two time seres are economically linked they are covariant stationary t tes are reliable to test for cointergration regress one variable on the other if they are cointergrated model is reliable pls add
Cointegration refers to a situation where two or more series are independently nonstationary, but some linear combination of them can be found to be stationary. Hence, they are tied to together in the long run by some mutual economic force. Regressing one variable on the other is one possible method to test, but a weak one. It only allows the cointegration to be found between two variables, and cannot be extended beyond one cointegrating relationship. A better (albeit quite a bit more complex) method is the Johansen method. The model is not necessarily “reliable”, but rather correctly specified if a cointegrating relationship is found. In other words, two random walks can be combined in a way such that their RELATIVE values revert to some mean. Often used in pairs arbitrage, spot vs. futures pricing, term structures, and spread options. I actually recently wrote a paper on cointegration in exchange rates.
^ Fill in for my quant section, please?
Tag team? As long as you take care of my level 3 ethics…
Cointegration: Test each series for unit root by regressing each series on the other, possibilities: A) Neither has unit root, good to go errors will be covar stationary B) One has a unit root, errors will not be covar stationary (use first differencing to fix) C) Both have a unit root, check if series are cointegrated, if they are errors will be covar stationary, if not errors will not be covar stationary (use first differencing to fix) Like wy says above though this only works when you are working with two series, i.e. a simple single variable regression model, but if we see this at all on the exam I highly doubt it will be in the context of a multivariate problem.