Hi all! Sashiburi.. Long time no post :)
This time I'm gonna explain briefly about cointegration analysis, which is very useful in many research. This method is kind of regression analysis but under nonstationary condition. Remember, when we regress for example 2 variables with each of them nonstationary, then what we will have is spurious regression, a regression that means nothing. Hence Professor Engle-Granger found out that, I heard both of them got Nobel prize for this, when 2 variables nonstationary but the residual of the regression is stationary then it means that those two variables are having long term equilibrium, or we call it cointegrated. For example in the PPP (Purchasing Power Parity) case. We regress 2 variables, exchange rate and inflation. No doubt that these two variables data set are nonstationary. Hence we can still make linear regression between those two and check the residual, is it stationary or not. You can test it by Dickey-Fuller test. If it's stationary then PPP holds, otherwise it is spurious regression. So if the residuals are stationary, what's the model? We then use the error correction model, regressing the stationary term of exchange rates and the stationary term of inflation rates and the error correction term, the lag 1 of residual. That's it! So no more worries of the nonstationarity..lol..:))
This time I'm gonna explain briefly about cointegration analysis, which is very useful in many research. This method is kind of regression analysis but under nonstationary condition. Remember, when we regress for example 2 variables with each of them nonstationary, then what we will have is spurious regression, a regression that means nothing. Hence Professor Engle-Granger found out that, I heard both of them got Nobel prize for this, when 2 variables nonstationary but the residual of the regression is stationary then it means that those two variables are having long term equilibrium, or we call it cointegrated. For example in the PPP (Purchasing Power Parity) case. We regress 2 variables, exchange rate and inflation. No doubt that these two variables data set are nonstationary. Hence we can still make linear regression between those two and check the residual, is it stationary or not. You can test it by Dickey-Fuller test. If it's stationary then PPP holds, otherwise it is spurious regression. So if the residuals are stationary, what's the model? We then use the error correction model, regressing the stationary term of exchange rates and the stationary term of inflation rates and the error correction term, the lag 1 of residual. That's it! So no more worries of the nonstationarity..lol..:))
