Jan Fidrmuc Jarko Fidrmuc

Brunel University, CEDI, CEPR and WDI

University of Munich, Comenius University and CESifo

2nd FIW-Research Conference „International Economics“Vienna University of EconomicsDecember 12, 2008

Foreign Languages and Trade

Introduction

Do languages affect trade? Easier communication lower

transaction costs greater trade Trade analysis (gravity model)

typically accounts for common official languageE.g. Rose (2000): common

language increases trade by 50%

Introduction (cont’d)

Gravity models: official languages only

Dummy variables, not proficiency Proficiency varies across countries

E.g. French in France, Belgium, Luxembourg, Switzerland, Canada,…

Other languages besides official ones matter tooNon-official indigenous languagesForeign languages

Introduction (cont’d)

Rauch (1999, 2001), Rauch and Trindade (2002), Bandyopadhyay, Coughlin and Wall (2008)Ethnic-networks increase tradeRauch and Trindade (2002):

ethnic Chinese networks in SE Asia increase trade by at least 60%

Introduction (cont’d)

Mélitz (2008)Official and non-official

indigenous languagesLanguage impact measured using

dummy variables (if official or spoken by more than 20%) or communicative probability

Only indigenous languages (Ethnologue database)

Our Contribution

First to study effect of native and foreign (learned) languages alikeTrade often relies on

communication in non-native languages

Unique extensive dataset on language proficiency in the EU

Non-linearity Old vs new Europe Role of English

Data

Special Eurobarometer 255: Europeans and their Languages, November - December 2005

Nationally representative surveys; only EU nationals included

Mother’s tongue(s) and up to 3 other languages that they speak well enough to have a conversation

Self-assessed proficiency: basic, good, very good

Trade flows: 2001-07

English (good/very good skills)

French (good/very good skills)

German (good/very good skills)

Russian (good/very good skills)

Spanish (good/very good skills)

Italian (good/very good skills)

Gravity Model

Gravity model methodology following Baldwin and Taglioni (2006)

Trade between i and j, Tijt, and output of i and j, yit and yjt,, measured in nominal US$

Distance between i and j: dij Common border and common

history dummies: bij and fij

ijt

F

fijff

D

dijddijijijjtitjtitijt PLfbdyyT ,,4321

Gravity Model (cont’d)

Common official-language dummies: Ldij

Communication probabilities: Pfij Time-varying country dummies:

Country-specific time-invariant and time-varying omitted variables

Country-specific measurement problems

ijt

F

fijff

D

dijddijijijjtitjtitijt PLfbdyyT ,,4321

Communicative Probability

Probability that two random individuals from two different countries speak the same language

1. English2. Languages spoken by at least 10%

of population in at least 3 countries German, French, Russian

3. Languages spoken by at least 4% of population in at least 3 countries

Italian, Spanish, Hungarian, Swedish

Communicative Probability

EU15 NMS/ACs EU29

English 22 11 17

German 7 2 5

French 5 1 3

Results: EU15

Common official language and communicative probability raise trade

English especially important Accounting for proficiency in English

lowers official-language effect French/German: weak/mixed results Spanish/Italian/Swedish: seemingly

strong effects Most country pairs’ at/close to zero

Results: EU 15Variable (1) (2) (3) (4) (5) (6) GDP 1.004 *** 0.897 *** 0.885 *** 0.880 *** 0.895 *** 1.007 ***

Distance -0.772 *** -0.748 *** -0.761 *** -0.750 *** -0.668 *** -0.754 ***

Contiguity 0.499 *** 0.471 *** 0.491 *** 0.364 *** 0.157 *** 0.478 ***

Official Languages

English 0.908 *** 0.543 *** 0.570 *** 0.662 *** 0.775 *** 0.786 ***

German 0.556 *** 0.581 *** 0.853 *** 0.841 *** 0.667 *** 0.336 ***

French 0.150 ** 0.186 ** 0.101 0.295 0.788 *** -0.033

Swedish 0.158 0.279 *** 0.235 ** 0.323 *** -2.974 *** 0.218 **

Dutch -0.344 *** -0.263 *** -0.340 *** -0.180 ** 0.150 *** -0.287 ***

Proficiency English 1.152 *** 1.074 *** 0.944 *** 1.022 *** French 0.080 0.065 -0.321 German -0.408 *** -0.274 *** 0.102 Italian 8.724 *** 11.687 *** Spanish 8.938 *** 12.071 *** Swedish 19.793 *** Cumulative EFG 0.396 ***

