Dynamics of export share of products in the international trades Matthieu Barbier and Deok-Sun Lee Dept. Physics, Inha University Dr. Matthieu Barbier Now at Dept. Ecology & Evol. Biol., Princeton Univ. USA International trades • if there are comparative advantage of importing rather than producing a product given factors of production, politics, culture, history, etc. D. Ricardo, On the Principles of Political Economy and Taxation (London: John Murray, 1817; retrieved 2012-12-07 via Google Books) • What products are the whole world producing and exporting? • Any fundamental “laws” there? Products that Korea is producing and exporting 1962 2000 http://atlas.media.mit.edu/ Products the whole world is producing and exporting 1962 3330 Petrol.oils & crude oils obt.from bitumin.minerals 0711 Coffee,whether or not roasted or freed of caffeine 2631 Cotton (other than linters),not carded or combed 2320 Natural rubber latex; nat.rubber & sim.nat.gums 2681 Seeps or lambswool,greasy or fleece-washed 2000 3330 Petrol.oils & crude oils obt.from bitumin.minerals 7810 Passenger motor cars,for transport of pass.& good 9310 Special transactions & commod.,not class.to kind 5417 Medicaments(including veterinary medicaments) 7788 Other elect.machinery and equipment Statistics and dynamics of “Export share” of products - empirical observation - what are we interested in? NBER-UN world trade data from 1962 to 2000 • R.C. Feenstra, R. E. Lipsey, H. Deng, A. C. Ma, and H. Mo, World Trade Flows: 19622000, NBER Working Paper No. 11040 (2005) year icode importer ecode . . . 132627 1962 218400 USA 454100 132628 1962 218400 USA 454100 132629 1962 218400 USA 454100 . . . . . . exporter . . . Korea Rep. Korea Rep. Korea Rep. sitc4 unit dot . . . 8310 1 8420 1 8471 1 . . . value quantity 1 NA 564 NA 1 NA • Product ID 4-digit Standard International Trade Classification (SITC), revision 2 • Mainly based on the importers’ reports • Curated and supplemented by the available data of trades of individual countries Export share of a product p in year t • Export volume (value in dollars) of a product p in year t : ππ π‘ • Export share (relative export volume) of a product p in year t : π΄π π‘ = ππ (π‘) π′ ππ′ π‘ This quantity is what we study here • 508 products maintain non-zero volume from 1962 to 2000 • Normalization • π π΄π π‘ = 1, Mean π΄π = 1 508 = 2 × 10−3 π΄π π‘ | π = 1,2,3, … , 508, π‘ = 1962, 1963, … , 2000 Uneven distribution of export share πΆπ‘ π΄ = ππππ. a product has π΄π π π΄ ∼ π΄−πΌ π‘ ≥π΄ ππ΄2 = π΄2π − π΄π 2 SITC 3330 Crude oil: A = 0.14 Power law behavior with exponent between 2 and 3 is observed for π΄ βΏ 10−3 for all years. The functional form of the distribution does not change with time The second moment increases slightly with time with anomalous peaks after oil shock (1973) Growth or decay for 39 years • Increase or decrease? How much? • Growth rate π΄π 2000 πΊπ = log10 π΄π 1962 • Skewed distribution Coal and water gas(3415) 8.0 × 10−5 → 2.9 × 10−7 Invalid carriages (7853) 5.7 × 10−6 → 2.1 × 10−3 • What products increase share and what do the opposite in the international trades? What is the law governing it? Variation of share for one year - relatively microscopic view Loss : Δπ΄π π‘ < 0 Gain : Δπ΄π π‘ = π΄π π‘ + 1 − π΄π π‘ > 0 Δπ΄π (π‘) Δπ΄π (π‘) π΄π (π‘) π΄π (π‘) Δπ΄π π‘ β ππ π΄π π‘ πΌπ = 1.0 ± 0.06 ππ = 0.09 ± 0.02 πΌπ Δπ΄π π‘ β ππ π΄π π‘ πΌπ = 0.98 ± 0.03 ππ = 0.07 ± 0.01 Linear scaling both for gain and loss of share ! πΌπ Our viewpoint, strategy, and goal Hidalgo et al., Science (2007) • Construct a particle-hopping model consistent with the linear scaling between Δπ΄ and π΄ • Can the model