R help - Jan 2005 - recursive penalized regression

Hi,
 
Few days ago I posted a question to r-sig-finance, which I thought would
be an easy one. To my surprise I have received no replies, which makes
me think that it is either harder than I thought, or that it makes no
sense. I am reposting the message (with some modifications) on the
R-help in a hope to get some leads, suggestions for alternatives, etc.
My apologies to those who had seen this on r-sig-finance.
 
 
I want to do a univariate no frills autoregression. The major
non-standard requirements are:
1. When estimating model parameters at time t the algorithm can only use
data up to time t.
2. The weights of the past observations should decay with time, e.g
exponentially
3. ability to apply some penalty, e.g. L2 (ridge), L1 (lasso), etc., to
model coefficients (or any other regularization technique)
 
Thanks in advance for your help,
Vadim

	[[alternative HTML version deleted]]

Seems to me that your requirements 1 and 2 could be met by casting the
autoregressive model in state-space form and using the Kalman filter.
I am not sure about the kind of regularization that you want to  apply
to model coefficients, but it could also likely be accommodated: see
Durbin and Koopman "Time Series Analysis by State Space Methods", sec.
6.5.

Best, ft.
-- 
Fernando TUSELL                                e-mail:
Departamento de Econometr?a y Estad?stica           etptupaf at bs.ehu.es 
Facultad de CC.EE. y Empresariales             Tel:   (+34)94.601.3733
Avenida Lendakari Aguirre, 83                  Fax:   (+34)94.601.3754
E-48015 BILBAO  (Spain)                        Secr:  (+34)94.601.3740