Object Reference : Object View and Procedure Reference : Equation
  
 
ardl
Estimate an equation with autoregressive distributed lags using linear and nonlinear least squares or quantile regression.
Syntax
equation.ardl(options) linear_regs [@ static_regs] [@asy dual_asymmetric_regs] [@asylr long_run_asymmetric_regs] [@asysr short_run_asymmetric_regs]
The linear_regs specification is required:
The linear_regs list should be the dependent variable followed by a list of linear distributed-lag regressors.
The remaining specifications are optional
static_regs should be a list of static regressors, not including a constant or trend term.
dual_asymmetric_regs are distributed-lag regressors which are asymmetric both in the short-run and long-run.
long_run_asymmetric_regs regressors are distributed lag-regressors which are asymmetric in the long-run but symmetric in the short-run.
short_run_asymmetric_regs are asymmetric regressors which are distributed lag-regressors which are asymmetric in the short-run but symmetric in the long-run.
You may specify the lag for an individual distributed-lag variable using the “@fl(variable, lag)” syntax. For instance, if the variable X should use 3 lags, irrespective of the fixed or automatic lag settings, you may specify this by entering “@fl(x, 3)” in the regressor list.
Options
Least Squares ARDL Options
 
method=arg (default = “ls”)
Set the method of estimation: "ls" (least-squares regression, default) or "qreg" (quantile regression).
determ=arg (default = “rconst”)
Johansen deterministic trend type: “none” (no deterministics), “rconst” (restricted constant and no trend), “uconst” (unrestricted constant and no trend), “rtrend” (unrestricted constant and restricted trend, “utrend” (unrestricted constant and unrestricted trend).
trend=arg (deprecated)
Johansen deterministic trend type: “none” (no deterministics), “const” (restricted constant and no trend, default), “uconst” (unrestricted constant and no trend), “linear” (unrestricted constant and restricted trend, “ulinear” (unrestricted constant and unrestricted trend).
Note: this is a deprecated s option which handles a subset of cases covered by the “determ=” option
fixed
Do not use automatic selection for lag lengths. This option must be used with the “deplags=” and “reglags=” options.
deplags=int (default = 4)
Set the number of lags for the dependent variable to int. If automatic selection is used, this sets the maximum number of possible lags. If fixed lags are used (the fixed option is set), this fixes the number of lags.
reglags=int (default = 4)
Set the number of lags for the explanatory variables (dynamic regressors) to int. If automatic selection is used, this sets the maximum number of possible lags. If fixed lags are used (the fixed option is set), this fixes the number of lags for each regressor.
ic=key (default =“aic”)
Set the method of automatic model selection. key may take values of “aic” (Akaike information criterion, default), “bic” (Schwarz criterion), “hq” (Hannan-Quinn criterion) or “rbar2” (Adjusted R-squared, not applicable in panel workfiles).
cov=arg
Covariance method: “ordinary” (default method based on inverse of the estimated information matrix), “huber” or “white” (Huber-White sandwich method), “hac” (Newey-West HAC, available for nonlinear least squares or ARMA estimated by CLS)..
nodf
Do not perform degree of freedom corrections in computing coefficient covariance matrix. The default is to use degree of freedom corrections.
covlag=arg (default=1)
Whitening lag specification: integer (user-specified lag value), “a” (automatic selection).
covinfosel=arg (default=“aic”)
Information criterion for automatic selection: “aic” (Akaike), “sic” (Schwarz), “hqc” (Hannan-Quinn) (if “lag=a”).
covmaxlag=integer
Maximum lag-length for automatic selection (optional) (if “lag=a”). The default is an observation-based maximum of .
covkern=arg (default=“bart”)
Kernel shape: “none” (no kernel), “bart” (Bartlett, default), “bohman” (Bohman), “daniell” (Daniel), “parzen” (Parzen), “parzriesz” (Parzen-Riesz), “parzgeo” (Parzen-Geometric), “parzcauchy” (Parzen-Cauchy), “quadspec” (Quadratic Spectral), “trunc” (Truncated), “thamm” (Tukey-Hamming), “thann” (Tukey-Hanning), “tparz” (Tukey-Parzen).
covbw=arg (default=“fixednw”)
Kernel Bandwidth: “fixednw” (Newey-West fixed), “andrews” (Andrews automatic), “neweywest” (Newey-West automatic), number (User-specified bandwidth).
covnwlag=integer
Newey-West lag-selection parameter for use in nonparametric kernel bandwidth selection (if “covbw=neweywest”).
covbwint
Use integer portion of bandwidth.
coef=arg
Specify the name of the coefficient vector (if specified by list); the default behavior is to use the “C” coefficient vector.
prompt
Force the dialog to appear from within a program.
p
Print results.
Quantile ARDL Options
 
