NEWS

BurStFin 1.3 (2022-04-18)

The 'slideWeight' function is added. This creates a vector suitable to use as time weights -- for example, the 'weights' argument to 'var.shrink.eqcor' or 'factor.model.stat'.

if in 'factor.model.stat' the 'constant.returns.okay' argument is 'TRUE', then any columns with constant values now get a variance of zero rather than the default variance for an all missing column. The previous behavior could arguably be considered a bug.

the default value of 'tol' in 'var.shrink.eqcor' is now 1e-4 instead of 1e-3. This is still a guess, but it seems to be a better guess.
equal time weights can be specified with 'weight=NULL' in 'factor.model.stat' and 'var.shrink.eqcor'.
'cumulative.variance.fraction' is a new component of the result of 'factor.model.stat' when its output is the factor model.
'constant.names' is a new component of the result of 'factor.model.stat' when its output is the factor model.

The 'x' in 'factor.model.stat' is immediately coerced with 'as.matrix' to avoid subsetting problems with some data types, 'timeSeries' for instance.
A second attempt is made with 'svd' inside 'factor.model.stat' if the first attempt fails to converge. It also does a sanity check on the result of 'svd'.

The 'threeDarr' function is added. This creates three-dimensional arrays out of matrices.

'var.shrink.eqcor' and 'factor.model.stat' have a new argument 'verbose' that controls whether some warnings are given. Both functions can warn if there are no negative values in the input 'x' -- an indication in finance that prices rather than returns are given. Warnings in 'factor.model.stat' about constant columns in 'x' and negative specific variances are also controlled.
'var.add.benchmark' has a new argument 'sum.to.one' that allows a "benchmark" to have weights that sum to something other than one. An example is to give portfolio weights minus benchmark weights.