A comparison of machine‐learning classifiers for selecting money managers

Machine‐learning algorithms have performed well on noisy datasets that are typical of financial data. This paper compares the performance of three types of machine‐learning classifier for selecting money managers. Naïve Bayes, neural network and decision tree learners were applied to a dataset of US equity managers. Although other studies have suggested that the performance of classifiers appears to be highly dependent on the nature of the problem and the dataset, the learning algorithms each had similar predictive accuracy and all outperformed by a substantial margin simple manager selection rules that are typical of the ways in which money managers and mutual funds are selected by investors. The results indicate that machine learners can be used as a decision‐support aid to improve the selection of money managers. Copyright © 2005 John Wiley & Sons, Ltd.     

Data preprocessing and data parsimony in corporate failure forecast models: evidence from Australian materials industry

The present study, based on data for delisted and active corporations in the Australian materials industry, is an attempt to develop a systematic way of selecting corporate failure‐related features. We empirically tested the proposed procedure using three datasets. The first dataset contains 82 financial economic factors from the corporation's financial statement. The second dataset comprises 73 relevant financial ratios, which either directly or indirectly measure a corporation's propensity to fail, and are conciliated from the first dataset. The third dataset is a parsimonious dataset obtained from the application of combining a filter and a wrapper to preprocess the first dataset. The robustness of this preprocessed dataset is tested by comparing its performance with the first and second datasets in two statistical (logistic regression and naïve‐Bayes) and two machine learning (decision tree, neural network) classes of prediction models. Tests for prediction accuracies and reliabilities, using the computational (ROC curve, AUC) and the statistical (Cochran's Q statistic) criteria show that the third dataset outperforms the other two datasets in all four predicting models, achieving various accuracies ranges from 81 per cent to 84 per cent.     

INTEGRATION OF RANDOM SAMPLE SELECTION,SUPPORT VECTOR MACHINES AND ENSEMBLES FOR FINANCIAL RISK FORECASTING WITH AN EMPIRICAL ANALYSIS ON THE NECESSITY OF FEATURE SELECTION

Financial risk forecasting (FRF) is an effective tool to help people forecast whether or not a company will fail in future. Among all techniques of FRF, the support vector machine (SVM) is the most newly developed, and one of the most accurate and effective techniques. This study is devoted to investigating an ensemble model of FRF by integrating bagging with an SVM to generate a data‐driven SVM ensemble. Bagging is used to produce diverse training datasets on which multiple SVM classifiers are trained to make FRF for a target company. Simple voting is employed to produce a final decision from the SVM model committee. The empirical study has two objectives. One is to verify whether the data‐driven SVM ensemble can produce a more dominating performance than the most frequently used techniques in the area of FRF, i.e. multivariate discriminant analysis, logistics regression and a single SVM. The other is to verify whether feature selection is necessary to help the SVM make more precise FRF, although the SVM can handle high‐dimensional data. The results indicate that the data‐driven SVM ensemble significantly improves the predictive ability of SVM‐based FRF. Meanwhile, feature selection can effectively help the SVM achieve better predictive performance, which means that use of feature selection is necessary in SVM‐based FRF. Copyright © 2012 John Wiley & Sons, Ltd.     

A quarterly time-series classifier based on a reduced-dimension generated rules method for identifying financial distress

Predicting financial distress has been and will remain an important and challenging issue. Many methods have been proposed to predict bankruptcies and detect financial crises, including conventional approaches and techniques involving artificial intelligence (AI). Financial distress information influences investor decisions, and investors depend on analysts’ opinions and subjective judgements in assessing such information, which sometimes results in investors making mistakes. In the light of the foregoing, this paper proposes a novel quarterly time series classifier, which reduces the sheer volume of high-dimensional data to be analysed and provides decision-makers with rules that can be used as a reference in assessing the financial situation of a company. This study employs the following six attribute selection methods to reduce the high-dimensional data: (1) the chi-square test, (2) information gain, (3) discriminant analysis, (4) logistic regression (LR) analysis, (5) support vector machine (SVM) and (6) the proposed Join method. After selecting attributes, this study utilises the rough set classifier to generate the rules of financial distress. To verify the proposed method, an empirically collected financial distress data-set is employed as the experimental sample and is compared with the decision tree, multilayer perceptron and SVM under Type I error, Type II error and accuracy criteria. Because financial distress data are quarterly time series data, this study conducts non-time series and time series (moving windows) experiments. The experimental results indicate that the LR and chi-square attribute selection combined with the rough set classifier outperform the listing methods under Type I, Type II error and accuracy criteria.     