N 1470 1470 1470 1470 1470 1470 Adjusted R2 0.972 0.974 0.974 0.975 0.980 0.973

Results: EU15, magnitude Consider column (5)

Average effect in EU15: 25% increase due to English proficiency (22% average communicative probability)

UK-IRL trade increased 2.2 times because English is official language and 2.7 times because of English proficiency 5.8 fold increase overall

NL-S trade increased 1.7 times and NL-UK trade more than doubled

Results: NMS/AC

English proficiency raises trade Large coefficient estimate but

proficiency is relatively low Average impact: 77% increase

(11% average communicative probability)

German and Russian also significant Average impact of German: 30%

Results: NMS/AC

Variable (1) (2) (3) (4) (5)

GDP 0.571 *** 0.573 *** 0.566 *** 0.566 *** 0.574 **

Distance -1.039 *** -1.024 *** -0.817 *** -0.820 *** -1.001 ***

Former Federations 2.278 *** 2.292 *** 1.478 *** 1.471 *** 2.299 ***

Contiguity 0.543 *** 0.531 *** 0.650 *** 0.654 *** 0.538 ***

Proficiency:

English 5.074 *** 5.182 *** 5.188 ***

German 13.381 * 13.239 *

Russian 3.748 *** 3.745 ***

Hungarian -0.309

Cumulative 4.978 ***

N 1254 1254 1254 1254 1254

Adjusted R2 0.847 0.850 0.858 0.858 0.850

Results: EU29

Weaker results English significant but impact less

powerful than in either EU15 or NMS/ACAverage English proficiency (17%)

raises trade by 11% French not significant and German

negative impact Remaining languages significant

Results: EU29

Variable (1) (2) (3) (4) (5) (6)

GDP 0.987 *** 0.767 *** 0.769 *** 0.773 *** 0.773 *** 0.843 ***

Distance -1.038 *** -1.029 *** -1.035 *** -1.025 *** -1.019 *** -1.028 ***

Former Federations 2.460 *** 2.455 *** 1.961 *** 1.988 *** 2.043 *** 2.466 ***

Contiguity 0.320 *** 0.325 *** 0.339 *** 0.328 *** 0.270 *** 0.317 ***

EU 0.263 *** 0.235 *** 0.216 *** 0.218 *** 0.210 *** 0.246 ***

Official languages

English 0.931 *** 0.715 *** 0.739 *** 0.752 *** 0.786 *** 0.802 ***

German 0.566 *** 0.571 *** 0.910 *** 0.893 *** 0.838 *** 0.337 ***

French 0.037 0.056 0.230 0.240 0.295 -0.160

Greek 2.319 *** 2.333 *** 2.316 *** 2.330 *** 2.359 *** 2.333 ***

Swedish 0.140 ** 0.162 *** 0.134 ** 0.153 ** -2.156 *** 0.162 **

Dutch -0.621 *** -0.622 *** -0.638 *** -0.610 *** -0.543 *** -0.614 ***

Proficiency:

English 0.664 *** 0.569 *** 0.595 *** 0.508 ***

French -0.315 -0.283 -0.276

German -0.470 *** -0.436 *** -0.330 **

Russian 1.603 *** 1.559 *** 1.557 ***

Italian 1.637 *** 1.706 ***

Spanish 2.645 ** 3.444 ***

Swedish 12.635 ***

Hungarian 3.577 ***

Cumulative 0.386 ***

N 5634 5634 5634 5634 5634 5634

Adjusted R2 0.930 0.930 0.931 0.931 0.931 0.930

Results: Discussion

Possible explanations for weaker EU29 results:

