Microsoft Word - 24-2311_s1 Engineering, Technology & Applied Science Research Vol. 8, No. 5, 2018, 3432-3438 3432 www.etasr.com Sun et al.: Analysis of Bilateral Trade Flow and Machine Learning Algorithms for GDP Forecasting Analysis of Bilateral Trade Flow and Machine Learning Algorithms for GDP Forecasting Jingwen Sun School of International and Public Affairs, Columbia University, New York, USA jingwensun321@gmail.com Yuan Suo School of Social Sciences, University of California, Irvine, USA suoyuandl@163.com Seeha Park Hankuk Academy of Foreign Studies Yongin, South Korea seehalemon@gmail.com Tianze Xu College of Liberal Arts and Sciences, University of Connecticut, Storrs, USA constans_x@163.com Yizhu Liu Pius XI Catholic High School, Milwaukee, USA yizhuliupius@yeah.net Weiqi Wang Lawrence Woodmere Academy, New York, USA weiqiwanglwa@163.com Abstract—The terms imports and exports describe goods and services traded between countries. Countries import goods they cannot produce domestically or can obtain at a lower cost from another country. According to the World Trade Organization (WTO) reports, the U.S. is the world’s largest importer based on capital investment, followed by the E.U., China, Germany, and Japan. For exports, China leads the world with an official trade amount of $1.904 trillion in 2013. E.U. ranks second, followed by U.S., Germany, and Japan. Trade in goods and services is defined as a change in ownership of material resources and services between economies. Trade indicators include the sale of goods and services as well as barter transactions or goods exchanged and are measured in million USD, the percentage of GDP for net trade, and the annual export and import growth. This study analyzes imports and exports of all countries for the 1960-2017 period and evaluates the correlations in trade statistics to predict future imports and exports. Since the GDP for any country depends mainly on trade, this study examines trade data and compares various machine learning algorithms to forecast a country’s GDP. Keywords-imports; exports; GDP; trade statistics; GDP forecast I. INTRODUCTION Trade data include exports, imports, and trade balance. In general, international trade represents the economic activity of a country related to some economic relationship with another country. A series of such activities forms a country’s trade balance. The trade volume of a country indicates the country’s collective effect of macroeconomic policies. Analyzing this effect from international trade policies requires data on the country’s exports and imports and the long-run equilibrium relationship between these two variables. Here, there always is the effect of time lag on trade volume since any change in a country’s import/export demand does not happen quickly, and therefore, trade data require a deep analysis [1]. One factor in explaining trade remains the comparative advantage. However, some new factors such as consumer preferences, advantages in the economy of scale, and the use of global production have emerged, and such factors may alter international trade patterns. In addition, bilateral trade is important in understanding the consequences of international trade [2]. Economies that are oriented mainly toward trade must carefully analyze their bilateral trade data since international trade flow and its direction, composition, and linkages require bilateral as well as multilateral analyses. Therefore, data should always be carefully examined in any empirical analysis [3]. Several studies have analyzed exports, imports, the economy, and their relationships, including cointegration between exports and imports [4, 5], long-run trade elasticity in less developed countries [6], cointegration in specific groups of countries [7], and related technical issues [8]. Some studies have investigated the importance of exports and imports in forming the economy and improving the quality of life [9]. There are various methods for determining the importance of exports and imports in the economy and analyzing their