regression.holiday {bsts}  R Documentation 
Add a regressionbased holiday model to the state specification.
AddRegressionHoliday( state.specification = NULL, y, holiday.list, time0 = NULL, prior = NULL, sdy = sd(as.numeric(y), na.rm = TRUE)) AddHierarchicalRegressionHoliday( state.specification = NULL, y, holiday.list, coefficient.mean.prior = NULL, coefficient.variance.prior = NULL, time0 = NULL, sdy = sd(as.numeric(y), na.rm = TRUE))
state.specification 
A list of state components that you wish to add to. If omitted, an empty list will be assumed. 
holiday.list 
A list of objects of type 
y 
The time series to be modeled, as a numeric vector
convertible to 
prior 
An object of class 
coefficient.mean.prior 
An object of type

coefficient.variance.prior 
An object of type

time0 
An object convertible to 
sdy 
The standard deviation of the series to be modeled. This
will be ignored if 
The model assumes that
y[t] = beta[d(t)] + observation_error
The regression state model assumes vector of regression coefficients beta contains elements beta[d] ~ N(0, sigma).
The HierarchicalRegressionHolidayModel assumes beta is composed of holidayspecific subvectors beta[h, ] ~ N(b0, V), where each beta[h,] contains coefficients describing the days in the influence window of holiday h. The hierarchical version of the model treats b0 and V as parameters to be learned, with prior distributions
b0 ~ N(b.bar, Omega)
and
V ~ IW(nu, S).
where IW represents the inverse Wishart distribution.
Returns a list with the elements necessary to specify a local linear trend state model.
Steven L. Scott steve.the.bayesian@gmail.com
Harvey (1990), "Forecasting, structural time series, and the Kalman filter", Cambridge University Press.
Durbin and Koopman (2001), "Time series analysis by state space methods", Oxford University Press.
bsts
.
RandomWalkHolidayStateModel
.
SdPrior
NormalPrior
trend < cumsum(rnorm(730, 0, .1)) dates < seq.Date(from = as.Date("20140101"), length = length(trend), by = "day") y < zoo(trend + rnorm(length(trend), 0, .2), dates) AddHolidayEffect < function(y, dates, effect) { ## Adds a holiday effect to simulated data. ## Args: ## y: A zoo time series, with Dates for indices. ## dates: The dates of the holidays. ## effect: A vector of holiday effects of odd length. The central effect is ## the main holiday, with a symmetric influence window on either side. ## Returns: ## y, with the holiday effects added. time < dates  (length(effect)  1) / 2 for (i in 1:length(effect)) { y[time] < y[time] + effect[i] time < time + 1 } return(y) } ## Define some holidays. memorial.day < NamedHoliday("MemorialDay") memorial.day.effect < c(.3, 3, .5) memorial.day.dates < as.Date(c("20140526", "20150525")) y < AddHolidayEffect(y, memorial.day.dates, memorial.day.effect) presidents.day < NamedHoliday("PresidentsDay") presidents.day.effect < c(.5, 2, .25) presidents.day.dates < as.Date(c("20140217", "20150216")) y < AddHolidayEffect(y, presidents.day.dates, presidents.day.effect) labor.day < NamedHoliday("LaborDay") labor.day.effect < c(1, 2, 1) labor.day.dates < as.Date(c("20140901", "20150907")) y < AddHolidayEffect(y, labor.day.dates, labor.day.effect) ## The holidays can be in any order. holiday.list < list(memorial.day, labor.day, presidents.day) ## In a real example you'd want more than 100 MCMC iterations. niter < 100 ## Fit the model ss < AddLocalLevel(list(), y) ss < AddRegressionHoliday(ss, y, holiday.list = holiday.list) model < bsts(y, state.specification = ss, niter = niter) ## Plot all model state components. plot(model, "comp") ## Plot the specific holiday state component. plot(ss[[2]], model) ## Try again with some shrinkage. With only 3 holidays there won't be much ## shrinkage. ss2 < AddLocalLevel(list(), y) ## Plot the specific holiday state component. ss2 < AddHierarchicalRegressionHoliday(ss2, y, holiday.list = holiday.list) model2 < bsts(y, state.specification = ss2, niter = niter) plot(model2, "comp") plot(ss2[[2]], model2)