Indian Journal of Finance and Banking Vol. 5, No. 2; 2021 ISSN 2574-6081 E-ISSN 2574-609X Published by CRIBFB, USA 1 NEURAL NETWORKS IN FINANCE: A DESCRIPTIVE SYSTEMATIC REVIEW Dr. K. Riyazahmed Assistant Professor Shri Dharmasthala Manjunatheshwara Institute for Management Development Mysore, Karnataka, India E-mail: riyazahmed@sdmimd.ac.in ABSTRACT Traditional statistical methods pose challenges in data analysis due to irregularity in the financial data. To improve accuracy, financial researchers use machine learning architectures for the past two decades. Neural Networks (NN) are a widely used architecture in financial research. Despite the wider usage, NN application in finance is yet to be well defined. Hence, this descriptive study classifies and examines the NN application in finance into four broad categories i.e., investment prediction, credit evaluation, financial distress, and other financial applications. Likewise, the review classifies the NN methods used under each category into standard, optimized and hybrid NN. Further, accuracy measures used by the research work widely differ, in turn, pose challenges for comparison of a NN under each category and reduces the scope of formalizing a theory to choose optimum network model under each category. Keywords: Neural Networks, ANN, Analytics, Machine Learning. JEL Classification Codes: G1, G17, M150. INTRODUCTION Financial data are immensely available, yet the innate nature of big data shows uncertainty, incompleteness, and inconsistency which pose challenges in using traditional statistical methods for financial data analysis (Brooks et al., 2019; Hariri et al., 2019). Financial researchers try to overcome the traditional statistical limitations by using machine learning architecture like Neural Networks (NN). NN imitates the human brain by using nodes and layers of connections, which pass signals with a set of associated weights and bias adjustments (Figure 1). NN results are not easily interpretable and so the analysis is in black-box nature. Irrespective of the non-interpretability of results, NN received importance in financial research due to its computing efficiency in handling financial big data. For the last two decades, financial researchers are using NN in various analyses like risk classification (Altman, Marco, & Varetto, 1994), bankruptcy and share price prediction (Barr & Mani, 1994). Despite the attention, NN application in finance is yet to be well defined. The last decade saw an increase in the finance research using NN in an analysis like forecasting of share prices (Chang, Liu, Fan, Lin, & Lai, 2009; Sapna & Argente, 2003; Safer, 2002), option prices mailto:E-mail:%20riyazahmed@sdmimd.ac.in https://www.cribfb.com/journal/index.php/ijfb Indian Journal of Finance and Banking Vol. 5, No. 2; 2021 2 (Lin & Yeh, 2009; Kohler, Krzyzak, & Todorovic, 2010), and the future prices (Dunis, Laws, & Evans, 2008; Laws & Dunis, 2013). Figure 1. standard Neural Network The Literature reviews of Feldman and Kingdon (1995), Wong and Selvi (1998), Vellido, Lisboa, and Vaughan (1999), Krishnaswamy, Gilbert, and Pashley (2000), Coakley and Brown (2000), Fadlalla and Lin (2001), Wei, Nakamori, Wang, and Yu (2007), Cavalcante et al. (2016), and Huang, Chai, and Cho (2020) are the existing works in this regard. Yet, the review works show significant limitations. Firstly, no review follows a protocol-based review process which is essential for reproducibility. Secondly, several reviews are not examining financial applications entirely or they focus on aspects like soft computing, and computational intelligence, instead of NN architectures (Table 1). Even though the studies analyze the NN applications to a certain extent, non-reproducibility is a serious concern. Further, the absence of a systematic review method results in serious drawbacks in the quality of review findings (Karunananthan, Maxwell, & Welch, 2020). Systematic reviews have greater potential than other research designs leading to the reproducibility of research (Shokraneh & Adams, 2019). Since the computing efficiency doubles every two years (Gustafson, 2011) which improves the efficiency of handling complex data sets, exploring the research works to date with the scientific review methods will help to understand the existing status of NN in analyzing the financial data. Table 1. Summary of existing reviews Study Period of review Nature of study Focus of the study Summary of conclusion Feldman and Kingdon (1995) 1988-1996 (*Authors‟ estimation) Descriptive, Non- Systematic review. Advantages of MLP, BPNN, and SOM. Generalization, architecture selection, and application of selected NN. Wong and Selvi (1998) 1990 – 1996 Descriptive, Review process disclosed. Classification of NN application in finance. The implication to NN developers. Velido, Lisboa, 1992 – 1998 Descriptive, Application of Comprehensively https://www.cribfb.com/journal/index.php/ijfb Indian Journal of Finance and Banking Vol. 5, No. 2; 2021 3 and Vaughan (1999) Review process disclosed. NN in business. reported the most quoted advantages and disadvantages of NN in various business applications. Krishnaswamy et al. (2000) 1989 – 1996 (*Authors‟ estimation) Descriptive, Non – systematic review Description of NN and its finance application. Backpropagation NN has proven robust. Supervised and unsupervised NN is used in finance. Coakley and Brown (2000) 1988 – 1997 (*Authors‟ estimation) Descriptive, Non – systematic review Financial Application, development of ANN models. ANN researchers face a challenge that there are no formal theories for determining optimal network model Fadlalla and Lin (2001) 1986 – 1997 Descriptive, Non – systematic review Financial Application, focus on feedforward and feed backward NN models. NN has great promise for financial applications and combinations of two approaches should be investigated. Calderon and Cheh (2002) 1993 – 1999 Descriptive, sources of review disclosed. NN in auditing and risk management. NN shows promising performance in Preliminary analytical procedures in the auditing process. Wei Huang et al. (2007) NA Descriptive, Non – systematic review Focus on input variables, NN models applied in forex, stock market, and economic forecasting. The prediction performance of neural networks can be improved by being integrated with other technologies. Cavalcante et al. (2016) 2009 -2015 Descriptive, review process disclosed Computational intelligence in finance (NN is a part of the study) Categorized studies into preprocessing, forecasting, and text mining. Huang et al. (2020) 2014 – 2018 Descriptive, review collection process disclosed. Deep learning applications in finance and banking. Reports about data inputs, preprocessing and evaluation rules of deep learning in finance and banking https://www.cribfb.com/journal/index.php/ijfb Indian Journal of Finance and Banking Vol. 5, No. 2; 2021 4 Hence, this structured review reveals an interpretable pattern of NN architectures and their application in finance research. The study follows the systematic review process provided by Moher, Liberati, Tetzlaff, Altman, and The PRISMA Group (2009) and Gupta, Chauhan, and Jaiswal (2019). The primary aim of the study is to classify the research papers based on their NN application in finance research. This is done by,  Classification of major topics and sub-topics, and  Identification of various NN architectures used in the classified subtopics. METHOD A literature review starts with searching for quality research papers from prominent journals (Ngai & Wat, 2002). Further collecting research papers from the online