Explainable Artificial Intelligence (XAI) in auditing

Artificial Intelligence (AI) and Machine Learning (ML) are gaining increasing attention regarding their potential applications in auditing. One major challenge of their adoption in auditing is the lack of explainability of their results. As AI/ML matures, so do techniques that can enhance the interpretability of AI, a.k.a., Explainable Artificial Intelligence (XAI). This paper introduces XAI techniques to auditing practitioners and researchers. We discuss how different XAI techniques can be used to meet the requirements of audit documentation and audit evidence standards. Furthermore, we demonstrate popular XAI techniques, especially Local Interpretable Model-agnostic Explanations (LIME) and Shapley Additive exPlanations (SHAP), using an auditing task of assessing the risk of material misstatement. This paper contributes to accounting information systems research and practice by introducing XAI techniques to enhance the transparency and interpretability of AI applications applied to auditing tasks.     

Predicting accounting fraud using imbalanced ensemble learning classifiers – evidence from China

The current research aims to launch effective accounting fraud detection models using imbalanced ensemble learning algorithms for China A-Share listed firms. Based on a sample of 33,544 Chinese firm-year instances from 1998 to 2017, this research respectively established one logistic regression and four ensemble learning classifiers (AdaBoost, XGBoost, CUSBoost, and RUSBoost) by 12 financial ratios and 28 raw financial data. Additionally, we divided the sample into the train and test observations to evaluate the classifiers' out-of-sample performance. In detail, we applied two metrics, namely, Area under the ROC (receiver operating characteristic) curve (AUC) and Area under the Precision-Recall curve (AUPR), to evaluate classifiers' discriminability. In the supplement test, this study put forward an algebraic fused model on the basis of the four ensemble learning classifiers and introduced the sliding window technique. The empirical results showed that the ensemble learning classifiers can detect accounting fraud for the imbalanced China A-listed firms far more effectively than the logistic regression model. Moreover, imbalanced ensemble learning classifiers (CUSBoost and RUSBoost) effectively performed better than the common ensemble learning models (AdaBoost and XGBoost) in average. The algebraic fused model in the supplement test also obtained the highest average AUC and AUPR among all the employed algorithms. Our results offer firm support for the potential role of Machine Learning (ML)-based Artificial Intelligence (AI) approaches in reliably predicting accounting fraud with high accuracy. Similarly, for the Chinese settings, our ML-based AI offers utmost advantage in forecasting accounting fraud. Finally, this paper fills the research gap on the applications of imbalanced ensemble learning in accounting fraud detection for Chinese listed firms.     

Artificial intelligence activities and ethical approaches in leading listed companies in the European Union

This study explores the information regarding Artificial Intelligence (AI) included by European listed companies in their annual and/or sustainability reports. The study mainly focuses on (1) the development and use of AI systems/projects reported by companies, (2) the extent to which companies disclose ethical principles or guidelines regarding AI and (3) the factors explaining these practices.The study analyses the reports of 200 companies listed in the major indexes of Germany, Sweden, Finland, France, Spain, and Italy, both from qualitative and quantitative perspectives. All reports are analysed, using content analysis methodology, to identify expressions such as ‘artificial intelligence’, ‘machine learning’, ‘deep learning’, and ‘big data’, and then classified accordingly. The study’s findings suggest a growing interest in the above-mentioned technologies, although 41.5% of companies do not report any activity in the field of AI. The adoption of ethical approaches to AI is at a very preliminary stage, and<5% of companies report on that issue. The quantitative analysis shows that larger companies, companies in the Technology and Telecommunications industries, and companies based in Southern countries are more likely to disclose information on AI activity. The majority of companies that develop ethical principles are listed in the Northern region and belong to the Technology and Telecommunications industries.The study provides evidence of AI disclosure, a type of non-financial disclosure that has not been explored yet in the literature. Unlike existing studies, we propose a first definition of the topic and a taxonomy that can be used in further research on AI disclosure and can contribute to the development of KPIs in the field. Furthermore, this study provides a theoretical framework integrating some traditional theories, such as Voluntary disclosure theory, Signalling theory, and Legitimacy theory, specifically drawn to interpret AI disclosure practices, which can help with a further in-depth exploration of AI disclosure combining concurrent perspectives. The study’s results may serve as a starting point for researchers and companies interested in the topic.     

