Testing the Early Warning System: Common Determinants of Currency Crisis in Turkey

Qiu, Xixi (2013) Testing the Early Warning System: Common Determinants of Currency Crisis in Turkey. [Dissertation (University of Nottingham only)] (Unpublished)

[thumbnail of 9Testing_the_Early_Warning_System-_Common_Determinants_of_Currency_Crisis_in_Turkey.pdf] PDF - Registered users only - Requires a PDF viewer such as GSview, Xpdf or Adobe Acrobat Reader
Download (425kB)
[thumbnail of 9Testing_the_Early_Warning_System-_Common_Determinants_of_Currency_Crisis_in_Turkey.pdf] PDF - Registered users only - Requires a PDF viewer such as GSview, Xpdf or Adobe Acrobat Reader
Download (425kB)

Abstract

This study focuses on examining the predictive power of both signal approach developed by Kaminsky, Lindo and Reinart (1998) and probit model developed by Frankel and Rose (1996). This study tests both two models in Turkey between 1990 and 2003 when there are two main serious currency crises happened in 1994 and 2001. Then, the study compares the results suggested by both two models and examines whether the same model has different predictive power in terms of different type of crises. Finally, two models give quite different results. The Signal approach has a satisfactory overall performance while only minority of individual variables works well. Only export, equity index and foreign currency reserves have good performance as noise-to-signal ratio is lower than 1 (0.34, 0.99 and 0.99), given the 90% as the threshold. As the threshold decreases, the number of variables also decline to only two (85%) and then one (80%), which is export with 0.4 as noise-to-signal ratio. However, in the probit model, most variables that have significant power in predicting currency crisis but the overall performance is not as good as signal approach. The increased exports, overnight rate and terms of trade have increased the probability of currency crisis. The decreased business confidence and industrial growth increase the probability of currency crisis. Interestingly, two models give quite different results, as most variables that have good performance in signal approach do not have significant predictive power in probit model. The export is only the indicator that only is agreed by both models.

Item Type: Dissertation (University of Nottingham only)
Depositing User: EP, Services
Date Deposited: 07 Mar 2014 10:05
Last Modified: 05 Jun 2018 07:26
URI: https://eprints.nottingham.ac.uk/id/eprint/26845

Actions (Archive Staff Only)

Edit View Edit View