iStarSensitivity: I-Star model sensitivity analysis

In braverock/blotter: Tools for Transaction-Oriented Trading Systems P&L

I-Star model sensitivity analysis

Description

An helper function to provide I-Star model parameters sensitivity analysis.

Usage

iStarSensitivity(object, paramsBounds, paramSteps, ...)

Arguments

object	An object of class 'iStarEst' from the iStarPostTrade function # TODO: the class is not defined at present, nor checks are in place here
paramsBounds	A matrix providing model parameters bounds to pass to nls. The same used in iStarPostTrade and defaults to the same values dicussed there
paramSteps	A vector of named elements representing each parameter step size to build the parameters sequences with. See 'Details'
...	Any other passthrough parameter

Details

The sensitivity analysis provided is a local one, with I-Star model paramaters being fixed one at a time. Then, for each fixed sequence of values provided for a parameter, the nonlinear problem is solved to estimate the remaing four paramaters by means of nonlinear regression.

Results of the analysis are reported along with the corresponding residul standard error (RSE) of each model being fitted. This quantity is expressed in the same dependent variable unit and best fit paramaters should be such that this quantity is minimized.

Of course, paramSteps is related to paramsBounds. In particular, it should be stress that, provided step sizes will be used to built parameters sequences from the lower bound specified in paramsBounds until the last multiple of the upper bound provided in paramsBounds is reached, for each paramater and its respective bound values. In other words, paramSteps is not allowed to have a sequence value that goes beyond the paramsBounds specified upper bound.

paramSteps default is 50 for a_1, 0.1 for a_2 and a_3, 0.05 for a_4 and 0.01 for b_1.

Value

A list with elements:

Params.Seqs:

A list of parameters sequences evaluated

nls.impact.fits:

A list of each model fitted with nls

Params.Sensitivity:

A matrix contining the results of the sensitivity analysis

Note

If paramaters fixed sequences values lead to nonlinear least squares estimation failures, an NA is put in place of the other paramaters being estimated and of the residual sum of squares, as they cannot be provided.

Author(s)

Vito Lestingi

References

The Science of Algorithmic Trading and Portfolio Management (Kissell, 2013), Elsevier Science.

See Also

iStarPostTrade, nls

braverock/blotter documentation built on Feb. 13, 2023, 1 p.m.