Fast Running Variance Computation

Source: R/core_func.R

Computes running sample variance of a time-series x in a fixed length window.

RunningVar(x, W, circular = FALSE)

Arguments

x

A numeric vector.
W

A numeric scalar; length of x window over which sample variance is computed.
circular

Logical; whether running sample variance is computed assuming circular nature of x time-series (see Details).

Value

A numeric vector.

Details

The length of output vector equals the length of x vector. Parameter circular determines whether x time-series is assumed to have a circular nature. Assume \(l_x\) is the length of time-series x, W is a fixed length of x time-series window.

If circular equals TRUE then

  • first element of the output time-series corresponds to sample variance of x[1:W],

  • last element of the output time-series corresponds to sample variance of c(x[l_x], x[1:(W - 1)]).

If circular equals FALSE then

  • first element of the output time-series corresponds to sample variance of x[1:W],

  • the \(l_x - W + 1\)-th element of the output time-series corresponds to sample variance of x[(l_x - W + 1):l_x],

  • last W-1 elements of the output time-series are filled with NA.

See runstats.demo(func.name = "RunningVar") for a detailed presentation.

Examples

x <- rnorm(10)
RunningVar(x, W = 3, circular = FALSE)
#>  [1] 0.3438391 0.5557440 1.1028621 3.2661092 3.2470397 2.1266755 0.3405289
#>  [8] 0.2836473        NA        NA
RunningVar(x, W = 3, circular = TRUE)
#>  [1] 0.34383915 0.55574395 1.10286211 3.26610922 3.24703967 2.12667546
#>  [7] 0.34052892 0.28364729 0.02847671 0.03175456