1

I have a set of data which histogram looks like this:

enter image description here

The values are not normally distributed since the distribution is not simetric.

For another set of data that is normally distributed, I have used a Ornstein-Uhlenbeck stochastic process to generate representative values.

Since OU processes only work with normally distributed data, what can I use to represent the depicted data set?

Thanks.

Santi Peñate-Vera
  • 161
  • 10
  • 1
    Ornstein, not Örnstein. – Did Nov 28 '16 at 12:07
  • Changed, now, any answer? – Santi Peñate-Vera Nov 30 '16 at 13:56
  • Your question is very unclear. Where does this histogram come from? What all this has to do with stochastic processes and with Ornstein-Uhlenbeck in particular? – zhoraster Nov 30 '16 at 20:19
  • The histogram represents the agregation of power plants generation forecast error. I want to be able to generate random meaningfull error series that follow that distribution instead of a normal distribution. – Santi Peñate-Vera Dec 01 '16 at 10:08
  • You've made even less clear. (By the way, you should add a nickname when responding to a comment, otherwise there's no notification.) There is a "process" in the title. Where is a process in your question? – zhoraster Dec 04 '16 at 06:55
  • I'm guessing you have some sort of a forecasting model and you're examining its errors.

    If the forecasts are generated separately and independently, then you shouldn't try to model the errors with a time series. You should simply fit a distribution to them - perhaps a composite normal or something.

     – zab Dec 06 '16 at 14:53
  • Both errors are weakly dependent, that why using OU is appealing. But since one of the errors is not estrictly normal, I wonder if I can modify the OU random generator to use the empirical distribution instead of a normal distriution and still be correct. – Santi Peñate-Vera Dec 07 '16 at 07:29

0 Answers0