simple.R contains a set of simple working examples about how to deal with time series.

############### Simple ###############

Mean : mean of historical data

forcastMean = meanf(myTs, h=10) plot(forcastMean)

Naive : last observed value

forcastNaive = naive(myTs, h=10) plot(forcastNaive)

Seasonal naive : last value from same season (uses frequency which is the nb of observation per periode of time. Ex: frequency=25)

forcastSeasonalNaive = snaive(myTs, h=50) plot(forcastSeasonalNaive)

Drift : last value plus average change

forcastDrift = rwf(myTs, h=10) plot(forcastDrift)

############### STLF (1990) ###############

STLF : applying a non-seasonal forecasting method to the seasonally adjusted data and re-seasonalizing using the last year of the seasonal component.

forcastSTLF = stlf(myTs, h=50) plot(forcastSTLF)

############### ETS (2008) ###############

ETS Exponential smoothing forecasts : ses method

forcastSES = ses(myTs, h=10) plot(forcastSES)

ETS Exponential smoothing forecasts : holt method (N,N), (A,N), (A,A) and (A,M).

forcastHolt = holt(myTs, h=10) plot(forcastHolt)

ETS Exponential smoothing forecasts : hw method

forcastHw = hw(myTs, h=50) plot(forcastHw)

ETS Exponential smoothing forecasts : custom model (Error, Trend, Seasonal) : "N"=none, "A"=additive, "M"=multiplicative and "Z"=automatically selected.

fitETS = ets(myTs, model="ZAA", damped=FALSE) forcastETS = forecast(fitETS, h=50) plot(forcastETS) accuracy(forcastETS, tail(myTs, 25)) # secound argument is the values to compare

ETS Exponential smoothing forecasts : Box-Cox to stabilize the variance (data show different variation at different levels of the series)

lam = BoxCox.lambda(myTs) # = 1.08781474274726 fitETSBoxCox = ets(myTs, additive=TRUE, lambda=lam) forcastETSBoxCox = forecast(fitETSBoxCox, h=10) plot(forcastETSBoxCox)

############### Arima (2008) ###############

Arima : custom

modelArima <- auto.arima(myTs) forcastArima <- forecast(modelArima, h=50) plot(forcastArima)

############### Auto ###############

Forecast : Automatically uses ETS (if period < 13) else STLF

forcastAuto = forecast(myTs, h=50) plot(forcastAuto) forcastAuto$model # view the selected model

############### Others ###############

Croston (1972) : simple exponential smoothing (SES)

forcastCroston = croston(myTs, h=10) plot(forcastCroston)

Cubic Spline Forecast (2005) : uses cubic smoothing splines

forcastSplinef = splinef(myTs, h=10) plot(forcastSplinef)

Theta method forecast (2000) : Equivalent to simple exponential smoothing with drift (2000)

forcastSplinef = thetaf(myTs, h=10) plot(forcastSplinef)

Double-Seasonal Holt-Winters (2003) : Multiple seasonal periods

forcastSplinef = dshw(myTs, period1=24, period2=3) plot(forcastSplinef)

############### TimeSeries Datasets examples ############### library(fma) data(package="fma") plot(BoxCox(dowjones,1))

ArnaudC / timeSeries