auxiliary quantlet for adedis. It defines the Nadaraya-Watson estimate of the link as a function of the (estimated) index of continuous explanatory variables. In adedis, the quantlet simpsonint is used to integrate over the function.
nelmin searchs for a minimum of a function. In each iteration step the function is evaluated at a simplex consisting of p+1 points. The simplex contracts until the variance of the evaluated function values is less than eps (or the maximal number of iterations is reached).
This quantlet computes different networks of the form single layer feedforward perceptron. The quantlet can be used alone or in connection with the library ISTA. The standalone version also needs the parameter data. Just choose 0 for the input. It is possible to split the data in a training and a t
This macro computes different networks of the form single layer feedforward perceptron. The macro can be used alone or in connection with the library ISTA. The standalone version also needs the parameter data. Just choose 0 for the input. It is possible to split the data in a training and a test se
Calculation of the Newey and West Heteroskedastic and Autocorrelation Consistent estimator of the variance. The first argument of the quantlet represents the series and the second optional argument the vector of truncation lags of the autocorrelation consistent variance estimator. If the second opt
computes the hessian matrix of a function func at a point x0 using the difference with a step h: d_(xy) f(x,y) = [f(x+h,y+h) - f(x+h,y-h) - f(x-h,y+h) + f(x-h,y-h)] / (4*h^2)
modified Newton-Raphson method for solving system func(x)=0 with backtracking (guarantees to decrease value of func in every iteration); compared with original Newton-Raphson method, it is less problematic to deal with highly oscillating functions
Inverse parabolic interpolation: finds the point x that is minimum/maximum of a parabola through three points (a,fa), (b,fb), (c,fc). INF is returned, if the three points are linear dependent (i.e. lying on the same line).
trains a one hidden layer feed forward network. The optional parameter param consists of 8 values. Boolean values for linear output, entropy error function, log probability models and for skip connections. The fifth value is the maximum value for the starting weights, the sixth the weight decay, th
trains a one hidden layer feed forward network. The optional parameter param consists of 8 values. Boolean values for linear output, entropy error function, log probability models and for skip connections. The fifth value is the maximum value for the starting weights, the sixth the weight decay, th
fits a nonparametric GARCH(1,1) process e[t+1] = s[t+1]*Z[t+1], s[t+1]^2 = g(e[t]^2,s[t]^2), where Z[t] are iid Gaussian, by inverting deconvolution kernel estimators.
Auxiliary routine for rICfil: calculates for dimension p=2 diag(E[ YY' u min(b/|aIhY|,u) ]) and diag(E[ YY' min(b/|aIhY|,u)^2 ]) for u square root of a Chi^2_2-variable, and Y~ufo(S_2) indep of u by using a polar representation of Lambda:= I^{1/2} Y u, u = | I^{-1/2} Lambda |, Y=I^{-1/
Auxiliary routine for rICfil: calculates for dimension p=2 (E[ YY' u min(b/|aIhY|,u) ]) and (E[ YY' min(b/|aIhY|,u)^2 ]) for u square root of a Chi^2_2-variable, and Y~ufo(S_2) indep of u by using a polar representation of Lambda:= I^{1/2} Y u, u = | I^{-1/2} Lambda |, Y=I^{-1/2} Lambd