uroot |

Carries out (panel) unit root tests on a group of series.

When used on a group of series, the procedure will perform panel unit root testing. The panel unit root tests include Levin, Lin and Chu (LLC), Breitung, Im, Pesaran, and Shin (IPS), Fisher - ADF, Fisher - PP, and Hadri tests on levels, or first or second differences.

Syntax

group_name.uroot(options)

Options

Basic Specification Options

You should specify the exogenous variables and order of dependent variable differencing in the test equation using the following options:

const (default) | Include a constant in the test equation. |

trend | Include a constant and a linear time trend in the test equation. |

none | Do not include a constant or time trend (only available for the ADF and PP tests). |

dif=integer (default=0) | Order of differencing of the series prior to running the test. Valid values are {0, 1, 2}. |

You may use one of the following keywords to specify the test:

sum (default) | Summary of the first five panel unit root tests (where applicable). |

llc | Levin, Lin, and Chu. |

breit | Breitung. |

ips | Im, Pesaran, and Shin. |

adf | Fisher - ADF. |

pp | Fisher - PP. |

hadri | Hadri. |

Sample Option

balance | Use balanced (across cross-sections or series) data when performing test. |

Lag Difference Options

Specifies the number of lag difference terms to be included in the test equation. Applicable in “Summary”, LLC, Breitung, IPS, and Fisher-ADF tests. The default setting depends on whether you choose to balance the samples across cross-sections.

If you do not include the “balance” option, the default is to perform automatic lag selection using the Schwarz criteria (“lagmethod=sic”).

Alternately, if you include the “balance” option, the default setting is a common, observation-based fixed lag (“lag=default”) where:

(1.1) |

lagmethod=arg (default=“sic”) | Method for selecting lag lengths (number of first difference terms) to be included in the Dickey-Fuller test regressions: “aic” (Akaike), “sic” (Schwarz), “hqc” (Hannan-Quinn), “tstat” (Ng-Perron first backward significant t-statistic). |

lag=arg | Specified lag length (number of first difference terms) to be included in the regression: integer (user-specified common lag length), vector_name (user-specific individual lag length, one row per cross-section). |

maxlag=arg | Maximum lag length to consider when performing automatic lag length selection: integer (common maximum lag length), or vector_name (individual maximum lag length, one row per cross-section). The default setting produces individual maximum lags of, default= where is the length of the cross-section. |

lagpval=arg (default=0.1) | Probability value for use in the t-statistic automatic lag selection method (when “lagmethod = tstat”). |

Kernel Options

Specifies options for computing kernel estimates of the zero-frequency spectrum (long-run covariance). Applicable to “Summary”, LLC, Fisher-PP, and Hadri tests.

hac=arg (default=“bt”) | Method of estimating the frequency zero spectrum: “bt” (Bartlett kernel), “pr” (Parzen kernel), “qs” (Quadratic Spectral kernel), |

band = arg, b=arg (default=“nw”) | Method of selecting the bandwidth: “nw” (Newey-West automatic variable bandwidth selection), “a” (Andrews automatic selection), number (user-specified common bandwidth), vector_name (user-specified individual bandwidths, one row for each cross-section). |

Other options

prompt | Force the dialog to appear from within a program. |

p | Print output from the test. |

Examples

The command:

Grp1.uroot(llc,exog=trend)

performs the LLC panel unit root test with exogenous individual trends and individual effects on series in GRP1.

Grp2.uroot(is,exog=const,maxlag=4,lagmethod=AIC)

performs the IPS panel unit root test on series in group GP2. The test includes individual effects, lag will be chosen by AIC from maximum lag of three.

Grp3.uroot(sum,exog=const,lag=3,hac=pr,b=2.3)

performs a summary of the panel unit root tests on the series in group GP3. The test equation includes a constant term and three lagged first-difference terms. The frequency zero spectrum is estimated using kernel methods (with a Parzen kernel), and a bandwidth of 2.3.

Cross-references

See “Unit Root Testing” for discussion of standard unit root tests performed on a single series, and “Panel Unit Root Testing” for discussion of unit roots tests performed on panel structured workfiles, groups of series, or pooled data.

See Series::uroot.

References

MacKinnon, James G., Alfred A. Haug, and Leo Michelis (1999), “Numerical Distribution Functions of Likelihood Ratio Tests For Cointegration,” Journal of Applied Econometrics, 14, 563-577.

Osterwald-Lenum, Michael (1992). “A Note with Quantiles of the Asymptotic Distribution of the Maximum Likelihood Cointegration Rank Test Statistics,” Oxford Bulletin of Economics and Statistics, 54, 461–472.