Machine learning methods in finance: Recent applications and prospects

We study how researchers can apply machine learning (ML) methods in finance. We first establish that the two major categories of ML (supervised and unsupervised learning) address fundamentally different problems than traditional econometric approaches. Then, we review the current state of research on ML in finance and identify three archetypes of applications: (i) the construction of superior and novel measures, (ii) the reduction of prediction error, and (iii) the extension of the standard econometric toolset. With this taxonomy, we give an outlook on potential future directions for both researchers and practitioners. Our results suggest many benefits of ML methods compared to traditional approaches and indicate that ML holds great potential for future research in finance.     

Audit‐firm group appointment: an artificial intelligence approach

Auditor appointment can be regarded as a matter of pursued audit quality and is driven by several factors. The adoption of an effective auditor procurement process increases the likelihood that a company will engage the right auditor at a fair price. In this study, three techniques derived from artificial intelligence (AI) are used to propose models capable of discriminating between cases where companies appoint a Big 4 or a Non‐Big 4 auditor. These three AI methods are then compared with the broadly used method of logistic regression. The results indicate that two of the AI techniques outperform logistic regression. In addition, one method further improves its performance by applying bagging. Finally, significant factors associated with auditor appointment are revealed. Copyright © 2009 John Wiley & Sons, Ltd.     

Can machine learning models save capital for banks? Evidence from a Spanish credit portfolio

We study the impact of machine learning (ML) models for credit default prediction in the calculation of regulatory capital by financial institutions. We do so by using a unique and anonymized database from a major Spanish bank. We first compare the statistical performance of five models based on supervised learning like Logistic Lasso, Trees (CART), Random Forest, XGBoost and Deep Learning, with a well-known model like Logit. We measure the statistical performance through different metrics, and for different sample sizes and features available. We find that ML models outperform, even when relatively low amount of data is used. We then translate this statistical performance into economic impact by estimating the savings in capital when using an advanced ML model instead of a simpler one to compute the risk-weighted assets following the Internal Ratings Based (IRB) approach. Our benchmark results show that implementing XGBoost instead of Logistic Lasso could yield savings from 12.4% to 17% in terms of regulatory capital requirements.     

The role of attribute selection in Deep ANNs learning framework for high‐frequency financial trading

In financial trading, technical and quantitative analysis tools are used for the development of decision support systems. Although these traditional tools are useful, new techniques in the field of machine learning have been developed for time‐series forecasting. This paper analyses the role of attribute selection on the development of a simple deep‐learning ANN (D‐ANN) multi‐agent framework to accomplish a profitable trading strategy in the course of a series of trading simulations in the foreign exchange market. The paper evaluates the performance of the D‐ANN multi‐agent framework over different time spans of high‐frequency (HF) intraday asset time‐series data and determines how a set of the framework attributes produces effective forecasting for profitable trading. The paper shows the existence of predictable short‐term price trends in the market time series, and an understanding of the probability of price movements may be useful to HF traders. The results of this paper can be used to further develop financial decision‐support systems and autonomous trading strategies for the financial market.     

Audit Quality in New Zealand – Early Evidence from the Regulator

This study examines the impact of audit regulation in New Zealand, using audit quality reviews undertaken by the Financial Markets Authority (FMA) between 2013 and 2017. Regulation has more than halved the number of registered audit firms indicating that it imposes costs on audit firms. The results show that audit quality is improving, but a high proportion of audits do not meet the FMA's requirements. A key area for improvement is the consistency of quality across audits performed by the same audit firm. The FMA advises audit firms to investigate the underlying cause of audit deficiencies to ensure that internal quality control systems are effective in producing high‐quality audits on a regular basis. A comparison of audit file ratings with the United Kingdom shows that New Zealand's audit quality is much lower. The variation in audit quality across countries concerns the International Forum of Independent Audit Regulators (IFIAR) as it has the potential to undermine stakeholder confidence in the audit industry. Thus monitoring progress in improving audit quality in New Zealand is important. This paper provides insights into audit quality at a country level, whereas most research focuses on the firm level.     

