Intelligent Systems in Accounting, Finance and Management最新文献

Although numerous scientific papers have been written on deep learning, very few have been written on the exploitation of such technology in the field of accounting or bookkeeping. Our scientific study is oriented exactly toward this specific field. As accountants, we know the problems faced in modern accounting. Although accountants may have a plethora of information regarding technology support, looking for errors or fraud is a demanding and time-consuming task that depends on manual skills and professional knowledge. Our efforts are oriented toward resolving the problem of error-detection automation that is currently possible through new technologies, and we are trying to develop a web application that will alleviate the problems of journal entry anomaly detection. Our developed application accepts data from one specific enterprise resource planning system while also representing a general software framework for other enterprise resource planning developers. Our web application is a prototype that uses two of the most popular deep-learning architectures; namely, a variational autoencoder and long short-term memory. The application was tested on two different journals: data set D, learned on accounting journals from 2007 to 2018 and then tested during the year 2019, and data set H, learned on journals from 2014 to 2016 and then tested during the year 2017. Both accounting journals were generated by micro entrepreneurs.

This study aims to compare the performances of logistic regression and random forest classifiers in a balanced oversampling procedure for the prediction of households that will face catastrophic out-of-pocket (OOP) health expenditure. Data were derived from the nationally representative household budget survey collected by the Turkish Statistical Institute for the year 2012. A total of 9,987 households returned valid surveys. The data set was highly imbalanced, and the percentage of households facing catastrophic OOP health expenditure was 0.14. Balanced oversampling was performed, and 30 artificial data sets were generated with sizes of 5% and 98% of the original data size. The balanced oversampled data set provided accurate predictions, and random forest exhibited superior performance in identifying households facing catastrophic OOP health expenditure (area under the receiver operating characteristic curve, AUC = 0.8765; classification accuracy, CA = 0.7936; sensitivity = 0.7765; specificity = 0.8552; F₁ = 0.7797).

Through an international Delphi study, this article explores the new electronic human resource management regimes that are expected to transform internal staffing. Our focus is on three types of information systems: human resource management systems, job portals, and talent marketplaces. We explore the future potential of these new systems and identify the key challenges for their implementation in governments, such as inadequate regulations and funding priorities, a lack of leadership and strategic vision, together with rigid work policies and practices and a change-resistant culture. Tied to this vision, we identify several areas of future inquiry that bridge the divide between theory and practice.

This study proposes an investment recommendation model for peer-to-peer (P2P) lending. P2P lenders usually are inexpert, so helping them to make the best decision for their investments is vital. In this study, while we aim to compare the performance of different artificial neural network (ANN) models, we evaluate loans from two perspectives: risk and return. The net present value (NPV) is considered as the return variable. To the best of our knowledge, NPV has been used in few studies in the P2P lending context. Considering the advantages of using NPV, we aim to improve decision-making models in this market by the use of NPV and the integration of supervised learning and optimization algorithms that can be considered as one of our contributions. In order to predict NPV, three ANN models are compared concerning mean square error, mean absolute error, and root-mean-square error to find the optimal ANN model. Furthermore, for the risk evaluation, the probability of default of loans is computed using logistic regression. Investors in the P2P lending market can share their assets between different loans, so the procedure of P2P investment is similar to portfolio optimization. In this context, we minimize the risk of a portfolio for a minimum acceptable level of return. To analyse the effectiveness of our proposed model, we compare our decision-making algorithm with the output of a traditional model. The experimental results on a real-world data set show that our model leads to a better investment concerning both risk and return.

This paper investigates the way that minimum tick size affects market quality based on an agent-based artificial stock market. Our results indicate that stepwise and combination systems can promote market quality in certain aspects, compared with a uniform system. A minimal combination system performed the best to improve market quality. This is the first study to analyse tick size systems that remain at the theory stage and compare four types of system under the same experimental environment. The results suggests that a minimal combination system could be considered a new direction for market policy reform to improve market quality.

This research developed and tested machine learning models to predict significant credit card fraud in corporate systems using Sarbanes-Oxley (SOX) reports, news reports of breaches and Fama-French risk factors (FF). Exploratory analysis found that SOX information predicted several types of security breaches, with the strongest performance in predicting credit card fraud. A systematic tuning of hyperparamters for a suite of machine learning models, starting with a random forest, an extremely-randomized forest, a random grid of gradient boosting machines (GBMs), a random grid of deep neural nets, a fixed grid of general linear models where assembled into two trained stacked ensemble models optimized for F1 performance; an ensemble that contained all the models, and an ensemble containing just the best performing model from each algorithm class. Tuned GBMs performed best under all conditions. Without FF, models yielded an AUC of 99.3% and closeness of the training and validation matrices confirm that the model is robust. The most important predictors were firm specific, as would be expected, since control weaknesses vary at the firm level. Audit firm fees were the most important non-firm-specific predictors. Adding FF to the model rendered perfect prediction (100%) in the trained confusion matrix and AUC of 99.8%. The most important predictors of credit card fraud were the FF coefficient for the High book-to-market ratio Minus Low factor. The second most influential variable was the year of reporting, and third most important was the Fama-French 3-factor model R² – together these described most of the variance in credit card fraud occurrence. In all cases the four major SOX specific opinions rendered by auditors and the signed SOX report had little predictive influence.