Incorporating textual and management factors into financial distress prediction: A comparative study of machine learning methods

Financial distress prediction (FDP) has been widely considered as a promising approach to reducing financial losses. While financial information comprises the traditional factors involved in FDP, nonfinancial factors have also been examined in recent studies. In light of this, the purpose of this study is to explore the integrated factors and multiple models that can improve the predictive performance of FDP models. This study proposes an FDP framework to reveal the financial distress features of listed Chinese companies, incorporating financial, management, and textual factors, and evaluating the prediction performance of multiple models in different time spans. To develop this framework, this study employs the wrapper-based feature selection method to extract valuable features, and then constructs multiple single classifiers, ensemble classifiers, and deep learning models in order to predict financial distress. The experiment results indicate that management and textual factors can supplement traditional financial factors in FDP, especially textual ones. This study also discovers that integrated factors collected 4 years prior to the predicted benchmark year enable a more accurate prediction, and the ensemble classifiers and deep learning models developed can achieve satisfactory FDP performance. This study makes a novel contribution as it expands the predictive factors of financial distress and provides new findings that can have important implications for providing early warning signals of financial risk.     

Predicting multistage financial distress: Reflections on sampling,feature and model selection criteria

This research proposes a prediction model of multistage financial distress (MSFD) after considering contextual and methodological issues regarding sampling, feature and model selection criteria. Financial distress is defined as a three‐stage process showing different nature and intensity of financial problems. It is argued that applied definition of distress is independent of legal framework and its predictability would provide more practical solutions. The final sample is selected after industry adjustments and oversampling the data. A wrapper subset data mining approach is applied to extract the most relevant features from financial statement and stock market indicators. An ensemble approach using a combination of DTNB (decision table and naïve base hybrid model), LMT (logistic model tree) and A2DE (alternative N dependence estimator) Bayesian models is used to develop the final prediction model. The performance of all the models is evaluated using a 10‐fold cross‐validation method. Results showed that the proposed model predicted MSFD with 84.06% accuracy. This accuracy increased to 89.57% when a 33.33% cut‐off value was considered. Hence the proposed model is accurate and reliable to identify the true nature and intensity of financial problems regardless of the contextual legal framework.     

A Robust Data‐Mining Approach to Bankruptcy Prediction

In this study, new variants of genetic programming (GP), namely gene expression programming (GEP) and multi‐expression programming (MEP), are utilized to build models for bankruptcy prediction. Generalized relationships are obtained to classify samples of 136 bankrupt and non‐bankrupt Iranian corporations based on their financial ratios. An important contribution of this paper is to identify the effective predictive financial ratios on the basis of an extensive bankruptcy prediction literature review and upon a sequential feature selection analysis. The predictive performance of the GEP and MEP forecasting methods is compared with the performance of traditional statistical methods and a generalized regression neural network. The proposed GEP and MEP models are effectively capable of classifying bankrupt and non‐bankrupt firms and outperform the models developed using other methods. Copyright © 2011 John Wiley & Sons, Ltd.     

The Role of Financial Variables in predicting economic activity

Previous research found that the US business cycle leads the European one by a few quarters, and can therefore be useful in predicting euro area gross domestic product (GDP). In this paper we investigate whether additional predictive power can be gained by adding selected financial variables belonging to either the USA or the euro area. We use vector autoregressions (VARs) that include the US and euro area GDPs as well as growth in the Rest of the World and selected combinations of financial variables. Out‐of‐sample root mean square forecast errors (RMSEs) evidence that adding financial variables produces a slightly smaller error in forecasting US economic activity. This weak macro‐financial linkage is even weaker in the euro area, where financial indicators do not improve short‐ and medium‐term GDP forecasts even when their timely availability relative to GDP is exploited. It can be conjectured that neither US nor European financial variables help predict euro area GDP as the US GDP has already embodied this information. However, we show that the finding that financial variables have no predictive power for future activity in the euro area relates to the unconditional nature of the RMSE metric. When forecasting ability is assessed as if in real time (i.e. conditionally on the information available at the time when forecasts are made), we find that models using financial variables would have been preferred in several episodes and in particular between 1999 and 2002. Copyright 2011 John Wiley & Sons, Ltd.     

