Thursday, February 14, 2013

A wonkish complaint about gravity models

. Thursday, February 14, 2013

Recently, I've been thinking a lot about how to apply gravity models of trade to FDI flows, especially to infer "missing FDI."[And, if you also interested in gravity models, Will (aka Kindred) has a recent post about quantum gravity models of trade.] One thing I've learned is that several economists are trying to create comprehensive gravity models of FDI that have solid theoretical and empirical footing, but there currently is no FDI gravity model that is as widely accepted as the standard gravity models of trade. Firm level locational decisions that drive FDI are theoretically more complex than trade decisions. Perhaps the most straightforward example of this is that while some argue FDI is a theoretical substitute for trade, bilateral FDI and trade are actually positively correlated. There are some new models that are much better equipped to estimate firm entry as a function firm productivity, building off work by Melitz as well as Helpman. We seem to be getting closer to a standard gravity model for FDI. I want to make two points about issues I see with this agenda.

1) Any gravity model (of trade, of FDI, of any flow) is necessarily retrospective. We establish the veracity of models by how well they predict previous flows. But, what happens when the decision rules firms use to determine economic activity fundamentally change? The gravity model can well predict economic exchange that follow previous patterns, but it runs into trouble when production networks follow new logics. A quick look at patterns of sales of overseas affiliates of US MNCs in 1999 and 2009 illustrates the logic behind FDI is shifting. About 60% of overseas affiliate sales were local. This is relatively consistent across regions, although local sales by overseas affiliates were as low as 51% in Africa. Overseas affiliate sales back to the US were just under 9%, indicating vertical FDI is less prevalent than most assume. Overseas affiliate sales to other markets, whether to other affiliates or to unaffiliated buyers was about 30%. These sort of sales indicate FDI locational decisions based on export platform models and complex supply chains. What is particularly interesting is how the composition of sales by US overseas affiliates have changed over the past 10 years. In 1999, local market, home market, and third market sales stood at 67% 10% and 13% respectively. Thus, FDI over the past 10 years has experienced an important change. MNCs, at least from the US, are shifting strategies from horizontal FDI to FDI motivated by export platform strategies and by the increasing complexity of supply chains.

The take away from these changes is that gravity models of FDI that are built on patterns of FDI flows in the 1990s may not generate appropriate predictions for FDI today. And, as complex supply chains and export platform models drive a greater percentage of MNC's investment decisions, estimation techniques that are fundamentally bi-lateral may significantly lose their theoretical appeal as well as their predictive capabilities (heteroskedasticity in the error term is going to be increasing difficult to fix!).

2) All these changes in patterns of FDI probably have important implications for gravity models of trade. After all, intrafirm trade accounts for half of global trade.

I guess the implications of all this are threefold. In a world in which we love data sets with long time series, we need to rethink when it is appropriate to pool temporally. And, we will need to continually go back to the structure of gravity models to ascertain whether they remain good predictors of economic flows. Finally, we must also consider when modeling economic exchange in a bilateral context misses so much of the underlying structural dynamics that dyadic estimation techniques will be wrong.


A wonkish complaint about gravity models




Add to Technorati Favorites