Exports in Context
Anybody who follows forecasts of GDP growth for 2011Q1 will notice that over time, estimates have been revised down (this is true for Macroeconomic Advisers, for instance). The dimmed prospects for GDP growth throws in high relief the importance of net exports. From the WSJ, “Foreign Shocks Temper America’s Export-Led Rebound”:
To an extent unique in post-World War II history, the U.S. economy’s climb out of recession has been led by selling crops, natural resources and manufactured goods to the rest of the world.
Now that important engine for U.S. growth—the rest of the world—is damping the improving outlook. The world’s No. 3 economy, Japan, is reeling from an earthquake and nuclear crisis. Unrest in the Middle East has sent oil prices—and global anxiety—soaring. Fast-growing China, anxious about inflation, and other emerging markets are trying to tap the brakes. And fiscal strain looms over Europe.
The contribution of export spending to GPD growth in an accounting sense (q/q, SAAR) is illustrated in Figure 1.
The importance of exports was also highlighted in Chapter 4 of the recently released Economic Report of the President. The report notes (p. 102):
U.S. export growth also benefits from changes in relative prices caused by faster inflation in growing emerging markets because faster inflation abroad means U.S. goods are cheaper on world markets relative to goods from these countries. These price and growth relationships suggest that if the United States is to double exports, an overwhelming portion of that new export growth will come from faster-growing emerging and developing economies. Figure 4-12 shows the share of projected growth of U.S. nominal exports by region using IMF forecasts for GDP and price growth in different regions.
Here is Figure 4-12.
These projections depend on projections of growth abroad, and the elasticity of export demand with respect to income.
How Big Are Export Elasticities Now?
These projections are based upon elasticites of exports with respect to growth of around 2, and one of the cited sources is Chinn (2005). Those estimates were based upon time series data up to 2001Q2. My subsequent work (see  and ) has taken into account potentially important supply side factors (that is, the quantity of exports depends on foreign demand as well as domestic supply). I obtain some slightly lower estimates of income elasticities imposing equality on supply and demand factors, but that result is suspect given the co-trending nature of rest-of-world growth and proxy measures for US capacity. Moreover, the fact that the elasticity estimates cited in the ERP are around two for cross section-based estimates does suggest that the elasticities are fairly high (the ERP notes that dropping Singapore pushes up the income elasticity to three).
In any case, I wanted to re-investigate this question, using the most recent data, up through 2010Q4. Figure 2 shows log real exports and log real rest-of-world GDP, rebased to 0 in 1999Q1.
Notice that real exports are 18.4% above their trough in 2009Q2 (and 1.1% above their pre-recession peak in 2008Q2), while rest-of-world GDP was up by 15.2% relative to that same period. Nominal exports are 26.7% above trough levels; this figure is of interest given the President’s goal of doubling exports, as discussed in this post.
In order to more formally assess the export-income relationship, I disaggregated into goods and services. For each, I estimated the following error correction model with up to two lags. (The approach follows that outlined in this post.)
Δ exp t = β 0 + φ exp t-1 + β 1 y t-1 + β 2 r t-1 + γ 0 Δ exp t-1 + γ 1 Δ exp t-2 + γ 2 Δ y *t-1 + γ 3 Δ y *t-2 + γ 4 Δ r t-1 + γ 5 Δ r t-2 + u t
For goods exports, only one lag of each variable was necessary.
I find that the long run elasticity of goods exports over the 1975Q1-2010Q4 period is 1.84 (while the long run price elasticity is 0.92). Both coefficients are statistically significantly different from zero at conventional levels, using robust standard errors. The error correction specification seems adequate, with the adjusted-R2 = 0.34, the standard error of regression at 0.024, and the usual tests for serial correlation in residuals failing to reject at conventional levels (Q(4), Q(8), LM test with 2 lags).
The long run elasticity of services exports over the 1975Q1-2010Q4 period is 1.59 (while the long run price elasticity is 0.49). In this case, only the long run income coefficient is statistically significantly different from zero at conventional levels, using robust standard errors. The error correction specification seems a bit less successful, with the adjusted-R2 = 0.13, the standard error of regression at 0.028, and the usual tests for serial correlation in residuals failing to reject at conventional levels (Q(4), Q(8), LM test with 2 lags).