Enhancing momentum investment strategy using leverage

Previous studies examine investment strategies based on leverage and momentum; none investigates both variables jointly as an investment strategy. This paper is the first incorporating leverage and momentum together. We show that low past returns (losers) forecast future negative abnormal returns only among stocks with high leverage levels, but not among stocks with low leverage levels. However, high past returns (winners) forecast future positive abnormal returns independently of leverage level. As a result, the negative relation between leverage and future abnormal returns is only observed among loser stocks, and the positive relation between past returns and future abnormal returns is only shown among non‐low leverage stocks. Our results are important in achieving better investment strategies: buying winners' stocks (independently of their level of leverage) and short‐selling losers' stocks with high leverage yield higher abnormal returns than strategies based on only one of these variables. Our two‐dimensional strategy yields risk‐adjusted abnormal returns of 15.66% per annum, whereas the single leverage or momentum strategies yield 7.70% and 7.96% per annum, respectively. The difference is nearly 8% and economically significant. If leverage is considered as proxy for default risk, our results, contrary to previous evidence, show that momentum profits are not exclusive of default stocks, and that momentum returns are not only driven by negative returns yielded by distress stocks.     

Macroeconomic forecasting with mixed data sampling frequencies: Evidence from a small open economy

The aim of this study was to forecast the Singapore gross domestic product (GDP) growth rate by employing the mixed‐data sampling (MIDAS) approach using mixed and high‐frequency financial market data from Singapore, and to examine whether the high‐frequency financial variables could better predict the macroeconomic variables. We adopt different time‐aggregating methods to handle the high‐frequency data in order to match the sampling rate of lower‐frequency data in our regression models. Our results showed that MIDAS regression using high‐frequency stock return data produced a better forecast of GDP growth rate than the other models, and the best forecasting performance was achieved by using weekly stock returns. The forecasting result was further improved by performing intra‐period forecasting.     

A Meta‐learning Framework for Bankruptcy Prediction

The implication of corporate bankruptcy prediction is important to financial institutions when making lending decisions. In related studies, many bankruptcy prediction models have been developed based on some machine‐learning techniques. This paper presents a meta‐learning framework, which is composed of two‐level classifiers for bankruptcy prediction. The first‐level multiple classifiers perform the data reduction task by filtering out unrepresentative training data. Then, the outputs of the first‐level classifiers are utilized to create the second‐level single (meta) classifier. The experiments are based on five related datasets and the results show that the proposed meta‐learning framework provides higher prediction accuracy rates and lower type I/II errors when compared with the stacked generalization classifier and other three widely developed baselines, such as neural networks, decision trees, and logistic regression. Copyright © 2011 John Wiley & Sons, Ltd.     

The detection of earnings manipulation: the three‐phase cutting plane algorithm using mathematical programming

The primary goal of this study was to propose an algorithm using mathematical programming to detect earnings management practices. In order to evaluate the ability of this proposed algorithm, the traditional statistical models are used as a benchmark vis‐à‐vis their time series counterparts. As emerging techniques in the area of mathematical programming yield better results, application of suitable models is expected to result in highly performed forecasts. The motivation behind this paper is to develop an algorithm which will succeed in detecting companies that appeal to financial manipulation. The methodology is based on cutting plane formulation using mathematical programming. A sample of 126 Turkish manufacturing firms described over 10 financial ratios and indexes are used for detecting factors associated with false financial statements. The results indicate that the proposed three‐phase cutting plane algorithm outperforms the traditional statistical techniques which are widely used for false financial statement detections. Furthermore, the results indicate that the investigation of financial information can be helpful towards the identification of false financial statements and highlight the importance of financial ratios/indexes such as Days' Sales in Receivables Index (DSRI), Gross Margin Index (GMI), Working Capital Accruals to Total Assets (TATA) and Days to Inventory Index (DINV). Copyright © 2009 John Wiley & Sons, Ltd.     

On the forecasting of high-frequency financial time series based on ARIMA model improved by deep learning

Through empirical research, it is found that the traditional autoregressive integrated moving average (ARIMA) model has a large deviation for the forecasting of high-frequency financial time series. With the improvement in storage capacity and computing power of high-frequency financial time series, this paper combines the traditional ARIMA model with the deep learning model to forecast high-frequency financial time series. It not only preserves the theoretical basis of the traditional model and characterizes the linear relationship, but also can characterize the nonlinear relationship of the error term according to the deep learning model. The empirical study of Monte Carlo numerical simulation and CSI 300 index in China show that, compared with ARIMA, support vector machine (SVM), long short-term memory (LSTM) and ARIMA-SVM models, the improved ARIMA model based on LSTM not only improves the forecasting accuracy of the single ARIMA model in both fitting and forecasting, but also reduces the computational complexity of only a single deep learning model. The improved ARIMA model based on deep learning not only enriches the models for the forecasting of time series, but also provides effective tools for high-frequency strategy design to reduce the investment risks of stock index.     

Predicting US bank failures: A comparison of logit and data mining models

Predicting bank failures is important as it enables bank regulators to take timely actions to prevent bank failures or reduce the cost of rescuing banks. This paper compares the logit model and data mining models in the prediction of bank failures in the USA between 2002 and 2010 using levels and rates of change of 16 financial ratios based on a cross‐section sample. The models are estimated for the in‐sample period 2002–2009, while data for the year 2010 are used for out‐of‐sample tests. The results suggest that the logit model predicts bank failures in‐sample less precisely than data mining models, but produces fewer missed failures and false alarms out‐of‐sample.     

