Huiwen Wang

An investigation and comparison of artificial neural network and time series models for Chinese food grain price forecasting

This paper compares the predictive performance of ARIMA, artificial neural network and the linear combination models for forecasting wheat price in Chinese market. Empirical results show that the combined model can improve the forecasting performance significantly in contrast with its counterparts in terms of the error evaluation measurements. However, as far as turning points and profit criterions are concerned, the ANN model is best as well as at capturing a significant number of turning points. The results are conflicting when implementing dissimilar forecasting criteria (the quantitative and the turning points measurements) to evaluate the performance of three models. The ANN model is overall the best model, and can be used as an alternative method to model Chinese future food grain price.

A hyperspherical transformation forecasting model for compositional data

Abstract Although Aitchison’s [Aitchison, J., 1986. The Statistical Analysis of Compositional Data, Chapman and Hall, London] method of logratio transformation of compositional data is widely used in various domains, it is limited by the assumption of a strict non-negativity of the components and the requirement of special treatments in practice of the zero components. We propose a dimension-reduction approach through a hyperspherical transformation that is capable of resolving the difficulty in maintaining non-negativity and unit-sum in forecasting compositional data over time. Applying the proposed model to a numerical simulation with a 4D compositional data embedded with zero components and forecasting the three production sectors in the Chinese economy both demonstrate the usefulness and validity of the new approach.

CIPCA: Complete-Information-based Principal Component Analysis for interval-valued data

Principal Component Analysis (PCA) has long been used as a tool in exploratory data analysis and for making predictive models. Recent years have witnessed the continuous emergence of huge-volume data from various computerized industries, which triggers the call for more efficient and effective PCA methods. In light of this, in this paper, we work on interval-valued data and propose a new PCA method called CIPCA. CIPCA discriminates itself from various well-established methods, e.g., VPCA and CPCA, in that it can capture the complete information in interval-valued observations. Taking a hypercube view with infinitely dense points uniformly distributed within the hypercubes, CIPCA defines the inner product of interval-valued variables, and transforms the PCA modeling into the computation of some inner products in the covariance matrix. Both comparative experiments with VPCA and CPCA on the synthetic data sets and applications on real-world data demonstrate the merits of CIPCA in modeling interval-valued data. In particular, CIPCA provides an efficient and effective way for conducting PCA for large-scaled numerical data, and can find the meaningful structure information hidden in massive data.

Interpretation of partial least-squares regression models with VARIMAX rotation

Abstract The VARIMAX rotation for factor analysis is used to orthogonally transform the factor subspace, resulting from partial least-square regression (PLSR). If the factors are nearly orthogonal, the transformation may help to interpret the physical meaning of each factor without altering the results of a PLSR model. A case study shows that after the VARIMAX rotation, the loading matrix satisfies “the simple structure criterion” and improves its explanatory ability.

Multiple linear regression modeling for compositional data

Compositional data, containing relative information, occur regularly in many disciplines and practical situations. Multivariate statistics methods including regression analysis have been adopted to model compositional data, but the existing research is still scattered and fragmented. This paper contributes to modeling the linear regression relationship for compositional data as both dependent and independent variables. First, some operations in Simplex space, such as the perturbation operation, the power transformation, and the inner product, are defined for compositional-data vectors. The regression models are then built by the original compositional data and transformed data, respectively, after the introduction of the Isometric Logratio Transformation (ilr). By theoretical inference, it turns out that the two models are equivalent in essence using the ordinary least squares (OLS) method. Two measures for testing goodness of fit, i.e., the observed squared correlation coefficient R^2 and the cross validated squared correlation coefficient Q^2, are also proposed to evaluate the regression models. Besides, the estimated regression parameters are explained to indicate the notion of relative elasticity. An empirical analysis finally illustrates the usefulness of the multiple linear regression models for compositional-data variables.

Updating Input–Output Tables with Benchmark Table Series

Numerous methods have been proposed to update input–output (I–O) tables. They rely on the assumption that the economic structure will not change significantly during the interpolation period. However, this assumption may not always hold, particularly for countries experiencing rapid development. This study attempts to combine forecasting with a matrix transformation technique (MTT) to provide a new perspective on updating I–O tables. Under the assumption that changes in the trend of an economic structure are statistically significant, the method extrapolates I–O tables by combining time series models with an MTT and proceeds with only the total value added during the target years. A simulation study and empirical analysis are conducted to compare the forecasting performance of the MTT to the Generalized RAS (GRAS) and Kuroda methods. The results show that the comprehensive performance of the MTT is better than the performance of the GRAS and Kuroda methods, as measured by the Standardized Total Percentage Error, Theils U and Mean Absolute Percentage Error indices.

Principal component analysis for probabilistic symbolic data: a more generic and accurate algorithm

In the symbolic data framework, probabilistic symbolic data are considered as those whose components are random variables with general probability distributions. Intervals (or uniform distributions), histograms (or empirical distributions), Gaussian distribution and Chi-squared distribution are all the special cases of them. The existing approaches devoted to the subject have a common shortcoming since they can not obtain the distributions of linear combinations (i.e., principal components) of random variables especially for not identically distributed ones. This paper will overcome the shortcoming by providing an exact probability density function for each principal component by using the inversion theorem. Further, the paper defines a covariance matrix for probabilistic symbolic data and presents a new principal component analysis based on this variance–covariance structure. The effectiveness of the proposed method is illustrated by a simulated numerical experiment, and two real-life cases including clustering of oils and fats data, and evaluation of indexed journals of Science Citation Index.

Principal Component Analysis for Normal-Distribution-Valued Symbolic Data

This paper puts forward a new approach to principal component analysis (PCA) for normal-distribution-valued symbolic data, which has a vast potential of applications in the economic and management field. We derive a full set of numerical characteristics and variance-covariance structure for such data, which forms the foundation for our analytical PCA approach. Our approach is able to use all of the variance information in the original data than the prevailing representative-type approach in the literature which only uses centers, vertices, etc. The paper also provides an accurate approach to constructing the observations in a PC space based on the linear additivity property of normal distribution. The effectiveness of the proposed method is illustrated by simulated numerical experiments. At last, our method is applied to explain the puzzle of risk-return tradeoff in Chinas stock market.

Impact of US financial crisis on Different Countries: Based on the Method of Functional Analysis of Variance

Abstract During the entire period of the 2007-2009 global financial crisis, different types of countries showed different characteristics on their economic development process. Comparing the economic development process between different types of countries contributes a lot to get an in-depth understanding of the different impacts of the crisis on national economy. In this paper, the method of Functional Analysis of Variance (FANOVA) is applied to make a comparative study on the economic development process of different types of countries, including the differences on the economic growth rate, the time of the economy recession, the extent of the recession and the recovery situation of the economy. Moreover, the paper performs a dynamic test on the significance of the difference on the economic growth rate during the whole stage.

Cone dominance and efficiency in DEA

This paper incorporates cones on virtual multipliers of inputs and outputs into DEA analysis. Cone DEA models are developed to generalize the dual of the BCC models as well as congestion models. Input-output data and/or numbers of DMUs for BCC models are inadequate to capture many aspects where judgments, expert opinions, and other external information should be taken into analysis. Cone DEA models, on the other hand, offer improved definitions of efficiency over general cone and polyhedral cone structures. The relationships between cone models and BCC models as well as those between cone models and congestion models are discussed in the development. Two numerical examples are provided to illustrate our findings.