TY - THES
T1 - Predicting Indonesian currency crises using early warning system models
AU - Syaifullah, -
PY - 2012
Y1 - 2012
N2 - Following the collapse of the Bretton Woods system of exchange rate management in the 1970s, the frequency of financial crises as well as the number of countries involved tends to increase. Even today, financial crises are still a major threat to many economies in the world and will undoubtedly continue in the future. Indonesia is no exception. With an open economy, Indonesia has experienced several financial crises. The 1997/98 Asian Financial Crisis was the worst in recent decades. It affected not only Indonesian macro economy but also the country’s social and political aspects. As a result this crisis is known as a multi-dimension crisis. The enormous impacts and huge recovery cost of financial crises encourage policy makers and economists to find ways to prevent these crises. This study aims to make a contribution in this field by constructing models to predict financial crises, in particular for Indonesia. It adopts and extends the signal model proposed by Kaminsky et al. (1998) as well as the discrete choice model proposed by Eichengreen et al. (1996) and Frankel and Rose (1996). In addition, as an alternative method, this study also applies the artificial neural network (ANN) model. The empirical findings indicate that these models perform well in predicting the Indonesian currency crises within the 24-month crisis window; however, the ANN model outperforms the other two models for both within and out of samples. Furthermore, in terms of consistency, sensitivity and prediction power of these models in predicting financial crises within three shorter crisis windows, namely the 6, 12 and 18 month crisis windows, the ANN model is also better than the other two models. Finally, the findings in this thesis support the argument that financial crises may be predicted and hence preventative measures may be implemented to deal with potential crises in the future.
AB - Following the collapse of the Bretton Woods system of exchange rate management in the 1970s, the frequency of financial crises as well as the number of countries involved tends to increase. Even today, financial crises are still a major threat to many economies in the world and will undoubtedly continue in the future. Indonesia is no exception. With an open economy, Indonesia has experienced several financial crises. The 1997/98 Asian Financial Crisis was the worst in recent decades. It affected not only Indonesian macro economy but also the country’s social and political aspects. As a result this crisis is known as a multi-dimension crisis. The enormous impacts and huge recovery cost of financial crises encourage policy makers and economists to find ways to prevent these crises. This study aims to make a contribution in this field by constructing models to predict financial crises, in particular for Indonesia. It adopts and extends the signal model proposed by Kaminsky et al. (1998) as well as the discrete choice model proposed by Eichengreen et al. (1996) and Frankel and Rose (1996). In addition, as an alternative method, this study also applies the artificial neural network (ANN) model. The empirical findings indicate that these models perform well in predicting the Indonesian currency crises within the 24-month crisis window; however, the ANN model outperforms the other two models for both within and out of samples. Furthermore, in terms of consistency, sensitivity and prediction power of these models in predicting financial crises within three shorter crisis windows, namely the 6, 12 and 18 month crisis windows, the ANN model is also better than the other two models. Finally, the findings in this thesis support the argument that financial crises may be predicted and hence preventative measures may be implemented to deal with potential crises in the future.
KW - Early warning system model
KW - Indonesian currency crisis
KW - Financial crisis
KW - Indonesian economy
KW - Signal model
KW - Crisis prediction model
KW - Artificial neural network model
KW - Probit model
M3 - Doctoral Thesis
ER -