The Effectiveness of International Trade Boycotts

Transcription

1 The Effectiveness of International Trade Boycotts Kilian Heilmann University of California San Diego June 2, 2014

2 Introduction Import boycotts against a whole nation (through embargoes or consumer boycotts) International trade boycotts are not a new phenomenon. They have been repeatedly used by countries to punish or coerce specific behavior of trading partners. Trade has surged over the last decades, potentially increasing the importance of trade boycotts However, the nature of trade has also changed. It has evolved from an exchange in final goods to a system of internationally integrated production of goods. This might make the disruption of trade more costly to the boycotting country

3 Research Question Do boycotts significantly disrupt trade at all? What is the dynamic structure of a trade boycott? Do boycotts have a temporary or long-lasting effect? Are there catch-up effects? Simple change in timing, but not in levels? Which products are affected? Are boycotts mainly consumer-driven and affect only well-known foreign brands? Are non-branded intermediate goods affected at all?

4 Literature Overview David & Meunier (2011) cannot find any significant effect of conflict on trade between US and France or China and Japan for period Pandya and Venkatesan (2013) and Clerides et al. (2012) find a drop in sales of American products in the Middle East and French products in the US during the Iraq War Fuchs & Klann (2013) find evidence that China is punishing trading partners who receive the Dalai Lama by reducing imports from them

5 Contribution New method to identify the effect of boycotts on trade (synthetic control group method) Monthly dataset (as compared to annual or quarterly data) Boycotts that have never been studied before, that are caused by factors unrelated to trade and can therefore be used as natural experiments to identify the impact of trade boycotts

6 Background Case Study 1: Mohammad Comics Crisis On September 30, 2005, the Danish newspaper Jyllands-Posten publishes a set of comics on Islamic prophet Muhammad. This caused protests by Muslims first in Denmark and then worldwide that partly turned violent and lead to several deaths. In January 2006, a consumer boycott of Danish products was organized in Muslim religious leaders in Saudi Arabia.

7 Background Case Study 2: Senkaku/Diaoyu Island Conflict Long-lasting conflict about sovereignty of uninhabited islets. Japan, Mainland China, and Taiwan all claim them as their national territory, but Japan effectively controls them Tensions rise in the 2000s as private activists (without government support) from China, Taiwan and Hong Kong repeatedly occupy the major island to express Chinese claims. In September 2012, the Japanese government announces its plan to buy the islands from their private owners leading to violent protests in many Chinese cities, violence against Japanese stores, and eventually a consumer boycott against Japanese products.

8 Data Muhammad Comic Crisis Monthly observations for Danish trade values (price times quantity) from SITC2 (67 industries) and SITC5 (3166 products) level 257 partner countries Source: Statistics Denmark Senkaku/Diaoyu Island Conflict Monthly observations for Japanese trade values from 1/2010-9/2013 (32 pre- and 13 post-boycott observations) All Harmonized System (HS) product categories (about 5,300 products) Source: Monthly Comtrade (still in beta phase) Data on bilateral distance, GDP from World Bank Indicators and CEFII

9 Features of the Data Highly volatile from month to month Strong seasonality Changing time trends

10 Construction of the Counterfactuals How to construct the counterfactual? Due to high volatility exploiting within variation does not seem to be feasible Using between variation (between countries) might get rid of time trends if they are common for different countries. Problem: Many potential control groups Synthetic Control Group Method constructs counterfactual by a weighted average of the control countries such that the synthetic control group replicates the boycotting country before the boycott

11 Synthetic Control Group For each country j, imports from the boycotted country in period t are determined by the factor model log Y j,t = δ t + θ t X j }{{} + λ t µ j }{{} +β t boycott j,t + ɛ j,t (1) observed unobserved δ t : Common time trend X j : Country-specific observed factors (e.g. bilateral distance, GDP) µ j : Country-specific unobserved factors (e.g. industry composition, non-tariff barriers) θ t, λ t : Time-specific factor loadings boycott j,t : Boycott indicator variable β t : Parameter of interest ɛ j,t : Idiosyncratic shock with zero mean

12 Synthetic Control Group Ideal experiment: Compare two countries that have exactly the same characteristics X and µ and only differ by carrying out the boycott. Infeasible because µ is not observed Synthetic Control Group: Match instead on observable characteristics and pre-treatment outcomes Abadie et al. (2010) show that for a large number of pre-treatment periods, the synthetic control group can only fit the treatment outcomes well if it also fits the unobserved characteristics

