1
$\begingroup$

I want to create a code which tests absolutely everything for time-series forecasting accuracy.

The current tests that I do are:

bptest() - tests against heteroskedasticity (to test if series should be transformed)

gqtest() - tests against heteroskedasticity (to test if series should be transformed)

RMSE - of in-sample sqrt(mean((data-fitted values)^2))

MAPE - of in-sample mean(abs(100*(data-fitted values)/data))

MAPE - of out-sample (removing the last 4 data points and running model) mean(abs(100*(last 4 data points-the 4 forecasted points)/last 4 data points))

MAPE- of forced out-sample (removing the last 4 data points and running model keeping the same parameters as the model of the full data)

ACF and PACF - of residuals data-fitted values

Here is a link that will show you my code so you can see what i have done so far and where/what i can add. I dont want to miss anything so as many suggestions as possible please! If you could explain them a little too that would be fab.

http://bit.ly/10iUY5Z

$\endgroup$
3
  • $\begingroup$ Are you asking about programming, or are you looking for specific statistics help? The best place to get a block of working code assessed is the Code Review Stack Exchange site, not here. $\endgroup$ Commented Nov 3, 2014 at 12:59
  • $\begingroup$ Both :) there is many statistical tests that can be done, and there may also be people who know how to implement the tests into R $\endgroup$ Commented Nov 3, 2014 at 13:00
  • 1
    $\begingroup$ I found this nice piece on Rob J Hyndman's blog and thought you could be interested. It's about automated analysis of time series. There are pieces of code ready for use and clear comments. $\endgroup$ Commented Nov 18, 2014 at 20:18

0

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Browse other questions tagged or ask your own question.