A Comparison of State-of-the-Art Classification Techniques for Expert Automobile Insurance Claim Fraud Detection

Several state‐of‐the‐art binary classification techniques are experimentally evaluated in the context of expert automobile insurance claim fraud detection. The predictive power of logistic regression, C4.5 decision tree, k‐nearest neighbor, Bayesian learning multilayer perceptron neural network, least‐squares support vector machine, naive Bayes, and tree‐augmented naive Bayes classification is contrasted. For most of these algorithm types, we report on several operationalizations using alternative hyperparameter or design choices. We compare these in terms of mean percentage correctly classified (PCC) and mean area under the receiver operating characteristic (AUROC) curve using a stratified, blocked, ten‐fold cross‐validation experiment. We also contrast algorithm type performance visually by means of the convex hull of the receiver operating characteristic (ROC) curves associated with the alternative operationalizations per algorithm type. The study is based on a data set of 1,399 personal injury protection claims from 1993 accidents collected by the Automobile Insurers Bureau of Massachusetts. To stay as close to real‐life operating conditions as possible, we consider only predictors that are known relatively early in the life of a claim. Furthermore, based on the qualification of each available claim by both a verbal expert assessment of suspicion of fraud and a ten‐point‐scale expert suspicion score, we can compare classification for different target/class encoding schemes. Finally, we also investigate the added value of systematically collecting nonflag predictors for suspicion of fraud modeling purposes. From the observed results, we may state that: (1) independent of the target encoding scheme and the algorithm type, the inclusion of nonflag predictors allows us to significantly boost predictive performance; (2) for all the evaluated scenarios, the performance difference in terms of mean PCC and mean AUROC between many algorithm type operationalizations turns out to be rather small; visual comparison of the algorithm type ROC curve convex hulls also shows limited difference in performance over the range of operating conditions; (3) relatively simple and efficient techniques such as linear logistic regression and linear kernel least‐squares support vector machine classification show excellent overall predictive capabilities, and (smoothed) naive Bayes also performs well; and (4) the C4.5 decision tree operationalization results are rather disappointing; none of the tree operationalizations are capable of attaining mean AUROC performance in line with the best. Visual inspection of the evaluated scenarios reveals that the C4.5 algorithm type ROC curve convex hull is often dominated in large part by most of the other algorithm type hulls.     

A typology of supplier relations: from determinism to pluralism in inter-firm empirical research

In a survey of 196 buyer–supplier relationships in the UK printing industry, eight constructs regarding performance, trust and governance were operationalised. The paper presents the results of a correlation analysis and a cluster analysis. Both yield meaningful results, but the typology of supplier relations obtained by cluster analysis gives more pluralistic and practically relevant insights than the deterministic findings of correlation analysis. Buyers are classified as one of three types: traditional wary traders, committed flexible partners or controlled routine partners. There is a significant positive correlation between performance and trust, of which variance in governance is independent. Implications for management and further research are discussed.     

R&D investments and corporate cash holdings

This paper presents evidence about how research and development (R&D) expenditures affect corporate cash holdings in European country groups that differ in their innovation capacity. In theory, one can expect intangible investments such as R&D to result in higher cash stocks than fixed investments, particularly because intangible capital is less suitable as collateral for obtaining external funds. The relationship can be expected to be particularly strong in innovative countries. These countries carry out a relatively high proportion of cutting-edge R&D, which tends to be particularly risky and may be associated with substantial gestation lags before becoming productive. These features tend to increase firms' precautionary cash holdings. To investigate this issue in a European context, we examine different groups of countries that are clustered based on differences in their innovative capacity. Our estimation results confirm a positive relation between changes in R&D investment and changes in cash holdings, whereas changes in fixed investment do not appear to be related to changes in cash positions. The impact of changes in R&D on cash tends to be higher for country groups characterized by a high level of innovative capacity than for countries with moderate levels of innovative capacity. However, the differences across country groups are less pronounced than expected.     

Ingmar Prucha’s contributions to economics and econometrics

We study the link between two recent approaches to modeling emission-generating technologies: the by-production approach and the axiomatic approach. The by-production approach models these technologies as intersections of two independent sub-technologies reflecting (1) the relations between goods in intended-output production designed by human engineers and (2) the emission-generating mechanism of nature governed by material-balance considerations. The axiomatic approach proposes a set of axioms that a pollution-generating technology should satisfy. We show that the by-production technology satisfies these axioms and that, conversely, any technology satisfying the axioms can be decomposed into two sub-technologies satisfying the by-production properties. In either approach, the technology can be functionally represented by two radial distance functions with well-defined properties. These distance functions can also serve as measures of technological and environmental efficiency. We exploit the link between the by-production and axiomatic approaches to offer preliminary suggestions about suitable functional forms for the empirical estimation of the two distance functions.     

