HAR Model Research Articles

What drives the uranium sector risk? The role of attention, economic and geopolitical uncertainty

Interest in nuclear energy has increased recently due to its low-carbon footprint, energy security concerns, and technological advances. Despite the recent surge in uranium stocks, there is a lack of research on uranium sector volatility. We fill this gap by analyzing the volatility of the Global X Uranium ETF (URA) from 2010 to 2024 using high-frequency data. Our analysis reveals that HAR models effectively capture URA volatility. Market-wide implied volatility and investor attention, captured by Google search volume, are found to contain valuable information for forecasting uranium sector volatility in an in-sample context. In contrast, economic and geopolitical uncertainty, as well as global financial risk, exhibit limited relevance. Although advanced models show some improvement in out-of-sample predictions, the basic HAR model remains a robust benchmark.

Heterogeneity, Jumps and Co-Movements in Transmission of Volatility Spillovers Among Cryptocurrencies

Abstract We analyze properties identified in the price volatility of Bitcoin and some of the leading cryptocurrencies namely Litecoin, Ripple, and Ethereum. We employ Heterogeneous Autoregressive models (HAR) in both a univariate and multivariate level of analysis. First, the significance of heterogeneity and jumps is examined, considering the ability of several univariate HAR models, to predict realized volatility of cryptocurrencies. Second, we examine the relevance of realized volatility jumps and covariances in the transmission of volatility spillovers among cryptocurrencies. We perform a comparative spillover analysis of the multivariate HAR models in two versions, considering variances only and covariances as well. Our results indicate that covariances and jumps inclusion lead to an increase in spillovers. The time-varying spillover analysis indicates higher dependency between Bitcoin and the other cryptocurrencies mostly at short frequencies.

Cross-Modal Federated Human Activity Recognition.

Federated human activity recognition (FHAR) has attracted much attention due to its great potential in privacy protection. Existing FHAR methods can collaboratively learn a global activity recognition model based on unimodal or multimodal data distributed on different local clients. However, it is still questionable whether existing methods can work well in a more common scenario where local data are from different modalities, e.g., some local clients may provide motion signals while others can only provide visual data. In this article, we study a new problem of cross-modal federated human activity recognition (CM-FHAR), which is conducive to promote the large-scale use of the HAR model on more local devices. CM-FHAR has at least three dedicated challenges: 1) distributive common cross-modal feature learning, 2) modality-dependent discriminate feature learning, 3) modality imbalance issue. To address these challenges, we propose a modality-collaborative activity recognition network (MCARN), which can comprehensively learn a global activity classifier shared across all clients and multiple modality-dependent private activity classifiers. To produce modality-agnostic and modality-specific features, we learn an altruistic encoder and an egocentric encoder under the constraint of a separation loss and an adversarial modality discriminator collaboratively learned in hyper-sphere. To address the modality imbalance issue, we propose an angular margin adjustment scheme to improve the modality discriminator on modality-imbalanced data by enhancing the intra-modality compactness of the dominant modality and increase the inter-modality discrepancy. Moreover, we propose a relation-aware global-local calibration mechanism to constrain class-level pairwise relationships for the parameters of the private classifier. Finally, through decentralized optimization with alternative steps of adversarial local updating and modality-aware global aggregation, the proposed MCARN obtains state-of-the-art performance on both modality-balanced and modality-imbalanced data.

Forecasting Crude Oil Volatility Using the Deep Learning‐Based Hybrid Models With Common Factors

ABSTRACTBased on empirical evidence of the Chinese commodity futures volatility dynamics, we propose a novel and flexible hybrid model, denoted as SAE‐HAR‐DL, which combines a supervised autoencoder (AE) with the deep learning‐based HAR model framework to capture essential common factor information and uses the reconstruction error of the AE component as a regularizer to enhance the generalization ability of the testing subsample. The empirical findings strongly support the effectiveness of this model in accurately forecasting crude oil futures volatility in the post‐COVID‐19 era, compared to the HAR, HAR‐PCA, and HAR‐DL models. Moreover, a robustness check also demonstrates the positive contribution of common factors to the volatility prediction of other commodity futures. Notably, we establish that these common factors act as effective regularizers, mitigating prediction losses within the HAR model in extreme risk events such as the COVID‐19 pandemic and the Russia–Ukraine conflict.

