Bias_Correc_VAR {MultiATSM}

Estimate an unbiased VAR(1) using stochastic approximation (Bauer, Rudebusch and Wu, 2012)

Description

Estimate an unbiased VAR(1) using stochastic approximation (Bauer, Rudebusch and Wu, 2012)

Usage

Bias_Correc_VAR(
  ModelType,
  BRWinputs,
  RiskFactors,
  N,
  Economies,
  FactorLabels,
  GVARinputs = NULL,
  JLLinputs = NULL,
  ev_restr = 1,
  nargout = 4
)

Arguments

ModelType

string-vector containing the label of the model to be estimated

BRWinputs

List containing the following necessary inputs for the estimation of the BRW model:

1. flag_mean: flag whether mean- (TRUE) or median- (FALSE) unbiased estimation is desired. Default is set to TRUE;

2. gamma: adjustment parameter. Value parameters should vary between 0 and 1. Default is set to 0.5;

3. N_iter: number of iterations used in the stochatic approximation algorithm after burn-in. Default is set to 5,000;

4. N_burn: number of burn-in iterations used in the stochatic approximation algorithm. Default is set to 0.15*N_iter;

5. B: number of bootstrap samples per iteration to calculate noisy measure of mean/median of the OLS estimator. Default is set to 50;

6. check: flag whether the user wishes to perform the closeness check. Default is set to TRUE;

7. B_check: number of bootstrap samples used in the closeness check. Default is set to 100,000.

RiskFactors

time series of the risk factors (T x F)

N

number of country-specific spanned factors (scalar)

Economies

string-vector containing the names of the economies which are part of the economic system

FactorLabels

string-list based which contains the labels of all variables present in the model

GVARinputs

inputs used in the estimation of the GVAR-based models (see "GVAR" function). Default is set to NULL

JLLinputs

inputs used in the estimation of the JLL-based models (see "JLL" function). Default is set to NULL

ev_restr

largest eigenvalue restriction under the P-measure. Default is set to 1

nargout

number of elements present in the list of outputs. Default is set to 4

Value

Bias-corrected VAR paramaters based on the framework of Bauer, Rudebusch and Wu (2012). The list contains:

1. Phi_tilde estimated coefficient matrix (F x F);

2. mu_tilde: estimated intercept (F x 1);

3. V_tilde: estimated variance-covariance matrix (F x F);

4. dist: root mean square distance (scalar);

5. Phi_sample: sample estimated variance-covariance matrix used in the checks (F x F x B_check) - this output is reported if nargout is set to 5.

References

Bauer, Rudebusch and, Wu (2012). "Correcting Estimation Bias in Dynamic Term Structure Models"
This function is based on the "est_unb_var" Matlab function available at Cynthia Wu's website (https://sites.google.com/view/jingcynthiawu/).

Examples

data(CM_Factors)
Factors <- t(RiskFactors[1:7,])

BRWinputs <- list()
BRWinputs$flag_mean <- TRUE
BRWinputs$gamma <- 0.4
BRWinputs$N_iter <- 1000
BRWinputs$N_burn <- 100
BRWinputs$B <- 10
BRWinputs$check <- 1
BRWinputs$B_check <- 5000

Economies <- "China"
N <- 3
ModelType <- "JPS"
FactorLabels <- NULL


BRWpara <- Bias_Correc_VAR(ModelType, BRWinputs, Factors, N, Economies, FactorLabels)