database has become an emerging culture in the information era (Petter & Lean 2009). So, this study uses the EBSCO Business Elite database, which is a repository of 525 peer-reviewed research journals, to collect research papers. The systematic review process prescribed by Moher et al. (2009) comprises defining a protocol for literature search, exclusion criteria of research papers, and final selection of papers (Figure 2). Figure 2. Selection process of research papers (Moher et al., 2009) Protocol directs the research paper selection based on the criteria. To obtain research papers the study followed an advanced search option in the EBSCO Business Elite database and used two keywords „Neural Networks‟ and „Finance‟. Research studies published in English under the subject areas of business, management, and finance are only considered. Empirical articles that are published in peer-reviewed academic journals are collected at the first level. Restricting the review only to published articles can strengthen quality control since many of the academic journals follow meticulous publishing criteria in terms of research contribution and robustness of the results (Light & Pillemer, 1984). The protocol process has helped to collect papers with high research quality. After collection, the studies with incomplete details, irrelevant context, and duplicate publications are excluded. Excluding research papers for incomplete details, duplicate publications, irrelevant context, native language Total N = 51 Scholarly Peer-reviewed articles [Excluding magazines, cover stories] N = 131 Primary search: EBSCO Host Business Elite N = 141 https://www.cribfb.com/journal/index.php/ijfb Indian Journal of Finance and Banking Vol. 5, No. 2; 2021 5 After exclusions, data extraction is done by carefully considering the title, abstract, and overall theme of the paper focuses on applying NN in finance. 141 papers are collected from the EBSCO Business Elite database and during the first level of screening 77 research papers are excluded. Finally, 51 research papers are considered for review. RESULTS The researcher scrutinized the collected research articles for their relevancy and suitability to be considered as a part of this review paper. When research papers fulfill the established criteria, the author read the full paper to find its contributions. Figure 3 represents the broader classification of research papers under major topics and sub-topics. IP – Investment Prediction; CE – Credit Evaluation; FD – Financial Distress. Figure 3. Classification of research papers based on major topics and sub-topics Based on the reading the author identified four main topics and several subtopics (Table 2). The next section discusses each main topic, inferences of the research work carried in the subtopics based on the NN methods. Further, NN with statistical and architectural advancements is classified as 'Optimized NN'. NNs incorporating financial theories and knowledge are classified under 'Hybrid NN'. A model-free NN is classified as 'Standard NN'. Table 2. Major research topics and subtopics Main topics Sub-topics Investment prediction Prediction of options prices, futures prices, share prices, Forex, indexes, bond yields, commodity spreads, trading patterns, shareholder wealth, and portfolio performance. Credit evaluation Predicting credit risk, and Estimating credit rating. Financial distress Evaluating financial distress Other financial applications Development of financial intelligent system, Detecting fraudulent financial reporting, Assessment of systematic risk, Evaluating operating performance, Assessing project portfolio performance. https://www.cribfb.com/journal/index.php/ijfb Indian Journal of Finance and Banking Vol. 5, No. 2; 2021 6 Figure 4. Framework of NN application in finance DISCUSSION Investment Prediction: Derivatives - a) Options Optimized NN – Back Propagation NN (BPNN) – BPNN minimizes the prediction error by giving the nodes with higher error rates lower weights and vice versa. Hence, BPNN is found to be more suitable for derivative prediction (Kaastra & Boyd, 1995). When used to predict Taiwan stock index options, BPNN demonstrated improved accuracy in support of hedging at in-the-money option (Lin & Yeh, 2009). Likewise, the research study of Hutchison et al. (1994) used BPNN to predict the prices of S & P 500 futures and options, yet the result is that BPNN did not show significantly better accuracy than other linear models like ordinary least squares. NN with Monte Carlo simulation – Monte Carlo, is a simulation technique to understand the impact of risk and uncertainty in prediction. It is used with linear NN for predicting American options (Kohler et al., 2010). Since it is a simulation based NN, the accuracy would widely differ in empirical prediction. Advanced Modular NN (AMNN) – AMNN is a series of independent NN which serves as a module and operates on separate inputs to accomplish a subtask. AMNN gives more accuracy than the standard NN when predicting European call option prices (Gradojevic et al., 2009). https://www.cribfb.com/journal/index.php/ijfb Indian Journal of Finance and Banking Vol. 5, No. 2; 2021 7 Feed forward NN (FFNN) – FFNN is used when the nature of financial data is neither sequential nor time-dependent. FFNN predicts European Index options and S&P 500 European call options more accurately than standard NN (Gencay & Gibson, 2007). Hybrid NN - NN with Black Scholes - The research study by Blynski, and Faseruk (2006) compared the effectiveness of option price forecasting using the traditional Black Scholes model with NN (hybrid NN) and standard NN. Likewise, the research study by Chen and Sutcliffe (2012) confirms that a hybrid NN along with Black Scholes predicts accurately than the standard NN or Black Scholes model individually. Similarly, Sperckelsen et al. (2014) used Black Scholes model variables in predicting option prices of currency futures (EUR/USD and concluded that the hybrid model is better than the theoretical option pricing model and MLP. NN with Black Scholes & Wavelet – Zapart (2003) in his research study uses Wavelet along with NN and Black Scholes. Wavelet is a mathematical advancement that addresses oscillations that decays quickly in a data set. The study found that Black Scholes NN with Wavelet predicts superior when analyzing option prices. Hybrid Black Scholes NN with stochastic volatility - Stochastic volatility represents the nature of volatility fluctuating over time. The research study by Gencay and Gibson (2007) found that the hybrid NN model with stochastic volatility predicts better than the standard NN Model while predicting European stock index options. b) Futures – Standard NN – Model-free NN, without the attributes of financial theories, also performs significantly in the case of predicting currency futures prices. A model-free NN is used to predict the high-frequency currency futures and the predictability power is better than the closed-form financial model (Sperckelsen et al., 2014). Optimized NN - Multilayer perceptron (MLP) – MLP commonly represents a feed-forward NN with three layers. MLP is used to predict the commodity futures to hedge against corn and ethanol spreads and found to be accurate in prediction (Dunis et al., 2015). Likewise, Karathanasapoulos et al. (2016) used MLP in gasoline futures contracts. Higher-order NN (HONN) -HONN utilizes a higher combination of NN inputs. The research study of Dunis et al. (2015) compared the performance of HONN with MLP. The study concludes that MLP and HONN are superior in predicting with leveraging