PROBABILISTIC APPROACHES FOR CREDIT SCREENING AND BANKRUPTCY PREDICTION

Three probabilistic neural network approaches are used for credit screening and bankruptcy prediction: a logistic regression neural network (LRNN), a probabilistic neural network (PNN) and a semi‐supervised expectation maximization‐based neural network. Using real‐world bankruptcy prediction and credit screening datasets, we compare the three probabilistic approaches using various performance criteria of sensitivity, specificity, accuracy, decile lift and area under receiver operating characteristics (ROC) curves. The results of our experiments indicate that the PNN outperforms the other two techniques for decile lift and specificity performance metric. Using the area under ROC curve, we find that for bankruptcy prediction data the PNN outperforms the other two approaches when false positive rates (FPRs) are less than 40 %. LRNN outperforms the other two techniques for FPRs higher than 40 % for bankruptcy data. We observe that the LRNN results are very sensitive to the ratio of examples belonging to two classes in training data and there is a tendency to overfit training data. Copyright © 2012 John Wiley & Sons, Ltd.     

Using a hybrid evolution approach to forecast financial failures for Taiwan-listed companies

Bankruptcy has been an important topic in finance and accounting research for a long time. Recent major bankruptcies have included seemingly robust companies such as Enron, Kmart, Global Crossing, WorldCom, and Lehman Brothers. These cases have become of serious public concern due to the huge influence these companies have on the real economy. This research proposes a hybrid evolution approach to integrate particle swarm optimization (PSO) with the support vector machine (SVM) technique for the purpose of predicting financial failures. The preparation phase collected an initial sample of 68 companies listed by the Taiwan Stock Exchange Corporation (TSEC). The financial datasets were constructed based on 33 financial ratios, four non-financial ratios and one combined macroeconomic index. To select suitable indicators for the input vector, the principle component analysis (PCA) technique was applied to reduce the data and determine how groupings of indicators measure the same concept. In the swarming phase, PSO was applied to obtain suitable parameters for SVM modeling without reducing the classification accuracy rate. In the modeling phase, the SVM model was used to build a training set that was used to calculate the model's accuracy and fitness value. Finally, these optimized parameters were used in the hybrid PSO–SVM model to evaluate the model's predictive accuracy. This paper provides four critical contributions. (1) Using the PCA technique, the statistical results indicate that the financial prediction performance is mainly affected by financial ratios rather than non-financial and macroeconomic ratios. (2) Even with the input of nearly 70% fewer indicators, our approach is still able to provide highly accurate forecasts of financial bankruptcy. (3) The empirical results show that the PSO–SVM model provides better classification accuracy (i.e. normal vs. bankrupt) than the grid search (Grid–SVM) approach. (4) For six well-known UCI datasets, the PSO–SVM model also provides better prediction accuracy than the Grid–SVM, GA–SVM, SVM, SOM, and SVR–SOM approaches. Therefore, this paper proposes that the PSO–SVM approach is better suited for predicting potential financial distress.     

An innovative feature selection method for support vector machines and its test on the estimation of the credit risk of default

Support vector machines (SVM) have been extensively used for classification problems in many areas such as gene, text and image recognition. However, SVM have been rarely used to estimate the probability of default (PD) in credit risk. In this paper, we advocate the application of SVM, rather than the popular logistic regression (LR) method, for the estimation of both corporate and retail PD. Our results indicate that most of the time SVM outperforms LR in terms of classification accuracy for the corporate and retail segments. We propose a new wrapper feature selection based on maximizing the distance of the support vectors from the separating hyperplane and apply it to identify the main PD drivers. We used three datasets to test the PD estimation, containing (1) retail obligors from Germany, (2) corporate obligors from Eastern Europe, and (3) corporate obligors from Poland. Total assets, total liabilities, and sales are identified as frequent default drivers for the corporate datasets, whereas current account status and duration of the current account are frequent default drivers for the retail dataset.     

