Application of Neural Network in Early Detection of Financial Crisis in Singapore
Fadia Mulyarti (a), Sugiyanto (a*), Sri Subanti (a)

a) Department of Statistics, Faculty of Mathematics and Natural Sciences, Universitas Sebelas Maret , Indonesia.
*sugiyanto61[at]staff.uns.ac.id

Abstract

The financial crises that occurred in 1997 and 2008 had a negative impact on several countries, including Singapore. A financial crisis can occur suddenly so that it can endanger a country^s economy if it is not prepared for it. Therefore, early detection of financial crises is needed as a form of crisis warning so that the government can anticipate and prepare appropriate policies. The independent variables used are monthly data of 11 key macroeconomic and financial indicators of Singapore^s economy from January 1990 to June 2021. Perfect signal is used as the dependent variable in the crisis early detection system. This study aims to build a model of a financial crisis detection system in Singapore using Multilayer Perceptron Backpropagation (MLPBP) as a neural network algorithm by comparing the optimization of Stochastic Gradient Descent (SGD) and Nesterov-accelerated Adaptive Moment Estimation (Nadam). The optimal hyperparameter value in the model was searched using the grid search method based on the accuracy and obtained the best model with 11-11-1 network architecture, best optimization is Nadam, hyperparameter learning rate = 0.1, \upsilon= 0.975, \nu=0.999, \epsilon=10^{-8}, batch size = 128, epoch = 100, and sigmoid activation function. The best model is then evaluated in the testing data and obtained 95.89% accuracy, 96.7% sensitivity, and 90% specificity. The results of the Perfect Signal prediction show that from January to June 2022 it is predicted that there will be no financial crisis in Singapore.

Keywords: financial crisis, early warning system, neural network, multilayer perceptron backpropagation

Topic: Computer Science