A hybrid data mining model of feature selection algorithms and ensemble learning classifiers for credit scoring

Data mining techniques have numerous applications in credit scoring of customers in the banking field. One of the most popular data mining techniques is the classification method. Previous researches have demonstrated that using the feature selection (FS) algorithms and ensemble classifiers can improve the banks' performance in credit scoring problems. In this domain, the main issue is the simultaneous and the hybrid utilization of several FS and ensemble learning classification algorithms with respect to their parameters setting, in order to achieve a higher performance in the proposed model. As a result, the present paper has developed a hybrid data mining model of feature selection and ensemble learning classification algorithms on the basis of three stages. The first stage, as expected, deals with the data gathering and pre-processing. In the second stage, four FS algorithms are employed, including principal component analysis (PCA), genetic algorithm (GA), information gain ratio, and relief attribute evaluation function. In here, parameters setting of FS methods is based on the classification accuracy resulted from the implementation of the support vector machine (SVM) classification algorithm. After choosing the appropriate model for each selected feature, they are applied to the base and ensemble classification algorithms. In this stage, the best FS algorithm with its parameters setting is indicated for the modeling stage of the proposed model. In the third stage, the classification algorithms are employed for the dataset prepared from each FS algorithm. The results exhibited that in the second stage, PCA algorithm is the best FS algorithm. In the third stage, the classification results showed that the artificial neural network (ANN) adaptive boosting (AdaBoost) method has higher classification accuracy. Ultimately, the paper verified and proposed the hybrid model as an operative and strong model for performing credit scoring.     

Grundsätze ordnungsgemäßer Bewertung durch Scoring

Scoring and comparable methods that are used to rate people’s future behaviour can provide useful information for suppliers and thereby reduce the asymmetrical information deficits and the corresponding risks (screening). Likewise, consumers can profit from scoring if they use the scores to reveal their creditworthiness (signalling). In a market economy where markets are not fully transparent, scoring agencies act as information intermediaries. They provide information on future behaviour at the meta level so that not every supplier or consumer has to rate themselves. This is why the reliability of scores and scoring agencies is so fundamental. This article examines the principles or minimum standards for valid scoring and discusses how to introduce, regulate and test them.     

Voting in Hiring Committees: Which “Almost” Rule is Optimal?

We determine the scoring rule that is most likely to select a high-ability candidate. A major result is that neither the widely used plurality rule nor the inverse-plurality rule are ever optimal, and that the Borda rule is hardly ever optimal. Furthermore, we show that only the almost-plurality, the almost-inverse-plurality, and the almost-Borda rule can be optimal. Which of the “almost” rules is optimal depends on the likelihood that a candidate has high ability and how likely committee members are to correctly identify the abilities of the different candidates.     

Country selection of emerging equity markets: benefits from country attribute diversification

We propose a multivariate scoring model based on four classes of variables to predict future returns of 23 emerging equity markets. For the periods 1986–1995 and 1996–2003, our long–short portfolio (11 top/bottom ranked countries) posts a quarterly significant average raw return and a quarterly significant average market risk-adjusted return. The all-classes model dominates the one-class-models. Results from this strategy are robust regardless of whether we concentrate on the 6 top/bottom countries, reduce the emerging market universe to the largest countries, eliminate the most rewarding country during the period, use different scores, or account for realistic implementation costs.     

Behavioral credit scoring model for technology-based firms that considers uncertain financial ratios obtained from relationship banking

The Korea government offers technology credit guarantee service to many technology-based small and medium enterprises (SMEs) suffering from funding problems. Many advanced application credit scoring models have been developed based on technology to reduce the high default rates of this service. However, a credit scoring model which can reflect changes in firms after a loan has been granted has not yet been developed. In the study reported here, we propose a behavioral credit scoring model that reflects the debt-paying ability of recipient firms, which is observed as a time series of financial ratios of firms via the relationship banking activities. We utilize this time series, as well as missing patterns of financial information, as additional predictors of loan defaults. We compare our proposed behavioral credit scoring models, fitted at different points of elapsed time, to the application credit scoring model. Finally, we suggest the best behavioral credit scoring model for technology-based SMEs. Our study can contribute to the reduction of the risk involved in credit funding for technology-based SMEs.     

