5,868 research outputs found
Save the best for last? The treatment of dominant predictors in financial forecasting
No full textWe study forecasting applications where the response variable is heavily correlated with one or a small set of covariates which we term dominant predictors. Dominant predictors commonly occur in financial forecasting where future market prices are heavily influenced by current prices, and to a much lesser degree, by many other, more subtle factors such as weather or calendar effects. We hypothesize that dominating predictors may mask the influence of the subtle factors, reducing forecasting accuracy. Consequently, we argue that it is crucial to find means of accurately accounting for the effect of the subtle factors on the response variable. To achieve this we present a two-stage modeling methodology which postpones the introduction of dominating predictors into the model building process until all predictive value from the other covariates has been extracted. To confirm our hypothesis and to test the effectiveness of the two-stage approach, we conduct an empirical study related to forecasting the outcome of sports events, which are well known to exhibit dominating predictors. Our results confirm that especially complex, nonlinear models are vulnerable to the masking effect and benefit from the two-stage paradigm. Our findings have important implications for forecasters who operate in environments where the influence of some predictors on the variable being forecast exceeds those of other covariates by a wide margin and we demonstrate appropriate ways to approach such forecasting tasks
The value of combining forecasts for profitable trading in the HK horserace betting market
No full textAdapting least-square support vector regression models to forecast the outcome of horseraces
No full textThis paper introduces an improved approach for forecasting the outcome of horseraces. Building upon previous literature, a state-of-the-art modelling paradigm is developed which integrates least-square support vector regression and conditional logit procedures to predict horses' winning probabilities. In order to adapt the least-square support vector regression model to this task, some free parameters have to be determined within a model selection step. Traditionally, this is accomplished by assessing candidate settings in terms of mean-squared error between estimated and actual finishing positions. This paper proposes an augmented approach to organise model selection for horserace forecasting using the concept of ranking borrowed from internet search engine evaluation. In particular, it is shown that the performance of forecasting models can be improved significantly if parameter settings are chosen on the basis of their normalised discounted cumulative gain (i.e. their ability to accurately rank the first few finishers of a race), rather than according to general purpose performance indicators which weight the ability to predict the rank order finish position of all horses equally
Identifying winners of competitive events: A SVM-based classification model for horserace prediction
No full textThe aim of much horserace modelling is to appraise the informational efficiency of betting markets. The prevailing approach involves forecasting the runners’ finish positions by means of discrete or continuous response regression models. However, theoretical considerations and empirical evidence suggest that the information contained within finish positions might be unreliable, especially among minor placings. To alleviate this problem, a classification-based modelling paradigm is proposed which relies only on data distinguishing winners and losers. To assess its effectiveness, an empirical experiment is conducted using data from a UK racetrack. The results demonstrate that the classification-based model compares favourably with state-of-the-art alternatives and confirm the reservations of relying on rank ordered finishing data. Simulations are conducted to further explore the origin of the model’s success by evaluating the marginal contribution of its constituent parts
Alternative methods of predicting competitive events: an application in horserace betting markets
No full textAccurately estimating the winning probabilities of participants in competitive events, such as elections and sports events, represents a challenge to standard forecasting frameworks such as regression or classification. They are not designed for modelling the competitive element, whereby a specific participant’s chance of success depends not only on his/her individual capabilities but also on those of his/her competitors. In this paper we consider this problem in the competitive context of horseracing and demonstrate how Breiman’s (2001) random forest classifier can be adapted in order to predict race outcomes. Several empirical experiments are undertaken which demonstrate the features of the adapted random forest procedure and confirm its effectiveness as a forecasting model. Specifically, we demonstrate that predictions derived from the proposed model can be used to make substantial profits, and that these predictions outperform those from traditional statistical techniques
Benchmarking classification models for software defect prediction: a proposed framework and novel findings
