mdl {mbreaks}

Comprehensive structural change estimation and testing

Description

mdl() calls main functions of the mbreaks package to execute the following estimation procedures:

• dotest() conducts Sup F tests of 0 versus m breaks and Double Max tests

• doseqtests() conducts the sequential Sup F tests of l versus l+1 breaks

• doorder() conducts the number of breaks selection from 1 to m breaks using information critera: ⁠{'KT','BIC','LWZ'}⁠

• dosequa() conducts the number of breaks selection by sequential tests from 1 to m breaks using sequential Sup F tests

• dofix() conducts structural break model estimation with fixn breaks

All the procedures automatically identify if the model is either i) pure structural breaks model or ii) partial structural breaks model

Usage

mdl(
  y_name,
  z_name = NULL,
  x_name = NULL,
  data,
  eps1 = 0.15,
  m = 5,
  prewhit = 1,
  robust = 1,
  hetdat = 1,
  hetvar = 1,
  hetomega = 1,
  hetq = 1,
  maxi = 10,
  eps = 1e-05,
  fixn = -1,
  fixb = 0,
  betaini = 0,
  printd = 0,
  const = 1,
  signif = 2,
  h = NULL
)

Arguments

y_name

name of dependent variable in the data set

z_name

name of independent variables in the data set which coefficients are allowed to change across regimes. default is vector of 1 (Mean-shift model)

x_name

name of independent variables in the data set which coefficients are constant across regimes. default is NULL

data

the data set for estimation

eps1

value of trimming (in percentage) for the construction and critical values. Minimal segment length h will be set at default = int(eps1*T) (T is total sample size).

• eps1=0.05 Maximal value of m = 10

• eps1=0.10 Maximal value of m = 8

• eps1=.15 Maximal value of m = 5

• eps1=.20 Maximal value of m = 3

• eps1=.25 Maximal value of m = 2

• eps1=0 This option allows users to explicitly specify minimum segment length h parameters. However, this option will not be allowed for testing and testing related functions

The default value is set at 0.15

m

Maximum number of structural changes allowed. If not specify, m will be set to default value matching eps1 input

prewhit

set to 1 to apply AR(1) prewhitening prior to estimating the long run covariance matrix.

robust

set to 1 to allow for heterogeneity and autocorrelation in the residuals, 0 otherwise. The method used is Andrews(1991) automatic bandwidth with AR(1) approximation with quadratic kernel. Note: Do not set to 1 if lagged dependent variables are included as regressors.

hetdat

option for the construction of the F tests. Set to 1 if want to allow different moment matrices of the regressors across segments. If hetdat = 0, the same moment matrices are assumed for each segment and estimated from the ful sample. It is recommended to set hetdat=1 if number of regressors x > 0.

hetvar

option for the construction of the F tests.Set to 1 if users want to allow for the variance of the residuals to be different across segments. If hetvar=0, the variance of the residuals is assumed constant across segments and constructed from the full sample. hetvar=1 when robust =1)

hetomega

used in the construction of the confidence intervals for the break dates. If hetomega=0, the long run covariance matrix of zu is assumed identical across segments (the variance of the errors u if robust=0)

hetq

used in the construction of the confidence intervals for the break dates. If hetq=0, the moment matrix of the data is assumed identical across segments

maxi

number of maximum iterations for recursive calculations of finding global minimizers.default = 10 (For partial change model ONLY)

eps

convergence criterion for recursive calculations (For partial change model ONLY)

fixn

number of pre-specified breaks. default = -1. It will be replaced automatically to 2 if no specification is given (For partial change model ONLY)

fixb

option to use fixed initial input \beta. If 1, the model will use values given in betaini. If 0, betaini is skipped

betaini

Initial beta_0 to use in estimation (Must be a ⁠p x 1⁠ matrix, where p is number of x variables)

printd

Print option for model estimation. default = 0, to suppress intermediate outputs printing to console

const

indicates whether the regression model include an intercept changing across regimes. Default value is 1

signif

significance level used to sequential test to select number of breaks.

• 4: 1% level

• 3: 2.5% level

• 2: 5% level

• 1: 10% level

h

Minimum segment length of regime considered in estimation. If users want to specify a particular value, please set eps1=0

Value

A list that contains the following:

• sbtests: class sbtests of Sup F tests of 0 versus m breaks and Double Max tests

• seqtests: class seqtests of sequential Sup F test of l versus l+1 breaks

• BIC: class model of structural break model with estimated number of breaks by BIC criterion

• LWZ: class model of structural break model with estimated number of breaks by LWZ criterion

• KT: class model of structural break model with estimated number of breaks by KT criterion

• sequa: class model of structural break model with estimated number of breaks by sequential tests

• fix: class model of structural break model with pre-specified fixn number of breaks

Note: All default values of error assumptions (robust, hetdat, hetvar, hetq) are set to 1. The implications on the structure of model's errors related to individual settings are explained within the arguments section for each option.

See Also

dotest(), doseqtests(), doorder(), dosequa(), and dofix() which are functions called by mdl().

Examples

 US_rate = mdl('rate',data=real)
 nkpc_lbs = mdl('inf',c('inflag','lbs','inffut'),data=nkpc,prewhit = 0)