1. Currency Unions and Trade: A Post-EMU Mea Culpa Reuven Glick, FRBSF Andrew K. Rose, Berkeley-Haas, ABFER, CEPR,NBER

2. Motivation Glick-Rose (2002) used panel approach to investigate effect of currency union on trade, using data through 1997 Assumption: Symmetry between currency union exit and entry Couldn’t test: 16 entries, 130 exits in 1948-1997 Can test now with EMU Caveat: is EMU similar to other currency unions? Other CUs involve small/poor countries, often unilateral Finding: insensitivity Earlier work stressed robustness, insensitivity True now, especially given modeling advances? 2

3. Preview of Findings 1.Symmetry looks OK (but uninteresting) 2.EMU different from other CUs 3.Econometric methodology matters a lot Consequentially, find it difficult to make strong statements with confidence about size of CU (EMU) effect 3

4. Old Methodology: Gravity Equation ln(T ijt ) =  0 +  1 ln(Y i Y j ) t +  2 ln(Y i Y j /Pop i Pop j ) t +  3 lnD ij +  4 Lang ij +  5 Cont ij +  6 FTA ijt +  7 Landl ij\ +  8 Island ij +  9 ln(Area i Area j ) +  10 ComCol ij +  11 CurCol ijt +  12 Colony ij +  13 ComNat ij +  CU ijt + {δ t } +  ijt 4

5. Data Set IMF DoTS trade: >200 “countries” 1948-2013 (with gaps) Trade is average of four potential bilateral flows Population, real GDP: WDI > PWT > IFS Country Characteristics: World Factbook RTAs: WTO Currency Unions: Glick-Rose updated 1:1 par for extended period of time (not hard fixes) Transitive: x—y and y-z imply x-z 5

6. Table 1: Pooled LS, Log Trade EER 2002 New with EMU dummy through 1997 Currency Union 1.30 (.13).92 (.09) 1.12 (.11) 1.09 (.11) EMU -1.10 (.13) Log Distance -1.11 (.02) -1.06 (.02) -1.06 (.02) -.94 (.02) Log Product Real GDPs.93 (.01) 1.03 (.01) 1.03 (.01).96 (.01) 6

7. Significance? Old: γ = 1.30 (robust standard error of.13) Pair of countries joined by a common currency trade over three times as much with each other (e 1.3  3.7) New, >200,000 extra observations: γ =.92 (.09) Smaller, but substantive (e.92  2.51), t-ratio exceeding 9 Tiny net EMU effect:.02 (.08), essentially nil Can raise easily though to.41 (.05) … by adding (plausible) fixed effects 7

8. Table 2: Add Dyadic FE, Log Trade EER 2002 New with EMU dummy through 1997 Currency Union.65 (.05).63 (.07).75 (.10).68 (.10) EMU -.33 (.11) Log Product Real GDPs.05 (.01).69 (.04).69 (.04).41 (.05) 8

9. Significance? Old: γ =.65 (.05) New: γ =.63 (.07) Now EMU effect has big net effect:.41 (.05) Still, much lower than other CUs, where γ =.75 (.10) Is EMU different from other currency unions? Or is post-1997 different from pre-1998? Answer: both 9

10. Table 3: Chow Tests, Log Trade Least SquaresDyadic FE Post-1997 versus all89. (.00) 134. (.00) EMU versus all27. (.00) 10. (.00) 10

11. Test for Symmetry (post-entry = - post- exit) Add (14) lags after currency union exit/entry and re-estimate Add Σ k θ k ENTRY ijt-k + Σ k φ k EXIT ijt-k above Can also split entry into EMU/non-EMU and test for symmetry Can do with LS or with dyadic FE Symmetry works well, perhaps surprisingly so 11