No full textSoftware defect prediction strives to improve software quality and testing efficiency by constructing predictive classification models from code attributes to enable a timely identification of fault-prone modules. Several classification models have been evaluated for this task. However, due to inconsistent findings regarding the superiority of one classifier over another and the usefulness of metric-based classification in general, more research is needed to improve convergence across studies and further advance confidence in experimental results. We consider three potential sources for bias: comparing classifiers over one or a small number of proprietary datasets, relying on accuracy indicators that are conceptually inappropriate for software defect prediction and cross-study comparisons, and finally, limited use of statistical testing procedures to secure empirical findings. To remedy these problems, a framework for comparative software defect prediction experiments is proposed and applied in a large-scale empirical comparison of 22 classifiers over ten public domain datasets from the NASA Metrics Data repository. Our results indicate that the importance of the particular classification algorithm may have been overestimated in previous research since no significant performance differences could be detected among the top-17 classifiers
A new methodology for generating and combining statistical forecasting models to enhance competitive event prediction
No full textForecasting methods are routinely employed to predict the outcome of competitive events (CEs) and to shed light on the factors that influence participants’ winning prospects (e.g., in sports events, political elections). Combining statistical models’ forecasts, shown to be highly successful in other settings, has been neglected in CE prediction. Two particular difficulties arise when developing model-based composite forecasts of CE outcomes: the intensity of rivalry among contestants, and the strength/diversity trade-off among individual models. To overcome these challenges we propose a range of surrogate measures of event outcome to construct a heterogeneous set of base forecasts. To effectively extract the complementary information concealed within these predictions, we develop a novel pooling mechanism which accounts for competition among contestants: a stacking paradigm integrating conditional logit regression and log-likelihood-ratio-based forecast selection. Empirical results using data related to horseracing events demonstrate that: (i) base model strength and diversity are important when combining model-based predictions for CEs; (ii) average-based pooling, commonly employed elsewhere, may not be appropriate for CEs (because average-based pooling exclusively focuses on strength); and (iii) the proposed stacking ensemble provides statistically and economically accurate forecasts. These results have important implications for regulators of betting markets associated with CEs and in particular for the accurate assessment of market efficiency
Benchmarking state-of-the-art classification algorithms for credit scoring: an update of research
No full textMany years have passed since Baesens et al. published their benchmarking study of classification algorithms in credit scoring [Baesens, B., Van Gestel, T., Viaene, S., Stepanova, M., Suykens, J., & Vanthienen, J. (2003). Benchmarking state-of-the-art classification algorithms for credit scoring. Journal of the Operational Research Society, 54(6), 627–635.]. The interest in prediction methods for scorecard development is unbroken. However, there have been several advancements including novel learning methods, performance measures and techniques to reliably compare different classifiers, which the credit scoring literature does not reflect. To close these research gaps, we update the study of Baesens et al. and compare several novel classification algorithms to the state-of-the-art in credit scoring. In addition, we examine the extent to which the assessment of alternative scorecards differs across established and novel indicators of predictive accuracy. Finally, we explore whether more accurate classifiers are managerial meaningful. Our study provides valuable insight for professionals and academics in credit scoring. It helps practitioners to stay abreast of technical advancements in predictive modeling. From an academic point of view, the study provides an independent assessment of recent scoring methods and offers a new baseline to which future approaches can be compared
A multi-objective approach for profit-driven feature selection in credit scoring
No full textIn credit scoring, feature selection aims at removing irrelevant data to improve the performance of the scorecard and its interpretability. Standard techniques treat feature selection as a single-objective task and rely on statistical criteria such as correlation. Recent studies suggest that using profit-based indicators may improve the quality of scoring models for businesses. We extend the use of profit measures to feature selection and develop a multi-objective wrapper framework based on the NSGA-II genetic algorithm with two fitness functions: the Expected Maximum Profit (EMP) and the number of features. Experiments on multiple credit scoring data sets demonstrate that the proposed approach develops scorecards that can yield a higher expected profit using fewer features than conventional feature selection strategies
- …