Dealing with the supply capacity is difficult. One could try to measure the size of the goods exports capacity using the capital stock. I’ve tried to capture the trend using manufacturing output. This variable is of course endogenous, but if one is trying to get the long run effect, might not be a problem. One way to see if statistically manufacturing production is endogenous with respect to the cointegrating vector is to see if log manufacturing production is weakly exogenous. I estimate a VEC(3) (consistent with a VAR specification with 4 lags), and check the statistical significance of the reversion coefficient on manufacturing production. A likelihood ratio test restricting the reversion coefficient to be zero fails to reject.
Re-estimating the goods export equation yields a slightly lower income elasticity, of 1.3. On the other hand, while the supply elasticity is positive (as anticipated) at 0.62, the long run coefficient is not statistically significant. (The adjusted R2 rises only slightly to 0.36.).
Do these results imply an elasticity of 2 is too high? Using this error correction specification’s point estimates, the top of the one standard error band places the long run coefficient at 2.3, and the bottom at 0.7; in other words, a demand elasticity of 2 is still quite plausible given the imprecision of the estimates. Cross-checking, I used dynamic OLS (allowing in a linear time trend, two leads and lags). This leads to a point estimate of 1.74, with (long run) standard error of 0.80, once again easily encompassing the 2.0 estimate.
Limitations of the Preceding Analysis
Even though I’ve broken down exports into goods and services, clearly one could disaggregate further. In this paper, I do; the conclusions are not particularly clear, given the sensitivity of results to trends. The income and price elasticities vary over categories (and income elasticities can easily exceed 2, depending on the assumptions regarding time trends).
In addition, the composition of exports is changing over time, which implies changes in the sensitivity of trade flows to both income and exchange rates. Figure 3 illustrates the changes in the (nominal) flows of exports by category, since 2009Q2.
It’s hard to say what this changing composition of exports implies for the aggregate income elasticity. To the extent that services are a rising share of total exports, the income elasticity might decline in aggregate (goods and services). On the other hand, the composition of services is shifting toward other private services (and away from travel and passenger fares). If service sectors abroad are liberalized, one could see measured income elasticities rise.
Global Rebalancing and Partial Equilibrium Analysis
There is a tendency to dismiss the partial equilibrium/static approach that is central when discussing the impacts of policies on trade flows. I agree that these types of elasticities are only part of the story, when one wants to discuss the prospects for global rebalancing. Aggregate saving (both public and private) and investment also matter; however, it would be wrong to take those as variables as strictly exogenous. Focus on those aggregates can also lead to analytical mis-steps. As Manoj Pradhan (Morgan Stanley) and Alan Taylor (Morgan Stanley and UC Davis) noted in an provocative blogpost yesterday:
we see lasting consequences beyond those unfolding in the immediate aftermath of the financial crisis:
- A huge rise in demand for capital in EMs with a more moderate increase in DMs. Talk of a savings glut or an investment drought may recede. The global real interest rate is likely to rise.
- Less saving flows out of EM economies. Growth prospects are the main driver but risk premia for newly resilient EMs may fall. If investment demand is muted in DMs, and saving flat, the shift is weaker in DMs. Global imbalances moderate, reinforcing the trends after the crisis.
- These current account shifts cannot be an “immaculate transfer” without real exchange rate adjustment. Recent real appreciation of EMs took the form of relative inflation and managed currencies (the latter creating political distractions). But EMs are likely to absorb further adjustment through nominal appreciation, given a triple whammy of cyclical reflation, growth differentials pushing nontradable inflation and oil/commodity price shocks. [Emphasis added — mdc]
Tracing out how income and relative price changes translates concretely into changes in flows requires some insight on these elasticities, even when acknowledging the partial nature of the analysis.
In addition, to the extent that there are nominal rigidities, monetary shocks could induce short run movements in trade flows that are partially unrelated to changes in savings and investment.