The relationship between renewable energy attention and volatility: A HAR model with markov time-varying transition probability

This study investigated whether a renewable energy attention (REA) index, based on natural language processing, Google search volume data, and dimensionality reduction methodology, can predict crude oil volatility. Considering the possible non-linear and time-varying effects, we adopted the time-varying transition probability Markov switching heterogeneous autoregressive-realized volatility (TVTP-MS-HAR-realized volatility (RV)) model. To further represent the impacts of REA, we developed an asymmetric TVTP-MS-HAR-RV model based on this model framework, i.e., the ASTVTP-MS-HAR-RV model. According to the results, the in-sample estimates indicated that West Texas Intermediate (WTI) volatility is more affected by negative REA shocks than by positive ones. Moreover, REA predicted WTI volatility better during low-volatility periods than in high ones. According to the out-of-sample findings, the ASTVTP-MS-HAR-RV-F model outperformed other competing models, indicating that time-varying transition probabilities and REA information can significantly improve volatility forecasting performance.

Comparative Analysis of Financial Market Volatility and Correlation Risk During the Great Recession and the COVID-19 Pandemic

This paper offers a starting point for reflection on the similarities and differences of the impact on financial markets of the Great Recession of 2008 and of the Covid-19 pandemic of 2020 in terms of volatility and correlation risk among the most significant financial indexes in Europe. More precisely, the dataset employed includes the daily returns of Ftse100, CaC40, Dax30 and FtseMib40, with reference to the two time periods in which the two major crises manifested their effects on the markets. We use two different methodological approaches: the analysis of the daily conditional variance using various families of GARCH models and the study of the weekly realized volatility using HAR models. Furthermore, the estimation of the dependence structure of the GARCH residuals using copula functions is performed.

AutoAugHAR: Automated Data Augmentation for Sensor-based Human Activity Recognition

Sensor-based HAR models face challenges in cross-subject generalization due to the complexities of data collection and annotation, impacting the size and representativeness of datasets. While data augmentation has been successfully employed in domains like natural language and image processing, its application in HAR remains underexplored. This study presents AutoAugHAR, an innovative two-stage gradient-based data augmentation optimization framework. AutoAugHAR is designed to take into account the unique attributes of candidate augmentation operations and the unique nature and challenges of HAR tasks. Notably, it optimizes the augmentation pipeline during HAR model training without substantially extending the training duration. In evaluations on eight inertial-measurement-units-based benchmark datasets using five HAR models, AutoAugHAR has demonstrated superior robustness and effectiveness compared to other leading data augmentation frameworks. A salient feature of AutoAugHAR is its model-agnostic design, allowing for its seamless integration with any HAR model without the need for structural modifications. Furthermore, we also demonstrate the generalizability and flexible extensibility of AutoAugHAR on four datasets from other adjacent domains. We strongly recommend its integration as a standard protocol in HAR model training and will release it as an open-source tool1.

Predicting expected idiosyncratic volatility: Empirical evidence from ARFIMA, HAR, and EGARCH models

We investigate the performances of the ARFIMA, HAR, and EGARCH models in capturing the time-varying property of idiosyncratic volatility (IVOL). We find that the expected IVOL predictions by HAR are superior. In diverse portfolio scenarios, a greater degree of judgment is required to assess the pricing ability of expected IVOLs. For the lowest value-weighted quintiles and the expected IVOL estimated by the HAR model, the IVOL-return relationship is negative. Conversely, the IVOL-return relationship is positive for the expected IVOL estimated by the EGARCH model. Further evidence suggests a complicated and mixed relationship between the expected IVOL estimated by the ARFIMA model and stock returns.

Forecasting realized volatility of crude oil futures prices based on machine learning

AbstractExtending the popular HAR model with additional information channels to forecast realized volatility of WTI futures prices, we show that machine learning‐generated forecasts provide better forecasting quality and that portfolios that are constructed with these forecasts outperform their competing models resulting in economic gains. Analyzing the selection process, we show that information channels vary across forecasting horizon. Variable selection produces clusters and provides evidence that there are structural changes with regard to the significance of information channels.