Credit Scoring and Loan Default

A metric of credit score performance is developed to study the usage and performance of credit scoring in the loan origination process. We examine the performance of origination FICO scores as measures of ex ante borrower creditworthiness using loan‐level data on ex post performance of subprime mortgages. Parametric and nonparametric estimates of credit score performance reveal different trends, especially on originations with low credit scores. The data suggest a trend of increased emphasis on higher credit scores accompanying a trend of increased riskiness in other origination attributes. Over time, this increased emphasis on credit scoring coincided with deterioration in FICO performance largely because of the fact that higher credit score originations of later cohorts were more likely to have riskier attributes. However, controlling for other attributes on originations and changes in economic conditions, we find that, as measures of borrower ranking, FICO performance on subprime loans over the years remains fairly stable.     

A Distance-Based Consensus Model with Flexible Choice of Rank-Position Weights

In this paper we propose a preference aggregation procedure for those cases in which the decision-makers express their preferences by means of a ranking of alternatives. Among the most applied methods for this purpose are those inspired by the Borda–Kendall rule, which attach to each alternative an aggregated value of the votes received in the different rank positions, and those based on distance measures between individual and collective preferences, which look for the solution that maximizes the consensus. The main idea here is to integrate these two approaches. Taking into account that the information about the values of weights or utilities assigned to each rank position is imprecise, we propose an evaluation of the alternatives using that vector of weights that minimizes the disagreement between DMs. In order to solve the problem, mixed-integer linear programming models are constructed. Two numerical examples are examined to illustrate the applicability of the proposed procedure.     

Assessing new empirical industrial organization (NEIO) methods: The cases of five industries

The methods proposed in the new empirical industrial organization (NEIO) literature have made significant contributions to our understanding of competitive behavior. However, these methods have yet to be compared with each other for their performance in explaining and diagnosing competitive market conduct. This inter-method comparison is important because conclusions about competitive behavior based on these methods have significant strategic as well as policy implications for firms. Our objective in this paper is to examine the performance of these different NEIO methods in terms of their discriminatory power, ability to identify strategic variables, and robustness in estimation. For empirical demonstration, we use data from diverse industries such as microprocessors, personal computers, facial tissue, disposable diapers and automobiles. Our results suggest that two commonly used NEIO methods-conjectural variation and non-nested model comparison-exhibit quite good convergence with each other and are consistent with a traditional time series method. This suggests that simpler methods such as conjectural variations deserve more credit. We also find that using these methods in tandem provides valuable additional information that may not be available when using any one method alone. While the emphasis in this study is on comparing different methods of analyzing competitive interaction, the findings also reveal some substantive insights about each market studied.     

A Comparison of Group Model Building and Strategic Options Development and Analysis

A distinctive tradition within group decision support uses models to structure managerial problems. In this tradition, stakeholders jointly construct a model on their issue of concern in facilitated workshops. In the past decades a wide variety of theoretical insights into and techniques for model-based decision support have been proposed and tested in practical applications. Methods are designed and used by experts; guidelines on their use are not completely spelled out in the literature. This lack of transparency may lead to difficulties in showing the value of methods to researchers in other fields, limit transferability of methods and complicate recombining elements of methods into a multimethodology. In this paper we aim to contribute to transparency by contrasting two model-driven methods: group model building (GMB) and Strategic Options Development and Analysis (SODA). We first develop a framework for comparing methods on a theoretical and practical level. Second, we describe the separate use of each approach, on one and the same issue, with a similar group of participants. By contrasting the choices made in a practical application we clarify process and results in different phases of problem analysis. Our conclusion is that theoretical assumptions of both approaches are more similar than expected. Each method captures different aspects of the problem and in this sense methods may supplement one another: where SODA focuses on the future and identification of actions, GMB aims to create insight into the relation between (past) behavior and structure of the problem. In choosing which element of the methods to use, it is important to realize that each element strikes a particular balance between costs (e.g. time taken from participants or modelers) and benefits (e.g. level of involvement or model verification). For instance, some elements speed up the process but do so at the cost of lowering participants’ involvement. A practical combination of elements of GMB and SODA thus requires the user to assess the relative importance of insight and action as project deliverables, weigh costs and benefits of elements of either method and string these together in a logical sequence that creates the outcomes required.     

Credit Where None Is Due? Authorized‐User Account Status and Piggybacking Credit

“Piggybacking credit” is a new practice that helps consumers improve their credit scores by paying to become “authorized users” on established accounts. Authorized users are not liable for paying an account, but because of Regulation B (which implements the 1974 Equal Credit Opportunity Act), the account's history factors into their credit scores. As a result piggybacking can be used to manipulate the signal of creditworthiness that scores provide and may help borrowers obtain credit for which they would not have otherwise qualified. This article investigates the policy questions raised by piggybacking. First, we evaluate whether the credit history disparities that motivated these provisions of Regulation B have persisted since they were written. Then, we assess the potential for score improvement through piggybacking. Finally, we evaluate the likely score effects of allowing credit scoring models to exclude authorized‐user accounts, the most widely proposed policy response to the emergence of piggybacking.     

