Indian Journal of Science and Technology

Abstract

Objectives: The main of study is to predict the volatility trend with high accuracy. The improved accuracy can help profitable trades. Outliers in dataset can reduce prediction accuracy; an innovative dynamic approach is proposed to filter outliers from the dataset. Methods: In this study, an intelligent methodology is proposed to predict volatility trend with hybrid regression based outlier removal technique and advanced deep learning Long Short-Term Memory (LSTM) techniques. Regression techniques are used for prediction and classification problems, the same principle is applied here for identifying the outliers in the data. The random data set is trained and tested multiple trials with regression technique. The algorithm generates standard errors, residual error and ‘p’ value for all the predictions, these values are compared with standard threshold across all the trials. The similar errors with multiple occurrences are identified as outliers and removed from the dataset. The LSTM techniques trained and tested with different epochs and network configuration till the prediction results improve. Result: This study uses India Volatility Index (VIX) data for predicting next day volatility. The results show significant improvement in accuracy with the proposed approach. The results in this paper demonstrate the LSTM techniques out performs regression, decision trees, random forest, SVM, boosting techniques and neural network based techniques. Application: The paper also shows that usage of regression techniques for removal of outlier further improves the forecasting accuracy by 10%. The proposed approach can be applied for forex and option trading. The proposed approach can also used in other predictive modeling problems such as CRM, Healthcare, Credit risk and Auto insurance.

Keywords: Deep Learning, learning Long Short-Term Memory (LSTM), Prediction, Regression, Trend, Volatility