impacts on economic growth. One uses simple (and sometimes multiple) regression between these three variables and other factors, and another employs causality, determining the factors and causes as well as inverse causality. Some studies have used VAR and VEC models to exclude this causality problem. A cointegration test has verified a long-run equilibrium relationship between GDP, exports, and imports in Tunisia [10], finding these two variables to influence economic growth and thus highlighting their importance. A study of the Nigerian economy has suggested economic growth to be determined by exports, imports, labor, and the exchange rate and posited their cointegration, also showing positive relationships of exports and labor to economic growth and negative effects of imports and the exchange rate on the economy [11]. A study of the Ethiopian economy has revealed that the growth rate of real exports has a positive relationship with the rate of economic growth, analyzing that, although the effect is non-significant in the short run, there is a strong long-run relationship between Engineering, Technology & Applied Science Research Vol. 8, No. 5, 2018, 3432-3438 3433 www.etasr.com Sun et al.: Analysis of Bilateral Trade Flow and Machine Learning Algorithms for GDP Forecasting these two variables [12]. These studies show that exports and imports have great impacts on economic growth. Economic growth is typically represented with GDP, GDP per capita, trade balance, consumption, investment, and government spending. In addition to the impact on economic growth, exports and imports are important for other reasons. An example is the model of innovation and trade, which predicts positive shocks in exports can facilitate more productive and innovative firms. Here the accompanying rent from the innovation effort of companies increases with the firm’s market size, known as the market size effect [13]. A country’s GDP is a primary indicator of the country’s economy. GDP represents the worth (in dollars) of all goods and services produced and imported/exported over a period of time. This is measured relative to previous GDP. An increase in GDP shows the growth of a country’s economy, and this figure is considered a key benchmark for the country’s economy. The U.S. has the largest GDP in the world, followed by China, Japan, Germany, the U.K., France, and India. In most cases GDP measurement is complicated. In simple terms, GDP is the income earned (gross profit) and expenditure spent by a country, and since it is considered relative to countries, exports and imports play a vital role in deciding a country’s GDP. GDP can be calculated as the total spending on all final goods and services (Consumption of goods and services (C) + Gross Investments (I) + Government Purchases (G) + (Exports (X) - Imports (M)), i.e. GDP=C+I+G+(X-M). II. WORLD TRADE DATA STATISTICS Trade statistics represent a unique dataset for modern economies. Trade data statistics can indicate economic geography and provide insights into economic development and globalization. International trade plays an important role in a wage-based economy. International trade data provide insights into global relationships between countries as well as information on the relationship between local and regional economies [14]. Statistical data confirm that trade has continued to support economic growth and development, helping to reduce poverty around the world. World merchandise exports have increased in value by about 32% since 2006, reaching USD 16 trillion in 2016. At the same time, world exports of commercial services have accelerated by about 64%, reaching a total of USD 4.77 trillion [15]. The importance of world trade data statistics is clear, but its usefulness depends on various factors such as data availability, analysis requirements, and data accuracy. Therefore, measuring trade statistics must adhere to the Standard International Trade Classification [16]. This is important since insufficient data or analysis may cause misinterpretation of