## How do I perform Particle Filtering of a time series

suppose Yt=f(Xt ,Zt) .Here only Yt is observable but Xt and Zt are not observable-however both are random variables for a given t( and hence{Xt} and {Zt} can be considered as two unobservable time series .)Therefore the problem is to disentangle the effect of Xt and Zt from Yt .  {Xt } also has a decreasing trend and sometimes there might be a structural break.

I want to use PARTICLE FILTERING so that the forecasting is good  as the distribution of {Yt} is far from naormality ..how to do that? Also How best to use UCM/SSM for this type of problem?

I have SAS 9.4.

If someone is interested to know more I can provide the time series Yt and codes.

2 REPLIES 2

## How do I perform Particle Filtering of a time series

suppose Yt=f(Xt ,Zt) .Here only Yt is observable but Xt and Zt are not observable-however both are random variables for a given t( and hence{Xt} and {Zt} can be considered as two unobservable time series .)Therefore the problem is to disentangle the effect of Xt and Zt from Yt .  {Xt } also has a decreasing trend and sometimes there might be a structural break.

I want to use PARTICLE FILTERING so that the forecasting is good  ..how to do that? Also How best to use UCM/SSM for this type of problem?

I have SAS 9.4.

If someone is interested to know more I can provide the time series Yt and codes.

## Re: How do I perform Particle Filtering of a time series

To my knowledge there is no convenient support for particle filtering in SAS yet.  Particle filtering is used when data follows nonlinear or non-Gaussian state space model.  If a problem can be formulated as a linear state space model, you can analyze it using PROC UCM or PROC SSM.  PROC UCM is restricted to the analysis of univariate response series---it permits a variety of ways to handle predictor series.  For univariate analysis PROC UCM is often easy to use and often quite adequate.  If you want to analyze more general types of sequential data (multivariate time series, panel data, or longitudinal data) then you can consider PROC SSM.  The documentation of both procedures (and SGF papers) contain many illustrative examples of their use.

Discussion stats
• 2 replies
• 654 views
• 0 likes
• 2 in conversation