I'd like to apply a Kalman filter, using Octave, to financial data but due to the nature of the data it will be difficult to impossible to specific the process error in advance of applying the filter. Can anyone suggest a way to specify the process error online from the incoming data simultaneously with the calculation of the filter itself? I'm trying to adapt the code from this website to suit my purposes.

  • 1
    $\begingroup$ Do you mean that you need the covariance matrix for the process noise? It's pretty common to not know that exactly. $\endgroup$
    – Jason R
    Mar 21 '12 at 23:03
  • $\begingroup$ @Jason R - that right. I'd like to be able to initialise the covariance matrix with some meaningful values. $\endgroup$ Mar 22 '12 at 13:34
  • $\begingroup$ What are your state variables that you're trying to estimate? You can't decide on a covariance matrix without specifying what your underlying system model is. Note that the easiest way to introduce process noise is to assume that it is white with some variance that you select in each component (i.e. the covariance matrix is diagonal). $\endgroup$
    – Jason R
    Mar 22 '12 at 13:40
  • $\begingroup$ I'm taking the code as is from the above linked website. There are two kalman filters I'd like to apply; a staightforward linear model with position, velocity and acceleration; and an Extended kalman filter apllied to a sine wave with frequency and amplitude. $\endgroup$ Mar 23 '12 at 12:51

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.