then it prints paper reference.
Periods in a season of the time series at the sampled frequency. If insample is a ts object then this is taken from its frequency, unless overriden.
Forecast horizon. Default = ppy.
Lowest aggregation level to use. Default = 1, maximumAL>1.
Highest aggregation level to use. Default = ppy.
Combination operator. This can be: "mean"; "median"; "wght" - where each aggregation level is weighted inversly to aggregation; "w.mean" - level and trend components are averaged, but seasonal and xreg follow the wght combination; "w.median" - as w.mean, but with median. It is suggested that for data with high sampling frequency to use one of the "w.mean" and "w.median".
Use parallel processing. 0 = no; 1 = yes (requires initialised cluster); 2 = yes and initialise cluster. Default is 0.
Display calculation progress in console. 0 = no; 1 = yes. Default is 0.
Provide output plot. 0 = no; 1 = time series and forecast only; 2 = time series, forecasts and components. For the components the spectral colouring scheme is used. Dark red is aggregation level 1. Default is 1.
Provide hybrid forecasts, as in Kourentzes et al. paper. If minimumAL > 1 then the minimumAL ETS forecasts are used. Default is TRUE.
Allow only that type of ETS at each aggregation level. This follows similar coding to the ets function. The first letter refers to the error type ("A", "M" or "Z"); the second letter refers to the trend type ("N","A","Ad","M","Md", "Z", "X" or "Y"); and the third letter refers to the season type ("N","A","M", "Z", "X" or "Y"). The letters mean: "N"=none, "A"=additive, "M"=multiplicative, "Z"=automatically selected, "X"=automatically select between none and additive and "Y"=automatically select between none and multiplicative. A "d" for trend implies damped. "X" and "Y" supported only by type=="es". If used with type=="ets" a warning will be given and they will default to "Z". By default model="ZZZ". If due to sample limitation ETS cannot be calculated at an aggregation level for the selected model, then no estimation is done for that specific level. For aggregation levels that seasonality becomes 1 then a non-seasonal model is estimated.
What type of exponential smoothing implementation to use. "es" = use from the smooth package; "ets" = use from the forecast package. Default is "es"
Vector or matrix of exogenous variables to be included in the MAPA. If matrix then rows are observations and columns are variables. Must be at least as long as in-sample. Additional observations are unused. Note that including xreg will force type="es".
MAPAx can use principal component analysis to preprocess xreg. When comp is -1 then the number of retained components is chosen automatically. When comp=0 then no pre-processing is performed and the original xreg is used. Any other value represents the number of principal components retained.
Pass additional arguments to es or ets.
Nikolaos Kourentzes, nikolaos@kourentzes.com; Fotios Petropoulos.
This function calls mapaest and mapacalc internally.
Kourentzes N., Petropoulos F., Trapero J.R. (2014) Improving forecasting by estimating time series structural components across multiple frequencies. International Journal of Forecasting, 30(2), 291--302.
Kourentzes N., Petropoulos F. (2015) Forecasting with multivariate temporal aggregation: The case of promotional modelling. International Journal of Production Economics.
You can find more information about MAPA at Nikos' blog.
mapa.