Intelligent Systems in Accounting, Finance and Management最新文献

Google’s LaMDA, Open AI’s GPT-3, and Meta’s BlenderBot are artificial intelligence (AI)-based chatbots, that have been trained on billions of documents creating the notion of “massive data.” These systems use human-generated documents to capture words and relationships between words that people use when they communicate. This paper examines some of the similarities of these systems and the emerging issues regarding these massive data language models, including whether they are sentient, the use and impact of scale, information use and ownership, and explanations of discussions and answers. This paper also directly investigates some artifacts of Google’s LaMDA and compares them with Meta’s BlenderBot. Finally, this paper examines emerging issues and questions deriving from our analysis.

Small and medium-sized enterprises are the pillars of an economy, and their poor performance has a negative impact on living standards of population and country development. This study analyzes real-life data of 89,851 small and medium-sized enterprises, out of which 295 have declared bankruptcy. The analysis is performed via 27 financial ratios. The study framework combines seven classifications and three resampling and seven feature selection methods. Out of all classification methods applied, CatBoost has achieved the best results for all combinations of resampling and feature selection methods. CatBoost surpassed the results of other classification methods for the area under curve parameter, achieving a value of 99.95%. The application of resampling methods on different classification models has not identified a statistically significant level of improvement in any of the resampling methods. This finding has also been observed for feature selection methods. Based on these findings, we assume that individual resampling and feature selection methods do not improve model performance compared with the original imbalanced sample's results. Our results suggest that, even though the data sample may be significantly imbalanced with a minority of bankrupt companies, most classification algorithms can handle this imbalance and achieve interesting results. Moreover, our findings provide broad practical application for all stakeholders who could need to detect bankrupting companies.

Prediction and the stock market go hand in hand. Due to the inherent limitations of traditional forecasting methods and the pursuit to uncover the hidden patterns in stock market data, stock market prediction using data mining techniques has caught the fancy of academicians, researchers, and investors. Based on a systematic review of more than 143 research studies spanning 25 years, the present paper brings to light the major issues concerning forecasting of stock markets based on data mining techniques, such as usage of data mining techniques in the stock market, input data types, single versus hybrid techniques, instruments and stock markets researched, types of software and algorithms used, measures of forecast accuracy, and performance of various data mining techniques. Emerging patterns related to various dimensions have been critically analyzed by highlighting the existing limitations and suggesting future research paradigms. This analysis can be useful for academicians, researchers and investors looking for futuristic directions in a given research domain.

Facial expression recognition (FER) has drawn the interest of practitioners and researchers due to its potential in opening new business opportunities. One critical aspect of any successful FER system is a feature extraction method that can efficiently find sufficient facial features and characterize facial expressions. This paper proposes an appearance-based feature extraction method by introducing a local feature descriptor, a multilayer-neighbor local binary pattern (LBP), for recognizing facial expressions. This new LBP operator is an extension of the original one-layer-neighbor LBP to two-layer-neighbor and three-layer-neighbor LBPs. We extract features by comparing new center points with neighborhood points. In addition, based on facial landmark locations, we extract active facial blocks during emotional stimulations. These prominent facial blocks utilize facial symmetry to improve the accuracy and speed of expression recognition. After using principal component analysis to reduce the dimensionality of features, we use a support vector machine to assign expressions to seven categories. We evaluate the proposed method by comparing it with other commonly used methods, and the proposed method is more accurate. Implications for business researchers are discussed.

Agricultural commodity price forecasting represents a key concern for market participants. We explore the usefulness of neural network modeling for forecasting problems in datasets of daily prices over periods of greater than 50 years for coffee, corn, cotton, oats, soybeans, soybean oil, sugar, and wheat. By investigating different model settings across the algorithm, delay, hidden neuron, and data-splitting ratio, we arrive at models leading to a decent performance for each commodity, with the overall relative root mean square error ranging from 1.70% to 3.19%. These results have small advantages over no-change models due to particular price adjustments in the prices considered here. Our results can be used on a standalone basis or combined with fundamental forecasts in forming perspectives of commodity price trends and conducting policy analysis. Our empirical framework should not be diffucult to implement, which is a critical consideration for many decision-makers and has the potential to be generalized for price forecasts of more commodities.

Financial statement fraud is a global problem for investors, audit firms, regulators, and other stakeholders. Fraud detection can be regarded as a binary classification problem with a false negative being more expensive than a false positive. Although existing studies have made great efforts to detect fraud using various data-mining techniques, the difference in misclassification costs is seldom considered. In this study, we propose a cost-sensitive cascade forest (CSCF) for fraud detection, which places heavy penalty on false negative prediction and self-adjusts the depth of a cascade forest according to the classifier’s recall (i.e. the classifier’s sensitivity). As missing values are ubiquitous in fraud research, we also explore the effect of selected missing data treatments on prediction performance, including complete case analysis, three selected classic statistical mechanisms (zero, mean, and modified mean imputation), and two machine learning (K-nearest neighbor [KNN] and random forest [RF]) approaches. The experimental results show that the proposed CSCF significantly improves the fraud prediction in comparison with one of the latest fraud detection models using the RUSBoost algorithm. Comparing different missing value treatments, even though RUSBoost and CSCF perform well when using complete case analysis, we find that the best performance is achieved when CSCF is used with missing data imputed as zero. Such treatment further improves the performance, and results in an area under curve (AUC) score of 0.82 compared to the highest AUC (0.71) from the baseline model. Supplementary analysis further reveals that the low AUC of complete case analysis for the two examined models persists under different training sizes. Thus, our findings shed light on the potential benefits of missing value imputation for the model’s performance for fraud detection.

In this study, we address the topic of credit risk stemming from central governments from a technical point of view. First, we explore various econometric and machine learning techniques to build an enhanced sovereign rating system that effectively differentiates the risk of default among countries. Our empirical results indicate that the machine learning method of XGBOOST has a superior out-of-sample and out-of-time predictive performance. Then, we use the models developed to calibrate a sovereign rating system and provide useful insights into the set-up of a parsimonious early warning system. Our results provide a more concise view of the most robust method for classifying countries’ default risk with significant regulatory implications, given that the efficient assessment of sovereign debt is crucial for effective proactive risk measurement.

This study investigates the advantage of combining the forecasting abilities of multiple generalized autoregressive conditional heteroscedasticity (GARCH)-type models, such as the standard GARCH (GARCH), exponential GARCH (eGARCH), and threshold GARCH (tGARCH) models with advanced deep learning methods to predict the volatility of five important metals (nickel, copper, tin, lead, and gold) in the Indian commodity market. This paper proposes integrating the forecasts of one to three GARCH-type models into an ensemble learning-based hybrid long short-term memory (LSTM) model to forecast commodity price volatility. We further evaluate the forecasting performance of these models for standalone LSTM and GARCH-type models using the root mean squared error, mean absolute error, and mean fundamental percentage error. The results highlight that combining the information from the forecasts of multiple GARCH types into a hybrid LSTM model leads to superior volatility forecasting capability. The SET-LSTM, which represents the model that combines forecasts of the GARCH, eGARCH, and tGARCH into the LSTM hybrid, has shown the best overall results for all metals, barring a few exceptions. Moreover, the equivalence of forecasting accuracy is tested using the Diebold–Mariano and Wilcoxon signed-rank tests.