DISCRETE CHOICE MODELS WITH MULTIPLE UNOBSERVED CHOICE CHARACTERISTICS*

Since the pioneering work by Daniel McFadden, utility‐maximization‐based multinomial response models have become important tools of empirical researchers. Various generalizations of these models have been developed to allow for unobserved heterogeneity in taste parameters and choice characteristics. Here we investigate how rich a specification of the unobserved components is needed to rationalize arbitrary choice patterns in settings with many individual decision makers, multiple markets, and large choice sets. We find that if one restricts the utility function to be monotone in the unobserved choice characteristics, then up to two unobserved choice characteristics may be needed to rationalize the choices.     

Optimal simple rules and the lower bound on the nominal interest rate in the Christiano–Eichenbaum–Evans model of the US business cycle

Schmitt-Grohé and Uribe (NBER wp 10724, 2004b) analyzes the optimal, simple and implementable monetary policy rules in a medium-scale macromodel, as the one proposed by Christiano et al. (J Polit Econ 113:1–45, 2005). In doing so, they use a sensible, but somewhat arbitrary constraint to account for the lower bound condition on the nominal interest rate. In this work, we check the robustness of their main results to such a criteria. We find that the optimal policies are actually absolutely robust to the easing of this criterion for all the different cases considered.

Unemployment, growth and taxation in industrial countries

To the layman, the upward trend in European unemployment is related to the slowdown of economic growth. We argue that the layman's view is correct. The increase in European unemployment and the slowdown in economic growth are related, because they stem from a common cause: an excessively rapid growth in the cost of labour. In Europe, labour costs have gone up for many reasons, but one is particularly easy to identify: higher taxes on labour. If wages are set by strong and decentralized trade unions, an increase in labour taxes is shifted onto higher real wages. This has two effects. First, it reduces labour demand, and thus creates unemployment. Secondly, as firms substitute capital for labour, the marginal product of capital falls; over long periods of time, this in turn diminishes the incentive to invest and to grow. The data strongly support this view. According to our estimates, the observed rise of 14 percentage points in labour tax rates between 1965 and 1995 in the EU could account for a rise in EU unemployment of roughly 4 percentage points, a reduction of the investment share of output of about 3 percentage points, and a growth slowdown of about 0.4 percentage points a year.     

Labor turnover before plant closure: “Leaving the sinking ship” vs. “Captain throwing ballast overboard”

Involuntary job loss in administrative data is commonly identified by exploiting the exogenous nature of mass layoffs or plant closures. However, prior knowledge can lead to selection in the labor turnover of distressed firms. This paper proposes a simple method to determine if and when selective turnover sets in. Based on a rich administrative data set for Austria, we show that separations up to two quarters before plant closure should be included in the treatment group. Moreover, we find that early leavers are associated with significantly lower costs of job loss due to plant closure.     

Random incentive systems in a dynamic choice experiment

Experiments frequently use a random incentive system (RIS), where only tasks that are randomly selected at the end of the experiment are for real. The most common type pays every subject one out of her multiple tasks (within-subjects randomization). Recently, another type has become popular, where a subset of subjects is randomly selected, and only these subjects receive one real payment (between-subjects randomization). In earlier tests with simple, static tasks, RISs performed well. The present study investigates RISs in a more complex, dynamic choice experiment. We find that between-subjects randomization reduces risk aversion. While within-subjects randomization delivers unbiased measurements of risk aversion, it does not eliminate carry-over effects from previous tasks. Both types generate an increase in subjects’ error rates. These results suggest that caution is warranted when applying RISs to more complex and dynamic tasks.     

Taxing a monopolist

We consider a simple, self-financing and informationally undemanding scheme to reduce the deadweight loss due to a monopolist's market power. Essentially, we propose taxing the monopolist and applying the tax revenue to generate a public demand for his output. It turns out that a favorable scenario for such a reform to generate an ‘efficiency increase’ (i.e. to increase total output) is an elasticity of market demand with an absolute value of less than 3 (a seemingly ‘realistic’ condition). We also consider the case for the implementation of the first best, and compare specific and ad-valorem taxes as a way to finance the public demand.