variables. According to the World Fact book, the world’s top exporter is China with the total export of $2.157 trillion in 2017, followed by the E.U. with $1.929 trillion in 2016 and the U.S. with $1.576 trillion in 2017 [17]. For top import countries, the World Fact book ranks U.S. as the leader with $2.352 trillion of total imports in 2017, followed by E.U. with $1.895 trillion in 2016 and China with $1.731 trillion in 2017 [18]. Germany and Japan followed. The total import for Germany was $1.104 trillion in 2017, and Japan was $625.7 billion in 2017. For exports, Germany exported $1.401 trillion in 2017, and Japan, $683.3 in 2017 [17, 18]. The top 18 world imports and exports can be seen in [19]. Cars top the list, with Germany as the largest exporter and the U.S. as the top importer. Refined petroleum follows, with the U.S as a sole leader in both exports and imports. The list also includes goods such as pharmaceuticals, gold, crude petroleum, telephone, broadcasting equipment, diamonds, petroleum gas, and aircrafts [20]. III. LITERATURE REVIEW Authors in [23] proposed a machine learning model for predicting agricultural commodity prices over one-, two-, and three-month periods ahead. They used the multivariate relevance vector machine based on Bayesian learning for regression and compared the performance of the MVRVM model to that of multiple-output artificial neural networks. Authors in [24] applied data mining to detect relationship patterns in customs administration data with market prices and current exchange rates in Ethiopia and discovered association rules to generate note worthy import/export patterns. They used datasets from the Ethiopian Revenue and Customs Authority, Central Statistics Agency, and the National Bank of Ethiopia and applied the WEKA tool for data analysis purposes, and the results verified that imported textile was significantly related to the market price and the currency exchange rate. They also concluded the Apriori algorithm as the fastest one in discovering association rules. Authors in [25] predicted bilateral trade flow, an important economic indicator, by using the gravity model of trade with a fully connected feed-forward neural network. They experimented with machine learning models by varying hidden layers and neurons in each hidden layer and found that fully connected feed-forward neural networks can improve the gravity model’s prediction performance. They also proposed that the LSTM model may yield better results than fully connected neural networks for time series data. Author in [26] used machine learning and data mining techniques for publicly available commodity data and forecasted country GDP, finding a correlation between export- import data and GDP. He considered commodity trade and GDP as inputs to the algorithm and designed a model to predict GDP for another day with the given commodity trade for the new day. He used a multi-class support vector machine with a genetic algorithm based on a fuzzy set and artificial neural network to predict GDP. Authors in [27] implemented a back- propagation neural network based on a genetic algorithm for port throughput forecasting. They used a 12-year dataset such that 11 years of data was used for simulations/training and the final year was used for forecasting. They verified the proposed hybrid model, the GA-BP forecasting model, to show better accuracy but it took longer to converge. Authors in [28] proposed a new machine learning approach for price modeling using a neural network with an advanced signal-processing tool. They used the proposed model to forecast prices of commodities such as coal, crude oil, and electricity and employed a mixture of a Gaussian neural network, showing significant improvements relative to other popular models. Authors in [29] constructed a GARCH model using an artificial neural network and evaluated its ability to forecast stock market volatility. They compared the Engineering, Technology & Applied Science Research Vol. 8, No. 5, 2018, 3432-3438 3434 www.etasr.com Sun et al.: Analysis of