13 Example: Saudi Arabia

14 Problems General Equilibrium effects: Control countries might take up some of the trade that the boycotting countries refuse, thus the control group is not independent of the treatment. This will overestimate the treatment effect. We can interpret the estimates as an upper bound to the impact of boycotts on trade. Overfitting: If the pool of control countries is very large compared to the pre-treatment periods, the fit will be mechanically high. Reduce control pool to countries similar Works best if treatment country is in the convex hull of the controls, that is the boycotting countries should not be at either end of the distribution for characteristics or outcomes. (Restrict weight to be between 0 and 1 to avoid out extrapolation bias) Statistical inference: No inference worked out, instead relies on placebo tests.

15 An Idea for Inference In practice, the pre-treatment fit between the treatment and the synthetic control will not be perfect, but subject to a prediction error. We can calculate the correlation coefficient and the root mean squared prediction error in the pre-treatment period to asses the goodness of fit. If we think of the prediction error as a stable process with mean zero, we can construct standard errors for the post-treatment estimates.

16 Case Study 1: Muhammad Comics Conflict There are many Muslim countries. How to choose treatment group? Idea: Choose by Muslim population as share of the total population Treatment group: more than 75% Muslim population (34 treatment countries) Control group: less than 10% Muslim population (100 control countries) Exclude all other countries with intermediate Muslim population Muslim world makes up only small share of Danish exports (2.6% of total exports in 2004), with Saudi Arabia being the biggest export partner (0.48%)

17 Table: Estimated Trade Disruption Country 3 Months 12 Months 24 Months in DKK in % in DKK in % in DKK in % Egypt -123, % -280, % -453, % (48,952) (17.4%) (97,905) (10.2%) (138,458) (6.3%) Indonesia 39, % 28, % -13, % (14,580) (11.5%) (29,160) (5.3%) (41,239) (3.6%) Iran -17, % -378, % -675, % (54,829) (20.2%) (109,658) (10.0%) (155,081) (7.2%) Kuwait -30, % -293, % -659, % (21,764) (14.4%) (43,527) (6.9%) (61,557) (4.6%) Libya -44, % -231, % -313, % (18,427) (25.6%) (36,855) (12.8%) (52,121) (9.3%) Maldives 10, % 16, % 31, % (5,461) (177.6%) (10,923) (105.4%) (15,447) (71.0%) Pakistan 123, % 167, % 264, % (21,538) (22.8%) (43,076) (12.4%) (60,918) (9.3%) Saudi Arabia -222, % -1,206, % -2,026, % (38,332) (5.4%) (76,663) (2.7%) (108,418) (1.8%) Syria -7, % -59, % -59, % (8,698) (16.5%) (17,395) (8.4%) (24,601) (6.0%) Tunesia -7, % -16, % 71, % (10,761) (30.3%) (21,522) (15.0%) (30,437) (10.4%) Turkey -65, % -89, % -2, % (55,172) (8.8%) (110,344) (4.4%) (156,051) (2.9%) Total -510, % -2,858, % -4,277, % (125,131) (3.6%) (250,262) (1.8%) (353,924) (1.3%) in thousands of DKK. Prediction errors in parentheses.

18 Spatial Distribution Legend Estimated Treatment Effect less than -55% -40 to -55% -20 to -40% -15 to -20% -5 to -15% -5 to 5% 5 to 10% 10 to 30% 30-50% 50-99% Controls Excluded No Data >100%

19 Treatment over Time x 106 Estimated Cumulative Treatment 95% Confidence Interval in billion DKK Boycott officially announced Months since Boycott

20 Validation 12.8 Log Total Danish Exports to Saudi Arabia Estimation Period Months since Boycott [ Validation Period ] Actual Treatment

21 Validation Randomly assigned treatment periods should not be significant I assign the treatment time three years prior to the actual treatment and calculate the estimated trade loss for the following three months Estimated Treatment

22 Identifying Consumer Products UN Statistics Division classifies products into Broad Economic Categories (BEC) that allow us to identify consumer, intermediate, and capital goods. (I recoded about 1200 products by my own judgment) Merge all treatment countries into one treatment country, split up the export values into these three categories and redo the estimation for each group Composition of pre-boycott trade: 28.3% consumer goods, 45.3% intermediate goods, 25.3% capital goods, and 1% others (low value shipments, donations, special transactions)

