Time series that are observed neither regularly nor contemporaneously pose problems for most multivariate analyses. Common and intuitive solutions to these problems include linear and step interpolation or other types of imputation to a higher, regular frequency. However, interpolation is known to cause serious problems with the size and power of statistical tests. Due to the diculty in measuring stochastically varying paleoclimate phenomena such as CO2 concentrations and surface temperatures, time series of such measurements are observed neither regularly nor contemporaneously.
It is widely accepted that long-run elasticities of demand for electricity are not stable over time. We model long-run sectoral electricity demand using a time-varying cointegrating vector. Specifically, the coefficient on income (residential sector) or output (commercial and industrial sectors) is allowed to follow a smooth semiparametric function of time, providing a flexible specification that allows more accurate out-of-sample forecasts than either fixed or discretely changing regression coefficients.
We show how temporal aggregation affects the size and power of the DOLS residual-based KPSS test of the null of cointegration. Size is effectively controlled by setting the minimum number of leads equal to one -- as opposed to zero -- when selecting the lag/lead order of the DOLS regression, but at a cost to power in finite samples. If high-frequency data for one or more series are available, we show how to effectively utilize the high-frequency data to increase power while controlling size.