Usually, you would impose sign restrictions when the sign of the response at impact isn't the question. You *have* to make *some* assumptions to disantangle the effects of various shocks in a VAR. Now, you can make assumptions that are less theoretically motivated (e.g., changes in variance of the shocks), but then you will face the issue of not knowing exactly what you identified in the data. Fortunately, there's a way to go work in both lanes at once: you can impose *more* restrictions than needed to separare the effects of shocks, so you can use the more agnostic statistical methods alongside the more theoretically motivated one like sign restrictions. The benefit? If you have too many restrictions, you can test a subset -- in your case, if you're more inclined to believe the statistical methods, you use them to test the more theoretical ones. That way, you can get a sense of whether your sign restrictions, say, are asking too much of the data -- and, if they don't, then you get the benefit of shocks you can interpret easily (e.g., positive supply shocks should lower prices and increase quantity in a competitive market model). So, you make a good point -- one that generated really interesting research.
...but sometimes economists manipulate the data to fit the theory. VAR models with sign restrictions, for example...
Nice seminar though...
Usually, you would impose sign restrictions when the sign of the response at impact isn't the question. You *have* to make *some* assumptions to disantangle the effects of various shocks in a VAR. Now, you can make assumptions that are less theoretically motivated (e.g., changes in variance of the shocks), but then you will face the issue of not knowing exactly what you identified in the data. Fortunately, there's a way to go work in both lanes at once: you can impose *more* restrictions than needed to separare the effects of shocks, so you can use the more agnostic statistical methods alongside the more theoretically motivated one like sign restrictions. The benefit? If you have too many restrictions, you can test a subset -- in your case, if you're more inclined to believe the statistical methods, you use them to test the more theoretical ones. That way, you can get a sense of whether your sign restrictions, say, are asking too much of the data -- and, if they don't, then you get the benefit of shocks you can interpret easily (e.g., positive supply shocks should lower prices and increase quantity in a competitive market model).
So, you make a good point -- one that generated really interesting research.
不错
Christopher Sims...
top kek