Bootstrap Replacement to Validate the Influence of the Economic Cycle on the Structure and the Accuracy Level of Business Failure Prediction Models

The aim of this study was to answer the question of how the economic cycle affects the stability and efficiency of business failure prediction models, using bootstrap replacement method for validation. We analyse 2228 Spanish small and medium‐sized enterprises for the period 2001–2009, and divide it into three different phases of the economic cycle (growth, crisis, recession). We find that the structure and the ability of business failure prediction models are different according to the economic cycle. These findings are relevant for the debate on the most suitable financial ratios when developing business failure prediction models and to pose their accuracy level in these prediction models. Copyright © 2015 John Wiley & Sons, Ltd.     

Short-term forecasting of industrial production by means of quick indicators

This paper reports results on building transfer function models with linear combinations of quick indicators as inputs for very short-term prediction of the monthly time series of the volume of industrial production in Finland. The number of input variables in the transfer function models is reduced in two alternative ways: by replacing the original indicators by their two first principal components and by omitting certain indicators. The prediction accuracy of the transfer function models is checked outside the sample and found superior to that of corresponding ARIMA models. Neither of the two ways of reducing the number of input variables leads to consistently more accurate forecasts than the other. It is also found that the prediction accuracy of the transfer function models compares rather favourably with the preliminary values of the volume of industrial production published by the Central Statistical Office during the periods of rapid growth.     

水利水电类公司融资结构与公司绩效研究——基于2007~2013年面板数据的实证分析

针对我国水利水电类上市公司2007~201 3年的财务数据构建了面板数据模型,分析其融资结构与公司绩效之间的关系。研究发现,水利水电类公司的资产负债率与公司绩效存在显著的倒U型关系,债务融资的公司治理效应仅在较低的债务水平才明显;与通常理论相反,在水利水电行业与国家投资行为关系密切的现状下,水利水电类公司的国有股比例与公司绩效正相关;第一大股东持股比例和短期债务比例与公司绩效的关系不显著。     

Methods for backcasting,nowcasting and forecasting using factor‐MIDAS: With an application to Korean GDP

We utilize mixed‐frequency factor‐MIDAS models for the purpose of carrying out backcasting, nowcasting, and forecasting experiments using real‐time data. We also introduce a new real‐time Korean GDP dataset, which is the focus of our experiments. The methodology that we utilize involves first estimating common latent factors (i.e., diffusion indices) from 190 monthly macroeconomic and financial series using various estimation strategies. These factors are then included, along with standard variables measured at multiple different frequencies, in various factor‐MIDAS prediction models. Our key empirical findings as follows. (i) When using real‐time data, factor‐MIDAS prediction models outperform various linear benchmark models. Interestingly, the “MSFE‐best” MIDAS models contain no autoregressive (AR) lag terms when backcasting and nowcasting. AR terms only begin to play a role in “true” forecasting contexts. (ii) Models that utilize only one or two factors are “MSFE‐best” at all forecasting horizons, but not at any backcasting and nowcasting horizons. In these latter contexts, much more heavily parametrized models with many factors are preferred. (iii) Real‐time data are crucial for forecasting Korean gross domestic product, and the use of “first available” versus “most recent” data “strongly” affects model selection and performance. (iv) Recursively estimated models are almost always “MSFE‐best,” and models estimated using autoregressive interpolation dominate those estimated using other interpolation methods. (v) Factors estimated using recursive principal component estimation methods have more predictive content than those estimated using a variety of other (more sophisticated) approaches. This result is particularly prevalent for our “MSFE‐best” factor‐MIDAS models, across virtually all forecast horizons, estimation schemes, and data vintages that are analyzed.     

Forecasting growth and inflation in an enlarged euro area

We compare models for forecasting growth and inflation in the enlarged euro area. Forecasts are built from univariate autoregressive and single‐equation models. The analysis is undertaken for both individual countries and EU aggregate variables. Aggregate forecasts are constructed by both employing aggregate variables and by aggregating country‐specific forecasts. Using financial variables for country‐specific forecasts tends to add little to the predictive ability of a simple AR model. However, they do help to predict EU aggregates. Furthermore, forecasts from pooling individual country models usually outperform those of the aggregate itself, particularly for the EU25 grouping. This is particularly interesting from the perspective of the European Central Bank, who require forecasts of economic activity and inflation to formulate appropriate economic policy across the enlarged group. Copyright © 2008 John Wiley & Sons, Ltd.     

