BUS 215 Lecture Notes - Lecture 13: Autocorrelation, Autoregressive Model, Capacity Utilization
Document Summary
Because autocorrelation is primarily a phenomenon of time series data, it is convenient to represent the linear regression model using t as a subscript to represent time: (13b. 1) where t = 1, 2, , n. We assume that there are observations covering n periods of time. Autocorrelation (also called serial correlation) exists when the error terms 1, 2, , n are not independent of one another. There are many ways we might envision non-independence among the errors. The first-order autoregressive model (sometimes called the ar1 model) is a common way of thinking about correlated errors: (13b. 2) where 1 +1 tu. 1 t t where is the autocorrelation parameter and ut is a well-behaved (i. e. , normally distributed, homoscedastic, non-autocorrelated) random disturbance with mean zero and constant variance. As you can see from equation (13b. 2), if = 0, then t is also well-behaved because t = ut. = 0 then t is unaffected by error t-1.