Prev: negative power flow in simpowersystems
Next: Questions on shared library in Mac
From: maheswaran Rathinasamy on 1 Jun 2010 23:46
Dear Sir,

I am working with forecasting using wavelets. The problem i am facing is with the boundary distortion. Since, the wavelet decomposition requires the futures values for calculating the coeff at time 't'. But at the end there is no availability of future data. If I use a causal filter then again the there is shifting the features in the time series and reduces the forecasting accuracy.


Can any one give some suggestions

maheswaran
From: TideMan on 2 Jun 2010 00:36
On Jun 2, 3:46 pm, "maheswaran Rathinasamy" <maheswara...(a)gmail.com>
wrote:
> Dear Sir,
>
> I am working with forecasting using  wavelets. The problem i am facing is with the boundary distortion. Since, the wavelet decomposition requires the futures values for calculating the coeff at time 't'. But at the end there is no availability of future data. If I use a causal filter then again the there is shifting the features in the time series and reduces the forecasting accuracy.
>
> Can any one give some suggestions
>
> maheswaran

Unfortunately, wavelet decomposition suffers from end effects just
like decomposition using the Fourier transform.
And for wavelets, the end effects get worse the higher the level of
decomposition (i.e., longer timescales), and since it is the long
timescales that we usually want to forecast, I think this means you
are screwed.
Put simply, wavelets are not suitable for forecasting.
Anyway, that's my experience.
 | 
Pages: 1
Prev: negative power flow in simpowersystems
Next: Questions on shared library in Mac