A Look Back at Merton Miller's “Financial Markets and Economic Growth”

The author begins by agreeing with Miller's characterization of the fragility of U.S. banks and of the shortcomings of the Asian model of bank finance‐driven growth. The article also expresses “emphatic agreement” with Miller's arguments that the protection of banks through deposit insurance, regulatory forbearance, and other forms of “bailout” have created costly moral‐hazard problems that encourage excessive risk‐taking. And the author endorses, at least in principle, Miller's main argument that the development of capital markets that do not require the direct involvement of banks should make economies if not less prone to financial crises, then at least more resilient in recovering from them. But having acknowledged the limitations of bank‐centered systems and the value of developing non‐bank alternatives for savers and corporate borrowers, the author goes on to point to the surprising durability of some banking systems outside the U.S.—notably Canada's, which has not experienced major problems since the 1830s. And even more important, the author views banks and capital markets not as “substitutes” for one another, but as mutually dependent “complements” whose interdependencies and interactions must be recognized by market participants and regulators alike.     

An Experimental Study on the Effect of Budget Information on Balanced Scorecard Preparer Individual Learning

We examine whether budgets affect individual learning in balanced scorecard (BSC) preparers for the purposes of scorecard target setting. Control systems research has called for studies examining the impact of multiple controls on common decision‐making phenomena. Given this, are there other cybernetic controls (budgets) that might influence the decisions of BSC preparers? From an experimental study involving 235 postgraduate university candidates, our findings suggest that the awareness of progressively greater budget information amongst BSC users in high uncertainty environments engenders greater individual learning about the organisation, altering BSC preparer target‐setting choices. Interestingly, this learning does not necessarily lead to better budget‐actual outcomes, but informs BSC preparers of the constraints facing the organisation from a funding ‘supply side’ perspective. The oft‐criticised budget, even within high uncertainty conditions, facilitates learning in a BSC system originally purported to replace or advance the traditional system. Finally, we contribute more broadly to a growing literature evidencing the appropriateness of budgets in flexible environments, by arguing for its impact on other performance management systems.     

Entropic approximate learning for financial decision-making in the small data regime

Financial decision-making problems based on relatively few observations and several explanatory variables can be problematic for the common machine learning (ML) tools, since they cannot efficiently discriminate the relevant information. To investigate the challenges of this “small data” regime, we employ several state-of-the-art ML methods for predicting whether three selected stocks from the Swiss Market Index will outperform the market, by using, as classification features, a set of commonly used technical indicators. We show that the recently introduced entropic Scalable Probabilistic Approximation (eSPA) algorithm significantly surpasses its competitors in both prediction accuracy and computational cost. We then discuss the interpretability of the employed ML methods and suggest some statistically derived heuristics to select the most appropriate and parsimonious financial decision-making candidate model.     

Predicting credit card fraud with Sarbanes‐Oxley assessments and Fama‐French risk factors

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.     

Social impacts of selected artificial intelligence applications : The Canadian context

Artificial intelligence (AI) will be a transforming technology because it will allow old things to be done in a dramatically different way-whether cheaper, faster, or simply better. This article looks at the social impacts of computerization and discusses natural language processing, machine translation, expert systems and the overall effect of AI applications on employment. It is concluded that AI applications are likely to develop in an evolutionary sequence rather than through one or more sudden breakthroughs. However, the sum of the changes which will result from the sequence of these suboptimal systems will almost certainly transform a wide range of human activities.     

Performance assessment of ensemble learning systems in financial data classification

Financial data classification plays an important role in investment and banking industry with the purpose to control default risk, improve cash and select the best customers. Ensemble learning and classification systems are becoming gradually more applied to classify financial data where outputs from different classification systems are combined. The objective of this research is to assess the relative performance of existing state‐of‐the‐art ensemble learning and classification systems with applications to corporate bankruptcy prediction and credit scoring. The considered ensemble systems include AdaBoost, LogitBoost, RUSBoost, subspace, and bagging ensemble system. The experimental results from three datasets: one is composed of quantitative attributes, one encompasses qualitative data, and another one combines both quantitative and qualitative attributes. By using ten‐fold cross‐validation method, the experimental results show that AdaBoost is effective in terms of low classification error, limited complexity, and short time processing of the data. In addition, the experimental results show that ensemble classification systems outperform existing models that were recently validated on the same databases. Therefore, ensemble classification system can be employed to increase the reliability and consistency of financial data classification task.     

Dynamic efficiencies of the 1997 Boeing‐McDonnell Douglas merger

We evaluate the welfare effects of the 1997 Boeing‐McDonnell Douglas merger in the medium‐sized, wide‐body aircraft industry. We find that the merger led to lower prices. To explain the price drop, we develop a dynamic oligopoly game with learning‐by‐doing. We quantify the welfare effects of the merger by incorporating both increased market power and merger efficiencies from accelerated learning‐by‐doing. Our dynamic analysis indicates that net consumer surplus increased by as much as $5.14 billion, whereas a static model ignoring efficiencies of learning‐by‐doing predicts a $0.92 billion loss.     