quant=number (default = 0.5)
Quantile to be fit (where number is a value between 0 and 1).
w=arg
Weight series or expression.
Note: we recommend that, absent a good reason, you employ the default settings Inverse std. dev. weights (“wtype=istdev”) with EViews default scaling (“wscale=eviews”) for backward compatibility with versions prior to EViews 7.
wtype=arg (default=“istdev”)
Weight specification type: inverse standard deviation (“istdev”), inverse variance (“ivar”), standard deviation (“stdev”), variance (“var”).
wscale=arg
Weight scaling: EViews default (“eviews”), average (“avg”), none (“none”).
The default setting depends upon the weight type: “eviews” if “wtype=istdev”, “avg” for all others.
cov=arg (default=“sandwich”)
Method for computing coefficient covariance matrix: “iid” (ordinary estimates), “sandwich” (Huber sandwich estimates), “boot” (bootstrap estimates).
When “cov=iid” or “cov=sandwich”, EViews will use the sparsity nuisance parameter calculation specified in “spmethod=” when estimating the coefficient covariance matrix.
bwmethod=arg (default = “hs”)
Method for automatically selecting bandwidth value for use in estimation of sparsity and coefficient covariance matrix: “hs” (Hall-Sheather), “bf” (Bofinger), “c” (Chamberlain).
bw =number
Use user-specified bandwidth value in place of automatic method specified in “bwmethod=”.
bwsize=number (default = 0.05)
Size parameter for use in computation of bandwidth (used when “bw=hs” and “bw=bf”).
spmethod=arg (default=“kernel”)
Sparsity estimation method: “resid” (Siddiqui using residuals), “fitted” (Siddiqui using fitted quantiles at mean values of regressors), “kernel” (Kernel density using residuals)
Note: “spmethod=resid” is not available when “cov=sandwich”.
btmethod=arg (default= “pair”)
Bootstrap method: “resid” (residual bootstrap), “pair” (xy-pair bootstrap), “mcmb” (MCMB bootstrap), “mcmba” (MCMB-A bootstrap).
btreps=integer (default=100)
Number of bootstrap repetitions
btseed=positive integer
Seed the bootstrap random number generator.
If not specified, EViews will seed the bootstrap random number generator with a single integer draw from the default global random number generator.
btrnd=arg (default=“kn” or method previously set using rndseed).
Type of random number generator for the bootstrap: improved Knuth generator (“kn”), improved Mersenne Twister (“mt”), Knuth’s (1997) lagged Fibonacci generator used in EViews 4 (“kn4”) L’Ecuyer’s (1999) combined multiple recursive generator (“le”), Matsumoto and Nishimura’s (1998) Mersenne Twister used in EViews 4 (“mt4”).
btobs=integer
Number of observations for bootstrap subsampling (when “bsmethod=pair”).
Should be significantly greater than the number of regressors and less than or equal to the number of observations used in estimation. EViews will automatically restrict values to the range from the number of regressors and the number of estimation observations.
If omitted, the bootstrap will use the number of observations used in estimation.
btout=name
(optional) Matrix to hold results of bootstrap simulations.
k=arg (default=“e”)
Kernel function for sparsity and coefficient covariance matrix estimation (when “spmethod=kernel”): “e” (Epanechnikov), “r” (Triangular), “u” (Uniform), “n” (Normal–Gaussian), “b” (Biweight–Quartic), “t” (Triweight), “c” (Cosinus).
m=integer
Maximum number of iterations.
s
Use the current coefficient values in estimator coefficient vector as starting values (see also param).
s=number (default =0)
Determine starting values for equations. Specify a number between 0 and 1 representing the fraction of preliminary least squares coefficient estimates.
Note that out of range values are set to the default.
coef=arg
Specify the name of the coefficient vector (if specified by list); the default behavior is to use the “C” coefficient vector.
showopts / ‑showopts
[Do / do not] display the starting coefficient values and estimation options in the estimation output.
prompt
Force the dialog to appear from within a program.
p
Print estimation results.
Examples
wfopen http://www.stern.nyu.edu/~wgreene/Text/Edition7/TableF5-2.txt
opens example data from Greene (2008, page 685), containing quarterly US macroeconomic variables between 1950 and 2000.
The following command
equation eq01.ardl(deplags=8, reglags=8) log(realcons) log(realgdp) @ @expand(@quarter, @droplast)
creates an equation object and estimates an ARDL model with the log of real consumption as the dependent variable, and the log of real GDP as a dynamic regressor. Quarterly dummy variables are included as static regressors. Automatic model selection is used to determine the number of lags of LOG(REALCONS) and LOG(REALGDP).
The command
equation eq02.ardl(deplags=3, reglags=3, fixed) log(realcons) log(realgdp) @ @expand(@quarter, @droplast)
estimates a second model, replicating Example 20.4 from Greene, with a fixed three lags of the dependent variable and three lags of the regressor.
equation eq03.ardl(deplags=1, reglags=1, fixed) log(realcons) log(realgdp) @asy log(realgovt)
The line above estimates an ARDL(1,1,1) model with the log of real consumption as the dependent variable, the log of real GDP as a linear regressor, and log of real government expenditures as a dual asymmetric regressor.
equation eq04.ardl(deplags=1, reglags=1, fixed) log(realcons) log(realgdp) @asy log(realgovt) @asysr log(realinvs)
extends the previous model and estimates an ARDL(1,1,1,1) model by including the log of real investments as a long-run asymmetric regressor.
equation eq05.ardl(deplags=1, reglags=1, fixed) log(realcons) log(realgdp) @asy log(realgovt) @asysr log(realinvs) @asylr log(tbilrate)
The line above extends the previous model and estimates an ARDL(1,1,1,1,1) model by including the log of treasury bill rates as a short-run asymmetric regressor.
wfopen oecd.wf1
equation eq06.ardl(fixed, deplags=1, reglags=1) log(cons) log(inf) log(inc)
This example estimates a panel ARDL model using the workfile “OECD.wf1”. This model replicates that given in the original Pesaran, Shin and Smith 1999 paper. Model selection is not used to choose the optimal lag lengths, rather a fixed single lag of both the dependent variable and the regressor is employed.
equation eq07.ardl(method=qreg, ls=fixed, deplags=1, reglags=1, quant=0.4) log(realcons) log(realgdp)
This command estimates a QARDL(1,1) model where lag selection is fixed for both the dependent and independent regressors, and the quantile value is 0.4.
Cross-references
See “ARDL and Quantile ARDL” for further discussion.