npscoefGrad: Gradients for Smooth Coefficient Kernel Regression Models

In micEconNP: Microeconomic Analysis with Non-Parametric Methods

Description

Obtain gradients of the Z variables in non-parametric Smooth Coefficient Kernel Regression models by numeric finite-difference differenciation.

Usage

   npscoefGrad( x, eps = 1e-3, ... )

Arguments

x	an object of class smoothcoefficient (returned by npscoef).
eps	the step size for the numeric finite-difference differentiation.
...	further arguments are currently ignored.

Value

a three-dimensional array that contains the gradients of the (smooth) coefficients and the predicted values with respect to the Z variables. The first dimension of the array corresponds to the observations. The second dimension of the array corresponds to the (smooth) intercept, the (smooth) coefficients of the X variables, and the predicted values of the dependent (Y) variable (in that order). The third dimension of the array corresponds to the Z variables.

Author(s)

Arne Henningsen (with significant help from Jeff Racine who provided the idea of obtaining these gradients)

See Also

npscoef.

Examples

   n <- 250
   x1 <- runif(n)
   x2 <- rnorm(n)
   z1 <- runif(n)
   z2 <- runif(n, min=-2, max=2)
   y <- sin( 5 * z1 ) + x1 * exp( z2 ) + x2 * ( z1^2 + z2 ) + rnorm( n, sd = 0.2 )
   model <- npscoef( y ~ x1 + x2 | z1 + z2, beta = TRUE )
   
   print( model )
   summary( model )
   round( coef( model ), 3 )
   
   grad <- npscoefGrad( model )
   round( grad, 3 )
   
   all.equal( grad[ , 1, ] + grad[ , 2, ] * x1 + grad[ , 3, ] * x2, grad[ , 4, ] )
   
   ## Not run: 
      for( j in 1:2 ) {
         for( i in 1:3 ) {
            plot( zData[,j], grad[ , i, j ],
               main = paste( "d b", i-1, " / d z", j, sep = "" ) )
            points(  zData[,j], gradTrue[ , i, j ], col = "red" )
            cat( "Press <enter>\n")
            readLines( n = 1 )
         }
      }
   
## End(Not run)

micEconNP documentation built on May 2, 2019, 6:30 p.m.