Machine learning-based approaches to forecasting international trade
Název práce v češtině: | Prognózování mezinárodního obchodu s využitím metod strojového učení |
---|---|
Název v anglickém jazyce: | Machine learning-based approaches to forecasting international trade |
Klíčová slova: | machine learning, mezinárodní obchod, prognostika |
Klíčová slova anglicky: | machine learning, international trade, forecasting |
Akademický rok vypsání: | 2016/2017 |
Typ práce: | bakalářská práce |
Jazyk práce: | angličtina |
Ústav: | Institut ekonomických studií (23-IES) |
Vedoucí / školitel: | Ing. Vilém Semerák, M.A., Ph.D. |
Řešitel: | skrytý - zadáno vedoucím/školitelem |
Datum přihlášení: | 29.05.2017 |
Datum zadání: | 12.06.2017 |
Datum a čas obhajoby: | 29.01.2019 09:00 |
Místo konání obhajoby: | Opletalova - Opletalova 26, O105, Opletalova - místn. č. 105 |
Datum odevzdání elektronické podoby: | 04.01.2019 |
Datum proběhlé obhajoby: | 29.01.2019 |
Oponenti: | Mgr. Bc. Vít Macháček, Ph.D. |
Kontrola URKUND: |
Seznam odborné literatury |
Main literature:
[1] Keck, A., Raubold, A. and Truppia, A. (2010) ‘Forecasting International Trade: A Time Series Approach’, OECD Journal: Journal of Business Cycle Measurement and Analysis, 2009(2), pp. 157–176. doi: https://doi.org/10.1787/jbcma-2009-5ks9v44bdj32. [2] Nummelin, T. and Hänninen, R. (2016) Model for international trade of sawnwood using machine learning models. Helsinki. doi: 10.13140/RG.2.2.26478.41284. [3] Yildirim, I., Ozsahin, S. and Okan, O. T. (2014) Prediction of Non-Wood Forest Products Trade Using Artificial Neural Networks, J. Agr. Sci. Tech. Available at: http://41.204.187.24:8080/bitstream/handle/123456789/3798/JAST_Volume 16_Issue Supplementary Issue_Pages 1481-1492.pdf?sequence=1&isAllowed=y (Accessed: 11 September 2018). Supporting literature: theoretical: [4] Baier, S. L. ;, Kerr, A. ; and Yotov, Y. V (2017) Gravity, Distance, and International Trade. Available at: http://hdl.handle.net/10419/155599www.econstor.eu (Accessed: 11 September 2018). [5] Chaney, T. (2018) ‘The Gravity Equation in International Trade: An Explanation’, Journal of Political Economy, 126(1), pp. 150–177. doi: 10.1086/694292. [6] Cieslik, A. (2007) Bilateral trade volumes, the gravity equation and factor proportions. Available at: https://pdfs.semanticscholar.org/414c/b1dfcc266a8ed3bdbc599dffe88d17b2cccd.pdf (Accessed: 17 September 2018). [7] Egger, P. H. and Nigai, S. (2015) ‘Structural gravity with dummies only: Constrained ANOVA-type estimation of gravity models’, Journal of International Economics. North-Holland, 97(1), pp. 86–99. doi: 10.1016/J.JINTECO.2015.05.004. empirical studies: [8] Hendry, D. F. and Clements, M. P. (2003) ‘Economic forecasting: some lessons from recent research’, Economic Modelling. North-Holland, 20(2), pp. 301–329. doi: 10.1016/S0264-9993(02)00055-X. [9] Rossi, B. and Sekhposyan, T. (2011) ‘Understanding models’ forecasting performance’, Journal of Econometrics. North-Holland, 164(1), pp. 158–172. doi: 10.1016/J.JECONOM.2011.02.020. [10] Diebold, F. X. and Mariano, R. S. (1995) ‘Comparing Predictive Accuracy’, Journal of Business & Economic Statistics, 13(3), pp. 253–263. doi: 10.1080/07350015.1995.10524599. [11] Kroner, K. F. and Lastrapes, W. D. (1993) ‘The impact of exchange rate volatility on international trade: Reduced form estimates using the GARCH-in-mean model’, Journal of International Money and Finance. Pergamon, 12(3), pp. 298–318. doi: 10.1016/0261-5606(93)90016-5. |
Předběžná náplň práce v anglickém jazyce |
Research question and motivation
In recent years, the machine learning field has experienced a rapid development with many new techniques developed. Application of these techniques has shaken nearly every industry and international trade has been no exception. Most of the applications were, however, focused on a micro level - that includes things such as supply chain analysis, demand patterns analysis, or understanding of customer-retailer relationships. I believe there is a huge untapped potential to make use of machine learning even elsewhere - on a macro level. The aim of this bachelor thesis is to focus on the application of machine learning techniques in the field of international trade and compare their feasibilities for forecasting international trade to the long-established approaches. The gravity model has been used to model the import and export flow from one country to another since the 1960’s. It has been improved upon many times by adding additional variables such as proxies for regulatory standards, customs environment, cultural similarity, trade agreements, border costs and many others to capture more subtle and/or more complex behaviours. The gravity model was succeeded by different models from the family of general or partial equilibrium models many of which are generally regarded as superior to it. Nevertheless, despite certain disadvantages, it still remains widely popular due to its simplicity and decent predicting accuracy. Using RMSE the aim is to compare the gravity model to models based on machine learning techniques (SVM, Random forest, NN) to establish whether these techniques can be used to forecast international trade with higher accuracy. Contribution This thesis investigates how the gravity model compares with other approaches for numerical estimation and aims to answer the following hypothesis: Can machine learning-based techniques be used to provide better forecasts of international trade compared with the gravity model? The thesis will also contain a ‘cost-benefit analysis’ of each of the approaches, specifically with regard to interpretability of results. As it is a rather unexplored area, there are not many previous pieces of literature or academic publications covering the application of the aforementioned techniques in international trade, however, publications in other areas do exist. More on literature in the last paragraph. Methodology Firstly, a gravity model for a sample of approximately 200 countries and for a total of 16 industries is fitted and used to forecast imports/exports in different industries - it will serve as a baseline for comparison with other approaches. Secondly, I shall proceed with estimating the other models, namely an ARMA model, SVM model, random forest model and NN model using yearly export-import data for years 1993-2012, keeping the period 2013-2017 as a test set. Thirdly, I shall make use of so-called rolling forecasts to predict the value of import/export for every year of my test set to obtain an RMSE that will then be used to compare the performance accross models. This approach will simplify the analysis as it will allow avoiding the necessity to take into account the input-output relationships between industry sectors. Lastly, the results are compared with traditional scoring models and further ways of research are suggested. The data for the analysis is obtained from publicly available sources such as WITS, Comtrade, and OECD. Outline Introduction - description of the topic Literature review Dataset description Data manipulation Own empirical study - gravity model, time series model, random forest model, SVM model, NN model Results comparison Limitations Conclusion and suggestions for further research |