Open ghost opened 9 years ago
Have you (or someone else) figured this out yet? I might use autoregression on my project as well and would like some advice as well.
If you just want to do first-order MVAR, it's very doable to code it yourself: it's just a linear regression problem, where the coefficients are the autoregression A matrix, the "inputs/regressors" are x[t-1], and the "outputs/regressands" are x[t]. You have one of these observations for each time point t in your process. Higher-order models just "stack" the inputs and the matrices: so, you vertically concatenate A1, A2, ... and x[t-1], x[t-2], ... in the problem for each time point. You can apply any conditioning/regularization you want as part of solving the regression to suit your bias/variance needs.
Does anyone know how to use the MATLAB vector autoregressive models?