[R-SIG-Finance] Optimum bandwidth for Parzen's kernel using highfrequency

[Luis Damiano]

Hi all!

We're studying Barndorff-Nielsen, Hansen, Lunde and Shephard (2009) [1] in
depth. In particular, as a first step we aim at replicating the estimation
based on Parzen's kernel. After attempting to use highfrequency's
rKernelCov, which is based on Barndorff-Nielsen, Hansen, Lunde and Shephard
(2004) [2], we've got a few questions.

In equation 1.2 [3], H works as a bandwidth for the kernel and a covariance
lag. Equation 1.3 shows how to find the optimum bandwidth, which depends on
a constant that changes for each density function. We're trying to find
their counterparts in highfrequency:

1. Is rKernelCov's kernel.param argument related to the bandwidth H in
Equation 1.2?
2. Is rKernelCov's kernel.dofadj argument related to h / (H + 1), the inner
expression in Equation 1.2?
3. When using spotvol's kernel method, setting the est parameter to cv or
quarticity makes the function estimate the optimum bandwidth. Is any of the
two methods related to Equation 1.3? If spotvol computes the optimum
bandwidth, we're considering modifying kernelk [4] to allow for Parzen's
kernel.

We are very thankful for your support and sorry if any of the questions
makes little sense, we may be a bit confused!

[1] https://ideas.repec.org/a/ect/emjrnl/v12y2009i3pc1-c32.html
[2] https://ideas.repec.org/a/ect/emjrnl/v12y2009i3pc1-c32.html
[3] http://onlinelibrary.wiley.com/doi/10.1111/j.1368-423X.2008.00275.x/full
[4] https://github.com/jonathancornelissen/highfrequency/blob/master/R/
spotvol.r

	[[alternative HTML version deleted]]