Bayesian Analysis of a Threshold Stochastic Volatility Model

This paper proposes a parsimonious threshold stochastic volatility (SV) model for financial asset returns. Instead of imposing a threshold value on the dynamics of the latent volatility process of the SV model, we assume that the innovation of the mean equation follows a threshold distribution in which the mean innovation switches between two regimes. In our model, the threshold is treated as an unknown parameter. We show that the proposed threshold SV model can not only capture the time‐varying volatility of returns, but can also accommodate the asymmetric shape of conditional distribution of the returns. Parameter estimation is carried out by using Markov chain Monte Carlo methods. For model selection and volatility forecast, an auxiliary particle filter technique is employed to approximate the filter and prediction distributions of the returns. Several experiments are conducted to assess the robustness of the proposed model and estimation methods. In the empirical study, we apply our threshold SV model to three return time series. The empirical analysis results show that the threshold parameter has a non‐zero value and the mean innovations belong to two separately distinct regimes. We also find that the model with an unknown threshold parameter value consistently outperforms the model with a known threshold parameter value. Copyright © 2016 John Wiley & Sons, Ltd.     

Forecasting realized volatility of oil futures market: A new insight

In this study we propose several new variables, such as continuous realized semi‐variance and signed jump variations including jump tests, and construct a new heterogeneous autoregressive model for realized volatility models to investigate the impacts that those new variables have on forecasting oil price volatility. In‐sample results indicate that past negative returns have greater effects on future volatility than that of positive returns, and our new signed jump variations have a significantly negative influence on the future volatility. Out‐of‐sample empirical results with several robust checks demonstrate that our proposed models can not only obtain better performance in forecasting volatility but also garner larger economic values than can the existing models discussed in this paper.     

A Bayesian nonlinear support vector machine error correction model

The use of linear error correction models based on stationarity and cointegration analysis, typically estimated with least squares regression, is a common technique for financial time series prediction. In this paper, the same formulation is extended to a nonlinear error correction model using the idea of a kernel‐based implicit nonlinear mapping to a high‐dimensional feature space in which linear model formulations are specified. Practical expressions for the nonlinear regression are obtained in terms of the positive definite kernel function by solving a linear system. The nonlinear least squares support vector machine model is designed within the Bayesian evidence framework that allows us to find appropriate trade‐offs between model complexity and in‐sample model accuracy. From straightforward primal–dual reasoning, the Bayesian framework allows us to derive error bars on the prediction in a similar way as for linear models and to perform hyperparameter and input selection. Starting from the results of the linear modelling analysis, the Bayesian kernel‐based prediction is successfully applied to out‐of‐sample prediction of an aggregated equity price index for the European chemical sector. Copyright © 2006 John Wiley & Sons, Ltd.     

Spatial dependence model with feature difference

This paper presents a new spatial dependence model with an adjustment of feature difference. The model accounts for the spatial autocorrelation in both the outcome variables and residuals. The feature difference adjustment in the model helps to emphasize feature changes across neighboring units, while suppressing unobserved covariates that are present in the same neighborhood. The prediction at a given unit incorporates components that depend on the differences between the values of its main features and those of its neighboring units. In contrast to conventional spatial regression models, our model does not require a comprehensive list of global covariates necessary to estimate the outcome variable at the unit, as common macro-level covariates are differenced away in the regression analysis. Using the real estate market data in Hong Kong, we applied Gibbs sampling to determine the posterior distribution of each model parameter. The result of our empirical analysis confirms that the adjustment of feature difference with an inclusion of the spatial error autocorrelation produces better out-of-sample prediction performance than other conventional spatial dependence models. In addition, our empirical analysis can identify components with more significant contributions.     

Hybridizing kernel‐based fuzzy c‐means with hierarchical selective neural network ensemble model for business failure prediction

More and more ensemble models are used to forecast business failure. It is generally known that the performance of an ensemble relies heavily on the diversity between each base classifier. To achieve diversity, this study uses kernel‐based fuzzy c‐means (KFCM) to organize firm samples and designs a hierarchical selective ensemble model for business failure prediction (BFP). First, three KFCM methods—Gaussian KFCM (GFCM), polynomial KFCM (PFCM), and Hyper‐tangent KFCM (HFCM)—are employed to partition the financial data set into three data sets. A neural network (NN) is then adopted as a basis classifier for BFP, and three sets, which are derived from three KFCM methods, are used to build three classifier pools. Next, classifiers are fused by the two‐layer hierarchical selective ensemble method. In the first layer, classifiers are ranked based on their prediction accuracy. The stepwise forward selection method is employed to selectively integrate classifiers according to their accuracy. In the second layer, three selective ensembles in the first layer are integrated again to acquire the final verdict. This study employs financial data from Chinese listed companies to conduct empirical research, and makes a comparative analysis with other ensemble models and all its component models. It is the conclusion that the two‐layer hierarchical selective ensemble is good at forecasting business failure.