Forecasting Ethereum’s volatility: an expansive approach using HAR models and structural breaks

Forecasting Ethereum’s volatility: an expansive approach using HAR models and structural breaks

Forecasting day-ahead expected shortfall on the EUR/USD exchange rate: The (I)relevance of implied volatility

The existing literature provides mixed results on the usefulness of implied volatility for managing risky assets, while evidence for expected shortfall predictions is almost nonexistent. Given its forward-looking nature, implied volatility might be more valuable than backward-looking measures of realized price fluctuations. Conversely, the volatility risk premium embedded in implied volatility leads to overestimating the observed price variation. This paper explores the benefits of augmenting econometric models used in forecasting the expected shortfall, a risk measured endorsed in the Basel III Accord, with information on implied volatility obtained from EUR/USD option contracts. The day-ahead forecasts are obtained from several classes of econometric models: historical simulation, EGARCH, quantile regression-based HAR, joint VaR and ES model, and combination forecasts. We verify whether the resulting expected shortfall forecasts are well-specified and test the models’ accuracy. Our results provide evidence that the information provided by forward-looking implied volatility is more valuable than that in backward-looking realized measures. These results hold across multiple model specifications, are stable over time, hold under alternative loss functions, and are more pronounced during periods of higher market uncertainty when risk modeling matters most.

Introducing NBEATSx to realized volatility forecasting

This paper investigates the application of neural basis expansion analysis with exogenous variables (NBEATSx) in the prediction of daily stock realized volatility for various time steps. It compares NBEATSx’s forecasting accuracy and robustness with several commonly used models, namely Long-Short Term Memory (LSTM) network, Temporal Neural Network (TCN), HAR, GARCH, and GJR-GARCH models. In this research, a total of six distinct stock indexes, three error measures, and four statistical tests are used, while three robustness tests are conducted to verify the outcomes of this paper. The findings of this research show that NBEATSx consistently yields statistically more accurate and robust forecasts than the other considered models. On average, NBEATSx generates forecasts that are respectively 13% and 8% more accurate for medium-term and long-term forecasting. Additionally, it produces forecasts that are respectively 43%, 60%, and 59% more robust for short-term, medium-term, and long-term forecasting. Yet, it should be noted that the superiority of NBEATSx in terms of forecast accuracy is not evident when applied to stock indexes from developing countries.

GARMA, HAR and Rules of Thumb for Modelling Realized Volatility

This paper features an analysis of the relative effectiveness, in terms of the Adjusted R-Square, of a variety of methods of modelling realized volatility (RV), namely the use of Gegenbauer processes in Auto-Regressive Moving Average format, GARMA, as opposed to Heterogenous Auto-Regressive HAR models and simple rules of thumb. The analysis is applied to two data sets that feature the RV of the S&P500 index, as sampled at 5 min intervals, provided by the OxfordMan RV database. The GARMA model does perform slightly better than the HAR model, but both models are matched by a simple rule of thumb regression model based on the application of lags of squared, cubed and quartic, demeaned daily returns.

An optimized deep learning model for human activity recognition using inertial measurement units

AbstractHuman activity recognition (HAR) has recently gained popularity due to its applications in healthcare, surveillance, human‐robot interaction, and various other fields. Deep learning (DL)‐based models have been successfully applied to the raw data captured through inertial measurement unit (IMU) sensors to recognize multiple human activities. Despite the success of DL‐based models in human activity recognition, feature extraction remains challenging due to class imbalance and noisy data. Additionally, selecting optimal hyperparameter values for DL models is essential since they affect model performance. The hyperparameter values of some of the existing DL‐based HAR models are chosen randomly or through the trial‐and‐error method. The random selection of these significant hyperparameters may be suitable for some applications, but sometimes it may worsen the model's performance in others. Hence, to address the above‐mentioned issues, this research aims to develop an optimized DL model capable of recognizing various human activities captured through IMU sensors. The proposed DL‐based HAR model combines convolutional neural network (CNN) layers and bidirectional long short‐term memory (Bi‐LSTM) units to simultaneously extract spatial and temporal sequence features from raw sensor data. The Rao‐3 metaheuristic optimization algorithm has been adopted to identify the ideal hyperparameter values for the proposed DL model in order to enhance its recognition performance. The proposed DL model's performance is validated on PAMAP2, UCI‐HAR, and MHEALTH datasets and achieved 94.91%, 97.16%, and 99.25% accuracies, respectively. The results reveal that the proposed DL model performs better than the existing state‐of‐the‐art (SoTA) models.