option. Sermpinis et al. (2013) tested HONN in predicting index futures. Likewise, Karathanasapoulos et al. (2016), in their research study used HONN to predict gasoline futures contracts. Radical base function neural network (RBFNN)– Karathanasapoulos et al. (2016) used a radical basic function neural network (RBFNN) to improve the trading performance of futures. RBFNN transforms the input signal into another form, which can be then feed into the network to get linear separability. The study concludes that RBF NN is superior in both trading performance and statistical accuracy. https://www.cribfb.com/journal/index.php/ijfb Indian Journal of Finance and Banking Vol. 5, No. 2; 2021 8 Hybrid NN - Feature transformed NN – Kim (2004) used feature transformed NN, including domain-specific factors like relative strength index to predict the futures prices and found that feature transformed NN predicts accurately than the linear models and concludes that incorporating domain knowledge in NN architecture improves performance. Shares & Indices – Standard NN – Haefke and Helmenstien (2002) used NN in forecasting indices and inferred that applicability of information criteria is important in the selection of NN. In contrast, Moreno and Olmeda (2007) claim that NN is not superior in predicting the stock markets to the linear models. Optimized NN - NN with Statistical optimization – LV D et al. (2020) used principal component analysis (PCA), Least absolute shrinkage and selection operators (LASSO), classification and regression trees (CART), and Piecewise linear representation methods (PLR) to optimize NN. As a result, there no significant improvement in NN incorporating the features, however, NN with PLR, resulted in an improvement in profit through better forecast ability. Back propagation (BPNN) and Piecewise linear representation (PLR) – In a research study by Chang et al. (2009), it is found that BPNN along with PLR consistently created good results for predicting upward, steady, and downward trends of stock prices. Hybrid NN – NN with Fama French five-factor model - Besides customizing the NN, researchers have used financial models like Fama and French five-factor model with NN and found improvement in the profitability of investors in both linear and nonlinear data (Jan & Ayub, 2019). NN with GARCH model– Ozbey and Paksoy (2020) combined GARCH with NN and compared the performance of the hybrid model with the classic GARCH model. The study found that the hybrid model is superior in predicting volatility to the classic GARCH. NN with Top-down theory, technical analysis, and dynamic time series methods –Huang, G.,Huang,GB., Shiji, and Youa (2014) used integrated models using conventional top-down trading theory, technical analysis, and dynamic time series methods and concludes that the hybrid system gives remarkable investment returns and demonstrates promising potential tools for stock market forecasting. Other related financial forecasts – Optimized NN – Back propagation NN (BPNN) –Chiang et al. (1996) used BPNN to predict the Net Asset Value (NAV) of mutual funds and found that BPNN outperforms the regression model. Jain and Nag (1995) predicted the prices of initial public offering (IPO) using BPNN and found significant economic benefits in BPNN. https://www.cribfb.com/journal/index.php/ijfb Indian Journal of Finance and Banking Vol. 5, No. 2; 2021 9 Multi-layer perception (MLP) – Indroa et al. (1999) used MLP and compared it with stepwise linear regression. The results show that MLP is superior to the linear model. General regression NN (GRNN) –Barnes and Lee (2009) used GRNN in analyzing the effect of macroeconomic and firm-specific factors in determining shareholder wealth. Multivariate NN -Wie et al. (2004) used Multivariate NN to predict earning per share in comparison with univariate and multivariate linear models incorporating fundamental accounting variables and found that NN along with accounting variables predicts more accurately than linear forecasting models. Table 3. NN research in Investment prediction Author(s) Purpose NN Model Sample Period Predictors Comparison R Squared Value Accuracy 1 Kaastra and Boyd (1995) Forecastin g Economic time- series data Back Propagation NN Conceptu al Paper - - - - - 2 Lin and Yeh (2009) Forecastin g Option Prices Back Propagation NN 15 582 call option price data points 2003 - 2004 Black Scholes variables - MAPE 5.2534 3 Hutchison et al. (1994) Pricing and Hedging derivative securities Ordinary Least squares, Radical Basis functions network, Multi-layer Perceptron, Projection Pursuit S&P 500 Future and options 1987 - 1991 Black Scholes variables No significant difference between models 84.76 - 4 Kohler et al. (2010) Pricing of American Options Least square NN Monte Carlo Simulated Data - - - - - 5 Gradojevic et al. (2009) Pricing European Call options Modular NN & Black Scholes NN S&P-500 index European call option prices, Chicago Board Options Exchange 1987 - 1994 Black Scholes variables BS NN Model > Modular NN - MAPE 1.87 6 Gencay and Gibson (2007) Pricing European Stock Index options Feedforward NN S&P 500 index9 options from the Berkeley 1989 - 1991 Price of the underlying, strike price, volatility, interest FFNN > Stochastic volatility (SV) and stochastic - - https://www.cribfb.com/journal/index.php/ijfb Indian Journal of Finance and Banking Vol. 5, No. 2; 2021 10 Option Database rate, time to maturity volatility random jump (SVJ), 7 Blensky and Fasurek (2006) Comparin g Option prices forecast of NN with Black Scholes model Back Propagation NN 64, 280 OEX 100 Index call option 1986 – 1993 - NN > Black Scholes model - - 8 Chen and Sutcliffe (2012) Pricing and hedging short sterling options Standard NN, Modified Black Model NN, and Hybrid NN Short sterling futures traded on NYSE 2012 (Quarterl y expiry cycle) Ask, bid, trade, spread trade, and block trade, Hybrid NN > Modified Black model, Standard NN r = 98.6864 - 9 Sperckelsen et al. (2014) Realtime pricing of options on currency futures Model-free option pricing NN, Multi- Layer Perceptron EUR/US D option on currency future from Chicago Mercantil e Exchange (CME) 2012 Futures price, Strike price, Expiration time, Risk- free rate, Asset volatility Hybrid NN > theoretical option pricing model 99% MAPE 0.3146 1 0 Zapart (2003) Pricing European and American Call options NN with Binomial trees and Wavelets, NN with Genetic algorithm and Black Scholes model Options prices as quoted on the Chicago Board Options Exchange are used 2003 Time to expiry, current stock price, risk-free rate - - - 1 1 Dunis et al. (2015) Modeling corn/ethan ol crush spread MLP, HONN, GPA Ethanol futures contract traded in Chicago Board 2005 - 2010 Leverage GPA > HONN, MLP - - 1 3 Sermpinis et al. (2013) The forecastin g FTSE 100 futures Higher-order NN, Multi- Layer Perceptron, Recurrent neural networks FTSE 100 futures 2007 - 2008 Realized daily returns (21 days) HONN> MLP, RNN - 18.85% 1 4 Karathanaso poulos et al. (2016) Modeling crack spread RBF, PSO, MLP - 2005 - 2015 20 ARIMA and 10 GARCH models PSORBF > MLP - - 1 Kim (2004) Future Feature Korean May to Positive Feature - - https://www.cribfb.com/journal/index.php/ijfb Indian Journal of Finance and Banking Vol. 5, No. 2; 2021 11 5 price prediction transformed ANN based on domain knowledge stock index (KOSPI) Novemb er 1996 volume index, Rate of Change, Momentum , etc. transformed ANN > Linear ANN 1 6 Haefke and Helmenstien (2002) Index Forecastin g and Model Selection Feedforward NN Austrian Traded Index (ATX) 2 Novemb er 1992 to 14 October 1994 Geometric mean, arithmetic mean The proposed