Repo for the course Time Series Analysis - FMSN45 at LTH
This computer exercise treats identification, estimation, and model validation in ARMA- and SARIMA-processes. Your goal is to be proficient in simulating different kinds of stationary and non-stationary time series, and to master the ’craft’ of identification and modeling of such. You will use the autocorrelation function (ACF), the partial autocorrelation function (PACF), probability plots and different parameter estimators as your basic toolkit. In this exercise, you will use these on both simulated and measured time series data.
In this computer exercise, you will work with input-output relations, as well as prediction in time series models. Firstly, you will be acquainted with time series having an exogenous input, having to analyze the impulse respons of such a system and from it build a suitable model. Secondly, this computer exercise deals with prediction, perhaps the most important application of time series modeling. You will be expected to make predictions of all models introduced in this course, i.e., up to a SARIMAX model.
This computer exercise treats recursive parameter estimation using Kalman filtering and recursive least squares. We attempt to model dynamic systems of both the SARIMA-type, having time-varying A and C polynomials, as well as to allow for ARMAX processes which have a synthetic input signal and time-varying B polynomial.
Using a dataset containing the hourly measurments from a weather station just outside copenhagen, a stationary ARMA model and ARMAX model with the puropse of temperature prediction were fitted. Moreover a recursive parameter estimation method was applied using Kalman filtering for temperature prediction.