Model Confidence Sets

The R package modelconf lets you estimate model confidence sets (MCS). These sets account for the uncertainty surrounding model choice.

About

This repository offers an implementation written in R of the algorithms for estimating model confidence sets and which were developed by Hansen, Lunde and Nason (2011) in their Econometrica paper. Model confidence sets are sets of models that are expected to contain all best model(s) with a given probability. Interpretation of the confidence set is therefore analogous to confidence intervals for population parameters. The notion of a confidence set is particularly useful in situations where competing model specifications are available and it is uncertain which model will be appropriate in a certain context.

The algorithm comes in two flavours, an in-sample and an out-of-sample version. Both of these are implemented here.

Alternative

There is another implementation of these algorithms available on CRAN via the MCS package. Maybe some of the code from here can be merged to over there or results from the two packages could be compared for correctness and efficiency.

Installation

Straight from github via devtools::install_github("nielsaka/modelconf").

Local build

Locally, the package can be installed using R CMD INSTALL /path/to/modelconf on the command line. If the package has been installed before, it might be necessary to enable the pre-clean option. With devtools, use options(devtools.install.args = "--preclean") to remove previously built binary files or R CMD INSTALL --preclean /path/to/modelconf on the command line.

License

This project is licensed under the MIT License. See the LICENSE file for details.