2

I'm trying to get some sensible labels on a forecast.

Here's my code:

library("forecast")

t <- ts(
  c(
    4410.0, 6435.0, 
    4939.0, 6487.0, 25521.0, 18764.0, 
    12223.0, 18590.0, 36898.0, 28826.0, 
    20329.0
    )  
  , frequency = 4
  , start=c(2011, 7, 1)
)

cast <- meanf(t,h=4)

par(mfrow=c(1,1),xaxt="n")
plot(cast)

ticks <- seq(as.POSIXct("2011-07-01"), by = "3 month", length.out=length(cast) + 4) 
axis.Date(1, at = ticks, labels = ticks, format="%Y-%m")

There are no errors or warnings... but no axis gets added to the plot :(

If I use xaxt="s" then I do get axis labels, but they're not they're things like 2014.0 (i.e. not formatted as dates at all). Ideally I'd actually like something a bit more readable like "Q3 2014" but I'd settle for 2014-07 (the first month in the quarter).

Any idea what I'm doing wrong? All of the stuff I've found searching the internet suggests disabling the default axis and then using the axis or axis.Date functions to add a custom on... but I can't get an axis of any sort using either of those methods :(

James Crosswell
  • 648
  • 4
  • 13
  • Interestingly enough, if you do `plot(hw(USAccDeaths,h=48))`, which comes straight from [the help page to `plot.forecast`](http://www.inside-r.org/packages/cran/forecast/docs/plot.forecast), you don't get the time axis annotated, either. I'd say this qualifies as a bug. I'll point Rob Hyndman to this. – Stephan Kolassa Jul 09 '14 at 16:22
  • @Stephan Kolassa If I do `require(forecast); plot(hw(USAccDeaths,h=48))` I see labels in the time axis. Maybe you didn't reset the graphical parameters and still had the configuration `par(mfrow=c(1,1),xaxt="n")` from the example given in the question (that happened to me). – javlacalle Jul 11 '14 at 18:46

1 Answers1

1

I would not say there is a bug, rather the function forecast::plot.forecast is not designed to be used with axis.Date or axis.POSIXct (which are not used in the package forecast).

Before calling the functions axis.Date and axis.POSIXct, the time points must be explicitly passed to plot as a sequence of Date or POSIXct objects. If we plot the time series as plot(t) then the x-axis does not seem to be properly defined as to be used by the above functions.

See the code below and how the variable time is created and passed to plot. plot(time, x) is used instead of just plot(x). Thus the function axis will be able to display the labels of the time axis. (In this example, the confidence intervals will not look as nice as those displayed by forecast::plot.forecast.)

library("forecast")
require(zoo)

t <- ts(
  c(
    4410.0, 6435.0, 
    4939.0, 6487.0, 25521.0, 18764.0, 
    12223.0, 18590.0, 36898.0, 28826.0, 
    20329.0
    )  
  , frequency = 4
  , start=c(2011, 7, 1)
)
cast <- meanf(t, h=4)

x <- ts(c(t, cast$mean), start = c(start(t), 1), frequency = frequency(t))
time <- as.yearqtr(seq(start(x)[1], end(x)[1] + 1, 1/frequency(x)))
time <- time[seq_along(x)]
time <- as.POSIXct(time)
plot(time, x, type = "n", xaxt = "n", ylim = range(c(cast$lower, cast$upper)))
lines(time[seq_along(t)], t)
lines(time[-seq_along(t)], cast$mean, col = "blue")
lines(time[-seq_along(t)], cast$upper[,2], col = "red")
lines(time[-seq_along(t)], cast$lower[,2], col = "red")
ticks <- seq(as.POSIXct("2011-07-01"), by = "3 month", length.out=length(cast) + 4) 
axis(side = 1, at = ticks, labels = ticks)
javlacalle
  • 1,029
  • 7
  • 15