Command Reference : Command Reference
Equation Methods
Estimates an equation using robust least squares.
You may perform three different types of robust estimation: M-estimation, S-estimation and MM-estimation.
robustls(options) y x1 [x2 x3…]
Enter the robustls keyword, followed by the dependent variable and a list of the regressors.
method=arg (default=“m”)
Robust estimation method: “m” (M-estimation), “s” (S-estimation) or “mm” (MM-estimation).
cov=arg (default=“type1”)
Covariance method type: “type1”, “type2”, or “type3”.
Specify a value for the tuning parameter. If a value is not specified, EViews will use the default tuning parameter for the type of estimation and weighting function (if applicable).
Convergence criterion. The criterion will be set to the nearest value between 1e-24 and 0.2.
Specify the name of the coefficient vector (if specified by list); the default behavior is to use the “C” coefficient vector.
Maximum number the number of iterations.
Force the dialog to appear from within a program.
Print results.
M-estimation Options
fn=arg (default=“bisquare”)
Weighting function used during M-estimation: “andrews” (Andrews), “bisquare” (Bisquare), “cauchy” (Cauchy), “fair”, “huber”, “huberbi” (Huber-bisquare), “logistic” (Logistic), “median”, “tal” (Talworth), “Welsch” (Welsch).
scale=arg (default=“madzero”)
Scaling method used for calculating the scalar parameter during M estimation: “madzero” (median absolute deviation, zero centered), “madmed” (median absolute deviation, median centered), "huber" (Huber scaling).
Use the hat-matrix to down-weight observations with high leverage.
S and MM estimation options
compare = integer (default=4)
Number of comparison sets.
refine = integer (default= 2)
Number of refinements.
trials = integer (default=200)
Number of trials.
Specifies the size of the subsamples. Note, the default is number of coefficients in the regression.
Specifies the random number generator seed
Specifies the type of random number generator. The key can be; 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”).
MM estimation options
M-estimator tuning parameter.
Note the S-estimator tuning parameter is set with the “tuning=” option outlined above.
Use the hat-matrix to down-weight observations with high leverage during m-estimation.
The following examples use the “Rousseeuw and Leroy.wf1” file located in the EViews application data directory.
robustls salinity c lagsal trend discharge
This line estimates a simple M-type robust estimation, with SALINITY as the dependent variable, and a constant, LAGSAL, TREND and DISCHARGE as independent variables.
The line:
robustls(method=mm, tuning=2.937, mtuning=3.44, cov=type2) salinity c lagsal trend discharge
estimates the same model, but using MM-estimation, with an S tuning constant of 2.937, an M tuning constant of 3.44, and using Huber Type II standard errors.
See “Robust Least Squares” for discussion.