Timeline for What's the deal with autocorrelation?
Current License: CC BY-SA 3.0
8 events
| when toggle format | what | by | license | comment | |
|---|---|---|---|---|---|
| Feb 5, 2013 at 19:11 | comment | added | Néstor | That is a difficult task that I'm on too. Especially with periodic deterministic models, it gets really hard to know what kind of noise model to use. The big problem is that you don't know a-priori the number of coefficients of the fourier model, so they are random variables you have to model too. In the presence of a low number of datapoints, I would definetly go for a reversible jump MCMC in order to model this. I would try different noise models and compare the AIC/BIC between them. For large datasets, however, this is unfeasible. | |
| Feb 5, 2013 at 19:03 | comment | added | BenDundee | Ahh ok, I think I see. This ties in to what I was going to ask in regards to 2. How might one go about patching this model (generically) to better understand the correlation? Could you add a constraint about the correlation matrix of the Fourier coefficients? | |
| Feb 5, 2013 at 18:49 | vote | accept | BenDundee | ||
| Feb 5, 2013 at 18:12 | comment | added | Néstor | I added an example. Hope it helps. And yes, he is refering to the CV of the residuals. | |
| Feb 5, 2013 at 18:12 | history | edited | Néstor | CC BY-SA 3.0 |
added 1379 characters in body
|
| Feb 5, 2013 at 14:25 | comment | added | BenDundee | Thank you for your response, and if you're willing, I'd like to try and digest these one at a time. For 1.), is there an intuitive way to understand why including more Fourier coefficients reduces autocorrelation and increases CV (I assume this is CV of the residuals)? | |
| Feb 5, 2013 at 4:02 | history | edited | Néstor | CC BY-SA 3.0 |
added 86 characters in body
|
| Feb 5, 2013 at 3:45 | history | answered | Néstor | CC BY-SA 3.0 |