本研究目的是要透過變數的挑選來建立一套金融危機的預警系統，其中金融危機早期預警系統的目的是為了要降低金融危機帶來的損失，因為國家的政府、金融機構能夠提早做因應的準備，避免經濟體遭受過大的衝擊，所以許多經濟學家們多致力於建立能夠為金融危機預警的預防系統。傳統在經濟領域上，許多學者採用計量經濟學的方法進行挑選變數，其中必須符合許多計量模型上的假設與限制。本研究為了打破在建模前的種種限制，特別引進另一種可以挑選變數的隨機森林演算法，隨機森林的好處是能夠直接地從資料本身進行分析，從資料中探尋出風險因子對金融危機形成的影響。
本研究利用2008年金融風暴的橫斷面資料進行隨機森林挑選變數的分析。我們結果顯示隨機森林在預測上的準確度遠高於線性迴歸模型的預測，預測的準確度是將資料實際依變數數值與透過隨機森林的預測值差距進行均方差的比較，並且透過統計方法來針對線性迴歸模型與隨機森林得預測能力進行檢定，檢定結果顯示兩個模型的預測結果的預測誤差是顯著不同的，故隨機森林的預測準確率是顯著高於線性迴歸模型的預測結果。本研究我們發現透過隨機森林所挑選到的變數是與過去文獻有重疊的，此外我們更進一步挖掘出一些過去文獻未顯著的重要變數。在實證結果方面，本研究的結果發現隨機森林的預測值比一般線性迴歸模型的準確度高，然而我們利用迴歸模型中的模型選擇來為挑選顯著的變數來建立線性迴歸模型，從中我們發現到向前選取法的預測準確度略高於隨機森林，使我們發現在特定的資料集下，簡單的方法就能夠擊敗複雜的方法是個有趣的發現。雖然透過向前選取法的準確度略高於隨機森林，但是隨機森林演算法能夠從中進一步去了解傳統過去的計量方法無法深度挖掘出變數間的潛在動態關係。最後，我們發現經由隨機森林挑選的變數來建立一般線性迴歸模型是有效提高線性迴歸模型預測準確度，所以本研究對於利用挑選有用的領先指標來建立金融危機的預警系統仍保持著樂觀的態度。

In this thesis, we determine the important leading indicators of the early warning system by the ensemble method of Random Forest. The early warning system is designed to employ prompt measures to mitigate the impact of economic downturns during the financial crisis. As a result, much effort has been devoted to establishing the financial early warning system. Nevertheless, the traditional econometric methods generally impose various assumptions to propose feasible estimators. By contrast, the Random Forest method provides flexible modeling without restrictions of the functional relationship between the response and the explanatory variables, and hence underlying dynamics of economic structure can fully be explored and analyzed.
We adopt the Random Forest method to identify important leading indicators in the early warning system with cross-sectional data set for the 2008 global financial crisis. In the research, we find that the variables we selected through Random Forest overlap with those in the previous literature. Moreover, we further explore some extra important variables via Random Forest. However, when using the commonly-used model selection in the regression analysis, we find that the prediction accuracy of the forward selection method is slightly higher than that of the Random Forest. Thus, the implication of this interesting finding is that with a particular data set, a-simple-method-may-be-useful-than-a-complex-one. Finally, we discover that building up a linear regression model based on Random Forest can effectively improve the prediction accuracy of the-linear-regression-model. Thus, we remain optimistic about the feasibility of the early warning system for the financial crisis.

1. J. Aizenman, M. M Hutchison. Exchange market pressure and absorption by international reserves: Emerging markets and fear of reserve loss during the 2008–2009 crisis. Journal of International Money and Finance, vol. 31(5), pp. 1076-1091, 2012.
2. L. Alessi, C. Detken. Identifying excessive credit growth and leverage. Journal of Financial Stability, vol.31-33, p. 11, 2017.
3. A. Altmann, L. Toloşi, O. Sander, T. Lengauer. Permutation importance: a corrected feature importance measure. Bioinformatics, vol. 26(10), pp. 1340-1347, 2010.
4. R. Anderton, T. Tewolde. The global financial crisis: Understanding the global trade downturn and recovery. The World Economy, vol. 34(5), pp. 741-763, 2011.
5. P. Büchlmann, B. Yu. Analyzing bagging. Annals of Statistics, pp. 927-961, 2002.
6. J. Babecký, T. Havránek, J. Matějů, M. Rusnák, K. Šmídková, & B. Vašíček. Leading indicators of crisis incidence: Evidence from developed countries. Journal of International Money and Finance, vol. 35, pp. 1-19, 2013.
7. A. Berg, E. Borensztein, C. Pattillo. Assessing early warning systems: how have they worked in practice? IMF Staff Papers, vol. 52(3), pp. 462-502, 2005.
8. S. P. Berkmen, G. Gelos, R. Rennhack, J. P. Walsh. The global financial crisis: Explaining cross-country differences in the output impact. Journal of International Money and Finance, vol. 31(1), pp. 42-59, 2012.
9. I. Borchert, A. Mattoo. The crisis-resilience of services trade. The Service Industries Journal, vol. 30(13), pp. 2115-2136, 2010.
