Search

Financial institutions, investors, mining companies and related firms need an effective accurate forecasting model to examine gold price fluctuations in order to make correct decisions. This paper proposes an innovative approach to accurately forecast gold price movements and to interpret predictions. First, it compares six machine learning models. These models include two very recent methods: the eXtreme Gradient Boosting (XGBoost) and CatBoost. The empirical findings indicate the superiority of XGBoost over other advanced machine learning models. Second, it proposes Shapley additive explanations (SHAP) in order to help policy makers to interpret the predictions of complex machine learning models and to examine the importance of various features that affect gold prices. Our results illustrate that the utilization of XGBoost along with SHAP approach could provide a significant boost in increasing the gold price forecasting performance. [ABSTRACT FROM AUTHOR]

In this paper, we propose a novel hybrid model that extends prior work involving ensemble empirical mode decomposition (EEMD) by using fuzzy entropy and extreme learning machine (ELM) methods. We demonstrate this 3-stage model by applying it to forecast carbon futures prices which are characterized by chaos and complexity. First, we employ the EEMD method to decompose carbon futures prices into a couple of intrinsic mode functions (IMFs) and one residue. Second, the fuzzy entropy and K-means clustering methods are used to reconstruct the IMFs and the residue to obtain three reconstructed components, specifically a high frequency series, a low frequency series, and a trend series. Third, the ARMA model is implemented for the stationary high and low frequency series, while the extreme learning machine (ELM) model is utilized for the non-stationary trend series. Finally, all the component forecasts are aggregated to form final forecasts of the carbon price for each model. The empirical results show that the proposed reconstruction algorithm can bring more than 40% improvement in prediction accuracy compared to the traditional fine-to-coarse reconstruction algorithm under the same forecasting framework. The hybrid forecasting model proposed in this paper also well captures the direction of the price changes, with strong and robust forecasting ability, which is significantly better than the single forecasting models and the other hybrid forecasting models. [ABSTRACT FROM AUTHOR]

As stock data is characterized by highly noisy and non-stationary, stock price prediction is regarded as a knotty problem. In this paper, we propose new two-stage ensemble models by combining empirical mode decomposition (EMD) (or variational mode decomposition (VMD)), extreme learning machine (ELM) and improved harmony search (IHS) algorithm for stock price prediction, which are respectively named EMD–ELM–IHS and VMD–ELM–IHS. Furthermore, to demonstrate the efficiency and performance of the proposed models, the results were compared with those obtained by other methods, including EMD based ELM (EMD–ELM), VMD based ELM (VMD–ELM), autoregressive integrated moving average (ARIMA), ELM, multi-layer perception (MLP), support vector regression (SVR), and long short-term memory (LSTM) models. The results show that the proposed models have superior performance in terms of its accuracy and stability as compared to the other models. Also, we find that the sizes of sliding window and training set have a significant impact on the predictive performance. [ABSTRACT FROM AUTHOR]

This paper proposes a new decomposition approach by separating realized covariation into components based on signs (positive and negative) and magnitudes (continuous, small jump, and large jump). The motivation behind this decomposition is that certain variation components can be more useful in forecasting than others. Including only “information-rich” components in realized (co)-variation forecasting models can improve predictive accuracy. Using various machine learning models, the marginal predictive content of each variation components can be assessed. The empirical exercise is based on all constituent of S &P 500 stocks between 2010 and 2019. We find that standard machine learning methods without the more granular variation measures offer limited improvement to out-of-sample fit (likely due to low signal-to-noise ratios) relative to benchmark HAR-type forecasting models. However, sparse models, which are specified using predictors selected using a first “variable selection” yield significant improvements in predictive accuracy when the decomposed variation measures are included. These predictive gains can be traced to the identification of short-lived pricing signals associated with co-jumps. [ABSTRACT FROM AUTHOR]