AR model with many lags

For a time series of 216 monthly returns, the correlation of returns
to past SUMS of returns over the last n1:n2 periods are

ACF_SUM(01:12) 0.134
ACF_SUM(13:24) -0.025
ACF_SUM(25:36) -0.098
ACF_SUM(37:48) -0.252
ACF_SUM(49:60) -0.031

The pattern of positive short-term autocorrelations and negative
longer-term autocorrelations has been observed in various financial
time series. How should this be modelled statistically? I don't want
to fit an unconstrained AR model with 48 coefficients. I could create
predictors out of the sums 1:12, 13:24, etc., and then regress against
them, but I have no reason believe that the weights on past returns
suddenly jump at multiples of 12, as such a model would imply.
Essentially I want to fit an AR(48) model where the coefficients are
constrained to be "sensible", but the question is how to do this.


.

Relevant Pages

  • Re: AR model with many lags
    ... to past SUMS of returns over the last n1:n2 periods are ... The pattern of positive short-term autocorrelations and negative ... Essentially I want to fit an ARmodel where the coefficients are ...
    (sci.stat.math)
  • ARIMA Time Series Modeling
    ... When I try to compare my AR and MA coefficients to commercial ... time series as well as the SAS:JMP software. ... % function atSqSum = local_funcGen ... hence my ARMA model. ...
    (comp.soft-sys.matlab)
  • Re: Wavelet coefficients -- what do they mean exactly?
    ... have several time series of wind measurements. ... I'm trying to capture how those magnitudes ... get the coefficients for certain scales/frequencies, ... For orthogonal wavelets, using wrcoef, you can inverse transform the ...
    (comp.soft-sys.matlab)
  • Re: Interesting (?) Property of Least Square Line Fit to Evenly-Spaced Data.
    ... Scott Seidman wrote: ... based upon a LS fit to a time series is actually a fairly ... form of the filter has been the bane of older approaches to TSA. ...
    (sci.stat.math)
  • Re: significance of correlation of a bandpass filtered time series
    ... bandpass filtered by applying transformations in spectral domain. ... just set some Fourier coefficients to zero and retransformed it to ... of degrees of freedom. ... Someone who knows time series would probably have to ...
    (sci.stat.edu)