Estimating the connectedness of commodity futures using a network approach

Using a network approach of variance decompositions, we measure the connectedness of 18 commodity futures and characterize both static and dynamic connectedness. Our results show that metal futures are net transmitters of shocks to other futures, and agricultural futures are vulnerable to shocks from the others. Furthermore, almost two-thirds of the volatility uncertainty for commodity futures are due to the connectedness of shocks across the futures market. Dynamically, we find connectedness always increases in times of turmoil. An analysis of connectedness networks suggests that investors could be forewarned that the connectedness of various classes of futures could threaten their portfolios.     

The performance of event study approaches using daily commodity futures returns

Simulations are conducted to assess the inferential accuracy of statistical event study approaches using daily futures returns. Methods examined include constant mean return models and several regression models—OLS, GARCH(1,1), and a GARCH(1,1) model having an error term with a Student's t distribution. The simulations address four of the most commonly analyzed agricultural futures commodities—corn, soybeans, live cattle, and hogs. In terms of the size of the test statistics, constant mean return models with short normal periods perform poorly, leading to unacceptably high rejection rates of the null hypothesis. Test statistics from constant mean return models with longer normal periods, OLS, and GARCH specifications provide rejection rates largely consistent with those of a unit normal distribution. Test statistics from all models are powerful enough to detect abnormal performance levels below those that would trigger limit locks. At small levels of abnormal performance the GARCH(1,1) model with a t distribution was consistently the most powerful model. © 2004 Wiley Periodicals, Inc. Jrl Fut Mark 24:533–555, 2004     

An empirical comparison of neural network and logistic regression models

The purpose of this paper is to critically compare a neural network technique with the established statistical technique of logistic regression for modeling decisions for several marketing situations. In our study, these two modeling techniques were compared using data collected on the decisions by supermarket buyers whether to add a new product to their shelves or not. Our analysis shows that although neural networks offer a possible alternative approach, they have both strengths and weaknesses that must be clearly understood.     

International market selection and subsidiary performance: A neural network approach

How should multinational enterprises (MNEs) select international markets? We develop a model of international market selection that adds firm-specific advantages and transaction cost considerations to previously explored target market factors based on Dunning's Eclectic Framework. Results obtained using neural network (NN) analysis indicates that our model has strong predictive power in explaining international market selection. Further tests show that firms selecting international markets predicted by the model reported significantly higher subsidiary performance relative to firms whose investments were not predicted by the model. Our results provide strong initial evidence that a firm-level strategic approach to international market selection facilitates MNE success.     

Information transmission in electronic versus open‐outcry trading systems: An analysis of U.S. equity index futures markets

In this article the intraday price discovery process between regular index futures (floor trading) and E‐mini index futures (electronic trading) in the S&P 500 and Nasdaq 100 index futures markets is examined, using intraday data from the introduction of the E‐mini index futures to 2001. Using both information shares (Hasbrouck, J., 1995) and common long‐memory factor weights (Gonzalo, J., & Granger, C. W. J., 1995) techniques, we find that both E‐mini index futures and regular index futures contribute to the price discovery process. However, since September 1998, the contribution made by E‐mini index futures has been greater than that provided by regular index futures. Based on regression analysis, we have also found direct empirical evidence to support the hypothesis that the joint effects of operational efficiency and relative liquidity determine the greater contribution made towards price discovery by electronic trading relative to open‐outcry trading over time. © 2005 Wiley Periodicals, Inc. Jrl Fut Mark 25: 679–715, 2005     

采用深度卷积神经网络方法的毫米波图像目标检测

通过收集大量的毫米波图像并建立相应的人体数据集进行检测,提出基于Faster R-CNN深度学习的方法检测隐藏于人体上的危险物品。该方法将区域建议网络和VGG19训练卷积神经网络模型相结合,构建了面向毫米波图像目标检测的深度卷积神经网络。为了提高毫米波图像的处理能力,采用Caffe深度学习框架在图形处理单元上进行训练和测试。实验结果证明了基于Faster R-CNN深度卷积神经网络的目标检测方法能有效检测毫米波图像中的危险物品,并且目标检测的平均准确率约94%,检测速度约为6 frame/s,对毫米波安检系统的智能化发展有着极其重要的参考价值。     

Investigating nonlinear speculation in cattle,corn, and hog futures markets using logistic smooth transition regression models

This study explores nonlinearities in the response of speculators' trading activity to price changes in live cattle, corn, and lean hog futures markets. Analyzing weekly data from March 4, 1997 through December 27, 2005, the authors reject linearity in all of these markets. Using smooth transition regression models, the authors found a similar structure of nonlinearities with regard to the number of different regimes, the choice of the transition variable, and the value at which the transition occurs. © 2007 Wiley Periodicals, Inc. Jrl Fut Mark 28:719–737, 2007     

Prediction of hotel bankruptcy using support vector machine,artificial neural network,logistic regression,and multivariate discriminant analysis

The objectives of this paper are firstly, to provide an optimal hotel bankruptcy prediction approach to minimize the empirical risk of misclassification and secondly, to investigate the functional characteristics of multivariate discriminant analysis, logistic, artificial neural networks (ANNs), and support vector machine (SVM) models in hotel bankruptcy prediction. The performances were evaluated not only in terms of overall classification and prediction accuracy but also in terms of relative error cost ratios. The results showed that ANN and SVM were very applicable models in bankruptcy prediction with data from Korean hotels. When jointly measuring both type I and type II errors, especially allowing for the greater costs associated with type I errors, however, ANN was more accurate with smaller estimated relative error costs than SVM. Thus, if the objective is to find the best early warning technique that performs accurately with small relative error costs, then, it will be worth considering ANN method for hotel bankruptcy prediction.     

An exploration of the patterns underlying related and unrelated collaborative ventures using neural network: Empirical investigation of collaborative venture formation data spanning 1985–2001

Collaborative ventures—both equity-based partnerships as well as project-based alliances—have dominated the international business scene over the past two decades. By means of this study we investigate the patterns of related and unrelated collaborative venture formation. Using a large database of over 90,000 collaborative ventures formed during the 1985–2001 period, this study clusters collaborative ventures on the basis of the industry group and home country relatedness of the collaborating partners. Self-organizing map technique within neural network methodology is used to accomplish this objective. The clusters obtained from the self-organizing map form the basis for developing taxonomy of collaborative ventures in which the neurons underlying clusters are classified based on the country of origin and industry affiliations of the collaborating partners and the collaborative venture. The distinguishing characteristics of the clusters and the taxonomy help augment our current understanding of the formation of collaborative ventures.