carlo - generate noise with statistics similar to input data.
SYNOPSIS
carlo input_file [-v] [-w|-ar] [-n realizations] [-seed seeds]
[-wo components] [-ssa window] [-sig level]
[-cov2|-cov3]
DESCRIPTION
Carlo will generate a user-specified number of zero-mean realizations
of noise, each of which is the same length as the series in
input_file and each of which contains the same variance
and a memory that is selected by the user. The realizations
can comprise white noise with the same variance as the data,
noise that is reddened according to an autoregressive (AR)
model that is fit to the data, or a partial SSA reconstruction of the
data plus noise that is
white or reddened but with characteristics fit to a reconstruction
of all the remaining components. The reddened noise is
constructed from an AR(1) model fitted to the input series or its
partial SSA reconstruction. If the SSA reconstruction
is selected, then the user also must specify the ssa window size
and the list of components to be
eliminated prior to fitting the AR model. The selected SSA reconstructions
are removed, a noise model fitted, the selected number of time-series
realizations synthesized, and finally the SSA reconstructions
are added back in to each Monte Carlo series. Carlo writes a file
that contains the Monte Carlo realizations, and also provides
the option of using those realizations immediately to develop
error bars for the eigenspectrum plots from ssa. If such
error bars are desired, the user must specify the ssa window size
and a significance level for eigenvalues to be output.
Upon successful completion, carlo produces the following output files:
- mcarlo.out: contains the Monte Carlo realizations (of length
equal to the time series in input_file).
- mcsig.out: contains a list of ssa-component numbers and the
corresponding eigenvalue at the (sig) and (100-sig) percent exceedence
levels, where sig is the user-specified significance level.
input_file The file containing a scalar time series.
The data is assumed to have a constant sampling rate, i.e.,
there are no gaps in the data record.
OPTIONS
- -v
-
Selects verbose mode, which will send progress reports
to the standard output [Default runs "silently"].
- -w
-
Specifies that white-noise realizations be generated.
This is the default.
- -ar
-
Specifies that, instead, an autoregressive model with a single term
be fit to the series in input_file and used to generate
the Monte Carlo realizations. If both -w and
-ar are specified, carlo stops with an error message.
The AR model is always fit using the Burg algorithm.
- -n realizations
-
Specifies the number of Monte Carlo realizations to be generated. [Defaults to 200.]
- -seed seeds
-
Provides a list of seeds to the random number generator.
Up to four integers can be specified in the form i1,i2,i3,i4.
If any entries are left blank, that seed assumes the default value.
Use verbose mode to see the default seed values.
- -wo components
-
Compute realizations of noise plus an SSA reconstruction (RCs) of the
data. components is a list specifying the RCs. The noisy part
can be white or red, and has same statistics as the data
minus the RCs. This option allows the user to test whether the data
can be modelled as a "signal" plus "noise". If components is not
provided, this option will default to -ar. SSA window size used in
reconstruction is window as specified by -ssa.
- -ssa window
-
Subject each realization to the first steps of SSA,
and determine at which points the eigenvalues for SSA component
number are outside of the two-tailed confidence interval specified
by -sig. window specifies the SSA window size to be applied.
- -sig level
-
Specifies the width of the confidence interval used with the -ssa option.
level is an integer in the interval (0,100).
For each SSA component, the 50 - level/2 and 50 + level/2 percentiles
are written to mcsig.out.
Specifying -sig without -ssa has no effect.
[Defaults to 90]
- -cov2
-
Specifies that the Vautard et al. (1992) algorithm be used
to estimate the lag-covariance matrix used in the ssa
operations initiated by either the -wo or -ssa options.
Specifying -cov2 without -wo or -ssa has no effect.
[Defaults to Burg (1978) algorithm.]
- -cov3
-
Specifies that the Broomhead and King (1986) algorithm be used to estimate
the lag-covariance matrix.
EXAMPLES
Given a time series in tseries.dat, generate 500 Monte Carlo
series (of the same length as tseries.dat) of the default white noise
with zero mean and standard deviation equal to that of tseries.dat:
carlo -n 500 tseries.dat
Given a time series in tsery.dat, generate 200 Monte Carlo
series (of the same length as tsery.dat) of zero-mean noise
colored according to an AR(1) model fitted to the tsery.dat series,
in verbose mode:
carlo tsery.dat -v -ar
Given a time series in tseries.dat, generate 100 Monte Carlo
realizations of AR(1)-colored noise with variance and memory
similar to tseries.dat, and then calculate the 99 percent confidence
bound on SSA eigenvalues from those noise series (using the
Burg-algorithm approach to SSA and an SSA window size of 50):
carlo tseries.dat -ar -ssa 50 -n 100 -sig 99
Given a time series in tsery.dat, generate 100 Monte Carlo realizations
of white noise with the same variance as a reconstruction of
tsery.dat MINUS the first two SSA components (from an SSA with
the Vautard et al. (1992) algorithm for estimating covariance and with
a window size of 70). Then use those realizations to determine the
95 percent confidence interval on SSA eigenvalues:
carlo tsery.dat -w -n 100 -wo 1 2 -ssa 70 -cov2 -sig 95
or, utilizing the defaults,
carlo tsery.dat -wo '1 2' -ssa 70 -cov2
SEE ALSO
spectra(1), ssa(1)
REFERENCES
Burg, J. P., 1978: A new analysis technique for time series data.
In Modern Spectrum Analysis, D. G. Childers (ed.), pp. 42-48,
IEEE Press, N.Y.
Broomhead, D.S., and King, G., 1986: Extracting qualitative dynamics from experimental data, Physica D, 20, 217-236.
Vautard, R., Yiou, P., and Ghil, M., 1992: Singular Spectrum Analysis:
A toolkit for short, noisy, chaotic time series,
Physica D, 58, 95-126.
Marsaglia, G., Zaman, A., and Tsang, W.W., 1990: Toward a universal random number
generator, Stats & Prob Letters, 8, 35-39.