Statistics for Environmental Engineers

Скачать в pdf «Statistics for Environmental Engineers»

If we accept this model, the intervention estimation problem becomes how to determine the weighting factor в. Fortunately, an alternate formulation of the model makes this reasonably simple. The white noise-random walk model is equivalent to an ARIMA (0,1,1) model (Box et al., 1994):

yt = yt_ 1 + a, _ ва,_!

where 0 < в < 1, and a, = independent random noise distributed as N(0,o<2).

The white noise and the random walk error components are related to o2 as follows:

o2 = во2

o2 = (1 _ в)2о2

The equivalence of the two forms of the model and the derivations are given in Pallesen et al. (1985).

A recursive iteration is done separately for each section of data before and after the intervention to estimate в. The method is to:

1. Choose a starting value for в and use at= yt— yt-1+ ва,-1 to recursively calculate the residuals at t = 2, 3,…, T. Assume at=0= 0 to start the calculations.

2.    Calculate the residual sum of squares, RSS(e) = Y*a] for each section. Add these to get the total RSS for the entire series. If a gap has been used to account for a transition period, data in the gap are omitted from these calculations.

3.    Search over a range of в to minimize RSS(e).

4.    Use the minimum RSS to estimate al = RSS(e), where n is the total number of residuals


used to compute RSS(e).

Скачать в pdf «Statistics for Environmental Engineers»