Authors :
E.C. Ogwu; H.I. Ojarikre
Volume/Issue :
Volume 8 - 2023, Issue 3 - March
Google Scholar :
https://bit.ly/3TmGbDi
Scribd :
https://bit.ly/3zvkBU7
DOI :
https://doi.org/10.5281/zenodo.7797387
Abstract :
An important measure of the performance of
any statistical method is how well it performs in practice
especially using real life data. This work has compared
the performance of different approaches of kernel
density estimation (KDE) for several real data sets, the
rule of thumb (NRD0), unbiased cross validation (UCV),
and the Sheather Jones direct plug-in (SJ-DPI)
approaches were considered. The data set examined
includes, the daily closing prices of the Nigerian
exchange (NGE), Coca-cola (KO) and NASDAQ stock
data for a period of 1year, starting from October 31st
2021 to September 30th 2022. In this work we obtained
the kernel density estimation of the data sets using the
UCV, NRD0 and SJ-DPI approaches, we further
compared the performances of each of the above
approaches in terms of the resulting KDE plots and the
integrated square error (ISE). We found out that the SJDPI approach produced the best KDE plot for the
studied data sets and that the SJ-DPI and NRD0
approaches has the best performance for small and large
samples sizes respectively in terms of ISE for the studied
data sets. The study therefore suggests that the rule of
thumb, unbiased cross validation and the plug-in
approaches of KDE should be applied to other real data
sets to compare the performance of these approaches.
We further suggested that, other approaches of KDE
such as the Bayesian approach, the solve the equation
approach and the biased cross validation approaches be
applied to NASDAQ, NGE, and COCA-COLA data sets
in other to compare and investigate the performance of
these approaches
Keywords :
Kernel density estimation, smoothening parameter, rule of thumb, unbiased cross validation, direct plug-in,integrated square error
An important measure of the performance of
any statistical method is how well it performs in practice
especially using real life data. This work has compared
the performance of different approaches of kernel
density estimation (KDE) for several real data sets, the
rule of thumb (NRD0), unbiased cross validation (UCV),
and the Sheather Jones direct plug-in (SJ-DPI)
approaches were considered. The data set examined
includes, the daily closing prices of the Nigerian
exchange (NGE), Coca-cola (KO) and NASDAQ stock
data for a period of 1year, starting from October 31st
2021 to September 30th 2022. In this work we obtained
the kernel density estimation of the data sets using the
UCV, NRD0 and SJ-DPI approaches, we further
compared the performances of each of the above
approaches in terms of the resulting KDE plots and the
integrated square error (ISE). We found out that the SJDPI approach produced the best KDE plot for the
studied data sets and that the SJ-DPI and NRD0
approaches has the best performance for small and large
samples sizes respectively in terms of ISE for the studied
data sets. The study therefore suggests that the rule of
thumb, unbiased cross validation and the plug-in
approaches of KDE should be applied to other real data
sets to compare the performance of these approaches.
We further suggested that, other approaches of KDE
such as the Bayesian approach, the solve the equation
approach and the biased cross validation approaches be
applied to NASDAQ, NGE, and COCA-COLA data sets
in other to compare and investigate the performance of
these approaches
Keywords :
Kernel density estimation, smoothening parameter, rule of thumb, unbiased cross validation, direct plug-in,integrated square error
