Fixed-parameter Model Research Articles

Enhancing cryptocurrency market volatility forecasting with daily dynamic tuning strategy

This study proposes a novel parameter tuning strategy, daily dynamic tuning, and applies it to forecast volatility in the cryptocurrency market. Comparative analysis with HAR-RV and machine learning models based on fixed parameters demonstrates the superior predictive performance of the dynamic tuning-based model. These findings are corroborated in robustness tests. Using interpretable tools for machine learning model explanations, this study delves further into the driving factors behind the predictive performance of the dynamic tuning-based model. The results indicate that hash rate and the effective federal funds rate play significant roles among various forecasting factors. Further categorization of the forecasting factors reveals that blockchain indicators and macroeconomic indicators are more useful for predictions, while market indicators are relatively less informative. Moreover, a time-varying analysis of variable importance scores suggests a significant increase in the importance score of blockchain indicators from October 2022 onwards, coinciding with the EU passing landmark legislation on digital assets in October 2022. Finally, an economic value analysis indicates that the dynamic tuning-based model can achieve higher economic returns in practice compared to traditional fixed parameter models. These findings are significant for enhancing cryptocurrency market risk prediction and management and exploring the potential applications of machine learning models in volatility forecasting.

Application of grey system model with intelligent parameters in predicting regional electricity consumption

PurposeGiven the regional heterogeneity of economic development, electricity consumption in various regions exhibits a discrepant growth pattern. The purpose of this study is to construct a grey system forecasting model with intelligent parameters for predicting provincial electricity consumption in China.Design/methodology/approachFirst, parameter optimization and structural expansion are simultaneously integrated into a unified grey system prediction framework, enhancing its adaptive capabilities. Second, by setting the minimum simulation percentage error as the optimization goal, the authors apply the particle swarm optimization (PSO) algorithm to search for the optimal grey generation order and background value coefficient. Third, to assess the performance across diverse power consumption systems, the authors use two electricity consumption cases and select eight other benchmark models to analyze the simulation and prediction errors. Further, the authors conduct simulations and trend predictions using data from all 31 provinces in China, analyzing and predicting the development trends in electricity consumption for each province from 2021 to 2026.FindingsThe study identifies significant heterogeneity in the development trends of electricity consumption systems among diverse provinces in China. The grey prediction model, optimized with multiple intelligent parameters, demonstrates superior adaptability and dynamic adjustment capabilities compared to traditional fixed-parameter models. Outperforming benchmark models across various evaluation indicators such as root mean square error (RMSE), average percentage error and Theil’s index, the new model establishes its robustness in predicting electricity system behavior.Originality/valueAcknowledging the limitations of traditional grey prediction models in capturing diverse growth patterns under fixed-generation orders, single structures and unadjustable background values, this study proposes a fractional grey intelligent prediction model with multiple parameter optimization. By incorporating multiple parameter optimizations and structure expansion, it substantiates the model’s superiority in forecasting provincial electricity consumption.

An Adaptive Hammerstein Model for FES-Induced Torque Prediction Based on Variable Forgetting Factor Recursive Least Squares Algorithm.

Modeling the muscle response to functional electrical stimulation (FES) is an important step during model-based FES control system design. The Hammerstein structure is widely used in simulating this nonlinear biomechanical response. However, a fixed relationship cannot cope well with the time-varying property of muscles and muscle fatigue. In this paper, we proposed an adaptive Hammerstein model to predict ankle joint torque induced by electrical stimulation, which used variable forgetting factor recursive least squares (VFFRLS) method to update the model parameters. To validate the proposed model, ten healthy individuals were recruited for short-duration FES experiments, ten for long-duration FES experiments, and three stroke patients for both. The isometric ankle dorsiflexion torque induced by FES was measured, and then the test performance of the fixed-parameter Hammerstein model, the adaptive Hammerstein model based on fixed forgetting factor recursive least squares (FFFRLS) and the adaptive Hammerstein model based on VFFRLS was compared. The goodness of fit, root mean square error, peak error and success rate were applied to evaluate the accuracy and stability of the model. The results indicate a significant improvement in both the accuracy and stability of the proposed adaptive model compared to the fixed-parameter model and the adaptive model based on FFFRLS. The proposed adaptive model enhances the ability of the model to cope with muscle changes.