Commentary: Performance Audit of PPPs – Getting the Basics Right

Performance auditing (PA) is an important vehicle for assessing Value for Money (VFM) of Public‐Private Partnerships (PPPs), as well as providing assurance to the Parliament and to the public about the accountability for the supporting strategic and operational frameworks. Experience to date with PPPs in Australia has been limited and mixed in terms of results. Few projects have reached the mature stage, let alone been completed. It has been suggested that we can learn from the audit approaches and systems developed by the National Audit Office in the UK. Australian Audit Offices need to ensure that they have robust PA/VFM auditing systems, analytical methodologies and tools in place to undertake quality evaluations at various stages of a PPP, but experience to date at federal and state levels also indicates that there is still a lot to do to get the basis elements of PPPs “right”.     

A predictive system integrating intrinsic mode functions,artificial neural networks,and genetic algorithms for forecasting S&P500 intra‐day data

There is an abundant literature on the design of intelligent systems to forecast stock market indices. In general, the existing stock market price forecasting approaches can achieve good results. The goal of our study is to develop an effective intelligent predictive system to improve the forecasting accuracy. Therefore, our proposed predictive system integrates adaptive filtering, artificial neural networks (ANNs), and evolutionary optimization. Specifically, it is based on the empirical mode decomposition (EMD), which is a useful adaptive signal‐processing technique, and ANNs, which are powerful adaptive intelligent systems suitable for noisy data learning and prediction, such as stock market intra‐day data. Our system hybridizes intrinsic mode functions (IMFs) obtained from EMD and ANNs optimized by genetic algorithms (GAs) for the analysis and forecasting of S&P500 intra‐day price data. For comparison purposes, the performance of the EMD‐GA‐ANN presented is compared with that of a GA‐ANN trained with a wavelet transform's (WT's) resulting approximation and details coefficients, and a GA‐general regression neural network (GRNN) trained with price historical data. The mean absolute deviation, mean absolute error, and root‐mean‐squared errors show evidence of the superiority of EMD‐GA‐ANN over WT‐GA‐ANN and GA‐GRNN. In addition, it outperformed existing predictive systems tested on the same data set. Furthermore, our hybrid predictive system is relatively easy to implement and not highly time‐consuming to run. Furthermore, it was found that the Daubechies wavelet showed quite a higher prediction accuracy than the Haar wavelet. Moreover, prediction errors decrease with the level of decomposition.     

Machine Learning and Expert Judgement: Analyzing Emerging Topics in Accounting and Finance Research in the Asia–Pacific

In this paper, we focus on the question to what extent machine learning (ML) tools can be used to support systematic literature reviews. We apply a ML approach for topic detection to analyze emerging topics in the literature—our context is accounting and finance research in the Asia–Pacific region. To evaluate the robustness of the approach, we compare findings from the automated ML approach with the results from a manual analysis of the literature. The automated approach uses a keyword algorithm detection mechanism whereby the manual analysis uses common techniques for qualitative data analysis, that is, triangulation between researchers (expert judgement). From our paper, we conclude that both methods have strengths and weaknesses. The automated analysis works well for large corpora of text and provides a very standardized and non-biased way of analyzing the literature. However, the human researcher is potentially better equipped to evaluate current issues and future trends in the literature. Overall, the best results might be achieved when a variety of tools are used together.     

中国区域绿色全要素生产率的影响因素及其空间特征

运用非径向非角度的SBM方向性距离函数与ML指数,将环境污染和能源消耗纳入生产率分析框架,将绿色全要素生产率(GTFP)增长分解为技术进步和技术效率改善,测度中国30个省市区GTFP,利用空间计量模型考量GTFP的影响因素及其空间特征。结果表明：样本期内全国平均GTFP累积增长31%,技术进步贡献较大。各地区的GTFP均呈现增长趋势,而技术效率则有所下降。地理邻接是空间溢出的主要途径,各变量对GTFP、技术进步和技术效率改善的直接效应和空间溢出效应呈现不同的作用机制,R&D投入、外商投资和环境管制等因素对GTFP的增长均有显著影响。