binsmlr: binsmlr

In rrunner/binsmlr: Creates Heterogenous Bins across a range of Ordinal values

Description

binsmlr, as in bin similar, is a package to create bins that are assessed to be heterogenous (in terms of default risk) across a range of ordinal values. The binning can be performed automatically, interactively and manually.

Details

The objective is to create groups of data (bins) across a range of ordinal values. The bins produced should be homogenous within and heterogenous across bins in terms of the default risk. A binning approach, in the context of data transformation, is common in different credit risk applications and in particular to credit risk portfolios having a relatively large number of defaults. A prerequisite is that each observation is assigned an ordinal value such as score or rating etc., that indicates the relative default risk as proposed by the rank ordering model. Also, a default indicator is required.

The main process is the following:

1. Create a relatively large set of initial bins.

2. Perform homogeneity tests between all adjacent bins

3. Merge the pair of bins that are the most similar in terms of proportions of state of non-default and state of default.

4. Repeat step 2 and 3 until all bins are heterogeneous or if a minimum number of bins are obtained.

The process above should result in bins that are statistically different in terms of default risk across a range of ordinal data.

The binsmlr package provides the following functionality as listed below.

Automatic binning

See autobin for a fully automatic binning approach.

Interactive binning

See create_initial_bins and reduce_bins for how to perform binning in a slightly more interactive way.

Manual binning

See merge_list_of_bins on how to perform binning manually once the basic data structure is in place.

rrunner/binsmlr documentation built on July 19, 2020, 12:41 a.m.