pbsytest: Bera, Sosa-Escudero and Yoon Locally-Robust Lagrange...

In plm: Linear Models for Panel Data

Description

Test for residual serial correlation (or individual random effects) locally robust vs. individual random effects (serial correlation) for panel models and joint test of serial correlation and the random effect specification by Baltagi and Li.

Usage

pbsytest(x, ...)

## S3 method for class 'formula'
pbsytest(
  x,
  data,
  ...,
  test = c("ar", "re", "j"),
  re.normal = if (test == "re") TRUE else NULL
)

## S3 method for class 'panelmodel'
pbsytest(
  x,
  test = c("ar", "re", "j"),
  re.normal = if (test == "re") TRUE else NULL,
  ...
)

Arguments

x	an object of class formula or of class panelmodel,
...	further arguments.
data	a data.frame,
test	a character string indicating which test to perform: first–order serial correlation ("ar"), random effects ("re") or joint test for either of them ("j"),
re.normal	logical, only relevant for test = "re": TRUE (default) computes the one-sided "re" test, FALSE the two-sided test (see also Details); not relevant for other values of test and, thus, should be NULL,

Details

These Lagrange multiplier tests are robust vs. local misspecification of the alternative hypothesis, i.e., they test the null of serially uncorrelated residuals against AR(1) residuals in a pooling model, allowing for local departures from the assumption of no random effects; or they test the null of no random effects allowing for local departures from the assumption of no serial correlation in residuals. They use only the residuals of the pooled OLS model and correct for local misspecification as outlined in \insertCiteBERA:SOSA:YOON:01;textualplm.

For test = "re", the default (re.normal = TRUE) is to compute a one-sided test which is expected to lead to a more powerful test (asymptotically N(0,1) distributed). Setting re.normal = FALSE gives the two-sided test (asymptotically chi-squared(2) distributed). Argument re.normal is irrelevant for all other values of test.

The joint test of serial correlation and the random effect specification (test = "j") is due to \insertCiteBALT:LI:91;textualplm (also mentioned in \insertCiteBALT:LI:95;textualplm, pp. 135–136) and is added for convenience under this same function.

The unbalanced version of all tests are derived in \insertCiteSOSA:BERA:08;textualplm. The functions implemented are suitable for balanced as well as unbalanced panel data sets.

A concise treatment of the statistics for only balanced panels is given in \insertCiteBALT:13;textualplm, p. 108.

Here is an overview of how the various values of the test argument relate to the literature:

• test = "ar":

  • RS*_{ρ} in Bera et al. (2001), p. 9 (balanced)

  • LM*_{ρ} in Baltagi (2013), p. 108 (balanced)

  • RS*_{λ} in Sosa-Escudero/Bera (2008), p. 73 (unbalanced)

• test = "re", re.normal = TRUE (default) (one-sided test, asymptotically N(0,1) distributed):

  • RSO*_{μ} in Bera et al. (2001), p. 11 (balanced)

  • RSO*_{μ} in Sosa-Escudero/Bera (2008), p. 75 (unbalanced)

• test = "re", re.normal = FALSE (two-sided test, asymptotically chi-squared(2) distributed):

  • RS*_{μ} in Bera et al. (2001), p. 7 (balanced)

  • LM*_{μ} in Baltagi (2013), p. 108 (balanced)

  • RS*_{μ} in Sosa-Escudero/Bera (2008), p. 73 (unbalanced)

• test = "j":

  • RS_{μρ} in Bera et al. (2001), p. 10 (balanced)

  • LM in Baltagi/Li (2001), p. 279 (balanced)

  • LM_{1} in Baltagi and Li (1995), pp. 135–136 (balanced)

  • LM1 in Baltagi (2013), p. 108 (balanced)

  • RS_{λρ} in Sosa-Escudero/Bera (2008), p. 74 (unbalanced)

Value

An object of class "htest".

Author(s)

Giovanni Millo (initial implementation) & Kevin Tappe (extension to unbalanced panels)

References

\insertRef

BERA:SOSA:YOON:01plm

\insertRef

BALT:13plm

\insertRef

BALT:LI:91plm

\insertRef

BALT:LI:95plm

\insertRef

SOSA:BERA:08plm

See Also

plmtest() for individual and/or time random effects tests based on a correctly specified model; pbltest(), pbgtest() and pdwtest() for serial correlation tests in random effects models.

Examples

## Bera et. al (2001), p. 13, table 1 use
## a subset of the original Grunfeld
## data which contains three errors -> construct this subset:
data("Grunfeld", package = "plm")
Grunsubset <- rbind(Grunfeld[1:80, ], Grunfeld[141:160, ])
Grunsubset[Grunsubset$firm == 2 & Grunsubset$year %in% c(1940, 1952), ][["inv"]] <- c(261.6, 645.2)
Grunsubset[Grunsubset$firm == 2 & Grunsubset$year == 1946, ][["capital"]] <- 232.6

## default is AR testing (formula interface)
pbsytest(inv ~ value + capital, data = Grunsubset, index = c("firm", "year"))
pbsytest(inv ~ value + capital, data = Grunsubset, index = c("firm", "year"), test = "re")
pbsytest(inv ~ value + capital, data = Grunsubset, index = c("firm", "year"), 
  test = "re", re.normal = FALSE)
pbsytest(inv ~ value + capital, data = Grunsubset, index = c("firm", "year"), test = "j")

## plm interface
mod <- plm(inv ~ value + capital, data = Grunsubset, model = "pooling")
pbsytest(mod)

plm documentation built on Sept. 21, 2021, 3:01 p.m.