12. Figure 1 12

13. Figure 2: Seems sensible, symmetric 13

14. Table 4: Symmetry Tests, Log Trade Fixed Effects:TimeDyadic, Time After CU Entry = - After CU Exit? 2.8 (.00) 1.4 (.15) Before CU Entry = - Before CU Exit? 1.4 (.13) 1.8 (.04) Both2.6 (.00) 1.8 (.01) After non-EMU CU Entry = After EMU Entry? 1.6 (.09).8 (.73) Before non-EMU CU Entry = Before EMU Entry? 1.1 (.39) 1.3 (.17) Both1.4 (.07) 1.4 (.07) After non-EMU CU Exit = - After EMU Entry? 2.1 (.01) 1.1 (.36) 14

15. Do Results Hold Up to Extra Scrutiny? Much econometric progress in this area 15

16. Newer (Export) models Much work on “theory-consistent” gravity estimation Follow Head-Mayer (2014) survey, use Least Squares with Time- Varying Country Dummy Variables: ln(X ijt ) =  CU ijt +  3 lnD ij +  4 Lang ij +  5 Cont ij +  6 FTA ijt +  10 ComCol ij +  11 CurCol ijt +  12 Colony ij +  13 ComNat ij + {λ it } + {ψ jt } +  ijt 16

17. Table 5a: Exporter/Imp. x Time FE, log Exports Sample:Whole through 1997 Currency Union.51 (.02).76 (.02).88 (.02) EMU -1.41 (.04) Observations879,794 515,602 Country∙Time Fixed Effects22,438 16,029 R2R2.72.71 RMSE1.93 1.77 17

18. Negative EMU effect plausible? Net effect: γ = -.65 (.03) Turns out to be quite robust to sample Quite worrying, if believable! Follow Balwin and Taglioni (2007) by adding dyadic FE Add 33,886 FE (to the current 22,438)! 18

19. Table 5b: Add Dyadic FE, log Exports Sample:Whole through 1997 Currency Union.34 (.02).30 (.03).29 (.03) EMU.13 (.03) Observations879,794 515,602 Country∙Time Fixed Effects22,438 16,029 Dyadic Fixed Effects33,886 29,538 R2R2.86 RMSE1.42 1.25 19

20. Net EMU effect now big, positive Very significant (exp(.43)-1≈) 54% t-ratio>20 Symmetry also works well 20

21. Table 6: Symmetry Tests, Log Exports Fixed Effects:Exporter x year, Importer x year FE Dyadic, Exporter x year, Importer x year FE After CU Entry = - After CU Exit? 1.4 (.15).8 (.71) Before CU Entry = - Before CU Exit?.4 (.98).8 (.68) Both1.0 (.41) 1.0 (.49) After non-EMU CU Entry = After EMU Entry? 1.8 (.04) 1.3 (.17) Before non-EMU CU Entry = Before EMU Entry?.6 (.89) 1.4 (.16) Both1.2 (.27) 2.8 (.00) After non-EMU CU Exit = - After EMU Entry? 5.4 (.00).9 (.51) 21

22. Figure 3: Note especially effect of EMU entry 22

23. Figure 4 23

24. Handling Zeros and Heteroskedasticity Are all the LS estimates above biased because of: 1.Heteroskedasticity, and/or 2.Discarded observations of zero/missing trade? Both strongly prevalent in the data set! Santos Silva and Tenreyro propose Poisson pseudo-maximum likelihood to handle both We follow Head and Mayer to compare Poisson and LS Estimate in cross-section, since Poisson panel not feasible 24

25. Figure 5 25

26. Poisson vs. Least Squares? LS estimates of CU consistently positive, big, stable EMU effect very negative Poisson estimates of CU wander around Rarely overlap with LS Often of different sign! EMU effect much smaller and different sign! MaMu statistic varies but usually close to unity: Poisson efficient Just one manifestation of string of inconsistent, confusing results Each robust within class 26

27. Table 7: Net EMU Effect, Panel Log Trade Log Exports Fixed EffectsTimeTime, Dyadic (country-pair) Exporter x Time, Importer x Time Exporter x Time, Importer x Time, Dyadic Default.02 (.08).41 (.05) -.65 (.03).43 (.02) Data at Five-Year Intervals.03 (.08).34 (.06) -.50 (.07).51 (.05) Similarly-sized Countries.08 (.11).28 (.08) -.72 (.04).42 (.03) No Important Trade Relation.21 (.09).36 (.06) -.27 (.04).51 (.03) Drop post-2006 -.02 (.10).46 (.06) -1.12 (.05).19 (.03) Drop >|2σ| Residuals -.05 (.07).42 (.05) -.64 (.03).30 (.02) 27

