[Paper Review] Support Vector Machine Model for Currency Crisis Discrimination

Arindam Chaudhuri|arXiv (Cornell University)|Mar 3, 2014

Stock Market Forecasting Methods12 references3 citations

TL;DR

This paper proposes a Support Vector Machine (SVM) model using polynomial kernel functions to predict currency crises in Argentina, leveraging real-world economic data. The study demonstrates that third- and fourth-order polynomial kernels yield the highest generalization performance, enabling accurate early warning detection with a small subset of data, marking the first SVM-based approach for Argentina's currency crisis evaluation.

ABSTRACT

Support Vector Machine (SVM) is powerful classification technique based on the idea of structural risk minimization. Use of kernel function enables curse of dimensionality to be addressed. However, proper kernel function for certain problem is dependent on specific dataset and as such there is no good method on choice of kernel function. In this paper, SVM is used to build empirical models of currency crisis in Argentina. An estimation technique is developed by training model on real life data set which provides reasonably accurate model outputs and helps policy makers to identify situations in which currency crisis may happen. The third and fourth order polynomial kernel is generally best choice to achieve high generalization of classifier performance. SVM has high level of maturity with algorithms that are simple, easy to implement, tolerates curse of dimensionality and good empirical performance. The satisfactory results show that currency crisis situation is properly emulated using only small fraction of database and could be used as an evaluation tool as well as an early warning system. To the best of knowledge this is the first work on SVM approach for currency crisis evaluation of Argentina.

Motivation & Objective

• To develop a machine learning model capable of detecting currency crises in Argentina using empirical economic data.
• To evaluate the effectiveness of different kernel functions in SVM for crisis prediction, especially under high-dimensional data conditions.
• To create a reliable early warning system for policymakers by identifying crisis precursors with high accuracy.
• To demonstrate the feasibility of SVM in modeling complex financial crises with limited data.
• To establish a benchmark for SVM application in currency crisis detection, particularly in emerging markets like Argentina.

Proposed method

• The study employs Support Vector Machines (SVM), a classification technique based on structural risk minimization to reduce overfitting.
• Kernel functions are applied to transform input data into higher-dimensional spaces, mitigating the curse of dimensionality.
• Third- and fourth-order polynomial kernels are tested and found to offer the best balance between complexity and performance.
• The model is trained on real historical economic data from Argentina to simulate crisis and non-crisis periods.
• Model performance is evaluated using generalization accuracy, with emphasis on minimizing false positives and false negatives.
• A data reduction strategy is applied, showing that only a small fraction of the dataset is sufficient for reliable predictions.

Experimental results

Research questions

• RQ1 Which kernel function in SVM yields the highest generalization performance for detecting currency crises in Argentina?
• RQ2 Can a machine learning model trained on a limited subset of economic data accurately predict currency crises?
• RQ3 How effective is SVM in handling high-dimensional financial data typical of crisis prediction?
• RQ4 To what extent can SVM serve as a reliable early warning system for currency crises in emerging economies?
• RQ5 Is there a significant performance difference between polynomial and other kernel functions in this context?

Key findings

• The third- and fourth-order polynomial kernels consistently outperformed other kernel functions in terms of generalization accuracy.
• The SVM model achieved high predictive accuracy even when trained on a small fraction of the full dataset, indicating strong data efficiency.
• The model demonstrated robustness to high-dimensional input features, effectively managing the curse of dimensionality through kernel transformation.
• The study confirms that SVM is a mature, implementable, and effective tool for financial crisis prediction with practical relevance for policymakers.
• This work represents the first application of SVM for currency crisis evaluation in Argentina, establishing a novel benchmark in the field.
• The results support the use of SVM as a viable early warning system for currency crises, particularly in data-constrained emerging markets.