Bilateral Trade Flow and Machine Learning Algorithms for GDP Forecasting performance of their model to that of other popular volatility models for various international stock indices. Authors in [30] proposed a model to predict future gold rates based on 22 market variables using machine learning techniques. They collected data from various online sources and implemented a linear regression model using artificial neural networks with the rapid miner tool. Authors in [31] proposed a unified modeling framework to justify the empirical regularity in the international trade network and analyzed the international trade network each year with exports of the country with other countries. They constructed a basic model with a directed weighted network for unified modeling. IV. MACHINE LEARNING IN GLOBAL TRADE DATA Machine learning applies artificial intelligence (AI) to automatically learn and improve from experience without being specifically programmed. Machine learning focuses on the development of computer programs for accessing and learning from data. The process of learning begins with observations or data, including examples, direct experience, and instructions, to identify certain patterns in data for making better decisions. The main aim is to allow the computer to learn automatically without human intervention and engage in appropriate actions. AI and machine learning represent a good way for financial institutions to optimize margin valuation adjustment (MVA), which can be performed through the assistance of machine learning by reducing margins for derivatives through a combination of “executing pairs of offsetting derivative trades” and “executing offsetting strategies with the same dealer” [21]. By choosing the best combination of the initial margin, machine learning reduces trades in a given period of time, and the basis for this is the degree of initial margin reduction in the past from various combinations of trades [20]. Financial institutions emphasize cost-effective means for regulatory requirements, such as efficient trade execution, data reporting, and prudent regulation. In this regard, AI can be used to obtain information and process orders, and machine learning can help create “trading robots” that can respond quickly to market changes. Therefore, such innovations can be used by firms to estimate financial impacts more accurately and minimize trading costs [20]. V. METHODOLOGY A. Datasets The data for this research study is gathered from [32]. The data consisted of yearly import/export data from 217 countries for the 1960-2017 period. Data were collected from the BoP (Balance of Payments) statistics yearbook and presented in USD. Several missing data points were identified as a result of no substantial import/export participation by some countries during early years. Therefore, the data for the last 43 years were considered in this study (i.e., 1975-2017). The data containing the top 10 countries in imports and exports were extracted from the primary data source for the analysis. The data (in billion USD) were used for further analysis. Tables I and II present the import and export data, respectively, for the study period. Datasets show the U.S. as leading both imports and exports from 1975 to 2017. Therefore, it is evident that the U.S. had the highest GDP and the GDP was growing at a constant pace. For better understanding, the import and export datasets are plotted in Figures 1 and 2 respectively. Fig. 1. Increase in imports. TABLE I. IMPORTS (IN BILLON $) FOR 1975-2017 Year Canada Colombia Germany UK Israel India Netherlands Sweden US S. Africa 1975 42 3 94 64 7 7 40 21 120 12 1976 47 3 108 66 7 6 44 23 149 11 1977 49 3 123 74 8 7 51 24 180 10 1978 53 4 149 88 9 9 60 25 208 12 1979 63 4 193 115 11 12 76 35 248 15 1980 71 6 222 134 12 17 88 40 291 23 1981 79 7 195 122 13 18 76 36 310 26 1982 67 7 186 119 12 18 72 34 299 21 1983 74 6 181 118 12 18 70 32 323 18 1984 88 6 178 123 12 18 70 33 399 19 1985 93 6 184 128 12 19 73 35 410 13 1986 99 6 226 148 13 20 86 41 449 15 1987 107 6 271 