Trade and Openness. Econ 2840

Transcription

1 Trade and Openness Econ 2840

2 Background Economists have been thinking about free trade for a long time. This is the oldest policy issue in the eld. Simple correlations: Richer countries have higher trade/gdp maybe that is just because of demand or industry structure, rather than eect of trade policy Sachs and Warner (1995) and Wacziarg and Welch (2008) construct measures of trade openness (i.e. policy) based on multiple dimensions: taris, manipulation of the exchange rate, government monopolies on exports, etc. in cross-section, more open countries are richer putting openness in a growth regression, it has a positive coecient Also, episodes of trade liberalization are accompanied by faster growth, on average Wacziarg and Welch: growth 1.5% higher following liberalization

3 That Doesn't Prove Causality! Omitted Variable Bias: maybe good institutions lead to both free trade and higher income looking at changes in trade openness does not x this. Maybe same institutional change that led to more trade led to better economic outcomes for other reasons. We need instruments, natural experiments, etc.

4 Frankel and Romer Does Trade Cause Growth? (1999) Instrument for trade with predicted trade share from a gravity equaiton Gravity Model: old empirical specication for trade that ts the data well, even without a great theoretical foundation analogy to gravitational attraction For creating their instrument, they use data on bilateral trade for 63 countries (instrument then applied to more countries than that. dep. var. is log of this (see next slide) trade share is total bilateral trade (imports plus exports) between countries i and j, divided by GDP in country i regress this on distance, population in each country, area of each country, dummy for landlocked, and a common border dummy interacted with all of the above Estimates reasonable: distance lowers trade; common border raises it; bigger own population lowers trade share; bigger other population raises it; bigger area lowers trade.

5 The Bilateral Trade Equation

6 Zero Trade Shares One issue in F&R approach: in their regression they take the log of trade share, so they have to drop zeros. Helpman, Melitz, and Rubenstein (2008): this is not right. Zero trade is not missing it is no trade! No trade is in fact common see next slide. The right thing to do is estimate seperate equations for extensive margin (engage in trade or not) and intensive margin (volume of trade, conditional on trading). Eect of distance etc. on these margins may be dierent. I don't know if this matters much for F&R paper.

7 Helpman, Melitz, and Rubinstein (2008)

8 Predicted Trade Shares Use estimates from Table 1 to predict bilateral trade shares in larger sample of countries (150) Sum up within a country to get its own predicted trade share predicted trade share not a function of own income, institutions, etc. Figure 1 and Table 2 show that predicted trade share does predict actual trade share

9 Actual versus Constructed Trade Share

10 The Relation between Actual and Constructed Overall Trade

11 More Investigation of Constructed Trade Share F-stat on constructed trade share in column (3) above is 13.1 Figure on next page shows the scatter of actual and predicted trade shares conditional on population and area big outliers are Singapore and Luxembourg Dropping these (second panel) lowers the F-stat to 10.1

12 Partial Association between Actual and Constructed Trade Share

13 The Big Result

14 Findings Trade raises income. IV is bigger than OLS (see next slide). Why controlling for area and population? Do they aect GDP through other channels? Maybe because popualtion is endogenous? Might be interesting to think about.

15 OLS vs. IV Why is IV bigger? If you thought that trade openness biased OLS up, then IV should be smaller of course, measurement error biases the other way In any case, it is good to investigate Their approach: regress trade share on instrument. Then take tted values and residuals from that. For each of these, partial out other stu from RHS or regression and plot result against income rst panel gives scatter underlying IV coecient; second panel gives the part of OLS that is not in IV

16 Partial Associations Between Income and the Trade Share

17 Dierence Between IV and OLS They conclude: no smoking gun Maybe the problem is indeed measurement error: trade volume is a bad measure of total interactions among countries. Leaves out exchange of ideas, travel, etc. IV cleans up meausrement error and so raises coecient. Of course, if that is right, then then trade was not the whole story.

18 What are the Channels? A priori, seems like trade should aect productivity. But could also aect growth/income through capital accumulation, etc. Use methodology of development accounting Aggregate Production: Y i = K α i (A i h i L i ) 1 α Human Capital Aggregate: h i = exp(φ(s i )) where s is schooling Rewrite the production function this way (why: because K/Y is constant in a steady state): Y i = (K i /Y i ) α/(1 α) e φ(s i ) A i L i divide by L and take logs: ln(y i /N i ) = α 1 α ln(k i/y i ) + φ(s i ) + ln(a i ) Regress each of these terms on trade share. Coecients have to add up to coecient from Table 3! (neat trick see Wong, 2007, for more of this).

19 Trade and the Components of Income

20 result: biggest part of the eect is through productivity (A) but weak signcance

21 Rodriguez and Rodrik (NBER Macro Annual 2000) Critique of F&R what we should care about is trade policy, not trade volume F&R instrument is diculty in engaging in trade. This is like an indiscriminate trade policy actual trade policy is targetted at e.g. market failures (infant industry protection) This policy might be good for growth even if indiscriminate policy was bad. Of course, real world trade policy could also be worse than indiscriminate policy for example if trade restriction resulted from rent seeking; then we get not only the lost gains from trade, but also the direct costs of rent seeking Second critique is regarding instrument: It could be spuriously correlated with other geographic features that are bad for output (see table next slide).

22 Frankel-Romer Regressions with Additional Geographic Variables

23 Feyrer's Papers: F&R in the Time Dimension Problem with F&R is that geography is so correlated with history, tropics, etc. that it is hard to sort out causality Ideal natural experiment is to look at changes in trading distance or costs between countries Two natural experiments: switch from sea freight toward air freight in post WWII period some country pairs become much closer (Germany-Japan); others no change (US-UK) closing and re-opening of the Suez Canan (1967 and 1975). similarly some pair distances change a lot (Kenya-UK) while others don't Both cases, nd that trade aects income This could be looked at within countries as well probably more cool experiments out there.