Section9RegressionwithStationaryTimeSeries.pdfVIP

Section 9 Regression with Stationary Time Series How time-series regression differs from cross-section  Natural ordering of observations contains information o Random reshuffling of observations would obscure dynamic economic relationship, but leave traditional regression unchanged o How can we incorporate this dynamic information into our regression model?  We usually think of the data as being drawn from a potentially infinite data-generating process rather than from a finite population of observations.  Variables are often call “time series” or just “series” rather than variables o Index observations by time period t o Number of observations = T  Dynamic relationship means that not all of the effects of xt occur in period t. o A change in xt is likely to affect y t + 1 , y t + 2 , etc. o By the same logic, y depends not only on x but also on x , x , etc. t t t – 1 t – 2 o We model these dynamic relationships with distributed lag models, in which y t f x t ,x t 1,x t 2 ,.  We will need to focus on the dynamic elements of both the deterministic relationship between the variables and the stochastic relationship (error term)  The dynamic ordering of observations means that the error terms are usually serially correlated (or autocorrelated over time) o Shocks to the regression are unlikely to completely disappear before the following period  Exception: stock market returns, where investors should respond to any shock and make sure that next period’s return is not predictable o Two observations are likely to be more highly correlated if they are close to the same time than if they are more widely