Evaluating interpretable machine learning predictions for cryptocurrencies

This study explores various machine learning and deep learning applications on financial data modelling, analysis and prediction processes. The main focus is to test the prediction accuracy of cryptocurrency hourly returns and to explore, analyse and showcase the various interpretability features of the ML models. The study considers the six most dominant cryptocurrencies in the market: Bitcoin, Ethereum, Binance Coin, Cardano, Ripple and Litecoin. The experimental settings explore the formation of the corresponding datasets from technical, fundamental and statistical analysis. The paper compares various existing and enhanced algorithms and explains their results, features and limitations. The algorithms include decision trees, random forests and ensemble methods, SVM, neural networks, single and multiple features N-BEATS, ARIMA and Google AutoML. From experimental results, we see that predicting cryptocurrency returns is possible. However, prediction algorithms may not generalise for different assets and markets over long periods. There is no clear winner that satisfies all requirements, and the main choice of algorithm will be tied to the user needs and provided resources.     

Modelling small‐business credit scoring by using logistic regression,neural networks and decision trees

Previous research on credit scoring that used statistical and intelligent methods was mostly focused on commercial and consumer lending. The main purpose of this paper is to extract important features for credit scoring in small‐business lending on a dataset with specific transitional economic conditions using a relatively small dataset. To do this, we compare the accuracy of the best models extracted by different methodologies, such as logistic regression, neural networks (NNs), and CART decision trees. Four different NN algorithms are tested, including backpropagation, radial basis function network, probabilistic and learning vector quantization, by using the forward nonlinear variable selection strategy. Although the test of differences in proportion and McNemar's test do not show a statistically significant difference in the models tested, the probabilistic NN model produces the highest hit rate and the lowest type I error. According to the measures of association, the best NN model also shows the highest degree of association with the data, and it yields the lowest total relative cost of misclassification for all scenarios examined. The best model extracts a set of important features for small‐business credit scoring for the observed sample, emphasizing credit programme characteristics, as well as entrepreneur's personal and business characteristics as the most important ones. Copyright © 2005 John Wiley & Sons, Ltd.     

Performance of Neural Networks in Managerial Forecasting

This paper investigates the effectiveness of a multi-layered neural network as a tool for forecasting in a managerial time-series setting. To handle noisy data of limited length we adopted two different neural network approaches. First, the neural network is used as a pattern classifier to automate the ARMA model-identification process. We tested the performance of multi-layered neural networks with two statistical feature extractors: ACF/PACF and ESACF. We found that ESACF provides better performance, although the noise in ESACF patterns still caused the classification performance to deteriorate. Therefore we adopted the noise-filtering network as a preprocessor to the pattern-classification network, and were able to achieve an average of about 89% classification accuracy. Second, the neural network is used as a tool for function approximation and prediction. To alleviate the overfitting problem we adopted the structure of minimal networks and recurrent networks. The experiment with three real-world time series showed that the prediction by Elman's recurrent network outperformed those by the ARMA model and other structures of multi-layered neural networks, especially when the time series contained significant noise.     

Bankruptcy prediction of financially stressed firms: an examination of the predictive accuracy of artificial neural networks

This is an extension of prior studies that have used artificial neural networks to predict bankruptcy. The incremental contribution of this study is threefold. First, we use only financially stressed firms in our control sample. This enables the models to more closely approximate the actual decision processes of auditors and other interested parties. Second, we develop a more parsimonious model using qualitative ‘bad news’ variables that prior research indicates measure financial distress. Past research has focused on the ‘usefulness’ of accounting numbers and therefore often ignored non‐accounting variables that may contribute to the classification accuracy of the distress prediction models. In addition, rather than use multiple financial ratios, we include a single variable of financial distress using the Zmijewski distress score that incorporates ratios measuring profitability, liquidity, and solvency. Finally, we develop and test a genetic algorithm neural network model. We examine its predictive ability to that of a backpropagation neural network and a model using multiple discriminant analysis. The results indicate that the misclassification cost of the genetic algorithm‐based neural network was the lowest among the models. Copyright © 2001 John Wiley & Sons, Ltd.     

