Abstract
The sporadic large fluctuations seen in the stock market are due to different factors. These large fluctuations are termed extreme events (EE). We have identified fundamental, technical, and external factors and categorized positive or negative EE depending on the impact of these factors. During such events, the stock price time series is found to be nonstationary. Hence, the Hilbert–Huang transformation is used to identify EEs based on high instantaneous energy () concentration. The analysis shows that concentration in the stock price is very high during both positive and negative EE, surpassing a threshold of where and are the mean energy and standard deviation of energy, respectively. Further, support vector regression is used to predict the stock price during an EE, with the close price being found to be the most useful input than the open-high-low-close (OHLC) inputs. The maximum prediction accuracy for one step using close price and OHLC prices are 95.98% and 95.64%, respectively. Whereas, for the two step prediction, the accuracies are 94.09% and 93.58%, respectively. These results highlight that the accuracy of one-step predictions surpasses that of two-step predictions. Also, accuracy decreases when predicting stock prices closer to an EE. The EEs identified from predicted time series exhibit statistical characteristics similar to those obtained from the original data.