183 17 23 104 51 501 18 1988 128 7 298 222 18 26 114 58 546 21 1989 141 7 319 233 18 29 119 62 580 21 1990 148 7 412 264 21 30 142 71 616 22 1991 152 7 452 251 23 28 146 66 610 22 1992 157 9 485 267 24 30 157 68 656 23 1993 168 12 420 255 27 31 144 56 713 24 1994 183 14 464 284 31 38 159 66 802 27 1995 199 16 559 327 36 49 210 81 891 34 1996 209 17 553 355 38 55 211 85 956 34 Engineering, Technology & Applied Science Research Vol. 8, No. 5, 2018, 3432-3438 3435 www.etasr.com Sun et al.: Analysis of Bilateral Trade Flow and Machine Learning Algorithms for GDP Forecasting Year Canada Colombia Germany UK Israel India Netherlands Sweden US S. Africa 1997 237 19 537 380 38 59 206 85 1040 35 1998 241 18 565 396 36 60 217 90 1100 33 1999 259 14 579 419 41 63 223 91 1230 31 2000 287 15 595 440 47 74 231 97 1450 34 2001 268 17 587 439 44 72 231 87 1370 32 2002 271 16 589 471 43 76 243 91 1400 33 2003 295 17 726 529 45 93 286 113 1510 44 2004 337 21 858 625 53 131 363 134 1770 59 2005 385 26 934 686 59 182 392 149 2000 69 2006 430 31 1080 784 63 225 441 169 2220 84 2007 471 38 1250 841 75 279 522 204 2360 98 2008 508 46 1410 867 85 379 595 226 2550 108 2009 412 40 1130 677 64 328 486 167 1970 83 2010 500 49 1270 753 77 439 532 197 2350 103 2011 568 64 1500 840 93 553 615 233 2680 123 2012 587 70 1410 844 93 580 600 221 2760 124 2013 586 71 1480 869 92 560 618 224 2760 122 2014 585 76 1510 915 95 554 632 231 2870 116 2015 531 65 1310 840 85 492 552 200 2760 100 2016 513 55 1330 804 90 472 555 202 2710 90 2017 548 57 1460 838 97 561 619 222 2900 100 TABLE II. EXPORTS (IN BILLON $) FOR 1975-2017 Year Canada Colombia Germany UK Israel India Netherlands Sweden US S. Africa 1975 39 3 103 60 4 6 43 21 130 11 1976 45 3 119 63 5 7 47 22 142 10 1977 48 4 137 76 6 8 52 23 152 12 1978 54 4 166 92 7 9 60 27 178 15 1979 64 5 198 117 8 10 75 34 223 20 1980 75 6 217 146 9 12 87 39 272 29 1981 81 5 200 136 9 12 81 36 294 24 1982 80 5 201 128 9 13 78 34 275 21 1983 86 4 192 121 9 14 75 34 266 21 1984 100 6 193 122 10 14 76 36 291 20 1985 101 5 207 132 10 13 79 37 289 19 1986 102 7 272 144 12 14 92 44 310 20 1987 112 7 326 175 14 16 109 53 349 26 1988 132 7 358 191 15 18 122 60 431 27 1989 142 8 379 199 16 21 128 63 487 26 1990 149 9 457 239 18 23 153 71 535 28 1991 149 10 443 240 17 24 157 70 578 27 1992 155 10 473 254 20 25 169 72 617 28 1993 168 10 421 246 21 28 160 62 643 30 1994 189 11 468 277 24 32 178 74 703 31 1995 218 13 571 322 28 39 235 96 794 35 1996 233 14 573 352 30 41 236 101 852 36 1997 249 15 563 383 32 45 231 101 934 37 1998 253 14 594 384 33 46 240 105 933 35 1999 283 14 594 393 38 52 242 106 970 34 2000 328 16 601 409 47 60 247 112 1080 37 2001 310 16 622 401 41 63 248 105 1010 36 2002 304 15 680 421 40 71 260 107 979 37 2003 329 16 819 480 44 85 322 131 1020 48 2004 382 20 1000 563 54 116 410 161 1160 59 2005 431 25 1080 622 59 155 446 174 1290 69 2006 465 29 1240 719 63 193 499 199 1460 80 2007 502 35 1480 765 73 240 593 235 1650 94 2008 536 44 1640 781 84 305 673 257 1840 103 2009 392 39 1300 625 70 261 550 188 1580 84 2010 469 46 1440 689 82 348 603 220 1850 108 2011 547 64 1690 799 95 446 692 262 2130 127 2012 551 69 1630 792 93 444 679 249 2220 118 2013 556 68 1710 813 98 468 711 253 2290 113 2014 568 65 1780 854 100 486 727 257 2380 110 2015 492 46 1580 790 94 429 632 225 2260 97 2016 476 42 1600 749 97 430 641 225 2210 92 2017 511 48 1740 802 102 488 714 240 2330 104 Engineering, Technology & Applied Science Research Vol. 8, No. 5, 2018, 3432-3438 3436 www.etasr.com Sun et al.: Analysis of Bilateral Trade Flow and Machine Learning Algorithms for GDP Forecasting Fig. 2. Increase in exports. With the growth in imports and exports of all countries identified, as shown in these plots, data analysis was conducted using linear regression to identify the correlation coefficient and other performance measures. B. Data Analysis The GDP of any country is dependent on import and export data since these factors make the greatest contributions to the country’s economy, and several algorithms have been implemented