23 Log Consumer Exports Realized Consumer Synthetic Consumer Boycott Comics Log Intermediate Exports Realized Intermediate Synthetic Intermediate Boycott Comics Log Capital Exports Realized Capital Synthetic Capital Boycott Comics Months since Boycott Log Other Exports Realized Others Synthetic Others Boycott 14.5 Comics Months since Boycott

24 Table: Results by Product Type Country 3 Month 12 Month 24 Month in DKK in % in DKK in % in DKK in % Consumer 9, % -736, % -1,352, % (57,802) (8.7%) (115,604) (4.3%) (163,489) (3.0%) Intermed. -110, % -537, % -164, % (55,469) (4.5%) (110,939) (2.3%) (156,891) (1.6%) Capital -92, % -303, % -86, % (79,699) (11.5%) (159,398) (6.3%) (225,423) (4.4%) Others -11, % 93, % 150, % (3,844) (11.7%) (7,687) (4.4%) (10,871) (3.1%) in million DKK Prediction errors in parentheses

25 Conclusion Strong heterogeneity of the impact of boycotts on trade between different Muslim countries Estimated trade loss of 758 million USD (-15.1%). Relatively small when compared to overall Danish exports during the same period of 185 billion USD (-0.4%). Punishment effect likely not to be effective Trade disruption concentrated in consumer goods. Small, but temporary drop in intermediates. No significant effect on capital goods.

26 Case Study 2: Senkaku/Diaoyu Island Conflict China is the biggest export market for Japan (ahead of US). It is also geographically really close. This creates problems for the synthetic control group method since China is not in the convex hull of the controls, but at one end of the distribution Need to relax the restriction on weights to lie between 0 and 1 (allow extrapolation)

27 Pre-Treatment Fit - China 23.5 Realized China Synthetic China Log Japanese Exports Boycott Months since Boycott

28 Results - China Table: Estimated Trade Disruption in bn. USD in % 3 Months % (0.91) (2.5%) 6 Months % (1.29) (1.8%) 12 Months % (1.82) (1.4%) Prediction errors in parentheses

29 Treatment over Time - China 2 x Estimated Cumulative Treatment 95% Confidence Interval Loss in Trade (in bn USD) Boycott announced Months after the Boycott

30 How to Identify Consumer Products? Instead of applying BEC classification, I make use of publications of the boycott movement itself Collect most famous Japanese brands and determine which product categories they are in and repeat the analysis for these categories

31 Table: Japanese Brands Automotive Acura Bridgestone Daihatsu Denso Hino Honda Infiniti Isuzu Kawasaki Kubota Lexus Mazda Mitsubishi Nissan Subaru Suzuki Toyota Yokohama Electronics audio technica Brother Canon Capcom Casio Epson Fujifilm Fujitsu Hitachi JVC Kenwood Konami Konica Korg Kyocera Mitsumi NEC Nikon Nintendo OKI Olympus Panasonic Pentax Pioneer Ricoh Sansui Sanyo Sega Sharp Sony TDK Toshiba Vaio Yamaha Food Asahi Biore BOSS Glico Kewpie Kikkoman Kirin Meiji Nissin Pocari Sweat Suntory UCC Coffee UHA Yakult Yoshinoya Cosmetics DHC Dongyangzhihua Huawang Kanebo Kao Kose Shiseido Shu Uemura SK-II Sofina Clothing Asics Kenzo Uniqlo Other ANA Bandai Butterfly Citizen Daikin INAX Kato Komatsu Mild Seven Mitsukoshi Mizuno Muji Nippon Noritz Omron Orient Rinnai Roland Seiko Sogo Sumitomo Tadano Toray Toto Yasaka Yonex Zojirushi

32 Top Japanese Brands and their Product Code 22: Beverages, spirits & vinegar 1902: Pasta, prepared or not, couscous, prepared or not 3304: Beauty, make-up & skin-care prep, manicure etc 8508 Electromechanical tools, working in hand, parts 8521 Video recording or reproducing apparatus 8703: Motor cars & vehicles for transporting persons 9006: Photographic still cameras, flash apparatus etc

33 Table: Estimated one year trade disruption by HS code HS Code in USD in % HS Code in USD in % % % (5.330) (10.9%) (4.996) (28.2%) % , % (0.143) (32.5%) ( ) (3.8%) % % (6.754) (3.6%) (1.783) (10.1%) % (0.786) (29.8%) In million USD. Prediction errors in parentheses

34 Car Exports to China (HS 8703) 3.05 Realized Mainland China Synthetic Mainland China Boycott Log Car Imports China Japanese Earthquake Time since Boycott

