MARTIN needs its oil changed
The shock du jour is poorly modelled.
While the current surge in inflation is in part driven by the rock bottom unemployment rate, there is no doubt that the high oil price is also making a significant contribution. So I thought I would see what insights MARTIN can provide about the canonical supply side shock - an increase in the global oil price.
The results were not what I was expecting.
The figures below show the response to a one year 10% increase in the global oil price.
Most of the responses make little sense. Output rises? Unemployment falls? Trimmed-mean inflation initially decreases before rising (due to the tightening labour market)? It all seems very odd.
Digging around a bit I think most of the problem comes from how MARTIN models inflation. Oil prices feed directly into import prices (which explains the sharp fall in the quantity of imports as their price rises) and headline CPI (which ironically nobody cares about).
This graph shows the impact on headline CPI (blue), trimmed-mean inflation (orange) and growth in the import price deflator (green).
However by assumption it has zero impact on the short run dynamics of trimmed-mean inflation! It only impacts trimmed-mean inflation via its long-run cointegration and even then the effect is very small. Since MARTIN also assumes that inflation expectations are exogenous there is little scope for the oil price shock to feed through into lower output, increased firm costs or a higher cash rate.
Instead the main mechanism is to push import prices higher, which lowers the terms of trade, decreasing the real exchange rate, which in turn boosts net exports. The fall in imports combined with the rise in exports drives GDP higher which in turn lowers unemployment.
In short none of this seems particularly realistic and certainly not how oil price shocks are traditionally seen ie as worsening the inflation-output trade off that central banks face when setting policy!
This highlights the risk of the macro-econometric approach. If you haven’t devoted enough time to empirically estimating a given shock or sector in the model there is a risk that lack of either empirical or theoretical grounding will give results that are wrong.
Hopefully the RBA has checked in MARTIN for a service since it was published!