integrated model shows significant performanc 0.041 AMAPE = 1.862 1 7 Moreno and Olmeda (2007) Predictabi lity of emerging and developed stock markets using NN Standard NN 49 MSCI (Morgan Stanley Capital Internatio nal) indexes March 1995 to March of 2001 (1560 daily observati ons) index returns, daily and weekly NN is not superior to the linear models - - 1 8 LV D et al. (2020) Dimensio nality reduction in stock trading MLP, Deep Belief Network (DBN), Stacked Auto- Encoders (SAE), RNN, LSTM, Gated Recurrent Unit (GRU) US SPICS and the Chinese CSICS Past 2000 trading days of SPICS and CSICS before Decembe r 31, 2017 44 technical, Volatility, Psychologi cal, cash flow indicators. LASSO NN - - 1 9 Chang et al. (2009) Stock Trading Points Prediction BPNN, GA, PLR Stock Prices 2004/01/ 02 to 2006/04/ 12 Moving average, Bias, RSI, ninety days stochastic line, etc. PLR +GA improves Profitability - - 2 0 Jan and Ayub (2019) Improving the predictabil ity of Fama French five-factor model Standard NN Manufact uring companie s in Pakistan Stock Exchange 2000 to 2015 Market cap, BV/MV ratio, % in total assets, and EBIT NN improves FF model r = 0.99989 MSE = 0.0012 2 1 Ozbey and Paksoy (2020) Estimatio n of index returns with GARCH and NN Hybrid NN, Exp GARCH, and Nor. Distrn Borsa Istanbul 100 price Index 2017 - 2018 Borsa Istanbul 100 Index value Hybrid NN + GARCH > Hybrid NN + normal distribution - MSE = 0.015926 2 2 Huang, Huang, Shiji, and Youa (2014) Integrated data mining in stock Top-down trading theory + ANN + technical Taiwan Semicond uctor Manufact 2011 - 2013 Stochastic KD, William %R, RSI, Integrated NN model improves stock - True positive = 98.50% https://www.cribfb.com/journal/index.php/ijfb Indian Journal of Finance and Banking Vol. 5, No. 2; 2021 12 forecastin g analysis + dynamic time series + and Bayesian probability uring Company and Evergreen Marine Corporati on PSY line, ADX, MA, MACD forecasting 2 3 Chiang et al. (1996) Mutual Fund NAV forecastin g BPNN 6 Year economic variables and 101 US mutual funds 1981 - 1986 GNP, Consumpti on, Investment, CPI, Money supply, unemploy ment, T- bill, Long term rate BPNN > Linear & Non-Linear regression 0.989 MAPE = 8.76 2 4 Jain and Nag (1995) Predicting IPO pricing FFNN 552 IPOs in the United States 1980 - 1990 11 variables [Size, Underwrite r, sales, ROA, ROI, Assets turnover, etc.] - - - 2 5 Indroa et al. (1999) Predicting mutual fund performan ce MLP Morningst ar Mutual Funds On-Disc database 1993 - 1995 Annualized return, turnover, P/E, P/B, Mar.Cap MLP > Linear models - MAPE = 4.88 2 6 Barnes and Lee (2009) Effects of Macroeco nomic- Firm- Specific Factors on Sharehold er Wealth General regression NN (GRNN) Miscellan eous Industrials in the Australian Stock Market 2007 D/E, Gross margin, Debt to cash, EVA, EPS, WACC funds, ROIC ANN is effective in the prediction 0.0548 MAE = 37.649 2 7 Wie et al. (2004) NN model for EPS forecastin g Univariate- NN and multivariate NN Quarterly EPS of 283 companie s in SEC 1992 – 2002 Inventory, A/R, capital expenditure , gross margin, Sel.Adm exp, Tax rate, labour force NN models > Linear models - MAPE = 0.362 https://www.cribfb.com/journal/index.php/ijfb Indian Journal of Finance and Banking Vol. 5, No. 2; 2021 13 Credit Risk Prediction: Credit evaluation – Optimized NN – Bayesian regularized NN (BRNN) - Sariev and Germano (2020) used BRNN to Predict the Probability of Default and found BRNN superior in prediction. Back propagation NN (BPNN) – Loss-given default (LGD) is used in credit risk assessment. LGD is the share of an asset that is lost if a borrower default. Loterman et al. (2012) compared the nonlinear techniques with the linear counterparts in predicting LGD of major international banks. Feed forward NN (FFNN) – Qi and Zhao (2011) found that nonparametric method like FFNN and regression tree predicts the LGD accurately both in and out of sample than their parametric counterparts. Cifter et al. (2009) investigated the relationship between industrial production and credit defaults (nonperforming loans) using FFNN based on wavelet decomposition. Fuzzy mathematical model –Aiqun et al. (2020) applied NN in risk assessment of logistic finance using back propagation NN and Fuzzy mathematical model. The study found that NN with the fuzzy mathematical model is accurate in risk assessment. Further, Baesens et al. (2003) provided a table with a graphical format that facilitates easy consultation to interpret the NN results. Credit Scoring – Standard NN - Chikolwa and Chan (2008) compared NN with ordinal regression (OR) to study the determinants of Commercial Mortgage-Backed Securities (CMBS) and concluded that NN is superior in prediction to OR. Trinkle and Baldwin (2007) applied NN in credit evaluation for loan finance and concluded that NN can be used in the credit scoring process with caution because of its hidden nature. Optimized NN – Backpropagation NN– Hajek (2011) applied NN to rate the United States municipalities in the state of Connecticut and found a higher accuracy of NN in classifying the municipalities with a limited subset of variables. Zan et al. (2004) compared the performance of Support vector machines (SVM) with back propagation NN (BPNN) on the credit rating of companies and found that both SVM and BPNN have the same accuracy in predicting the credit rating. Table 4. NN literatures in Credit Risk analysis Author(s) Purpose NN Model Sample Output Variable Predictors Comparison R Squared Value Accuracy 1 Sariev and Germano (2020) Estimatio n of the probabilit y of default BRNN, BPNN East Europea n, German, and Polish 2007 - 2012 (EE), 2007 - 2013 (P) Payables turnover, ROA, cash ratio, sales/total assets, BRNN > BPNN - - https://www.cribfb.com/journal/index.php/ijfb Indian Journal of Finance and Banking Vol. 5, No. 2; 2021 14 data LA/TA, interest coverage 2 Lotterman et al. (2012)# Benchmar king regression algorithms for loss given default modeling NN, SVM, and OLS. six LGD datasets from internati onal banks - - SVM, NN > Linear Models 0.1295 MAE = 0.3118 3 Qi and Zhao (2011)# Comparis on of modeling methods for Loss Given Default OLS, fractional response regression (FRR), inverse Gaussian regression (IGR), and inverse Gaussian regression with beta transforma tion (IGR- BT), and regression tree (RT), NN 3751 defaulte d securitie s in the US, Moody‟ s Ultimate Recover y Databas e 1985 to 2008 - NN, regression tree > linear regression, fractional response, OLS 0.576 - 4 Cifter et al. (2009) Examine the relationshi p between industrial productio n and credit defaults FFNN using Wavelet decomposi tion 83 monthly observat ions Industria l producti on and credit default rates are from Central Bank of Turkey 2001 to 2007 Industrial production, Credit defaults all sectors, Wholesale, and retail trade Industrial cycle affects the sectoral credit-default cycles at different. time scales - MSE = 0.00022 5 Aigun et al. (2020) Risk assessmen t of logistic finance BPNN and Fuzzy mathemati cal - 2019 - NN +Fuzzy is accurate in the prediction - - 6 Baesens et al. (2003) Rule Extraction and Decision Tables MLP, Neuro rule, Trepan, and Nef German credit dataset from UCI - Term of loan, Purpose, savings account Extract very compact rule sets and trees for all data sets - - https://www.cribfb.com/journal/index.php/ijfb Indian Journal of Finance and Banking Vol. 5, No. 2; 2021 15 for Credit- Risk Evaluatio n class. repositor y, Bene1 and Bene2 datasets from Benelue x financial institutio ns balance, income, property, No. of years as a client, Economical sector 7 Chikolwa and Chan (2008) Determina nts of credit ratings Standard NN Vs