10. C. E. Borio, P. Disyatat. Global imbalances and the financial crisis: Link or no link? BIS Working Paper, vol. 346, p. 43, 2011.
11. L. Breiman. Bagging predictors. Machine learning, vol. 24(2), pp. 123-140, 1996.
12. L. Breiman. Manual on setting up, using, and understanding random forests v3.1. Statistics Department University of California Berkeley, CA, USA, 2002.
13. L. Breiman, J. Friedman, C. J. Stone, R. A. Olshen. Classification and regression trees. CRC press, 1984.
14. A. Buja, W. Stuetzle. Observations on bagging. Statistica Sinica, vol. 16(2), pp. 323-351, 2006.
15. G. Caggiano, P. Calice, L. Leonida. Early warning systems and systemic banking crises in low income countries: A multinomial logit approach. Journal of Banking & Finance, vol. 47, pp. 258-269, 2014.
16. J. J. Dabrowski, C. Beyers, J. P. de Villiers. Systemic banking crisis early warning systems using dynamic Bayesian networks. Expert systems with applications, vol. 62, pp. 225-242, 2016.
17. A. Demirgüç-Kunt, E. Detragiache. Cross-country empirical studies of systemic bank distress: a survey. National Institute Economic Review, vol. 192(1), pp. 68-83, 2005.
18. F. X. Diebold, R. S. Mariano. Comparing predictive accuracy. Journal of Business & economic statistics, vol. 20(1), pp. 134-144, 2002.
19. T. G. Dietterich. Ensemble learning. The handbook of brain theory and neural networks, vol. 2, pp. 110-125, 2002.
20. M. Feldkircher. The determinants of vulnerability to the global financial crisis 2008 to 2009: Credit growth and other sources of risk. Journal of International Money and Finance, vol. 43, pp. 19-49, 2014.
21. M. Feldkircher, R. Horvath, & M. Rusnak. Exchange market pressures during the financial crisis: A Bayesian model averaging evidence. Journal of International Money and Finance, vol. 40, pp. 21-41, 2014.
22. J. Frankel, & G. Saravelos. Can leading indicators assess country vulnerability? Evidence from the 2008–09 global financial crisis. Journal of International Economics, vol. 87(2), pp. 216-231, 2012.
23. J. Friedman, T. Hastie, & R. Tibshirani. The elements of statistical learning (vol. 1), Springer series in statistics New York, 2001.
24. S. R. Ghosh, & A. R. Ghosh. Structural vulnerabilities and currency crises. IMF Staff Papers, pp. 481-506, 2003.
25. M. Hermansen, & O. Röhn. Economic resilience: The usefulness of early warning indicators in OECD countries, OECD Economics Department Working Papers, vol. 1250 , pp. 30, 2015.
26. Z. Jones, & F. Linder. Exploratory data analysis using random forests. Paper presented at the Prepared for the 73rd annual MPSA conference, 2015.
27. G. Kaminsky, , S. Lizondo, , & C. M. Reinhart . Leading indicators of currency crises. Staff Papers, vol. 45(1), pp. 1-48, 1998.
28. K. Kauko, How to foresee banking crises? A survey of the empirical literature. Economic Systems, vol. 38(3), pp. 289-308, 2014.
29. M. Lang, & P. G. Schmidt. The early warnings of banking crises: Interaction of broad liquidity and demand deposits. Journal of International Money and Finance, vol. 61, pp. 1-29, 2016.
30. S. Li, & S. Wang. A financial early warning logit model and its efficiency verification approach. Knowledge-Based Systems, vol. 70, pp. 78-87, 2014.
31. J. Morgan. Classification and regression tree analysis. Report No1. Boston University School of Public Health, 2014.
32. A. K. Rose & M. M. Spiegel. Cross‐country causes and consequences of the 2008 crisis: International linkages and American exposure. Pacific Economic Review, vol. 15(3), pp. 340-363, 2010.
33. A. K. Rose & M. M. Spiegel. Cross-country causes and consequences of the crisis: An update. European Economic Review, vol. 55(3), pp. 309-324, 2011.
34. A. K. Rose & M. M. Spiegel. Cross-country causes and consequences of the 2008 crisis: early warning. Japan and the World Economy, vol. 24(1), pp. 1-16, 2012.
35. A. K. Rose, & M. M. Spiegel. Could an early warning system have predicted the crisis? VOX CEPR's Policy PortalNov, 2014.
36. B. Vašíček, D. Žigraiová, M. Hoeberichts, R. Vermeulen, K. Šmídková, & J. de Haan. Leading indicators of financial stress: New evidence. Journal of Financial Stability, vol. 28, pp. 240-257, 2017.
37. J. C. Wang & T. Hastie. Boosted varying-coefficient regression models for product demand prediction. Journal of Computational and Graphical Statistics, vol. 23(2), pp. 361-382, 2014.
38. F. Ward. Spotting the danger zone: forecasting financial crises with classification tree ensembles and many predictors. Journal of Applied Econometrics, vol. 32(2), pp. 359-378, 2017.
39. Q.S. Xu, & Y.-Z. Liang. Monte Carlo cross validation. Chemometrics and Intelligent Laboratory Systems, vol. 56(1), pp. 1-11, 2001.