35 Car Exports to Hong Kong and Macau (HS 8703) Log Car Imports Hong Kong Realized Hong Kong Synthetic Hong Kong 2.84 Boycott Time since Boycott Log Car Imports Macau Realized Macao Synthetic Macao Boycott Time since Boycott

36 Conclusion - China Sharp disruption of trade in the range of 5.8 billion USD within a single year, yet still rather small when compared to overall Japanese exports to China and total Japanese exports in the same period (0.8% of 732 bn USD) Slight evidence for a catch-up effect after about seven months Disruption concentrated in passenger cars. Negative effects on other consumer goods, but much lower absolute decline.

37 Conclusion International trade boycotts are disruptive to trade, but they do not halt trade completely Strong heterogeneity among boycotting countries with estimates between 4% and 15% of total trade disruption. Boycotts seem to be a one-time reduction of trade and trade levels seem to revert to pre-boycott levels after about 7-15 months. (Slight evidence for a catch-up effect in the case of China) Consumer goods and especially highly branded signature export goods show the largest relative decline suggesting that boycotts are mainly consumer-driven. Short-term decline in intermediate goods trade however suggests that there are other mechanisms of boycotts at work.

38 Thank you

39 Synthetic Control Group For each country j, imports from the boycotted country in period t are determined by the factor model log Y j,t = δ t + θ t X j }{{} + λ t µ j }{{} +β t boycott j,t + ɛ j,t (2) observed unobserved δ t : Common time trend X j : Country-specific observed factors (e.g. bilateral distance, GDP) µ j : Country-specific unobserved factors (e.g. industry composition, non-tariff barriers) θ t, λ t : Time-specific factor loadings boycott j,t : Boycott indicator variable β t : Parameter of interest ɛ j,t : Idiosyncratic shock with zero mean

40 Synthetic Control Group Let w.l.o.g. j = 1 be the treatment unit and j = 2... J + 1 the control units Let T 0 be the number of pre-treatment periods Ideal experiment: Compare two countries that have exactly the same characteristics X and µ and only differ by carrying out the boycott. Infeasible because µ is not observed Synthetic Control Group: Match instead on observable characteristics and outcomes

41 Synthetic Control Group Suppose that there exists a vector of weights w such that J+1 wj X j = X 1 (3) j=2 J+1 wj Y j,t = Y 1,t t = 1... T 0, (4) j=2 with the restrictions J+1 wj = 1 (weights sum up to one) j=2 0 < w j < 1 j = 2... J + 1 (no extrapolation)

42 Synthetic Control Group The model in (2) implies that for the synthetic control group it holds that J+1 J+1 J+1 J+1 wj Y j,t = δ t + θ t wj X j + λ t wj µ j + wj ɛ j,t j=2 j=2 and that the difference between the actual treatment unit and the synthetic control group in the pre-treatment period t = 1... T 0 is j=2 j=2 J+1 Y 1,t w J+1 j Y j,t = θ t X 1 w J+1 j X j +λ t µ 1 w J+1 j µ j + w ( ) j ɛ1,t ɛ j,t j=2 j=2 j=2 j=2 } {{ } } {{ } = 0 = 0

43 Synthetic Control Group Rearranging, summing over all time periods, and dividing by T 0 yields J+1 µ 1 wj µ j 1 T 0 J+1 1 T 0 λ t = (ɛ 1,t ɛ j,t ) T 0 j=2 t=1 wj T j=2 0 t=1 The right hand side goes to zero as T 0 gets large. This implies that for a long pre-boycott period, the synthetic control group can only fit the outcomes if it also fits the unobserved factors. This suggests to use the simple difference between realized and counterfactual value as the estimate for the treatment effect for each period: J+1 β t = Y 1,t wj Y j,t. j=2

44 Synthetic Control Group In practice however, one will not be able to find weights such that equations (2) and (3) hold exactly. In this case the weights are chosen such that the equations hold approximately. Define the column vector Z j = (Y j,1, Y j,2... Y j,t0, X j ) that stacks all export values Y j for the pre-treatment period 1... T 0 and the known characteristics X j of country j. Similarly, define the matrix Z C = [Z 2 Z 3... Z J+1 ] that collects these column vectors for all the potential control countries. The (J 1) vector w is then the solution to the following minimization problem given a weighting matrix V w = arg min Z 1 Z C w = arg min (Z1 Z C w) V (Z 1 Z C w) w w (5)