TV MSGARCH

Please note: This is a fork of MSGARCH package. More about MSGARCH available at http://keblu.github.io/MSGARCH/.

This fork includes time-varying transition probabilities for the methods FitML() and CreateSpec(). This fork was created for the purpose of my master thesis. I do not take any responsibility for the functionality of the package and advise anyone to use this fork at your own risk.

Installation & Setup

In order to install this fork of the MSGARCH package one needs to install the devtools package first.

install.packages("devtools")
require("devtools")

Make sure that MSGARCH is uninstalled on your machine. Then download the fork of the package from this repository.

devtools::install_github("mkaywins/MSGARCH", subdir="Package")

Load the package.

library(MSGARCH)

How to use?

Before we fit a (TV-)MSGARCH model to data, we need to define the model specification through CreateSpec. This step is analogous to the basic version of the package with the key difference that in the switch.spec argument the variable do.tvp must be set to TRUE.

# To fit the model with time-varying probabilities (TVP), we must set
# do.tvp = TRUE:
spec = CreateSpec(switch.spec = list(do.tvp=TRUE),
                  variance.spec = list(model = c("sGARCH", "sGARCH")),
                  distribution.spec = list(distribution = c("norm", "norm")))

# sample data
data("SMI", package = "MSGARCH")
data = SMI

# Constructing the covariate matrix Z
Z = matrix(1, nrow = length(data)-1, ncol = 2)
Z[,2] = data[1:2499]
data = data[2:2500]

Subsequently, we fit the model through the MLE method FitML. Note that the covariate matrix Z only needs to be supplied for time-varying switching. Thus, it only works if the object spec includes the correct specifications (do.tvp = TRUE and do.mix = FALSE).

# We fit the model via the maximum likelihood method:
fit = FitML(spec = spec, data = data,  Z = Z)

Which extensions were made?

The following extensions were made to the MSGARCH package:

• Incorporate time-varying transition probabilities into the ML estimation ✔️
• Enable one-step-ahead predictions for S3 method predict() ✔️
• Enable one-step-ahead predictions for method PredPdf()️ ✔️
• Enable one-step-ahead predictions for method Risk() ✔️
• Make ML-Fit object (estimated through time-varying switching) compatible for method State() ✔️
• S3 methods like print() and summary() indicate wheter time-varying MSGARCH was used ✔️
• TransMat() returns the initial transition matrix ✔️
• ExtractStateFit() is enabled for MLFit object which were estimated through time-varying switching ✔️
• S3 method plot() work for time-varying switching ✔️
• Volatility() works for time-varying switching ✔️
• UncVol() works for time-varying switching ✔️

The following functionalists are not yet implemented for time-varying switching:

• Time-varying Mixture modelling, i.e. do.tvp=TRUE and do.mix=TRUE is not yet supported ❌
• PIT() is not working for time-varying switching ❌
• simulate.MSGARCH_SPEC() simulate is work in progress ❌
• multi-step-ahead predictions are not supported for time-varying switching ❌

Important Info

I want to re-emphasize that I created this fork solely for the purpose of my master's thesis. I don't want to make any pull-request to the original package and I am not planing on maintain this repo in the future.

---

Please cite the package in publications!

By using MSGARCH you agree to the following rules:

1. You must cite Ardia et al. (2019) in working papers and published papers that use MSGARCH.
2. You must place the following URL in a footnote to help others find MSGARCH: https://CRAN.R-project.org/package=MSGARCH.
3. You assume all risk for the use of MSGARCH.

Ardia, D., Bluteau, K., Boudt, K., Catania, L., Trottier, D.-A. (2019).
Markov-switching GARCH models in R: The MSGARCH package.
Journal of Statistical Software, 91(4), 1-38.
https://doi.org/10.18637/jss.v091.i04

Other references

Ardia, D., Bluteau, K., Boudt, K., Catania, L. (2018).
Forecasting risk with Markov-switching GARCH models: A large-scale performance study
International Journal of Forecasting, 34(4), 733-747.
https://doi.org/10.1016/j.ijforecast.2018.05.004

Ardia, D., Bluteau, K., Ruede, M. (2019).
Regime changes in Bitcoin GARCH volatility dynamics.
Finance Research Letters, 29, 266-271.
https://doi.org/10.1016/j.frl.2018.08.009