for the analysis of trade data. Therefore, this study focused on analyzing import and export data. The study initially employed linear regression, followed by a comparison with other machine learning algorithms. Here, preprocessed data from 10 countries were used as the input for linear regression, and the results for export and import datasets were obtained as shown in Tables III and IV, respectively. TABLE III. EXPORTS ANALYSIS FOR TOP 10 COUNTRIES Country CC MAE RMSE RAE (%) RRS(%) Canada 0.9642 40.606 46.7947 25.1067 25.8104 Colombia 0.8346 8.3626 10.8224 49.9367 53.1576 Germany 0.9375 162.9892 195.2004 32.1998 33.721 India 0.8353 78.3047 90.7998 54.2975 53.1019 Israel 0.9484 9.2247 10.3887 31.6044 30.7659 Netherlands 0.933 72.6982 84.1967 34.3541 34.8714 S. Africa 0.8842 13.7084 16.464 43.1783 45.1946 Sweden 0.942 22.7929 27.3736 31.1841 32.5885 UK 0.9639 61.4828 70.191 25.9338 25.899 US 0.9566 184.4345 211.1925 29.5354 28.3296 CC: Correlation Coefficient, MAE: Mean Absolute Error, RMSE: Root Mean Square Error, RAE: Relative Absolute Error, RRSE: Root Relative Squared Error TABLE IV. IMPORTS ANALYSIS FOR TOP 10 COUNTRIES Country CC MAE RMSE RAE (%) RRS(%) Canada 0.9585 44.3818 51.9594 27.3267 27.6743 Colombia 0.8578 9.0713 11.2877 48.4398 49.8184 Germany 0.9447 130.7667 154.7405 31.125 31.8091 India 0.8272 94.3616 110.4496 54.8779 54.241 Israel 0.9562 7.6478 8.748 29.0447 28.426 Netherlands 0.9347 62.5513 71.9923 34.2352 34.4388 South Africa 0.8703 15.8974 18.4622 46.8866 47.7568 Sweden 0.9344 20.9784 25.3766 32.8641 34.5429 UK 0.9628 67.9066 77.7684 25.9899 26.2868 US 0.9639 221.3444 249.555 25.9518 25.8817 CC: Correlation Coefficient, MAE: Mean Absolute Error, RMSE: Root Mean Square Error, RAE: Relative Absolute Error, RRSE: Root Relative Squared Error From these import and export Tables, it can be observed that the correlation coefficient is close to 1, indicating a strong positive linear relationship. However, the RMSE is slightly high in some cases. Since the dataset consisted of only 43 instances (last 43 years of data), the linear regression model showed good correlations and high RMSE values. Therefore, further analysis using other machine learning algorithms was conducted with import and export datasets from the U.S. and Germany. VI. RESULTS AND DISCUSSION To investigate on the best-performing machine learning algorithms for a better analysis of trade data and to forecast country GDP, this study considered 5 machine learning algorithms: • Linear Regression (LR) • RBF Regressor (RBF) • Support Vector Machine (SVM) • Regression by Discretization (RD) • Reduced Error Pruning Tree(REP) The U.S. and Germany import and export datasets were used as input datasets. A 10-fold cross-validation method was implemented to select training and testing datasets. Here correlation coefficients, root mean squared error, relative absolute error, and root relative squared error were considered as performance measures. A two-tailed test with a 0.05 confidence level was conducted and the results for these four performance measures are respectively shown in Tables V- VIII. Table V shows the correlation coefficients obtained for 4 datasets in the experiment with 5 algorithms. The Table shows that the RBF algorithm generated better models with high correlation coefficients, followed by RD, LR, SVM, and REP. Table VI (relative absolute error analysis) shows that RBF generated models with much smaller errors than LR and SVM. The results for RD lie between RBF and REP results. Therefore, RBF can be considered to perform well with respect to the relative absolute error measure. TABLE V. CORRELATION COEFFICIENT ANALYSIS Data Set LR RBF SVM RD REP Germany_exports 0.97 0.99 0.97 0.97 0.96 Germany_imports 0.97 0.99 0.97 0.97 0.96 US_exports 0.98 0.99 0.98 0.99 0.97 US_imports 0.98 0.99 0.98 0.99 0.97 TABLE VI. RELATIVE ABSOLUTE ERROR ANALYSIS Data Set LR RBF SVM RD REP Germany_exports 35.46 13.78 35.85 20.99 22.45 Germany_imports 34.41 15.57 32.05 21.36 23.81 US_exports 34.21 12.02 33.16 16.60 22.30 US_imports 29.38 