explain the empirical observations such as the broad distribution of share? • Can the model predict the evolution of share of individual products? Urn model with quadratic preferential selection - Factorized steady state - Pseudo-condensation Urn model (or misanthrope process) as a model for the dynamics of export share • A complete graph of N sites, each site representing a product • A total of M particles, each particle representing the unit of share • Each site has ππ particles and {ππ |π = 1,2, … , π} specifies the system’s state (π΄π = • At each (microscopic) time step π, two sites i and j are selected with probability π’ ππ , ππ ππ ) π and a particle at site i is transferred to site j, where π’ π, π = Ππ π Ππ΄ π ∝ π2 π2 with ππ Ππ π = π π π=1,2≤ππ ≤π ππ 0 π2 π π2 (2 ≤ ππ ≤ π) ππ‘βπππ€ππ π ππ Ππ΄ π = β π2 β (1 ≤ ππ ≤ π − 1) π π 2 π π π π=1,1≤ππ ≤π−1 π 0 ππ‘βπππ€ππ π Example: N=14 M=35 Every site has at least one particle π2 = 5 π1 = 4 π4 = 2 π3 = 1 π14 = 3 π5 = 2 π6 = 1 π13 = 2 π7 = 1 π12 = 4 π11 = 3 π10 = 1 We consider the case where N is very large π9 = 2 π8 = 4 Why is it quadratic, not linear? • To be consistent with the linear relation between Δπ΄ and π΄ • The annual variation of a product’s share is the sum of plus and minus like random walk • Suppose that particle-hopping occurs Ω times for one year. Then m2π 2Ω N π2 Δππ = ππ π‘ + 1 − ππ π‘ = ππ ∼ π=1 1 or −1 • A parameter is introduced: π = Ω π2 Ω π π π2 π ~ 0.1 (empirical) ~5 • One year corresponds to π〈π2 〉 times of transfer of particles Urn model with quadratic preferential selection • Different from the zero-range process in that the particle-hop probability depends on the number of particles of the destination as well as of the source • Particle-hop probability is proportional to the square of the number of particles in the source and destination site • Each unit of share (particle) is likely to move with probability proportional to the share of the present product and that of the destination product • Our model is therefore capturing (only) the trend of a country’s economic policy towards enhancing the likelihood of profit beyond the different abundance/deficiency of factors of production from country to country. • Related works Godrèche, Bouchaud, Mezard, JPA 28, L603 (1995) – model A, B, C Godrèche and Luck, EPJB 23, 473 (2001) – zeta urn model Majumdar, Evans, Zia PRL (2005); Evans, Majumdar, and Zia, JSP 123, 357 (2006) -- condensation Barabasi and Albert, Science (1999) – (linear) preferential attachment of links Relation to empirical results • Distribution of share π π΄ corresponds to the distribution of the number π π¨=ππ΄ of particles π π = , which can be obtained analytically for the π΄ stationary state • Is the broad distribution of share caused by the linear scaling between Δπ΄ and π΄ ? • Can the model predict the trajectory of share of individual product? Urn model based on ? Factorized steady state • ππ‘ π : Probability of a specific particle configuration π at time t • πππ‘ ππ‘ = π’ ππ + 1 , ππ − 1 ππ‘ … , ππ + 1, … , ππ − 1 … − π’ ππ , ππ ππ‘ {π} π<π +π’ ππ + 1, ππ − 1 ππ‘ {… , ππ − 1, … , ππ + 1, … } − π’ ππ , ππ ππ‘ {π} • Factorized state assumed for the stationary state π∞ π ∝ with π π to be determined • Detailed balance condition π’ π + 1, π − 1 π π + 1 π π − 1 = π’ π, π π π π π • Function f: π π = π 1 π 2 π 1 π−1 π−1 π’ 2,β β=1 π’ β+1,1 =π 1 π 2 4 π 1 ππ ππ π−1 1 π2 where π β« 1 is used. π’ π, π ∝ π2 π2 Evans and Hanney, JPA 38, R195 (2005) Single-site particle-number distribution • π∞ π −1 = ππ,π π β=1 π • Partition function ππ,π = πβ π1 π2 β― ππ π β π πβ πΏ (π1 + π2 + β― + Evans and Hanney, JPA 