Time-resolved dynamic computational modeling of human EEG recordings reveals gradients of generative mechanisms for the MMN response.

Despite attempts to unify the different theoretical accounts of the mismatch negativity (MMN), there is still an ongoing debate on the neurophysiological mechanisms underlying this complex brain response. On one hand, neuronal adaptation to recurrent stimuli is able to explain many of the observed properties of the MMN, such as its sensitivity to controlled experimental parameters. On the other hand, several modeling studies reported evidence in favor of Bayesian learning models for explaining the trial-to-trial dynamics of the human MMN. However, direct comparisons of these two main hypotheses are scarce, and previous modeling studies suffered from methodological limitations. Based on reports indicating spatial and temporal dissociation of physiological mechanisms within the timecourse of mismatch responses in animals, we hypothesized that different computational models would best fit different temporal phases of the human MMN. Using electroencephalographic data from two independent studies of a simple auditory oddball task (n = 82), we compared adaptation and Bayesian learning models' ability to explain the sequential dynamics of auditory deviance detection in a time-resolved fashion. We first ran simulations to evaluate the capacity of our design to dissociate the tested models and found that they were sufficiently distinguishable above a certain level of signal-to-noise ratio (SNR). In subjects with a sufficient SNR, our time-resolved approach revealed a temporal dissociation between the two model families, with high evidence for adaptation during the early MMN window (from 90 to 150-190 ms post-stimulus depending on the dataset) and for Bayesian learning later in time (170-180 ms or 200-220ms). In addition, Bayesian model averaging of fixed-parameter models within the adaptation family revealed a gradient of adaptation rates, resembling the anatomical gradient in the auditory cortical hierarchy reported in animal studies.

Joint Analysis of Crash Frequency by Severity Based on a Random Parameters Approach

Unobserved heterogeneity is a major challenge in estimating reliable road safety models. The random parameters approach has been proven to be an effective way to account for unobserved heterogeneity but has rarely been used for crash frequency by severity level. In this paper, a fixed parameters model, a basic random parameters model, and an improved random parameters model, allowing for heterogeneity in the means and correlation of random parameters, are estimated and comparatively evaluated. To quantitatively analyze the impact of explanatory variables on the crash frequency of various severity levels, the calculating method of marginal effects for estimated models is proposed. The results indicate that (1) the basic random parameters model statistically outperforms the fixed parameters model, and the statistical fit can be further improved by introducing heterogeneous means and correlation of random parameters; (2) for the predictive performance, the basic random parameters model is more accurate than the fixed parameters model, and the improved random parameters model can further reduce the mean error, mean absolute error, and root mean square error by 40–100%, 3.7–8.3%, and 7.6–8.9%, respectively; (3) ignoring the unobserved heterogeneity or neglecting the heterogeneity in the means and correlation of random parameters may result in biased safety inferences, and the maximum bias of marginal effects can easily exceed 100 percent; and (4) the safety effects of explanatory variables are thoroughly discussed and the potential safety countermeasures are provided. The random parameters approach and the method for calculating marginal effects proposed in this study are expected to provide a new methodological alternative for the joint analysis of crash frequency by severity and should be helpful in uncovering the mechanism of crash occurrence and the resulting injury severity accurately.

Stablecoins as the cornerstone in the linkage between the digital and conventional financial markets

Do stablecoins mean anything to risk management across financial markets? We answer this question by examining the interrelationships between stocks, treasuries, stablecoins, and cryptocurrencies in nonparametric quantile-causality-in-means and quantile-on-quantile regression models. Differently from previous studies of stablecoins’ price volatility, we focus on stablecoins’ trade volumes. The results extend the evidence on stablecoins’ responses to other crypto assets, whose declining returns, through the phenomena designated by us as “flight-to-cryptosafety”, fuel the stablecoins’ market capitalization. Issuance of new stablecoins, because of their backing by conventional safe assets, raises US Treasuries’ demand. And then, decaying Treasury yields cause “flight-to-safety” from stocks to Treasuries. Results from the quantile-based fixed-parameter models are robust to the dynamic connectedness metric based on the time-varying parameter vector auto-regression (TVP-VAR) framework.

