| Process data in chemical industries are characterized by serial correlation in measured variables due to the presence of inertial elements such as tanks, reactors, and recycle streams. These data are often non-stationary, and the traditional approach to modeling involves transformation of the integrating series to achieve stationarity. When cointegration is present in non-stationary systems, it is known that the time series procedure of eliminating trends in the non-stationary data via differencing, may result in a loss of potentially valuable information about static long-run components. In order to compensate for differencing, econometricians have developed the theory of cointegration which models the non-stationarity explicitly using an error correction formulation.; Cointegration analysis addresses two issues: determination of the number of cointegrating vectors in non-stationary data, and estimation of the cointegrating parameters. In this thesis, the first topic is briefly introduced. The main focus falls on the various estimation techniques used to identify the cointegrating vectors. The three objectives of this thesis are: (i) to investigate methods for estimation in cointegrated systems, (ii) to investigate methods to uncover the cointegrating relationships, and (iii) to make preliminary evaluations as to whether or not cointegration analysis will be valuable in chemical engineering.; Results from simulations and analysis show that there is an improvement on forecasts for models that have imposed cointegrating relationships. The results also show that some cointegration techniques are system dependent, i.e. the results vary from system to system. |