Ordinal regression MBS credit ratings of Standard and Poor 1999 - 2005 Loan to value; Debt Service Coverage Ratio; issue size; bond tenure, property diversity, geographica l diversity, CMBS rating ANN > Ordinal regression Pseudo R squared = 0.018 Classificati on accuracy = 80% 8 Trinkle and Baldwin (2007) Interpreta ble credit model developm ent using NN ANN models created from previous research studies Two German consume r credit data sets, SAS data repositor y - Age, car, cards, Cash, etc. ANN > General credit models - Accuracy rate = 0.63084 9 Hajek (2011) Municipal credit rating modeling by neural networks FFNN, RBFNN, Probabilist ic NN, Cascade correlation NN, Group method of data handling (GMDH) polynomia l NNs, Support Vector Machines. Credit informat ion‟s of r 169 US municip alities (located in the State of Connect icut) 2003 - 2007 Population. Population growth, median family income, unemploym ent rate, total revenue to total expenditure, tax revenue to total revenue, tax collectibles, debt service, total debt to the total population, tax An accurate credit rating classification - PNN, Classificati on Accuracy test = 98.8% https://www.cribfb.com/journal/index.php/ijfb Indian Journal of Finance and Banking Vol. 5, No. 2; 2021 16 collection rate, form of income. 1 0 Zan et al. (2004) Credit rating analysis with support vector machines and neural networks backpropa gation neural network (BNN) Vs Support vector machines Taiwan Ratings Corporat ion, Securitie s and Futures Institute, S&P Compus tat data set. 1991 to 2000 TA, TL, DE, CR, ROA, ROE, EPS, NOI, NII, etc. BNN, SVM >Linear regression - Accuracy rate = 80% # Cross validation Financial Distress: Optimized NN – Multi-layer perceptron (MLP) –Loukeris and Eleftheriadis (2015) used MLP, hybrid MLP with neurogenetic, and voted perceptron algorithm (VPA). VPA is a method that linearly separates data with a larger margin to predict financial distress. Manel (2012) used five MLP and compared them with traditional financial analysis to predict financial distress and found that the MLP is superior in accuracy. Learning vector quantization (LVQ) – LVQ is a NN method with a supervised algorithm to let choose the number of training instances to hang on to. Brockett et al. (2006) compared the multiple discriminant analysis (MDA) and logistic regression with LVQ to analyze the solvency of life insurance companies and found NN architectures are superior in predictions. Hidden layer learning vector quantization – Like LVQ, hidden layer LVQ (HDLVQ) outperforms the traditional NN methods and financial techniques while evaluating the corporate solvency of life insurance companies. The research study by Neves and Vieira (2006) integrated HDLVQ to correct the outputs of MLP and found the technique is superior to traditional techniques like z core models and standard NN. Hybrid NN – Fuzzy analytical hierarchy and CAMEL model – CAMELS framework is the most widely applied methodology to study the financial position of banks. Wanke et al. (2016) used NN with a fuzzy analytical hierarchical model along the CAMELS framework to predict the financial distress of banks. Z score model –Z score model is a financial technique to evaluate the financial distress of a company. Pradhan (2011) used NN along with Z score and found it classifies accurately. Profitability index and capital structure variables –Willer et al. (2020) created a business insolvency forecasting model using NN and found that the predictable power of NN shows significant accuracy. Yang et al. (1998) and Atiya (2001) confirms the accuracy. https://www.cribfb.com/journal/index.php/ijfb Indian Journal of Finance and Banking Vol. 5, No. 2; 2021 17 Table 5. NN literatures in Financial distress Author(s) Purpose NN Model Sample Output Variable Predictors Comparison R Squared Value Accuracy 1 Loukeris and Eleftheria dis (2015) Credit portfolio selection process MLP, hybrid MLP with neurog enetic, and voted percept ron algorith m 1411 compan ies from Greek commer cial bank 1994 to 1997 16 Financial ratios - - MSE 0.034 2 Manel (2012) Predictio n of financial distress BPNN 528 Tunisia n firm (Central bank of Tunisia report) 1999 - 2006 26 financial ratios predicting distress - - Classificati on accuracy – 98.9% 3 Brockett et al. (2006) Compari son of NN and statistical models for life insurers' financial distress predictio n back- propag ation and learnin g vector quantiz ation (LVQ) Vs multipl e discrim inant analysi s and logistic regressi on analysi s Texas Depart ment of Insuran ce data 1991 to 1995 IRIS variables BNN, LVQ > MDA, Logistic Regression - Correct rate (1994) = LVQ (1.00), BP (0.971) 4 Neves and Vieira (2006)# Improvin g Bankrupt cy Predictio n with Hidden Layer Learnin g vector quantiz ation + Multi- layer percept ron 780,000 financia l stateme nts of French compan ies, Industri 1998 - 2000 Input consists of 30 financial ratios Generalisat ion error: MLP = 8.8%, HLVC-Q = 7.3% https://www.cribfb.com/journal/index.php/ijfb Indian Journal of Finance and Banking Vol. 5, No. 2; 2021 18 Learning Vector Quantiza tion al French firms, 583 bankrup t firms 5 Wanke et al. (2016) Predictin g performa nce in ASEAN banks Fuzzy analytic hierarc hy process NN Financi al ratios of 88 Associa tion of Southea st Asian Nations bank 2010 to 2013 CAMELS ratios Possible to explain the causes of inefficiency using NN - RMSE = 0.0248 6 Pradhan (2011) Predictio n of financial distress BPNN State Bank of India 2001 – 2010 Z score variables - - - 7 Willer et al. (2020) Forecasti ng business insolvenc y Standar d NN - - Profitability index, capital structure - - 91% Accuracy Dynamic Model 8 Yang et al. (1998) Probabili stic Neural Network s in Bankrupt cy Predictio n Fisher discrim inant analysi s, BPNN, PNN, PNN without patterns normali zed. 122 compan ies U.S. oil and gas industry 1984 to 1989 Net cash flow to total assets, TD/TA, CL/TD, etc Fischer discriminant analysis, PNN normalized> BP NN, Probabilistic NN without a pattern - Fisher discriminan t analysis = 87% correct classificati on, 9 Atiya (2001) Bankrupt cy Predictio n for Credit Risk Standar d NN Defaulte d and from solvent US firms, 716 solvent firms and 195 defaulted firm - Merton‟s asset- based model - - Correct rate = 85.50% # Cross validation Other Financial Applications: Standard NN Apart from the major topics of research, researchers have found NN is efficient to detect fraudulent reporting (Koskivaara & Back, 2007; Omar et al. 2017) and project portfolio management (Costantino et al., 2015). https://www.cribfb.com/journal/index.php/ijfb Indian Journal of Finance and Banking Vol. 5, No. 2; 2021 19 Optimized NN – A fuzzy analytical model is used for creating a financial information system (Wang et al., 2020). A self-organization map algorithm, an unsupervised learning NN that produces a low- dimensional, discretized representation of the input space of the training samples, is called a map. SOM is used to analyze the integration of EU capital markets (Horobet 2014), differences in world economies (Cimpoeru 2015). Hybrid NN – NN with the Ohlson model is used to predict operating performance (Ying-Hua & Shih-Chin, 2013). CONCLUSION A descriptive systematic review was conducted to find the application of neural networks in financial research. The study found a keen research interest to use NN for predicting financial data. This is obvious from the statistic that about 53% of the collected research studies applied NN in investment prediction. Credit evaluation and financial distress topics contribute to 20% and 17% of each of the collected papers. There are very few