Performing CUSUM transformation to the input matrix of multivariate time series. If the input is a vector, it is treated as a matrix of one row.
Usage
cusum.transform(x)
Arguments
x
input matrix
Value
The transformed matrix is returned. Note that the returned matrix has the same number of rows but one fewer columns compared with the input matrix.
Details
For any integers p and n, the CUSUM transformation \(T_{p,n}: R^{p\times n}\to R^{p\times (n-1)}\) is defined by
$$
[T_{p,n}(M)]_{j,t} := \sqrt{t(n-t)/n}\biggl(\frac{1}{n-t}\sum_{r=t+1}^n M_{j,r} - \frac{1}{t}\sum_{r=1}^t M_{j,r}\biggr).
$$