Selection of gait parameters during constrained walking

It is commonly thought that at prescribed speeds humans choose gait parameters that minimize the cost of transportation. However, it is unclear whether and how the relationship between step length and step frequency is affected by the additional physiological factors caused by constraints. We performed a series of experiments to understand the selection of gait parameters under different constraints from a probabilistic perspective. First, we show that the effect of constraining step length on step frequency (i.e., monotonically decrease, Experiment I) is different from the effect of constraining step frequency on step length (i.e., inverted-U, Experiment II). Using the results from Experiment I and II, we summarized the marginal distribution of step length and step frequency and built their joint distribution in a probabilistic model. The probabilistic model predicts the selection of gait parameters by achieving the maximum probability of joint distribution of step length and step frequency. In Experiment III, the probabilistic model could well predict gait parameters at prescribed speeds, and it is similar to minimizing the cost of transportation. Finally, we show that the distribution of step length and step frequency were completely different between constrained and non-constrained walking. We argue that constraints in walking are major factors determining how humans choose gait parameters due to their involvement of mediators, i.e., attention or active control. Using the probabilistic model to account for gait parameters has an advantage compared with fixed-parameter models in that it can still include the effect of hidden mechanical, neurophysiological, or psychological variables by grouping them into distribution curves.

Learning Motion Constraint-Based Spatio-Temporal Networks for Infrared Dim Target Detections

Efficient infrared dim object detection has been challenged by low signal-to-noise ratios (SNRs). Traditional methods rely on the gradient difference and fixed-parameter model. These methods fail to adapt to sophisticated and variable situations in the real world. To tackle the issue, a deep learning method based on the spatio-temporal network is proposed in this paper. The model is established by the Convolutional Long Short-Term Memory cell (Conv-LSTM) and the 3D Convolution cell (3D-Conv). It is trained to learn the motion constraint of moving targets (spatio-temporal constraint module, called STM) and to fuse the multiscale local feature between the target and background (deep spatial features module, called DFM). In addition, a variable interval search module (state-aware module, called STAM) is added to the inference. The submodule decides to conduct a global search for images only if the target is lost due to fast motion, uncertain obstruction, and frame loss. Comprehensive experiments indicate that the proposed method achieves better performance over all baseline methods. On the mid-wave infrared datasets collected by the authors, the proposed method achieves a 95.87% detection rate. The SNR of the dataset is around 1–3 dB, and the background of the sequence includes sky, asphalt road, and buildings.

A FREQUENCY-DOMAIN ANALYSIS OF MEDIUM-SCALE DSGE MODELS

This paper employs the local Bayesian likelihood methodology to estimate a medium-scale dynamic stochastic general equilibrium (DSGE) model on different frequencies and uses frequency-domain tools to evaluate the time-varying parameter model and the fixed-parameter model. These techniques yield fresh insights into theoretical and empirical implications conveyed by alternative models beyond what conventional time-domain approaches can offer. We show that parameter estimates are sensitive to frequencies, and goodness-of-fit varies substantially with the frequency bands. Overall, the estimated time-varying parameter model captures the properties of the U.S. data better in the business cycle frequency band, and beyond this band, the fixed-parameter model performs better. Additionally, our study also reveals the importance of structural shocks in improving the fit between models and data. Finally, we utilize the spectral representation of generalized forecast error variance decomposition to investigate the frequency dynamics of volatility connectedness. We find shocks to economic activity have an impact on variables at different frequencies with different strengths, and markets become more connected during crisis periods. Furthermore, this study provides insights into a question policymakers are much concerned with: which shocks are major sources of economic volatilities and which sectors serve as major recipients of shocks?

Development of a Climate-Sensitive Matrix Growth Model for Larix gmelinii Mixed-Species Natural Forests and Its Application for Predicting Forest Dynamics under Different Climate Scenarios

Larix gmelinii natural forests, which are of great ecological and economic importance, are mainly distributed in northeast China. Sustainable management of these forests play a vital role in ecological security in northeast China, especially in the context of climate change. Forest growth models, which support forest management decision-making, are lacking for Larix gmelinii natural forests, hampering the prescription of forest management strategies. In this study, we produced a climate-sensitive, transition-matrix model (CM) for Larix gmelinii natural forests. For comparison, a variable transition model without including climate change effects (NCM) and a fixed-parameter model (FM) were also built. We examined the performance of the CM, NCM, and FM by conducting short- (5 years) and long-term (100 years) simulations. The results showed that for short-term prediction, no significant difference was observed among the three predictive models. However, the long-term prediction ability of the CM under the three different RCPs was superior to that of the FM and NCM. The number of trees and basal area were predicted to increase under climate change, which might result in natural disasters, such as snow break, windthrow, and forest fire. Silvicultural practices, such as reducing the intermediate thinning interval and the enrichment planting of slow-growing trees, should be implemented to mitigate the deleterious effects of climate change.