9.58 30.72 16.36 20.82 TABLE VII. ROOT MEAN SQUARED ERROR ANALYSIS Data Set LR RBF SVM RD REP Germany_exports 187.53 86.55 195.54 118.76 139.53 Germany_imports 146.63 78.10 149.10 105.87 119.36 US_exports 205.34 83.24 237.78 111.97 155.62 US_imports 247.66 106.71 262.79 156.85 202.58 Engineering, Technology & Applied Science Research Vol. 8, No. 5, 2018, 3432-3438 3437 www.etasr.com Sun et al.: Analysis of Bilateral Trade Flow and Machine Learning Algorithms for GDP Forecasting TABLE VIII. ROOT RELATIVE SQUARED ERROR ANALYSIS Data Set LR RBF SVM RD REP Germany_exports 36.13 15.43 36.88 22.80 26.56 Germany_imports 33.85 16.69 33.01 23.69 26.95 US_exports 30.89 12.26 33.36 16.81 23.58 US_imports 28.26 11.45 29.47 17.66 23.25 Table VII (root mean squared error) and Table VIII (root relative squared error) also verify RBF as providing better results than the other algorithms. Although RMS values are high (with a smaller dataset), RBF obtains 60% better results than LR. The performance plots for the above measures using the RBF algorithm are shown in Figures 3-6. Fig. 3. Correlation coefficienet. Fig. 4. Mean absolute error. Fig. 5. Root mean squared error Given these measures, the RBF algorithm is shown to provide good performance with a gain of about 60%. The results demonstrate that RBF outperformed the other four algorithms and that the models generated using RBF produced better results in forecasting of country GDP. Fig. 6. Root relative squared error VII. CONCLUSIONS Trade analysis results show a strong positive linear relationship in trade statistics. Imports and exports increased linearly with the world engaging in increased trade activity. In particular, the U.S., China, Japan, and India have made remarkable progress in the past decade in both exports and imports. This study takes two different directions. The first direction is the analysis of the top 10 import/export countries to show the correlation in trade data, and the second is to investigate different machine learning algorithms to identify the best algorithm for the prediction of trade data and country GDP. Here the WEKA data analysis tool was used for data analysis, and the experiments were conducted using five machine learning algorithms. The results show that all five algorithms exhibited good correlations but that the RBF algorithm outperformed the other algorithms. This suggests that the RBF algorithm may perform well in forecasting trade data to predict a country’s GDP. REFERENCES [1] J. Uddin, “Time Series Behavior of Imports and Exports of Bangladesh: Evidence from Cointegration Analysis and Error Correction Model”, International Journal of Economics and Finance, Vol. 1, No. 2, pp. 156- 162, 2009 [2] I. S. Kim, S. Liao, K. Imai, “Measuring Trade Profile with Granular Product-level Trade Data”, available at: https://imai.fas.harvard.edu/ research/files/BIGtrade.pdf, 2018 [3] R. Sen, “Analyzing International Trade Data in a Small Open Economy: The Case of Singapore”, ASEAN Economic Bulletin, Vol. 17, No. 1, pp. 23-35, 2000 [4] M. Bahmani-Oskooee, H. J. Rhee, “Are imports and exports of Korea cointegrated?”, International Economic Journal, Vol. 11, No. 1, pp. 109- 114, 1997 [5] T. T. Cheong, “Are Malaysian exports and imports cointegrated? A comment”, Sunway Academic Journal, Vol. 2, pp. 101-107, 2005 [6] M. Bahmani-Oskooee, “Cointegration approach to estimate the long-run trade elasticities in LDCs”, International Economic Journal, Vol. 12, No. 3, pp. 89-96, 1998 [7] T. C. Tang, “Are imports and exports in the OIC member countries cointegrated? A reexamination”, IIUM Journal of Economics and Management, Vol. 14, No. 1, pp. 1-31, 2006 [8] C. K. Choong, S. C. Soo, Z. Yusop, “Are Malaysian exports and imports cointegrated?”, Sunway Academic Journal, Vol. 1, pp. 29-38, 2004 Engineering, Technology & Applied Science Research Vol. 8, No. 5, 2018, 3432-3438 3438 www.etasr.com Sun et al.: Analysis of Bilateral Trade Flow and Machine Learning Algorithms for GDP Forecasting [9] M. J. Sirgy, D. J. Lee, C. Miller, J. E. Littlefield, E. G. Atay, “The