works (10%) found on other financial aspects. The following are the reflections of the review. First, it is observed that the researchers have used arbitrary data partition and architecture selection in all the research works. Besides, the performance or evaluation metrics widely differ among the collected research studies, giving less scope for comparing the accuracy of an NN architecture in a particular subtopic. Hence, in this study, a meta-analytic comparison to generalize the NN architectures and formalizing a theory to choose a suitable NN method under a topic has serious limitations. Second, there are few studies (Neves & Vieira, 2006; Qi & Zhao, 2011) that have performed the cross-validation in NN models. Cross-validation increases efficiency in using financial data as every observation is used for both training and testing which results in a more accurate estimate of out-of-sample prediction. Further, overfitting and underfitting of data will be managed efficiently through cross-validation. Third, unlike prediction, in the research studies of classifying financial data, there is a scope for a meta-analysis based on generalizing Area Under Curve (AUC) that help to estimate the accuracy of classification on a particular topic. Further, the study observed NN architectures including domain-specific knowledge performs with more accuracy. Hence, more domain-based Hybrid NN architectures can be trained. Besides, the review has the following limitations. The descriptive systematic review has examined only the research papers published and available under the EBSCO database. Consequently, the works of the literature review are prone to publication bias, which occurs with publishing only statistically significant results. Beyond, the research works in conference proceedings and working papers are not reviewed. Hence some sub-topics and main research topics might have been remaining uncovered. Systematically including more research papers from other sources will improve the chance of a meta-analysis of NN in financial research. Since meta-analysis on machine learning by Krittanawong et al. (2020) and Roelofs et al. (2019) are the only works available, that too in the medical domain, a meta-analysis of NN architectures in financial research will be a significant contribution to the existing financial and NN literature. https://www.cribfb.com/journal/index.php/ijfb Indian Journal of Finance and Banking Vol. 5, No. 2; 2021 20 REFERENCES Altman, E., Marco, G., & Varetto, F. (1994). Corporate Distress Diagnosis: Comparisons Using Linear Discriminant Analysis and Neural Networks (the Italian Experience). Journal of Banking and Finance, 18(3), 505-529. Aiqun, W., Zicong, H., Yilin, W., Kolivand, H., Balas, V. E., Paul, A., & Ramachandran, V. (2020). Risk assessment of logistics finance enterprises based on BP neural network and fuzzy mathematical model. Journal of Intelligent & Fuzzy Systems, 39(4), 5915–5925. https://doi.org/10.3233/JIFS-189066 Atiya, A.F. (2001). Bankruptcy prediction for credit risk using neural networks: A survey and new results, IEEE Transactions on Neural Networks, 12(1), 929-935. Baesens, B., Setiono, R., Mues, C., & Vanthienen, J. (2003). Using Neural Network Rule Extraction and Decision Tables for Credit-Risk Evaluation. Management Science, 49(3), 312–329. https://doi.org/10.1287/mnsc.49.3.312.12739 Barr, D., & Mani. G. (1994). Using neural nets to manage investments, AI EXPERT, 34(3), 16– 22. Barnes, B, M., & Lee, C, S, V. (2009). Effects of Macroeconomic and Firm-Specific Factors on Shareholder Wealth: Some Australian Evidence. The Journal of wealth management, 12 (1), 41-61. https://doi.org/10.3905/jwm.2009.12.1.041 Brockett, P. L., Golden, L. L., Jang, J., & Yang, C. (2006). A Comparison of Neural Network, Statistical Methods, and Variable Choice for Life Insurers‟ Financial Distress Prediction. Journal of Risk & Insurance, 73(3), 397–419. https://doi.org/10.1111/j.1539- 6975.2006.00181.x Brooks, C., Andreas, G, F., Hoepner, McMillan, D., Vivian, A, & Simen, C, W. (2019) Financial data science: the birth of a new financial research paradigm complementing econometrics? The European Journal of Finance, 25(17), 1627-1636, https://doi.org/10.1080/1351847X.2019.1662822 Blynski, L., & Faseruk, A. (2006). Comparison of the effectiveness of option price forecasting: Black-Scholes vs. simple and hybrid neural networks. Journal of Financial Management & Analysis, 19(2), 46. Cavalcantea, R, C., Brailerio, C, R., Souza, L, V., Nobrega, P, J., Oliveria, L, A. (2016). Computational Intelligence and financial markets: A survey and future directions. Expert systems with applications, 55, 194 – 211. http://dx.doi.org/10.1016/j.eswa.2016.02.006 Chang, P. C., Liu, C. H., Fan, C. Y., Lin, J. L., & Lai, C. M. (2009). An Ensemble of Neural Networks for Stock Trading Decision Making. In: Huang DS., Jo KH., Lee HH., Kang HJ., Bevilacqua V. (eds) Emerging Intelligent Computing Technology and Applications. With Aspects of Artificial Intelligence. ICIC 2009. Lecture Notes in Computer Science, 5755. https://doi.org/10.1007/978-3-642-04020-7_1 https://doi.org/10.3233/JIFS-189066 https://doi.org/10.1287/mnsc.49.3.312.12739 https://doi.org/10.3905/jwm.2009.12.1.041 https://doi.org/10.1111/j.1539-6975.2006.00181.x https://doi.org/10.1111/j.1539-6975.2006.00181.x https://doi.org/10.1080/1351847X.2019.1662822 http://dx.doi.org/10.1016/j.eswa.2016.02.006 https://doi.org/10.1007/978-3-642-04020-7_1 https://www.cribfb.com/journal/index.php/ijfb Indian Journal of Finance and Banking Vol. 5, No. 2; 2021 21 Chen, F., & Sutcliffe, C. (2012). Pricing and Hedging Short Sterling Options Using Neural Networks. Intelligent Systems in Accounting, Finance & Management, 19(2), 128–149. https://doi.org/10.1002/isaf.336 Chikolwa, B., & Chan, F. (2008). Determinants of Commercial Mortgage-Backed Securities Credit Ratings: Australian Evidence. International Journal of Strategic Property Management, 12(2), 69–94. https://doi.org/10.3846/1648-715X.2008.12.69-94 Chiang, W.C., Urban, T. L. & Baldridge, G.W. (1996). A neural network approach to mutual fund net asset value forecasting, Omega - International journal of management science, 24, 205–215. Cifter, A., Yilmazer, S., & Cifter, E. (2009). Analysis of sectoral credit default cycle dependency with wavelet networks: Evidence from Turkey. Economic Modelling, 26(6), 1382–1388. https://doi.org/10.1016/j.econmod.2009.07.014 Cimpoeru, S. (2017). Using Self organizing maps for assessing systemic risk. Evidences form global economic crisis. The Bucharest University of Economic Studies. http://www.ecocyb.ase.ro/nr20152/05%20-%20Cimpoeru_Smaranda.pdf Coakley, J., & Brown. C. (2000). Artificial neural networks in accounting and finance: Modelling issues. Intelligent Systems in Accounting Finance & Management,9(2). Costnatino, F., Gravio, G, D., & Nonino, F. (2015). Project selection in project portfolio management: An artificial neural network model based on critical success factors. International journal of project management, 33, 1744 – 1754. http://dx.doi.org/10.1016/j.ijproman.2015.07.003 Dunis, C. L., Laws, J., Middleton, P. W., & Karathanasopoulos, A. (2015). Trading and hedging the corn/ethanol crush spread using time-varying leverage and nonlinear models. European Journal of Finance, 21(4), 352–375. https://doi.org/10.1080/1351847X.2013.830140 Fadlalla, A., & Lin, C. (2001) An analysis of the applications of neural networks in finance. Interfaces, 31(4), 112–122. Feldman, K, & Kingdon, J. (1995). Neural networks and some applications to finance, Applied Mathematical Finance, 2(1), 