Assessing the predictive performance of artifIcial neural network‐based classifiers based on different data preprocessing methods,distributions and training mechanisms

We analyse the implications of three different factors (preprocessing method, data distribution and training mechanism) on the classification performance of artificial neural networks (ANNs). We use three preprocessing approaches: no preprocessing, division by the maximum absolute values and normalization. We study the implications of input data distributions by using five datasets with different distributions: the real data, uniform, normal, logistic and Laplace distributions. We test two training mechanisms: one belonging to the gradient‐descent techniques, improved by a retraining procedure, and the other is a genetic algorithm (GA), which is based on the principles of natural evolution. The results show statistically significant influences of all individual and combined factors on both training and testing performances. A major difference with other related studies is the fact that for both training mechanisms we train the network using as starting solution the one obtained when constructing the network architecture. In other words we use a hybrid approach by refining a previously obtained solution. We found that when the starting solution has relatively low accuracy rates (80–90%) the GA clearly outperformed the retraining procedure, whereas the difference was smaller to non‐existent when the starting solution had relatively high accuracy rates (95–98%). As reported in other studies, we found little to no evidence of crossover operator influence on the GA performance. Copyright © 2005 John Wiley & Sons, Ltd.     

Technological bias at the exchange rate market

Prediction of exchange rates has been a topic for debate in economic literature since the late 1980s. The recent development of machine learning techniques has spurred a plethora of studies that further improves the prediction models for currency markets. This high-tech progress may create challenges for market efficiency along with information asymmetry and irrationality of decision-making. This technological bias emerges from the fact that recent innovative approaches have been used to solve trading tasks and to find the best trading strategies. This paper demonstrates that traders can leverage technological bias for financial market forecasting. Those traders who adapt faster to the changes in market innovations will get excess returns. To support this hypothesis we compare the performance of deep learning methods, shallow neural networks with baseline prediction methods and a random walk model using daily closing rate between three currency pairs: Euro and US Dollar (EUR/USD), British Pound and US Dollar (GBP/USD), and US Dollar and Japanese Yen (USD/JPY). The results demonstrate that deep learning achieves higher accuracy than alternate methods. The shallow neural network outperforms the random walk model, but cannot surpass ARIMA accuracy significantly. The paper discusses possible outcomes of the technological shift for financial market development and accounting conforming also to adaptive market hypothesis.     

Interpretable selective learning in credit risk

Forecasting credit default risk has been an important research field for several decades. Traditionally, logistic regression has been widely recognized as a solution because of its accuracy and interpretability. Although complex machine learning models may improve accuracy over simple logistic regressions, their interpretability has prevented their use in credit risk assessment. We introduce a neural network with a selective option to increase interpretability by distinguishing whether linear models can explain the dataset. Our methods are tested on two datasets: 25,000 samples from the Taiwan payment system collected in October 2005 and 250,000 samples from the 2011 Kaggle competition. We find that, for most of samples, logistic regression will be sufficient, with reasonable accuracy; meanwhile, for some specific data portions, a shallow neural network model leads to much better accuracy without significantly sacrificing interpretability.     

INSOLVENCY PREDICTION IN THE PRESENCE OF DATA INCONSISTENCIES

In this paper we use data inconsistencies as an indicator of financial distress. Traditional models for insolvency prediction normally ignore inconsistent data, either by removing or replacing it. Instead of removing that information, we propose a new variable to capture it; using it together with traditional accounting variables (based on financial ratios) for the purpose of insolvency prediction. Computational tests use three datasets based on the financial results of 2033 Brazilian Health Maintenance Organizations over 7 years (2001 to 2007). Sixteen classification methods were used to evaluate whether or not the new variable impacted solvency prediction. Tests show a statistically significant improvement in classification accuracy – average results improve 1.3 (p = 0.003) and 1.8 (p = 0.006) percentage points, for 10‐fold and leave‐one‐out cross‐validations respectively. In addition, the analysis of false positives and false negatives shows that the new variable reduces the potentially harmful misclassification of false negatives (i.e. financially distressed companies being classified as financially healthy) and also reduces the estimated overall error rate. Regarding the extensibility of the results, even though this work uses data from Brazilian companies only, the calculation of the financial ratios variables, as well as the inconsistencies, could be extended to most companies worldwide subject to governmental accounting regulations aligned with the International Financial Reporting Standards. Copyright © 2014 John Wiley & Sons, Ltd.     

