Become an expert in R — Interactive courses, Cheat Sheets, certificates and more!
Get Started for Free

genhol

Generate Holiday Regression Variables


Description

A replacement for the genhol software by the U.S. Census Bureau, a utility that uses the same procedure as X-12-ARIMA to create regressors for the U. S. holidays of Easter, Labor Day, and Thanksgiving. This is a replacement written in R, the U.S. Census Bureau software is not needed.

Usage

genhol(x, start = 0, end = 0, frequency = 12, center = "none")

Arguments

x

a vector of class "Date", containing the occurrences of the holiday. It can be generated with as.Date().

start

integer, shifts the start point of the holiday. Use negative values if start is before the specified date.

end

integer, shifts end point of the holiday. Use negative values if end is before the specified date.

frequency

integer, frequency of the resulting series

center

character string. Either "calendar", "mean" or "none" (default). Centering avoids a bias in the resulting series. Use "calendar" for Easter or Chinese New Year, "mean" for Ramadan. See references: Notes on centering holiday.

Details

The resulting time series can be used as a user defined variable in seas(). Usually, you want the holiday effect to be removed from the final series, so you need to specify regression.usertype = "holiday". (The default is to include user defined variables in the final series.)

Value

an object of class "ts" that can be used as a user defined variable in seas().

References

United States Census Bureau, Notes on centering holiday regressors: https://www.census.gov/srd/www/genhol/genhol_center.html

See Also

seas() for the main function of seasonal.

Examples

data(holiday)  # dates of Chinese New Year, Indian Diwali and Easter

### use of genhol

# 10 day before Easter day to one day after, quarterly data:
genhol(easter, start = -10, end = 1, frequency = 4)
genhol(easter, frequency = 2)  # easter is allways in the first half-year

# centering for overall mean or monthly calendar means
genhol(easter, center = "mean")
genhol(easter, center = "calendar")

### replicating X-13's built-in Easter adjustment

# built-in
m1 <- seas(x = AirPassengers,
regression.variables = c("td1coef", "easter[1]", "ao1951.May"),
arima.model = "(0 1 1)(0 1 1)", regression.aictest = NULL,
outlier = NULL, transform.function = "log", x11 = "")
summary(m1)

# user defined variable
ea1 <- genhol(easter, start = -1, end = -1, center = "calendar")

# regression.usertype = "holiday" ensures that the effect is removed from
# the final series.
m2 <- seas(x = AirPassengers,
           regression.variables = c("td1coef", "ao1951.May"),
           xreg = ea1, regression.usertype = "holiday",
           arima.model = "(0 1 1)(0 1 1)", regression.aictest = NULL,
           outlier = NULL, transform.function = "log", x11 = "")
summary(m2)

all.equal(final(m2), final(m1), tolerance = 1e-06)


# with genhol, its possible to do sligtly better, by adjusting the length
# of easter from Friday to Monday:

ea2 <- genhol(easter, start = -2, end = +1, center = "calendar")
m3 <- seas(x = AirPassengers,
           regression.variables = c("td1coef", "ao1951.May"),
           xreg = ea2, regression.usertype = "holiday",
           arima.model = "(0 1 1)(0 1 1)", regression.aictest = NULL,
           outlier = NULL, transform.function = "log", x11 = "")
summary(m3)


### Chinese New Year

data(seasonal)
data(holiday)  # dates of Chinese New Year, Indian Diwali and Easter

# de facto holiday length: http://en.wikipedia.org/wiki/Chinese_New_Year
cny.ts <- genhol(cny, start = 0, end = 6, center = "calendar")

m1 <- seas(x = imp, xreg = cny.ts, regression.usertype = "holiday", x11 = "",
           regression.variables = c("td1coef", "ls1985.Jan", "ls2008.Nov"),
           arima.model = "(0 1 2)(0 1 1)", regression.aictest = NULL,
           outlier = NULL, transform.function = "log")
summary(m1)

# compare to identical no-CNY model
m2 <- seas(x = imp, x11 = "",
           regression.variables = c("td1coef", "ls1985.Jan", "ls2008.Nov"),
           arima.model = "(0 1 2)(0 1 1)", regression.aictest = NULL,
           outlier = NULL, transform.function = "log")
summary(m2)

ts.plot(final(m1), final(m2), col = c("red", "black"))

# modeling complex holiday effects in Chinese imports
# - positive pre-CNY effect
# - negative post-CNY effect
pre_cny <- genhol(cny, start = -6, end = -1, frequency = 12, center = "calendar")
post_cny <- genhol(cny, start = 0, end = 6, frequency = 12, center = "calendar")
m3 <- seas(x = imp, x11 = "",
           xreg = cbind(pre_cny, post_cny), regression.usertype = "holiday",
           x11 = list())
summary(m3)


### Indian Diwali (thanks to Pinaki Mukherjee)

# adjusting Indian industrial production
m4 <- seas(iip,
x11 = "",
xreg = genhol(diwali, start = 0, end = 0, center = "calendar"),
regression.usertype = "holiday"
)
summary(m4)

# without specification of 'regression.usertype', Diwali effects are added
# back to the final series
m5 <- seas(iip,
x11 = "",
xreg = genhol(diwali, start = 0, end = 0, center = "calendar")
)

ts.plot(final(m4), final(m5), col = c("red", "black"))

# plot the Diwali factor in Indian industrial production
plot(series(m4, "regression.holiday"))


### Using genhol to replicate the regARIMA estimation in R

# easter regressor
ea <- genhol(easter, start = -1, end = -1, center = "calendar")
ea <- window(ea, start = start(AirPassengers), end = end(AirPassengers))

# estimating ARIMA model in R base
arima(log(AirPassengers), order = c(0,1,1), seasonal = c(0,1,1), xreg = ea)

summary(seas(AirPassengers, regression.variables = c("easter[1]"),
             regression.aictest = NULL))

# Note that R defines the ARIMA model with negative signs before the MA term,
# X-13 with a positive sign.

seasonal

R Interface to X-13-ARIMA-SEATS

v1.8.2
GPL-3
Authors
Christoph Sax [aut, cre] (<https://orcid.org/0000-0002-7192-7044>), Dirk Eddelbuettel [ctb] (<https://orcid.org/0000-0001-6419-907X>)
Initial release
2021-03-23

We don't support your browser anymore

Please choose more modern alternatives, such as Google Chrome or Mozilla Firefox.