17-42. https://doi.org/10.1080/13504869500000002 Gençay, R., & Gibson, R. (2007). Model Risk for European-Style Stock Index Options. IEEE Transactions on Neural Networks, 18(1), 193–202. https://doi.org/10.1109/TNN.2006.883005 Gradojevic, N., Gençay, R., & Kukolj, D. (2009). Option Pricing With Modular Neural Networks. IEEE Transactions on Neural Networks, 20(4), 626–637. https://doi.org/10.1109/TNN.2008.2011130 https://doi.org/10.1002/isaf.336 https://doi.org/10.3846/1648-715X.2008.12.69-94 https://doi.org/10.1016/j.econmod.2009.07.014 http://www.ecocyb.ase.ro/nr20152/05%20-%20Cimpoeru_Smaranda.pdf https://doi.org/10.1080/1351847X.2013.830140 https://doi.org/10.1080/13504869500000002 https://doi.org/10.1109/TNN.2006.883005 https://doi.org/10.1109/TNN.2008.2011130 https://www.cribfb.com/journal/index.php/ijfb Indian Journal of Finance and Banking Vol. 5, No. 2; 2021 22 Gupta, P., Chauhan, S. & Jaiswal, M.P. (2019). Classification of Smart City Research - a Descriptive Literature Review and Future Research Agenda. Information System Frontiers, 21, 661–685. https://doi.org/10.1007/s10796-019-09911-3 Gustafson, J., L. (2011). Moore‟s Law. In: Padua D. Encyclopedia of Parallel Computing. Springer, Boston, MA. https://doi.org/10.1007/978-0-387-09766-4_81 Hutchison A.W, Lo A.W, & Poggio, T. (1994) A nonparametric approach to pricing and hedging derivative securities via learning networks. The journal of Finance, 49(3), 851- 889. Hajek, P. (2011). Municipal credit rating modelling by neural networks. Decision support systems, 51 (1), 108 – 118. https://doi.org/10.1016/j.dss.2010.11.033 Hariri, R. H., Fredericks, E.M. & Bowers, K. M. (2019). Uncertainty in big data analytics: survey, opportunities, and challenges. Journal of Big Data,6. https://doi.org/10.1186/s40537-019-0206-3. Haefke, C., & Helmenstein, C. (2002). Index forecasting and model selection. International Journal of Intelligent Systems in Accounting Finance & Management, 11(2), 119–135. https://doi.org/10.1002/isaf.214 Horobet, A., Belascu, L., Ionita, I., & Serban - Oprescu, A.-T. (2014). A Neural Networks Perspective on the Financial integration of European Capital Markets. Economic Computation & Economic Cybernetics Studies & Research, 48(1), 1–12 Huang, G., Huang, G, B., Shiji, S., & Youa, K. (2014). Trends in extreme learning machines: A review. Neural Networks. 61, 34 – 45. http://dx.doi.org/10.1016/j.neunet.2014.10.001 Huang, J., Chai, J. & Cho, S. (2020). Deep learning in finance and banking: A literature review and classification. Frontiersof Business Research in China, 14. https://doi.org/10.1186/s11782-020-00082-6 Indro, D. C., Jiang, C. X., Patuwo, B. E., & Zhang, G. P. (1999). Predicting mutual fund performance using artificial neural networks. Omega, 27(3), 373-380. https://doi.org/10.1016/S0305-0483(98)00048-6 Jan, M. N., & Ayub, U. (2019). Do the Fama and French Five-Factor Model Forecast Well Using Ann? Journal of Business Economics & Management, 20(1), 168–191. https://doi.org/10.3846/jbem.2019.8250 Kaastra, I. & Boyd, M. (1996). Designing a Neural Network for Forecasting Financial and Economic Time Series. Neurocomputing, 10, 215-236. https://doi.org/10.1016/0925- 2312(95)00039-9 https://doi.org/10.1007/s10796-019-09911-3 https://doi.org/10.1007/978-0-387-09766-4_81 https://doi.org/10.1016/j.dss.2010.11.033 https://doi.org/10.1186/s40537-019-0206-3 https://doi.org/10.1002/isaf.214 http://dx.doi.org/10.1016/j.neunet.2014.10.001 https://doi.org/10.1186/s11782-020-00082-6 https://doi.org/10.1016/S0305-0483(98)00048-6 https://doi.org/10.3846/jbem.2019.8250 https://doi.org/10.1016/0925-2312(95)00039-9 https://doi.org/10.1016/0925-2312(95)00039-9 https://www.cribfb.com/journal/index.php/ijfb Indian Journal of Finance and Banking Vol. 5, No. 2; 2021 23 Karathanasopoulos, A., Dunis, C., & Khalil, S. (2016). Modeling, forecasting, and trading with a new sliding window approach: the crack spread example. Quantitative Finance, 16(12), 1875–1886. https://doi.org/10.1080/14697688.2016.1211796 Karunananthan, S., Maxwell, L, J., & Welch, V. (2020) Protocol: When and how to replicate systematic reviews. Campbell Systematic Reviews. 16. https://doi.org/10.1002/cl2.1087 Krishnaswamy, C.R., Gilbert, E.W., & Pashley, M.M. (2000) Neural network applications in finance: A practical introduction. Financial Practice and Education, 10(1), 75-84. Kriteenawong, C., Virk, H, H., Bangalore, S., Wang, Z., Johnson, W, K., Rachel. (2020). Machine learning prediction in cardiovascular diseases: a meta-analysis. Scientific reports, 10. https://doi.org/10.1038/s41598-020-72685-1 Kohler, M., Krzyżak, A., & Todorovic, N. (2010). Pricing of High-Dimensional American Options by Neural Networks. Mathematical Finance, 20(3), 383–410. https://doi.org/10.1111/j.1467-9965.2010.00404.x Koskivaara, E., & Back, B. (2007). Artificial Neural Network Assistant (ANNA) for Continuous Auditing and Monitoring of Financial Data. Journal of Emerging Technologies in Accounting, 4, 29–45. https://doi.org/10.2308/jeta.2007.4.1.29 Kim, K. (2004). Artificial neural networks with feature transformation based on domain knowledge for the prediction of stock index futures. Intelligent Systems in Accounting, Finance & Management, 12(3), 167–176. https://doi.org/10.1002/isaf.252 Light, R.J., & Pillemer, D. B. (1984). Summing up: The science of reviewing research. Cambridge, MA: Harvard University Press. Lin, C.-T., & Yeh, H.-Y. (2009). Empirical of the Taiwan stock index option price forecasting model - applied artificial neural network. Applied Economics, 41(15), 1965–1972. https://doi.org/10.1080/00036840601131672 Loterman, G., Brown, I., Martens, D., Mues, C., & Baesens, B. (2012). Benchmarking regression algorithms for loss given default modeling. International Journal of Forecasting, 28(1), 161–170. https://doi.org/10.1016/j.ijforecast.2011.01.006 Loukeris, N., & Eleftheriadis, I. (2015). Further Higher Moments in Portfolio Selection and A Priori Detection of Bankruptcy, Under Multi-layer Perceptron Neural Networks, Hybrid Neuro-genetic MLPs, and the Voted Perceptron. International Journal of Finance & Economics, 20(4), 341–361. https://doi.org/10.1002/ijfe.1521 LV, D., Wang, D., Li, M., & Xiang, Y. (2020). DNN models based on dimensionality reduction for stock trading. Intelligent data analysis, 24(1), 19-45. https://doi.org/10.3233/IDA- 184403 https://doi.org/10.1080/14697688.2016.1211796 https://doi.org/10.1002/cl2.1087 https://doi.org/10.1038/s41598-020-72685-1 https://doi.org/10.1111/j.1467-9965.2010.00404.x https://doi.org/10.2308/jeta.2007.4.1.29 https://doi.org/10.1002/isaf.252 https://doi.org/10.1080/00036840601131672 https://doi.org/10.1016/j.ijforecast.2011.01.006 https://doi.org/10.1002/ijfe.1521 https://doi.org/10.3233/IDA-184403 https://doi.org/10.3233/IDA-184403 https://www.cribfb.com/journal/index.php/ijfb Indian Journal of Finance and Banking Vol. 5, No. 2; 2021 24 Manel, H. (2012). Prediction of Financial Distress for Tunisian Firms: A Comparative Study between Financial Analysis and Neuronal Analysis. Business Intelligence Journal, 5(2), 374–382. Moher, D., Liberati, A., Tetzlaff, J., Altman, D, G., & The PRISMA Group. (2009). Preferred Reporting Items for Systematic Reviews and Meta-Analyses: The PRISMA Statement. PLoS Med. 6(7). https://doi.org/10.1371/journal.pmed1000097 Moreno, D., & Olmeda, I. (2007). Is the predictability of emerging and developed stock markets really exploitable? European Journal of Operational Research, 182(1), 436–454. https://doi.org/10.1016/j.ejor.2006.07.032 Ngai, E. W. T., & Wat, F. K. T. (2002) A Literature Review and Classification of Electronic Commerce Research. Information & Management, 39, 415-429. Neves, J. C., & Vieira, A. (2006). Improving bankruptcy prediction with Hidden Layer Learning Vector Quantization. European Accounting Review, 15(2), 253–271. https://doi.org/10.1080/09638180600555016 Omar, N., Johari, Z, A., & Smith, M. (2017). Predicting fraudulent financial reporting using artificial neural network. Journal of Financial Crime, 24(2), 362–387. https://doi.org/10.1108/JFC-11-2015-0061 Ozbey, F., & Paksoy, S. (2020). Estimation of the XU100 Index Return Volatility with the Integration of GARCH Family Models and ANN. Business and economics research journal, 11(2), 385 – 396. Petter, S., McLean, E. R. (2009). A meta-analytic assessment of the DeLone and McLean IS success model: An examination of IS success at the individual level. Information Management, 46(3), 159-166. Pradhan, R., Pathak, K. K., & Singh, V. P. (2011). Z Score Reveals Credit Capacity: A Case Study of SBI. International Journal of Financial Management, 1(3), 72–78. Qi, M., & Zhao, X. (2011). Comparison of modeling methods for Loss Given Default. Journal of banking and finance, 35(11). 