38, R195 (2005) Partition function in our model • Logarithmic singularity of the generating function πΉ(π ) at π = 1 ∞ π=1 π πΉ π = π π π = 1 π 2 π π ∞ π=1 π2 =1+ 1−π ln 1−π − 1−π π 2 + π( 1 − π 2 ln 1 − π ) (J.E. Robinson, Phys. Rev. 83, 678 (1951)) • Partition function ππ,π = ππ 2ππ Steepest descent path π = π ∗ + ππ¦ | − ∞ < π¦ < ∞ π −π−1 ππ π π = ln πΉ π − π ln π β 2π πππ ππ ππ ππ π =π π ππ π 0 2ππ = 1−π ln 1−π − 1−π π 2 = ∞ ππ¦ ππ π =π +ππ¦ ∗ π −∞ 2π +π 1−π +π 1−π 2 ln 1 − π π ′ π = 0 at the saddle point π ∗ , which exists within the radius of convergence (otherwise, condensate is formed) π π ∗ = 1 − π −ππ 2 in case π −ππ 2 βͺ 1 with particle density π = π • ππ,π = 1 π 2π π π−ππ 2 π 2 −ππ 2 − πΊ ππ −ππ 2 π 2 ∞ πΊ π₯ ≡ π −∞ π₯ 1−π π§ ln 1−π π§ +π π§ 2π π₯ ππ§ β ππππ π‘. 1 π₯ ln π₯ π₯β«1 π₯βͺ1 Single-site particle-number distribution in our model • π π = • π→0 πΊ • πΊ with π = π 2 π ππ 2 π πππ π = π π ππ 2 π βͺ ln π π ππ π β 2ππ πΊ 2ππ π ππ β π π β 1 2 π π 2 π − π πππ 2 1 π π ππ ∼ 2ππ πΊ π 2ππ π ππ ∼ π π ∼ 1 π2 π 2 π π π π π −1 − 2π π π → ∞ π β« ln π πΊ • π πππ π 1 πΊ π πππ −1 ππ− π π = π(1) π ∼ ln π πΊ • 1 π 1 1 π 2 ππ 1 π ∼ π ln π πΊ π ππ π = ππ −ππ ln ππ −ππ π π β 1 π2 π 2 π π π ππ − + π π ln π A bump is formed for π΅ β² π < π΄ for the last two cases Condensate-free π π= =2 π π = 50, 100, 200 π 2 π ππ π= π 2 = 0.88 π = 50 ~ 0.22 (π = 200) Approximate π 1 − ππ 2 π π β 2 π π π π 2 Condensate … π π = = 10 π π = 50, 100, 200 π 2 π ππ π= π 2 = 458180 π = 50 ~ 114545 (π = 200) Approximate π π β 1 π2 π 2 π π π π −1 − π 2π π Condensation? • Nature of condensation has been studied for the (continuum) mass transfer model in 1D (Majumdar, Evans, Zia PRL (2005); Evans, Majumdar, and Zia, JSP (2006)) • If the particle-hop probability is given by π’ π, π ∝ ππ ππ , the single-site particlenumber distribution is π π ∼ π−π • If π > π, it may happen that π = π π π ∼ • If π ≤ π, π ∼ π π1−π can π π 1−π π=1 π < π even for finite π. be infinite π(π 2−π ) and can be equated to any finite π by introducing a suitable cutoff leading to exponential decay. However, if π is infinite, we should compare π and π, both of which are large numbers, and depending on the relation between π and π, a pseudo-condensate can appear • International trade dynamics is at the edge of condensation-free phase. Application - Does this model explain the international trade at the aggregate level and individual level? Yes! it does at the aggregate level Simulate the model with π = 508, π = 2 × 105 , π = 5 and the initial values from the 1962 data Share distribution Growth-rate distribution Second moment Gain versus share Δπ΄π π‘ β ππ π΄π π‘ πΌπ = 0.9 ππ = 0.05 πΌπ Loss versus share Δπ΄π π‘ β ππ π΄π π‘ πΌπ = 0.9 ππ = 0.04 πΌπ Evolution of export share of individual products • An ensemble of Κ = 300 simulation results • The middle 80% of the simulation values for π΄π π‘ is shaded Bad… Not bad…. Quantifying the typicality of empirical observations among simulated trajectories • Values of return (one-year growth rate) π π π‘ = ln simulation for each p and t. That is, we have one 1 π π • (2) π΄π π‘+1 π΄π π‘ ππππ π π (π‘) are compared between real and and K=300 simulation values (πΎ) π‘ , π π π‘ , … , π π (π‘) Normalized rank ππππ ππ π‘ = rank of π π π‘ πππππ π‘βπ π ππ π‘ 1 1 1 − ∈ − , K+1 2 2 2 • If a product p is well predicted by the model, a total of T=39 such normalized ranks at different years ππ (π‘)|π‘ ∈ [1962,2000] should be uniformly distributed over [-1/2, 1/2] • Deviation from Uniformity : i) sort T=39 values of ππ π‘ ’s in increasing order from the smallest to the largest such that π₯1 = 1 π ππ π‘1 ≤ π₯2 = ππ π‘2 ≤ β― ≤ π₯π = ππ π‘π . If they are uniform, then one would find π₯π = − + 2 π+1 for 1 ≤ π ≤ π 1 π iii) Non-uniformity or Unpredictability of a product p is defined as ππ = π=1 π₯π − π₯π π • ππ > 0.1 is observed only with probability 0.05 for 39 uniformly-distributed numbers (Marhuenda, Morales, Pardo, Statistics 39, 315 (2005)) Classifying products 1 Mean rank ππ = π 2000 π‘=1962 ππ π‘ positive: higher returns (annual growth) than expected negative: lower returns than expected π π‘=1 Rank fluctuation πππ = ππ π‘ − ππ 2 Larger than 0.29 : higher variability of rank than expected Smaller than 0.29 :lower variability 1 Unpredictability ππ = π ππ=1 π₯π − π₯π Larger than 0.1 : deviate significantly from our model prediction Smaller than 0.1 : predictable by the model Predictability and mean-rank The most unpredictable products SITC 6589 7853 6880 2239 2640 6122 2652 6545 2231 2654 3231 9610 2742 2114 2814 2235 8996 2634 8983 7931 Description Other_made-up_articles_of_textile_materials,n.e.s Invalid_cariages,motorized_or_not,parts Uranium_depleted_in_u235_&_thorium,&_their_alloys Flours_or_meals/oil_seeds/oleag.fruit_non_defatted Jute_&_other_textile_bast_fibres,nes,raw/processed Saddlery_and_harness,or_any_material_for_animals True_hemp,raw_or_processed,not_spun;tow_and_waste Fabrics,woven,of_jute_or_of_other_textile_bast_fib Copra Sisal_&_other_fibres_of_agave_family,raw_or_proce. Briouet.ovoids_&_sim.solid_fuels,of_coal_peat_lig. Coin(other_than_gold)_not_being_legal_tender Iron_pyrites,unroasted Goat_&_kid_skins,raw_(fresh,salted,dried,pickled) Roasted_iron_pyrites,whether_or_not_agglomerated Castor_oil_seeds Orthopaedic_appliances,surgical_belts_and_the_like Cotton,carded_or_combed Gramophone_records_and_sim.sound_recordings Warships_of_all_kinds Ap_1962 mean rank rank fluctuation unpredictability 9.81E-06 0.19632 0.4202 0.25825 5.69E-06 0.075288 0.48974 0.2515 1.50E-05 0.049059 0.47185 0.22591 1.57E-05 -0.016992 0.45334 0.20349 0.0021461 -0.20094 0.27152 0.20094 2.48E-05 0.18517 0.34404 0.19679 0.00015328 -0.16539 0.36012 0.19637 0.0026562 -0.19545 0.26332 0.19545 0.0023281 -0.1843 0.32578 0.19152 0.0013954 -0.18264 0.33049 0.18585 0.00032608 -0.17393 0.34473 0.18585 0.0001946 -0.049669 0.44353 0.18522 0.00046455 -0.17933 0.3278 0.1828 0.00061788 -0.10918 0.41458 0.17901 0.00023743 -0.13716 0.39188 0.17896 0.00020929 -0.15999 0.35708 0.17857 0.00023142 0.1768 0.19731 0.17825 3.19E-05 0.060561 0.42361 0.17805 0.00061524 0.17707 0.20675 0.17707 0.0003461 -0.080777 0.43397 0.17263 Raw materials and agricultural products Mostly they have their share fall behind prediction. Products with unpredictably increased share SITC 6589 6122 8983 8996 6642 8710 8821 8959 5416 8720 7741 5530 8942 223 5161 8841 8310 1110 6647 6123 Description Other_made-up_articles_of_textile_materials,n.e.s Saddlery_and_harness,or_any_material_for_animals Gramophone_records_and_sim.sound_recordings Orthopaedic_appliances,surgical_belts_and_the_like Optical_glass_and_elements_of_optical_glass Optical_instruments_and_apparatus Chemical_products_&_flashlight_materials Other_office_and_stationery_supplies Glycosides;glands_or_other_organs_&_their_extracts Medical_instruments_and_appliances Electro-medical_apparatus Perfumery,cosmetics_and_toilet_preparations Childrens_toys,indoor_games,etc. Milk_&_cream,fresh,not_concentrated_or_sweetened Ethers,alcohol_peroxides,ether_perox.,epoxides_etc Lenses,prisms,mirrors,other_optical_elements Travel_goods,handbags,brief-cases,purses,sheaths Non_alcoholic_beverages,n.e.s. Safety_glass_consisting_of_toughened/laminat.glass Parts_of_footwear Ap_1962 mean rank rank fluctuation unpredictability 9.81E-06 