Semi‐supervised Robust Feature Selection with ℓq‐Norm Graph for Multiclass Classification

Flexible manifold embedding (FME) is a semi-supervised dimension reduction framework. It has been extended into feature selection by using different loss functions and sparse regularization methods. However, these kind of methods used the quadratic form of graph embedding, thus the results are sensitive to noise and outliers. In this paper, we propose a general semi-supervised feature selection model that optimizes an ℓ q -norm of FME to decrease the noise sensitivity. Compare to the fixed parameter model, the ℓ q -norm graph brings flexibility to balance the manifold smoothness and the sensitivity to noise by tuning its parameter. We present an efficient iterative algorithm to solve the proposed ℓ q -norm graph embedding based semi-supervised feature selection problem, and offer a rigorous convergence analysis. Experiments performed on typical image and speech emotion datasets demonstrate that our method is effective for the multiclass classification task, and outperforms the related state-of-the-art methods.

Heterogeneity assessment in incident duration modelling: Implications for development of practical strategies for small & large scale incidents

Accurate prediction of incident duration and response strategies are two imperative aspects of traffic incident management. Past research has applied various types of regression models for predicting incident durations and quantification of associated factors. However, an important methodological aspect is unobserved heterogeneity which may be present due to several unobserved/omitted factors. Incorporation of heterogeneity has significant potential to enhance predictive capabilities as well as obtain more robust insights for designing practical incident management strategies. This study uses two frequentist/quasi-Bayesian statistical techniques—simulation-assisted random parameter and quantile regression models, to address unobserved heterogeneity and to compare both methodologies with respect to the two aspects of incident management. By using 2015 Virginia incident data related to more than 45,000 incidents, the heterogeneous associations of incident durations with several factors including detection source, incident type, roadway type, temporal factors, and incident characteristics are explored. Specifically, as the name implies, quantile regression models associations between different quantiles of incident duration and explanatory factors. This facilitates designing more appropriate strategies for small, medium and large-scale incidents. Compared to quantile regression and fixed parameter models, random parameter models can potentially give more accurate predictions of incident durations. However, they do not (typically) capture different quantiles of incident durations. By characterizing and harnessing the unobserved heterogeneity, the study proposes the concept of Incident Reduction Factors that can assist traffic incident managers and practitioners in developing customized strategies for small- and large-scale incidents. The practical implications of results are discussed from the perspectives of travelers and incident managers.

Random parameters Bayesian hierarchical modeling of traffic conflict extremes for crash estimation

The use of Extreme Value Theory (EVT) models for traffic conflict-based crash estimation is becoming increasingly popular with considerable recent advances achieved. The latest advances include developing EVT models that combine several conflict indicators and the use of data from several sites to increase the sample size of conflict extremes. Nevertheless, one important issue while developing EVT models is accounting for the unobserved heterogeneity across different conflict observation sites and road user behaviours which can lead to biased and inefficient parameter estimates and erroneous inferences. This study proposes a random parameters (RP) Bayesian hierarchical extreme value modeling approach to account for the unobserved heterogeneity. The proposed approach is applied to estimate rear-end crashes from traffic conflicts collected from four signalized intersections in the city of Surrey, British Columbia. Traffic conflicts were characterized by four indicators: time to collision (TTC), modified TTC (MTTC), post-encroachment time (PET), and deceleration rate to avoid a crash (DRAC). MTTC was used to fit the generalized extreme value distribution, while the other three conflict indicators were treated as covariates. Six covariates including TTC, PET, DRAC, traffic volume, shock wave area, and platoon ratio were considered to account for non-stationarity in conflict extremes. Several RP, random intercepts (RI), and fixed parameters (FP) Bayesian hierarchical univariate extreme value models were developed. The results indicate that the RP model outperforms both the RI model and the FP model in terms of crash estimation accuracy and precision. Such superiority may be due to the ability of the RP model to better account for the unobserved heterogeneity.

