Go global or local - An issue of transfer learning

Go global or go local with pricing models? This is an important question when thinking about pricing assets outside of the US. Do you estimate models only using local information or do you take relationships that exist in the US as the basis for empirical analysis. If prices in local markets are driven by a global model with US coefficients, then the development of pricing models for local markets can be made much easier. The focus can be on building strong global models and then just applying to the diverse of markets around the world. Quant life would be much easier. 

This a problem in transfer learning and a recent paper suggests that using pricing models in the US as a basis for pricing in other countries makes sense.  See "How Global is Predictability? The Power of Financial Transfer Learning" 

While we focus on a simple global model, the researchers include a local component based on the deviation for the global component which they refer to as a generalized elastic net which can adjust to the global model. Again, this may be an easy way to build pricing models across the global by starting with a universal model. Easy may not be the right word for describing this approach but it has merit and places a focus on universal sensitivity of key variables.