Abstract
M.Sc.
A concern for astronomers is to determine whether the period of a variable star stays
constant over time. One aspect that has been neglected in the past when testing for constant
period is that the variance of the observed period decreases over time, at least for some
stars. In this short dissertation, two models that account for the change in variance will
be investigated. One model assumes an exponentially decreasing variance while the other
allows for an abrupt change in the variance. For each of these two models a test for a change
in variance is proposed.
The properties of the estimates in the time and frequency domain are discussed and we
show that for these models, estimation in the frequency domain fails.
The models also provide for the presence of a non-constant period. The possibly changing
period is modeled by a sum of low-frequency cosine terms. The number of cosine terms is
determined using an order selection test introduced by Aerts, Claeskens and Hart (1999).
We show that not accounting properly for a changing variance may result in a changing
period not being detected.