Are petrol retailers less responsive to changes in wholesale or crude oil prices when they face lower competition? The case of Greater Sydney

Market efficiency is distorted when retailers are unwilling to pass on changes in wholesale costs or the price of oil (as a major production input cost) to motorists. Existing approaches to measuring pass-through parameters may obscure such behaviour because of the use of fixed-parameter models and/or aggregated price data across different locations and brands over time. We contribute to the literature by examining if there is a mismatch between wholesale and retail petrol prices across different brands supplying unleaded petrol within the same geographical location. To do so, we examine the effects of upstream shocks, arising from changes in oil or wholesale petrol prices, on the retail petrol prices of 672 stations in Sydney, Australia's largest city. We find that the pass-through parameters are significantly lower for stations with fewer competitors in their immediate proximity. When there are fewer than 13 petrol stations within a 5-km radius, or 33 stations within 10 km, irrespective of their brand, they become increasingly less responsive to changes in the wholesale petrol price or the price of crude oil. Our findings enhance market transparency and provide regulators with a rule of thumb that allows them to monitor the unleaded petrol market more effectively.

Comparative analysis of alternative random parameters count data models in highway safety

Unobserved heterogeneity has been a major challenge in developing reliable road safety models. A number of statistical techniques have been developed to account for unobserved heterogeneity and of which, the random parameters approach is one of the most effective method and has been frequently used recently. In this study, to ascertain the performance of various methods that can account for unobserved heterogeneity, the following six models in the negative binomial framework were estimated and thoroughly evaluated: 1) a fixed parameters model; 2) a random parameters model; 3) a random parameters model with heterogeneity in means; 4) a random parameters model with heterogeneity in means and variances; 5) a correlated random parameters model; 6) a correlated random parameters model with heterogeneity in means. These models were thoroughly evaluated from several angles including statistical fit, predictability, causality, marginal effects, explanatory power and practicality.Results indicate that: 1) the simple fixed parameters model resulted in reduced statistical fit, relatively inaccurate predictions, restrictive inferences and added risk of biased marginal effects, though the model is simple for application and interpretive analysis; 2) introducing random parameters could improve the goodness-of-fit, prediction accuracy and the ability to uncover causality; 3) compared with a basic random parameters model, considering the heterogeneity in means/variances or correlation of random parameters brought further improvement in statistical fit, predictive performance and causal inferences; 4) estimating complex models with heterogeneity in means/variances or correlated random parameters provided more insights, but at the expense of significant increase in computational cost and less intuitive outputs; 5) using global means of random parameters rather than the simulated observation-specific parameters was more likely to result in biased marginal effects and erroneous safety measures; 6) new insights into safety factors and their interactions were derived.Results from this study are expected to provide safety analysts with additional guidance for choosing appropriate models when unobserved heterogeneity exists. Additionally, the concluded interactions of safety factors can potentially help develop more effective safety measures.

Indoor Positioning System Using Dynamic Model Estimation.

Indoor Positioning Systems (IPSs) are used to locate mobile devices in indoor environments. Model-based IPSs have the advantage of not having an exhausting training and signal characterization of the environment, as required by the fingerprint technique. However, most model-based IPSs are done using fixed model parameters, treating the whole scenario as having a uniform signal propagation. This might work for most small scale experiments, but not for larger scenarios. In this paper, we propose PoDME (Positioning using Dynamic Model Estimation), a model-based IPS that uses dynamic parameters that are estimated based on the location the signal was sent. More specifically, we use the set of anchor nodes that received the signal sent by the mobile node and their signal strengths, to estimate the best local values for the log-distance model parameters. Also, since our solution depends highly on the selected anchor nodes to use on the position computation, we propose a novel method for choosing the three best anchor nodes. Our method is based on several data analysis executed on a large-scale, Bluetooth-based, real-world experiment and it chooses not only the nearest anchor but also the ones that benefit our least-square-based position computation. Our solution achieves a position estimation error of 3 m, which is better than a fixed-parameters model from the literature.