2842-2855. https://doi.org/10.1016/j.jbankfin.2011.03.011 Roelofs, R., Shankar, V., Recht, B., Fridovich-Keil, S., Hardt, M., Miller, J & Schmidt, L. (2019). A Meta-Analysis of Overfitting in Machine Learning. Advances in Neural Information Processing Systems 32 (Neur IPS 2019), 1-11. Retrieved from https://openreview.net/pdf/98fdb4bb89fd868a5dda9f827133eec6e11bb4fa.pdf Safer, A. M. (2002). The application of neural networks to predict abnormal stock returns using insider trading data. Applied Stochastic Models in Business & Industry, 18(4), 381–389. https://doi.org/10.1002/asmb.466 https://doi.org/10.1371/journal.pmed1000097 https://doi.org/10.1016/j.ejor.2006.07.032 https://doi.org/10.1080/09638180600555016 https://doi.org/10.1108/JFC-11-2015-0061 https://doi.org/10.1016/j.jbankfin.2011.03.011 https://openreview.net/pdf/98fdb4bb89fd868a5dda9f827133eec6e11bb4fa.pdf https://doi.org/10.1002/asmb.466 https://www.cribfb.com/journal/index.php/ijfb Indian Journal of Finance and Banking Vol. 5, No. 2; 2021 25 Sapna, O., Botti, V., & Argente, E. (2003). Application of neural networks to stock prediction in “pool” companies. Application of Artificial Intelligence 17(7):661–673. Sariev, E., & Germano, G. (2020) Bayesian regularized artificial neural networks for the estimation of the probability of default. Quantitative Finance, 20(2), 311–328. https://doi.org/10.1080/14697688.2019.1633014 Sermpinis, G., Laws, J., & Dunis, C. L. (2013). Modeling and trading the realized volatility of the FTSE100 futures with higher-order neural networks. European Journal of Finance, 19(3), 165–179. https://doi.org/10.1080/1351847X.2011.606990 Shokraneh, F., & Adams, C. E. (2019). Study-based registers reduce waste in systematic reviewing: discussion and case report. Systematic Reviews, 8(1), 129. Sperckelsen, C., Mettenheim, H., & Breitner, M. H. (2014). Real-Time Pricing and Hedging of Options on Currency Futures with Artificial Neural Networks. Journal of Forecasting, 33(6), 419–432. https://doi.org/10.1002/for.2311 Trinkle, B. S., & Baldwin, A. A. (2007). Interpretable credit model development via artificial neural networks. Intelligent Systems in Accounting, Finance & Management, 15(3/4), 123–147. https://doi.org/10.1002/isaf.289 Vellido A., Lisboa, P. J., Vaughan, J. (1999). Neural networks in business: a survey of applications (1992–1998). Expert Syst Appl. 17(1), 51–70. Wie Zhang, A., Qing Cao, & Schniederjans, M. J. (2004). Neural Network Earnings Per Share Forecasting Models: A Comparative Analysis of Alternative Methods. Decision Sciences, 35(2), 205–237. https://doi.org/10.1111/j.00117315.2004.02674.x Wanke, P., Kalam Azad, M. A., Barros, C. P., & Hadi, V. A. (2016). Predicting performance in ASEAN banks: an integrated fuzzy MCDM-neural network approach. Expert Systems, 33(3), 213–229. https://doi.org/10.1111/exsy.12144 Wang, A., Liu, Y., Isaeva, E., & Rocha, Á. (2020). Intelligent financial management of company based on neural network and fuzzy volatility evaluation. Journal of Intelligent & Fuzzy Systems, 38(6), 7215–7228. https://doi.org/10.3233/JIFS-179798 Wei, K., Nakamori, Y., Wang, S., & Yu, L. (2007). Neural Networks in Finance and Economics Forecasting. International Journal of Information Technology &Decision Making, 6(1), 113-140. https://doi.org/10.1142/S021962200700237X Willer P.J., Giarola, V, A., Cardoso, C, A., & Pinheiro, N., T. (2020). Business Insolvency Forecasting Using Artificial Neural Networks. Management and Development, 17(2), 136–162. https://doi.org/10.25112/rgd.v17i2.1777 https://doi.org/10.1080/14697688.2019.1633014 https://doi.org/10.1080/1351847X.2011.606990 https://doi.org/10.1002/for.2311 https://doi.org/10.1002/isaf.289 https://doi.org/10.1111/j.00117315.2004.02674.x https://doi.org/10.1111/exsy.12144 https://doi.org/10.3233/JIFS-179798 https://doi.org/10.1142/S021962200700237X https://doi.org/10.25112/rgd.v17i2.1777 https://www.cribfb.com/journal/index.php/ijfb Indian Journal of Finance and Banking Vol. 5, No. 2; 2021 26 Wong, B. K., & Selvi, Y. (1998). Neural network applications in finance: A review and analysis of literature (1990-1996). Information and Management, 34(3), 129–139. Yang, Z. R., Platt, M. B., Platt, H. D (1999). Probabilistic neural networks in bankruptcy prediction. Journal of Business Research, 44(2), 67–74. Ying-Hua, C., & Shih-Chin, W. (2013). Integration of Evolutionary Computing and Equity Valuation Models to Forecast Stock Values Based on Data Mining. Asia Pacific Management Review, 18(1), 63–78. https://doi.org/10.6126/APMR.2013.18.1.04 Zan, H., Chen, H., Hsu, C., Chen, W., & Wu, S. (2004). Credit rating analysis with support vector machines and neural networks: a market comparative study. Decision Support Systems, 37(4), 543–558. https://doi.org/10.1016/s0167-9236(03)00086-1 Zapart, C. A. (2003). Beyond Black–Scholes: A Neural Networks-Based Approach to Options Pricing. International Journal of Theoretical & Applied Finance, 6(5), 469. https://doi.org/10.1142/S0219024903002006 APPENDICES Appendix A. Figure depicting NN architectures in Investment Prediction *Hybrid NN methods are developed by incorporating the financial theories (Black Scholes option pricing, Fama French five-factor model), advanced statistics (GARCH, PCA, LASSO, CART, PLR), advanced NN architectures (HONN, GPA, MLP), and domain-specific factors. 0 2 4 6 8 10 12 14 16 Standard NN Back Propagation Feed forward Polynomial Monte carlo Baysian regularisation Hybrid NN Wavelet Radical Based function NN Multi layer perceptron Investment Prediction https://doi.org/10.6126/APMR.2013.18.1.04 https://doi.org/10.1016/s0167-9236(03)00086-1 https://doi.org/10.1142/S0219024903002006 https://www.cribfb.com/journal/index.php/ijfb Indian Journal of Finance and Banking Vol. 5, No. 2; 2021 27 Appendix B. Figure depicting NN architectures in credit evaluation, financial distress, and other applications *Support Vector Machine is a similar machine learning algorithm like NN. Copyrights Copyright for this article is retained by the author(s), with first publication rights granted to the journal. This is an open-access article distributed under the terms and conditions of the Creative Commons Attribution license (http://creativecommons.org/licenses/by/4.0/) 0 2 4 6 8 10 12 14 Support Vector Machines Baysian regularisation Back Propagation Feed forward Standard NN Fuzzy system Multi layer perceptron Learning Vector quantization Self Organisation Map algorithm Credit Evaluation, Financial Distress and Other applications