Подробнее
The virgin non-parametric regression Computationally intensive. Requires a tonne of observations makes 2019 computers bleed for convergence Heavily depends on hyper-parameters that must be chosen optimally Choice of hyper-parameters needs orders of magnitude more computational power Does not have a closed-form solution Terribly inefficient, high-level implementations are slower than Crysis on Pentium III Dies a horrible death with 10 regressors from the curse of dimensionality Conditions yielding normality of the estimator are hard to verify Does not have a clear notion of efficient estimation Even after estimation, partial effects require extra computation Is rarely used in applied economic analysis Impossible to find simple instructions for inference Most courses spend hours just to get to Parzen-Rozenblatt. Nadaraya-Watson. maybe LOESS, and then stop implementations vary widely in software, same kernels have different scaling factors Fails miserably in regions of low density of explanatory variables Convergence so poor, has to be injected with parametric assumptions or index restrictions Behaves like a spoiled brat at boundaries, requires correction Even bootstrapped confidence intervals are not trusted at first glance THE CHAD LINEAR REGRESSION With 20 data points, ends up Can be made asymptotically being published in top efficient just in a couple of steps macroeconomic journals Works in one step without fine-tuning Can be computed analytically with paper and pencil Gives a nice, clean interpretation any grandma can comprehend Constant partial effect across the entire support Optimised to the Moon and beyond in linear algebra libraries Can handle hundreds of regressors with ease Built in every software package that can handle numbers Could not care less about error distribution, ends up being normal anyway No one bats an eye when parametric assumptions arc violated The support of explanatory variables does not matter Inference can be done with one subtraction and one division by a toddler Lots of online tutorials about versatile extensions accessible to beginners Gives the same consistent result in R. STATA. SAS, SPSS. Excel, and makeshift online ‘calculators' No questions asked with robust standard errors Can account for arbitrary non-linearity by including functions or regressors, remains analytical
anon
Еще на тему
