Rolling cross-correlation at given lags.

Rolling cross-correlation at given lags. Can be useful to show how the relationship between two time series changes over time, including out-by-one timing errors.

rollccf(
  DATA = data.frame(Q = , P = ),
  width = list(365, 90),
  by = 28,
  lags = base.lag + c(0, 1, -1),
  base.lag = estimateDelay(DATA, rises = rises, plot = FALSE),
  rises = FALSE,
  na.action = na.contiguous,
  na.max.fraction = 1/3
)

# S3 method for rollccf
xyplot(
  x,
  data = NULL,
  ...,
  with.data = TRUE,
  type = list(c("h", "b")),
  type.data = "l",
  par.settings = simpleTheme(pch = ".", cex = 2),
  layout = c(1, length(x$rolls) + with.data * 2),
  strip = strip.default,
  ylim = c(0, 1),
  xlab = NULL,
  as.table = TRUE
)

ccfForLags(DATA, lags = 0, na.action = na.contiguous, na.max.fraction = 1/3)

Arguments

DATA

a named list, data frame, time series or zoo object containing the two data series.

width

a list or number specifying the width of window(s), in time steps, in which to calculate cross correlations.

by

temporal resolution: cross correlation is calculated in windows spaced every by time steps.

lags, base.lag

lags for which to calculate the cross correlation. By default these are based on the overall maximum cross correlation, base.lag.

rises

if TRUE, compute the cross correlation with rises in streamflow. In this case the streamflow series must be named "Q".

na.action

function to handle missing data in each window (not the whole series). This is only applied when the number of missing values is less than na.max.fraction.

Could be na.exclude.

na.max.fraction

if the proportion of missing values in the moving window exceeds this value, the corresponding result will be NA.

x, data

x is an object produced by the rollccf function. data is ignored.

with.data

if TRUE, include the original data series in the plot.

type, type.data

drawing styles for the cross correlation series and input data series. See panel.xyplot.

par.settings, layout, strip, ylim, xlab, as.table, ...

passed to xyplot.

Value

rollccf returns a list of class "rollccf", with components:

rolls: a list of time series of cross correlations. One element for each value of width.
data: time series input data.
lags, width, call: values of arguments used.

Details

This is a fairly straightforward application of rollapply with the ccf function. It may be better to do a time-varying regression between the two series.

Author

Felix Andrews felix@nfrac.org

Examples


data(Canning)
foo <- rollccf(Canning)
xyplot(foo)