Assessing the impact of traffic signal performance on crash frequency for signalized intersections along urban arterials: A random parameter modeling approach

The recent development of Automated Traffic Signal Performance Measures (ATSPMs), has provided new opportunities and insights into traffic signal operations. As agencies begin to make decisions regarding investment in infrastructure and operation systems, it is imperative to understand the impacts these systems may have on safety. Past research has thoroughly investigated the impact of geometry and signal timing parameters on the safety of intersections, but little is understood on the relationship between improved signal performance and safety. This study uses vehicle trajectory data to create performance metrics for 121 signalized intersections on ten corridors near Columbus, Ohio. These metrics are used to understand the relationship between signal performance and safety. Two performance metrics, percent arrivals on green (POG) and level of travel time reliability (LOTTR), were used along with other volume and geometric data to model the total crash frequency on signalized mainline approaches. The crash data were modeled using a random parameters negative binomial approach. In consideration of potential unobserved heterogeneity between intersections, a correlated random parameters specification was tested alongside the traditional uncorrelated random parameters and fixed parameters model. Based on goodness of fit measures, the correlated random parameter model was chosen to interpret results because this model explains the complex cross-correlation among the estimates of random parameters. The elasticity values revealed a one percent increase in percent arrivals on green is associated with a reduction in total crashes by 1.12 %. The results of this study show the investment in signal operations and optimization result in an improvement in safety at signalized intersections. Further research should be explored to expand this study to additional intersections over a larger time period.

Radar Cross Section Near-Field to Far-Field Prediction for Isotropic-Point Scattering Target Based on Regression Estimation.

Radar cross section near-field to far-field transformation (NFFFT) is a well-established methodology. Due to the testing range constraints, the measured data are mostly near-field. Existing methods employ electromagnetic theory to transform near-field data into the far-field radar cross section, which is time-consuming in data processing. This paper proposes a flexible framework, named Neural Networks Near-Field to Far-Filed Transformation (NN-NFFFT). Unlike the conventional fixed-parameter model, the near-field RCS to far-field RCS transformation process is viewed as a nonlinear regression problem that can be solved by our fast and flexible neural network. The framework includes three stages: Near-Field and Far-field dataset generation, regression estimator training, and far-field data prediction. In our framework, the Radar cross section prior information is incorporated in the Near-Field and Far-field dataset generated by a group of point-scattering targets. A lightweight neural network is then used as a regression estimator to predict the far-field RCS from the near-field RCS observation. For the target with a small RCS, the proposed method also has less data acquisition time. Numerical examples and extensive experiments demonstrate that the proposed method can take less processing time to achieve comparable accuracy. Besides, the proposed framework can employ prior information about the real scenario to improve performance further.

Comparisons of deposit types and implications of the financial crisis: Evidence for U.S. banks

AbstractThis paper makes three diverse contributions. First, whereas the extant literature estimates a single elasticity of substitution/complementarity from an input distance function, we calculate a range of elasticities. Second, we make a substantive contribution to the literature on bank input substitution/complementarity because somewhat surprisingly there has been very little work on this issue. Third, our analysis of the substitutability/complementarity of deposit types for U.S. banks in 2008–2015 (crisis and beyond), vis‐à‐vis 1992–2007 (pre‐crisis), is, to the best of our knowledge, the first to consider the effect of structural change on elasticities of substitution/complementarity. To account for the extent of the heterogeneity in the U.S. banking industry we estimate random coefficients models, as opposed to standard fixed parameter models. The key empirical findings are the changes in the substitutability/complementarity of the quantities of particular pairs of deposit types between the two sample periods, which points to changes in depositors' preferences across banks' deposit portfolios. To illustrate, for savings deposits, which are characterized by flexibility and liquidity, and time deposits, which are less so and thus have higher interest rates, we find significantly lower quantity complementarity in 2008–2015. From this finding, we can conclude that savings and time deposits have become more distinct, which we suggest should be reflected in banks' strategic management of their deposit portfolios.

Volatility Prediction Using a Realized-Measure-Based Component Model

Abstract This article introduces a volatility model with a component structure allowing for a realized measure based on high-frequency data (e.g., realized variance) to drive the short-run volatility dynamics. In a joint model of the daily return and the realized measure, the conditional variance of the daily return has a multiplicative component structure: the first component traces long-run (secular) volatility trends, while the second component captures short-run (transitory) movements in volatility. Despite being a fixed-parameter model, its component structure implies time-varying parameters, which are “data-driven” to capture changing volatility regimes. We discuss the model dynamics and estimation by maximum likelihood. The empirical analysis reveals statistically significant out-of-sample gains compared to benchmark models, particularly for short forecast horizons and during the financial crisis.