MackNet_Parameters: MackNet_Parameters

In EduardoRamosP/MackNet: Mack-Net model: Blending Mack's model with Recurrent Neural Networks

Description

It calculated the predictive MackNet parameters.

Usage

MackNet_Parameters(Triangles, Error, Exposure, MackBias, ZeroMean)

Arguments

Triangles	Full triangles predicted by each RNN included within the ensemble.
Error	Test error of all the possible levels of weighted decay.
Exposure	Exposure measure. Written premiums is an appropriate measure to scale cumulative payments and incurred cost.
MackBias	If this variable is set to 0, the bias adjustment suggested by England y Verrall (2006) in "Predictive Distributions of Outstanding Liabilities in General Insurance" is applied, this means that residuals are multiplied by N/(N-p). In case this variable is set to 1, the adjustment suggested by Mack (1993) in "Distribution-free calculation of the standard error of chain ladder reserve estimates" is applied, this means that residuals are multiplied by n(i)/(n(i)-1). Finally, if this variable is set to 2, no adjustment is applied.
ZeroMean	If this variable is set to 0, residuals are not scaled to have zero mean. By default they are adjusted.

Value

The formula generates the following outputs:

• Factors Predictive development factors of the MackNet model.

• Alpha Predictive alpha by development year of the MackNet model. This variable is named as sigma in "Mack-Net model: Blending Mack's model with Recurrent Neural Networks".

• RNN_Residuals Residuals of the MackNet model. These residuals are used within the boostrap procedure.

• Error Optimum level of weighted decay.

EduardoRamosP/MackNet documentation built on Sept. 26, 2020, 9:21 a.m.