A Hierarchical Clustering Approach Based on Three-Dimensional Gray Relational Analysis for Clustering a Large Group of Decision Makers with Double Information

Two types of information, collectively referred to as double information, are usually required in management decision-making. The first is preference information expressed in a judgment matrix. The second is reference information expressed in a multi-attribute decision matrix. In this paper, we investigate large-scale group clustering problems with double information in group decision-making. We first establish a novel three-dimensional gray correlation degree index, which integrates the alternative decision-making vector, index vector and alternative preference vector, to fully excavate the correlation between decision makers with double information. We then develop a new clustering procedure combining three-dimensional gray relational analysis and the concept of hierarchical clustering. Moreover, a model for determining clustering centers is established on the basis of the maximum gray correlation degree within each cluster and minimum gray correlation degree among clusters. A heuristic algorithm for the model to identify the core decision maker in each cluster is proposed. Finally, we illustrate the applications of the developed procedures with a practical case. The rationality of the proposed method is demonstrated by comparing results with results obtained using other methods, including the traditional gray clustering method and hierarchical clustering method with single information; i.e., preference information or reference information.     

A Comparative Analysis of the Use of Corporate Advertising in the United States and Japan

This article outlines one possible approach that companies can use to identify candidate countries for standardized international advertising campaigns. By clustering countries based on their similarity economically, culturally, and in their media availability and usage, marketers can identify those in which similar campaigns may be successful. In this paper, 40 countries were clustered into six groups within each of which standardization could be attempted. Discriminant analysis provided validation for the appropriateness of this six cluster solution. The implications of these results are discussed and directions for future research based on this approach are suggested.     

Analyzing Ordinal Data for Group Representation

With n individuals ranking m objects, the exhaustive comparison approach, proposed in this paper, produces a list of order vectors sorted by the relative number of concordant pairs. The exhaustive comparison approach compares all possible order vectors instead of "an" optimal order vector to help the data analyst to consider "practical" solutions rather than a "desired" solution. An overall concordant order ratio is proposed to measure "how well" each order vector may represent the ranking structure of an ordinal data set. And the marginal concordant ratio evaluating the goodness of fit of each object in each order vector is also proposed in this paper. Comparisons among some popular ranking methods are discussed in this article. An empirical survey data regarding how travellers considered various factors for choosing travelling locations are used to illustrate the proposed method and calculations.     

Mean‐field games with differing beliefs for algorithmic trading

Even when confronted with the same data, agents often disagree on a model of the real world. Here, we address the question of how interacting heterogeneous agents, who disagree on what model the real world follows, optimize their trading actions. The market has latent factors that drive prices, and agents account for the permanent impact they have on prices. This leads to a large stochastic game, where each agents performance criteria are computed under a different probability measure. We analyze the mean‐field game (MFG) limit of the stochastic game and show that the Nash equilibrium is given by the solution to a nonstandard vector‐valued forward–backward stochastic differential equation. Under some mild assumptions, we construct the solution in terms of expectations of the filtered states. Furthermore, we prove that the MFG strategy forms an ε‐Nash equilibrium for the finite player game. Finally, we present a least square Monte Carlo based algorithm for computing the equilibria and show through simulations that increasing disagreement may increase price volatility and trading activity.     

RFM-based repurchase behavior for customer classification and segmentation

Customer behavior modeling and classification are well-studied areas for applications in retail. Past studies implemented the purchase behavior modeling based on the physical behavior of a subject. In this research, we apply the recency, frequency, and monetary (RFM) model and data modeling techniques to detect behavior patterns for a customer. Each transaction attributed to a customer is part of one's behavior, and an instance of the feature vector, it is modeled on a set of transactions to constitute repurchase behavior. The proposed scheme is validated by simulating a publicly accessible real-world data set with a need-tailored multi-layer perceptron (MLP) and also support vector machine (SVM) and decision tree classification (DTC) methods. The experiments yield a high customer classification rate of more than 97% for the different numbers of the customers. Empirical analysis shows that eight transactions are sufficient to classify a customer with high accuracy.     

