Breadcrumb
Locally stationary time series
Supervisor: Guy Nason
Theme: Time Series
A great deal of existing time series theory and practice relies on the theory of stationary time series. The stationarity assumption is quite restricting in that it forces the statistical properties (mean, variance, autocovariance, etc) of the series to remain constant over all time. This constancy assumption is often untenable in practice. For example, many financial time series are not stationary and their non-stationarities are caused by the underlying fluid political and economic factors. There are several interesting potential projects involving locally stationary time series (and multidimensional versions), discovering more about there theoretical properties and capabilities, applying them to new time series and novel situations. For example, can locally stationary processes be used to fuse multiple sources of data to obtain a combined and informative view. (See also Piotr Fryzlewicz's project on the same topic).