Cointegration Approach I want to perform a cointegration test between metal prices in USA and India. For USA prices are in dollars per pound and for India they are is in rupees per quintal (100 kilogram).
Before checking for cointegration do I need to convert one data set so that both are in similar currency (preferable rupee to dollar)?
 A: Yes, you need to convert the prices to have the same currency. 
If the prices are cointegrated when measured in the same currency, they will not necessarily be cointegrated when measured in different currencies -- because currency exchange rates may fluctuate a lot over time. 
It makes more sense to measure prices in the same currency if you expect that the metal is valued more or less the same (or at a fixed discount/surcharge) in U.S. as it is in India. Even if that is the case, diverging currency exchange rates may disturb the picture painted by nominal prices in different currencies. 
Meanwhile, you do not care about measuring in pounds versus kilograms versus quintals as the ratio of pound to kilogram or pound to quintal remains constant.
A: Strictly speaking, no you don't have to convert. The theory of cointegration doesn't mention that you should look for a cointegration relationship between apples and apples. 
If both series have a unit root and if they share a common trend on the long term, then a cointegration relationship exists between them, whatever the unit or the potential hidden relationship between those units.
Now, if you are in the context of some economic theory (as I think you are), the potential cointegration relationship would be easier to understand or interpret if you are using the same units as you wouldn't risk to miss an explanatory phenomenon such as the relative fluctuations of both currencies.