1. Heterogeneity: EU15 vs NMS/AC Trade between EU15 and NMS/AC

still below potential Different political, economic and

linguistic legacy NMS/AC have not yet reached

their new linguistic equilibrium

2. Effect of languages not linear

Results: Non-linear Effect

Add squared communicative probability Hump-shaped effect of English

diminishing returns Peak at around 70% probability Quadratic term not significant in

NMS/AC and EU29 French/German: weaker/negative effect Other languages: quadratic terms not

significant in NMS/AC and EU29Except Russian: U-shaped in NMS/AC

Results: Non-linear Effect EU15Variable (1) (2) (3) (4) (5)

Intercept GDP Distance Contiguity included but not reported

Official languages

English 0.908 *** 1.369 *** 1.672 *** 1.749 *** 1.601 ***

German 0.556 *** 0.661 *** 0.030 0.015 0.325 ***

French 0.150 * 0.292 *** 0.400 0.514 1.003 ***

Swedish 0.158 ** 0.362 *** 0.256 *** 0.279 *** 17.057 ***

Dutch -0.344 *** -0.283 *** -0.404 *** -0.286 *** 0.030

Proficiency:

English 5.157 *** 6.005 *** 6.008 *** 5.178 ***

French 1.119 *** 1.220 *** 0.040 **

German -2.633 *** -2.499 *** -1.108 **

Italian 46.564 *** 33.852 ***

Spanish 10.856 *** 11.446 ***

Swedish 80.606 ***

Proficiency (Quadratic):

English -3.580 *** -4.481 *** -4.580 *** -3.690 ***

French -1.552 *** -1.712 *** -0.872 **

German 3.230 *** 3.172 *** 1.571 ***

Italian -748.687 *** -461.089 ***

Spanish -75.874 ** -52.094

Swedish -857.98 ***

N 1470 1470 1470 1470 1470

Adjusted R2 0.972 0.975 0.977 0.978 0.983

Non-linear Effect: EU15

-1.0

-0.5

0.0

0.5

1.0

1.5

2.0

2.5

english

french

german

Robustness: EU15

Results potentially driven by outliersCountry pairs with especially

high/low trade Effect of English proficiency

highest around 50th percentile (median regression)

Effect of foreign languages not due to outliers

Results: EU 15, Quantile Regressions

-0.4

-0.2

0.0

0.2

0.4

0.6

0.8

1.0

Q5 Q15 Q25 Q35 Q45 Q55 Q65 Q75 Q85 Q95

Quantile Regression

OLS

Results: EU 15, Quantile Regressions

OLS Q10 Q25 Q50 Q75 Q90 Test

Income 0.895 *** 0.962 *** 0.931 *** 0.874 *** 0.836 *** 0.795 *** 26.15

Distance -0.694 *** -0.464 *** -0.695 *** -0.709 *** -0.787 *** -0.852 *** 0.94

Contiguity 0.643 *** 0.673 *** 0.483 *** 0.687 *** 0.591 *** 0.319 *** 7.06

Eng. off. lang. 0.488 *** 1.088 *** 0.890 *** 0.433 ** 0.426 *** 0.400 *** 5.10

Eng. proff. 0.549 *** 0.304 0.340 *** 0.697 *** 0.426 *** 0.272 *** 9.46

Intercept -21.313 *** -27.083 *** -23.557 *** -20.109 *** -17.209 *** -14.193 *** 22.42

N 1800 1800 1800 1800 1800 1800 1800

Pseudo R2 0.918 0.738 0.735 0.722 0.716 0.714 ND

Conclusions

Language has a strong effect on trade

Countries with common official language trade more with each other

Proficiency in foreign languages also increases trade

Effects of languages different in EU15 and NMS/AC

Effect of languages seems non-linear (diminishing returns)

Conclusions (cont’d)

Universal proficiency in English could raise trade up to 2.7 times (EU15)

Rose: monetary unions 2-3 fold increase in tradeCommon currency costly (OCA

theory) Improving English proficiency does

not require abandoning national languages

Large gains possible at little cost

Position of German in CEECs?