Liquidity and realized volatility prediction in Chinese stock market: A time-varying transitional dynamic perspective

Basing on the features of emerging Chinese stock market, this article discusses whether sharply deteriorating liquidity propels the stock market into a “crisis” state and investigates the dynamic impacts of the market liquidity on volatility forecasting. We construct the Markov-switching (MS) liquidity-adjusted HAR models with liquidity from the perspective of time-varying transition probabilities (TVTP). Empirical evidence suggests that a sharp deterioration in liquidity increases the probability of a “crisis” state for China's stock market. Out-of-sample forecasting results demonstrate that our proposed TVTP-MS-HAR-CJ-LIQ model, combining TVTP and MS-HAR-CJ with liquidity, substantially improves the predictive performance. Considering liquidity's impact from the TVTP perspective is suggested for the emerging but attention-attracting Chinese stock market.

Forecasting realized volatility through financial turbulence and neural networks

This paper introduces and examines a novel realized volatility forecasting model that makes use of Long Short-Term Memory (LSTM) neural networks and the risk metric Financial Turbulence (FT). The proposed model is compared to five alternative models, of which two incorporate LSTM neural networks and the remaining three include GARCH(1,1), EGARCH(1,1), and HAR models. The results of this paper demonstrate that the proposed model yields statistically significantly more accurate and robust forecasts than all other studied models when applied to stocks with middle-to-high volatility. Yet, considering low-volatility stocks, it can only be confidently affirmed that the proposed model yields statistically significantly more robust forecasts relative to all other models considered.

Integer-Valued HAR(p) model with Poisson distribution for forecasting IPO volumes

Integer-Valued HAR(<i>p</i>) model with Poisson distribution for forecasting IPO volumes

Sample frequency robustness and accuracy in forecasting Value-at-Risk for Brent Crude Oil futures

In this paper we examine how sensitive Value-at-Risk (VaR) forecasts based on simple linear quantile regressions are to the sampling frequency used to calculate realized volatility. We use sampling frequencies from one to 108 min for ICE Brent Crude Oil futures and test the out-of-sample performance of a set of quantile regression models using formal coverage tests. The results show that a one-factor model performs exceptionally well for most sampling frequencies used to calculate realized volatility. In comparison with the well-known Heterogeneous Auto-regressive Model of Realized Volatility (HAR-RV) and a quantile regression version of the HAR model (HAR-QREG), we also find that the one-factor model is much less sensitive to the sampling frequency used to calculate realized volatility.

Outlier-robust methods for forecasting realized covariance matrices

This paper proposes two new approaches to improve the estimation of the coefficients of the multivariate HAR (MHAR) model with the primary purpose of improving forecast performance. A robust estimator of the covariance matrix is adopted to replace the realized covariance matrix while estimating the MHAR model. The robustness to outliers of the new estimator makes the OLS estimation scheme for the MHAR model more reliable. In addition, a robust estimation scheme is developed for the MHAR model, which is based on the multivariate least-trimmed squares method. Both approaches provide significant improvements in forecasting performance based on both statistical loss and portfolio outcomes. The forecast performance of the multivariate HARQ model can also be improved with the proposed approaches, as evidenced by robustness checks.

Dorsal suspension in full-body harnesses: Comparison of perceived discomfort, pressure and physiological response

Falls from height are still a major occupational hazard, particularly in the construction industry, where it is the leading cause of fatal injuries. When opting for personal fall arrest systems, the use of a safety harness is mandatory. Following a fall arrest, motionless suspension in a full-body harness can be hazardous and lead to death. This paper investigates the effect of harness type, gender, and somatotype on perceived discomfort, the pressure exerted by the harness straps, and the evolution of physiological parameters in motionless suspension.For this study, 30 male and 30 female students (10 ectomorph, 10 mesomorph, and 10 endomorph for each gender) were suspended for a maximum of ten minutes from the dorsal D-ring, in four different harnesses. The selected harnesses for this study are commercially available, one of them is X-configuration, another is Y-configuration, and the other two are the more traditional H-configuration. The results show that the H-configuration harnesses, which are very common in the industry and often the cheapest models, are perceived as the least tolerable in suspension. Moreover, H-configuration harnesses lead to greater reduction in muscular blood oxygenation and greater increase in the heart rate, which could make this type of harness hazardous. The Y-configuration harness seems to be the best performer in suspension. The influence of the sub-pelvic strap on suspension tolerance is not very clear and could be studied further. Those results could be useful to develop new harness models that would increase the suspension time tolerance.