costat (version 1.1-1)

plottstosscan: Plot the results of a multiple stationarity test on a time series

Description

The tstosscan function applies multiple hypothesis tests to several consecutive sub-portions of a time series of varying width. This function plots the results of such a set of multiple hypothesis tests.

Usage

plottstosscan(x, inc, ttsobj, xlab = "Time", ylab = "x")

Arguments

x

The time series under analysis.

inc

Inc is a number describing the start index of the series. E.g. if you analyzed a portion whose starting index value was, t=1234, then you would want the plot to have a horizontal axis whose initial label was somewhere near 1234, or wherever the time s

ttsobj

The object from link{tstosscan} containing the results of the multiple hypothesis test.

xlab

Supplied as the xlab argument of the plot

ylab

Supplied as the ylab argument of the plot

Value

• Nothing

Details

The plot first draws a light grey time series plot of the underlying time series. The left-hand axis quantifies the values.

Then a series of ``horizontal lines'' is plotted, one for each span specified in the spans argument of the previous call to tstosscan. Each horizontal line is made up of a series of dashes or crosses. A dash indicates that the stationarity test applied to a subseries centred at that location, with the span (width) of time series values at the given span (given by the right hand axis), indicates no evidence for nonstationary (ie assume stationary here for now). A cross indicates that the series portion at that span, at that location is nonstationary.

Overall, this plot can give you an idea of how nonstationary (second order) a time series is. A lot of crosses at small spans can indicate highly non-stationary series. A few crosses, even at large spans, indicate near-stationarity.

It should be remembered that this plot produces a picture that summarizes the results of multiple hypothesis tests. This means that, e.g., for a size of test at 0.05 you would expect 1 in 20 of the results to indicate non-stationarity, even if the underlying time series was completely stationary.

References

`Costationary and stationarity tests for stock index returns' by Car dinali and Nason, 2008, University of Bristol Technical Report 08:08.

See Also

tstosscan

Examples

#
# Compute tests of stationarity on white noise Gaussian
# (not run on installation because its compute intensive)
#
v <- rnorm(128)
v.tstos <- tstosscan(v, spans=c(32,64))

plottstosscan(v, v.rstos, inc=0)