0.19632 0.4202 0.25825 2.48E-05 0.18517 0.34404 0.19679 0.00061524 0.17707 0.20675 0.17707 0.00023142 0.1768 0.19731 0.17825 7.58E-05 0.16225 0.29715 0.16343 0.00072931 0.1606 0.21861 0.16702 0.00012313 0.16025 0.18614 0.16258 0.00019894 0.14761 0.18156 0.16491 0.00048222 0.14543 0.20509 0.14688 0.0011123 0.14526 0.16567 0.15741 0.00023524 0.13707 0.24761 0.13707 0.0012252 0.13646 0.19143 0.14615 0.0025954 0.13611 0.22104 0.1435 0.00011694 0.13454 0.24441 0.13454 0.00031245 0.13332 0.20459 0.13814 0.00034402 0.13254 0.21238 0.14809 0.0010566 0.13254 0.18475 0.14231 0.00020163 0.13114 0.23304 0.13114 0.00019306 0.12949 0.24531 0.1322 0.00023262 0.12949 0.23223 0.1318 Medical appliances, toys, cosmetics They are not exclusively subject to economic demands Products with the largest fluctuation of rank SITC 7853 6880 2239 9610 7931 7911 2634 7913 6589 2714 6724 2860 2114 7914 7915 7187 7933 451 541 4235 Description Invalid_cariages,motorized_or_not,parts Uranium_depleted_in_u235_&_thorium,&_their_alloys Flours_or_meals/oil_seeds/oleag.fruit_non_defatted Coin(other_than_gold)_not_being_legal_tender Warships_of_all_kinds Rail_locomotives,electric Cotton,carded_or_combed Railway_&_tramway_coaches,vans,trucks_etc. Other_made-up_articles_of_textile_materials,n.e.s Potassium_salts,natural,crude Puddled_bars_and_pilings;ingots,blocks,lumps_etc. Ores_and_concentrates_of_uranium_and_thorium Goat_&_kid_skins,raw_(fresh,salted,dried,pickled) Railway_&_tramway_passenger_coaches_&_luggage_van Rail&tramway_freight_and_maintenance_cars Nuclear_reactors_and_parts Ships,boats_and_other_vessels_for_breaking_up Rye,unmilled Potatoes Olive_oil Ap_1962 mean rank rank fluctuation unpredictability 5.69E-06 0.075288 0.48974 0.2515 1.50E-05 0.049059 0.47185 0.22591 1.57E-05 -0.016992 0.45334 0.20349 0.0001946 -0.049669 0.44353 0.18522 0.0003461 -0.080777 0.43397 0.17263 0.00026424 -0.012199 0.4247 0.15095 3.19E-05 0.060561 0.42361 0.17805 0.00023165 0.045399 0.4203 0.15411 9.81E-06 0.19632 0.4202 0.25825 0.00030787 -0.046532 0.41845 0.15565 0.00075701 -0.039735 0.4175 0.14973 0.0026274 -0.0030498 0.41522 0.13535 0.00061788 -0.10918 0.41458 0.17901 0.00048688 -0.059167 0.41353 0.14772 0.00070004 -0.061258 0.41318 0.16395 3.72E-05 0.047839 0.40814 0.15278 0.00027311 -0.023789 0.40432 0.12718 0.00060056 -0.076246 0.40347 0.14527 0.0024874 -0.030063 0.40127 0.12806 0.0012997 0.016556 0.39846 0.13502 Railways, warships, uranium, Nuclear reactors Offer and demand are highly variable in time and historically determined Summary and Discussion • • • • • • • Time-evolution of the market share of products in the world trade has been studied by data analysis and model study Urn model with quadratic preferential selection reproduces linear scaling of annual gain and loss of share and the power-law distribution of share with exponent 2 The model represents the pressure of directing a country’s investment towards more popular products in the global economy The quadratic preferential selection leads the world trade market to the edge of condensation The condition for the emergence of pseudo condensate has been found. The model explains the empirical observations very successfully at the aggregate level The share trajectory of more than 60% products are predicted by the model capturing the pressure towards enhancing the likelihood of profit. • Nature of unpredictable products provides the reason of deviation from model prediction • For more realistic and predictive model, one should consider the network structure of product space– hopping in product space does not happen randomly but depending on the proximity of two products.