Impact of Imports and Exports on A Country’s Quality of Life”, Social Indicators Research, Vol. 83, No. 2, pp. 245-281, 2007 [10] A .A. J. Saaed, M. A. Hussain, “Impact of Exports and Imports on Economic Growth: Evidence from Tunisia”, Journal of Emerging Trends in Economics and Management Sciences, Vol. 6, No. 1, pp. 13-21, 2015 [11] D. Omotor, “The Role of Exports in the Economic Growth of Nigeria: The Bounds Test Analysis”, International Journal of Economic Perspectives, Vol. 2, No. 3, pp. 222-235, 2008 [12] F. E. Chemeda, “The Role of Exports in Economic Growth with Reference to Ethiopian Country”, Conference on Annual Meeting of American Agricultural Economics Association in Chicago, Chicago, USA, August 5-8, 2001 [13] P. Aghion, A. Bergeaud, M. Lequien, M. Melitz, The Impact of Exports on Innovation: Theory and Evidence, National Bureau of Economic Research, 2018 [14] L. Charles, G. Daudin, “Eighteenth-Century International Trade Statistics. Sources and Methods”, Revue de l’OFCE, Vol. 4, pp. 7-36, 2015 [15] WTO, World Trade Statistical Review, WTO, 2017 [16] M. Jerven, “On the accuracy of trade and GDP statistics in Africa: Errors of commission and omission”, Journal of African Trade, Vol. 1, No. 1, pp. 45-52, 2014 [17] The Central Intelligence Agency, The World Factbook, Country Comparison: Imports, available at: https://www.cia.gov/library/ Publications/the-world-factbook/rankorder/2087rank.html [18] The Central Intelligence Agency, The World Factbook, Country Comparison: Exports, available at: https://www.cia.gov/library/ publications/the-world-factbook/rankorder/2078rank.html [19] J. Desjardins, “The Top Importers and Exporters of the World’s 18 Most Traded Goods”, available at: http://www.visualcapitalist.com/top- importers-exporters-worlds-18-traded-goods/ [20] Teletrac Navman, “Top 18 Imports and Exports Around the World”, available at: https://www.teletracnavman.com/infographics/top-imports- exports [21] Financial Stability Board, Artificial Intelligence and Machine Learning in Financial Services: Market Developments and Financial Stability Implications, Financial Stability Board, 2017 [22] Info World, “An Intro to Genetic Algorithms”, available at: https://www. infoworld.com/article/3151009/software/an-intro-to-genetic-algorithms. html [23] A. M. Ticlavilca, D. M. Feuz, M. McKee, “Forecasting Agricultural Commodity Prices Using Multivariate Bayesian Machine Learning Regression”, NCCC-134 Conference on Applied Commodity Price Analysis, Forecasting, and Market Risk Management. St. Louis, USA, April 19-20, 2010 [24] M. Manaye, B. Borena, “Association Pattern Discovery of Import Export Items in Ethiopia”, HiLCoE Journal of Computer Science and Technology, Vol. 1, No. 2, pp. 82-88, 2013 [25] S. Circlaeys, C. Kanitkar, D. Kumazawa, “Bilateral Trade Flow Prediction”, available at: http://cs229.stanford.edu/proj2017/final- reports/5240224.pdf, 2017 [26] H. R. Joseph, “GDP Forecasting through Data Mining of Seaport Export-Import Records”, 9th International Conference on Data Mining, Las Vegas, USA, July 22-25, 2013 [27] F. F. Ping, F. X. Fei, “Multivariant forecasting mode of Guangdong province port throughput with genetic algorithms and Back Propagation neural network”, Procedia – Social and Behavioral Sciences, Vol. 96, pp. 1165-1174, 2013 [28] M. Panella, F. Barcellona, R. L. D’Ecclesia, “Forecasting Energy Commodity Prices Using Neural Networks”, Vol. 2012, Article ID 289810, 2012 [29] R. G. Donaldson, M. Kamstra, “An artificial neural network - GARCH model for international stock return volatility”, Journal of Empirical Finance, Vol. 4, No. 1, pp. 17-46, 1997 [30] I. ur Sami, K. N. Junejo, “Predicting Future Gold Rates using Machine Learning Approach”, International Journal of Advanced Computer Science and Applications, Vol. 8, No. 12, pp. 92-99, 2017 [31] S. Peluso, A. Mira, P. Muliere, A. Lomi, “International Trade: a Reinforced Url Network Model”, available at: https://arxiv.org/ abs/1601.03067, 2016 [32] World Bank, Indicators, available at: https://data.worldbank.org/ indicator