Risk assessment in internal auditing: a neural network approach

Risk assessment is a systematic process for integrating professional judgments about relevant risk factors, their relative significance and probable adverse conditions and/or events leading to identification of auditable activities (IIA, 1995, SIAS No. 9). Internal auditors utilize risk measures to allocate critical audit resources to compliance, operational, or financial activities within the organization (Colbert, 1995). In information rich environments, risk assessment involves recognizing patterns in the data, such as complex data anomalies and discrepancies, that perhaps conceal one or more error or hazard conditions (e.g. Coakley and Brown, 1996; Bedard and Biggs, 1991; Libby, 1985). This research investigates whether neural networks can help enhance auditors’ risk assessments. Neural networks, an emerging artificial intelligence technology, are a powerful non‐linear optimization and pattern recognition tool (Haykin, 1994; Bishop, 1995). Several successful, real‐world business neural network application decision aids have already been built (Burger and Traver, 1996). Neural network modeling may prove invaluable in directing internal auditor attention to those aspects of financial, operating, and compliance data most informative of high‐risk audit areas, thus enhancing audit efficiency and effectiveness. This paper defines risk in an internal auditing context, describes contemporary approaches to performing risk assessments, provides an overview of the backpropagation neural network architecture, outlines the methodology adopted for conducting this research project including a Delphi study and comparison with statistical approaches, and presents preliminary results, which indicate that internal auditors could benefit from using neural network technology for assessing risk. Copyright © 1999 John Wiley & Sons, Ltd.     

Time-varying neural network for stock return prediction

We consider the problem of neural network training in a time-varying context. Machine learning algorithms have excelled in problems that do not change over time. However, problems encountered in financial markets are often time varying. We propose the online early stopping algorithm and show that a neural network trained using this algorithm can track a function changing with unknown dynamics. We compare the proposed algorithm to current approaches on predicting monthly US stock returns and show its superiority. We also show that prominent factors (such as the size and momentum effects) and industry indicators exhibit time-varying predictive power on stock returns. We find that during market distress, industry indicators experience an increase in importance at the expense of firm level features. This indicates that industries play a role in explaining stock returns during periods of heightened risk.     

Explaining qualifications in audit reports using a support vector machine methodology

The verification of whether the financial statements of a firm represent its actual position is of major importance for auditors, who should provide a qualified report if they conclude that the financial statements fail to meet this requirement. This paper implements support vector machines (SVMs) to develop models that may support auditors in this task. Linear and non‐linear models are developed and their performance is analysed using training samples of different size and out‐of‐sample/out‐of‐time data. The results show that all SVM models are capable of distinguishing between qualified and unqualified financial statements with satisfactory accuracy. The performance of the models over time is also explored. Copyright © 2005 John Wiley & Sons, Ltd.     

Modeling default risk with support vector machines

Predicting default risk is important for firms and banks to operate successfully. There are many reasons to use nonlinear techniques for predicting bankruptcy from financial ratios. Here we propose the so-called Support Vector Machine (SVM) to predict the default risk of German firms. Our analysis is based on the Creditreform database. In all tests performed in this paper the nonlinear model classified by SVM exceeds the benchmark logit model, based on the same predictors, in terms of the performance metric, AR. The empirical evidence is in favor of the SVM for classification, especially in the linear non-separable case. The sensitivity investigation and a corresponding visualization tool reveal that the classifying ability of SVM appears to be superior over a wide range of SVM parameters. In terms of the empirical results obtained by SVM, the eight most important predictors related to bankruptcy for these German firms belong to the ratios of activity, profitability, liquidity, leverage and the percentage of incremental inventories. Some of the financial ratios selected by the SVM model are new because they have a strong nonlinear dependence on the default risk but a weak linear dependence that therefore cannot be captured by the usual linear models such as the DA and logit models.