28. Summary Glick-Rose (2002) concluded “a pair of countries which joined/left a currency union experienced a near- doubling/halving of bilateral trade.” Based on: 1.Assumption of symmetry between currency union exits and entries 2.Caveat: EMU might be different from other currency unions 3.Our results insensitive to precise econometric methodology Here, re-estimate using variety of models, annual panel >200 countries, 1948-2013, 15 EMU years 28

29. Conclusions Symmetry assumption seems reasonable but uninteresting Caveat warranted: EMU is sui generis, different Methodology matters a lot Pooled LS implies that EMU has no effect on trade (other CUs big) Dyadic FE implies EMU has big positive effect (lowers other CU effects) Exports with country x time FE implies EMU has big negative effect Other CUs big positive effect Adding dyadic FE imply smaller CU and big positive EMU effects Poisson vary over time, rarely overlap with LS, usually small, insignificant Conclude: cannot estimate currency union effects reliably 29

30. Log TradeLog Exports TimeeTime, DyadicExp/Imp x TimeAdd Dyadic Currency UnionEMUCurrency UnionEMUCurrency UnionEMUCurrency UnionEMU Default1.12 (.11) -1.10 (.13).75 (.10) -.33 (.11).76 (.02) -1.41 (.04).30 (.03).13 (.03) Drop Time Effects 1.30 (.11) -1.29 (.14).85 (.10) -.70 (.11) n/a n/a n/a n/a Data at Five-Year Intervals 1.16 (.11) -1.13 (.14).76 (.12) -.43 (.13).76 (.04) -1.26 (.08).37 (.06).14 (.08) Add Quadratic Output Terms.80 (.11) -1.07 (.14).53 (.10) -.31 (.11) n/a n/a n/a n/a No Industrial Countries.82 (.12) -.04 (.41).72 (.16) -.13 (.23).46 (.03).98 (.18) -.02 (.04) 1.19 (.14) Larger Countries)1.06 (.12) -1.08 (.14).66 (.11) -.32 (.12).70 (.02) -1.37 (.04).28 (.03).14 (.03) No Poor Countries 1.21 (.13) -1.26 (.15).47 (.11) -.07 (.12).81 (.02) -1.43 (.04).23 (.03).22 (.04) Similarly-sized Countries 1.25 (.14) -1.17 (.18).86 (.17) -.58 (.18).64 (.04) -1.36 (.06).43 (.06) -.01 (.06) No Important Trade Relation 1.11 (.12) -.90 (.15).72 (.12) -.36 (.13).82 (.03) -1.09 (.05).19 (.04).31 (.05) Drop pre-1960 1.15 (.11) -1.15 (.14).79 (.11) -.42 (.12).79 (.02) -1.46 (.04).25 (.03).20 (.04) Drop pre-1980 1.18 (.15) -1.37 (.17).26 (.18) -.08 (.19).79 (.03) -1.57 (.05).13 (.08).34 (.08) Drop post-2006 1.07 (.11) -1.09 (.15).70 (.10) -.24 (.12).79 (.02) -1.91 (.05).28 (.03) -.10 (.04) Drop CFA 1.11 (.13) -1.08 (.15).75 (.11) -.34 (.12).79 (.02) -1.43 (.04).26 (.03).16 (.03) Drop ECCB, US$, Fr…. 1.27 (.15) -1.24 (.17) 1.01 (.20) -.60 (.21).89 (.03) -1.53 (.04).25 (.05).16 (.05) Drop >|2σ| Residuals 1.17 (.10) -1.22 (.12).69 (.08) -.28 (.09).79 (.02) -1.43 (.03).57 (.02) -.28 (.03) 30