RSM412H1 Lecture Notes - Lecture 11: Time Series, Partial Correlation, Autocorrelation
Document Summary
April 2, 2020: time series: series of data points ordered in time, time is usually independent variable, goal is to make forecast for future, things to consider with time series: Time dependence: cannot use basic assumption of linear regression model that observations are independent. Homoscedasticity: covariance of ith term and (i+m)th term should not be function of time, can stationarize time series. Take difference in values rather than values themselves or other transformation. Gdp is equal to function of previous year gdp plus error term. Auto-regressive time series model: any shock to gdp will gradually fade off in future, expanding the equation, noise/shocks vanish quickly with time. Correlation between x(t) and x(t-n) always 0 for ma. Correlation gradually decreases with n for ar. Acf and pacf plots: for given time serives, need to answer, main difference is correlation between time series objects at different time points.