A Multi-Agent Model for Overlapping Negotiations

In the last few years, research on multi-agent systems has addressed different aspects of intelligent negotiations using methods developed in different domains including game theory, decision theory, and economic models. The research proposed in Andersson and Sandholm (1999), Sandholm (1993), Sandholm and Lesser (1995) and Aknine et?al. (2004), Aknine (2002) are significant examples. However, only some of this work focuses on problems related to complex negotiations, particularly those concerning new generation applications. This new research raises fundamental difficulties we have encountered, especially in overlapping negotiations and combined negotiations. This article is interested essentially in overlapping negotiations, which include several agent roles in a same negotiation. One or more agents may play each of these roles. This work shows that the high-level negotiation models are necessary in order to control the execution of overlapping negotiation processes, since, in these negotiations we are facing both classical problems of multi-agent negotiations based on two agents’ roles and the problems concerned with the interdependence of these negotiations. Synchronization of these different processes is thus necessary because of the multiplicity of the roles. Thus, this paper presents a formalized negotiation model, which deals with this problem. It gives a theoretical analysis of the suggested model and discusses the results of the experimental evaluation. To perform this evaluation, we use the application of intelligent service agencies on the Internet.     

React to the future: political visualization,emotional reactions and political behavior

Abstract

This study explores the mediating role political visualization – the process of imagining future political scenarios – plays in determining how political advertising affects voting behaviour. Specifically, we theorize that when partisans are exposed to political ads that are narrative (compared to non-narrative) in nature, they will engage in more political visualization. Partisans will then experience emotional reactions to these imagined futures – specifically, enthusiasm for the in-group candidate and anger towards the out-group candidate. These emotional reactions, in turn, will make a partisan more likely to vote for the in-group candidate and less likely to vote for the out-group candidate. We test this model by employing an experimental design where American partisans were presented a political ad (in the form of an email) that is either narrative or non-narrative. Results provide support for most of our expectations and suggest that visualization may play an important role in determining the influence of a political ad.     

The MARS Approach in the Verbal and Holistic Evaluation of the Negotiation Template

The negotiation template, which defines a set of potential negotiation offers, is traditionally evaluated by means of the simple additive weighting method (SAW). However, some recent research reports on the potential problems and inconsistencies in using and interpreting SAW-based scores. Thus, in this paper we consider the issue of evaluating negotiation offers when the negotiator’s preferences are expressed verbally. We present a new approach called Measuring Attractiveness near Reference Situations (MARS), which combines the algorithms of two multiple criteria decision making methods: ZAPROS and MACBETH. Applying the elements of ZAPROS allows identifying a small set of reference alternatives that consists of the best resolution levels for all the negotiation issues but one. In pair-wise comparisons of these alternatives negotiators need to evaluate trade-offs only, which means deciding which concessions are better to be made. Using the elements of MACBETH allows determining the strong interval scale based on verbal judgments defined by negotiators at the beginning of the preference elicitation process. We study in detail the legitimacy of hybridizing ZAPROS and MACBETH that differ in their philosophies of decision support as well as discuss the drawbacks of these two MCDM methods and propose some alternative solutions that make this approach applicable to supporting negotiators in the evaluation of negotiation offers. Finally, we present an example in which we indicate the differences in the negotiation offers’ scoring process conducted by means of MARS and the traditional ZAPROS and MACBETH procedures.     

A Multivariate Stochastic Volatility Model Applied to a Panel of S&P500 Stocks in Different Industries

We estimate a multivariate stochastic volatility model for a panel of stock returns for a number of S&P 500 firms from different industries. To directly compare our results with those from the univariate estimation literature on the same data, we use an efficient importance sampling (EIS) method to estimate the likelihood function of the given multivariate system that we analyze. As opposed to univariate methods where each return is estimated separately for each firm, our results are based on joint estimation that can account for potential common error term interactions based on industry characteristics that cannot be detected by univariate methods. Our results reveal that there are important differences in the industry effects, something that suggests that differential gains to portfolio allocations in the different industries that we examine. There are differences because of idiosyncratic factors and the common industry factors that suggest that each industry requires a separate treatment in arriving at portfolio allocations.     

文化创意企业智力资本对企业绩效影响的实证分析

运用 VAIC法对文化创意企业的智力资本进行测量,基于平衡计分卡的思想构建文化创意企业绩效评价指标体系,运用主因子赋权和分析法计算文化创意企业绩效,并以文化创意企业绩效作为因变量,以文化创意企业智力资本作为自变量,构建线性回归模型对文化创意企业智力资本对企业绩效的影响进行实证分析。分析结果表明：智力资本对文化创意企业绩效的影响作用已经大于物质资本,是企业绩效的主要驱动因素;智力资本的构成要素人力资本和结构资本对文化企业绩效发挥的作用程度有所不同,结构资本对企业绩效发挥了积极的促进作用,人力资本影响作用不大;文化创意企业各类资本之间具有相关性。