Integrating Financial and Textual Indicators for Enhanced Financial Risk Prediction: A Deep Learning Approach

ABSTRACT


INTRODUCTION
Predicting financial risk is a crucial problem in finance since it enables companies, investors, and governments to make wise choices and avert possible financial disasters.According to the study by Mashrur et al. (2022), the process of accurately predicting financial risk is complex.It depends on a number of variables, including textual data and conventional financial indicators (Mashrur et al., 2020).Al-Eitan et al. (2019) have highlighted that financial analysts have traditionally based their assessments of a company's financial health and risk on traditional financial indicators, including liquidity ratios, leverage ratios, and return on assets (ROA).However, Fridson & Alvarez, (2022), has noted that financial indicators sometimes give inconsistent signals in real-world situations, making risk prediction a challenging endeavor (Al-Eitan & Bani-Khalid, 2019).Another issue highlighted by Arnold et al. (2022), that threatens the stability of prediction models is the multicollinearity of financial indicators and worries about missing data.Indicators for cross-border risk assessment are gradually being standardized through the adoption of international accounting standards like IFRS (Arnold et al., 2022;Phan et al., 2018).Textual information, such as sentiment analysis and language from financial news articles, is increasingly important for predicting financial risk (Bawa, 2023).Textual data changes in regulatory stance and management tenor might be crucial in anticipating financial risk (Feyen, 2023).However, Humphreys & Wang, (2018), has stressed that issues like bias in sentiment analysis and mistakes in reporting must be resolved.Additionally, there is still little research on how textual indicators interact with certain financial metrics like ROA or solvency ratios (Feyen et al., 2023;Humphreys & Wang, 2018;Karas & Režňáková, 2020).Malekloo et al. (2022), has stated that these components' intricate interrelationships call for in-depth examination.The study further highlights that with the introduction of big data and advancements in artificial intelligence, the existing financial environment is changing quickly (Malekloo et al., 2022).Therefore, it is crucial to investigate cutting-edge methods for estimating financial risk that may make use of both financial and textual data (Xing et al., 2018).According to Mai et al. (2019), comparing the effects of both types of indicators on predicted financial performance, can close the gap between established textual data analysis and traditional financial analysis.Furthermore, it also stresses that the goal is to construct more reliable risk prediction models by using the synergy between these components as well as their separate contributions (Mai et al., 2019).The study Neale, B. (2021).The craft of qualitative longitudinal research: the craft of researching lives through time.Neale et al. (2021), has explored the dynamic nature of financial markets and the demand for cutting-edge instruments to negotiate their complexities serve as the driving forces behind this inquiry.The study has also provided insights at how deep learning can integrate financial and textual indicators, to provide insightful information for enhanced financial risk prediction techniques (Neale, 2021).

Financial Risk Indicators
The process of predicting financial risk is complex and involves a number of different indicators and variables (Henrique et al., 2019).Peng & Huang (2020) state the financial risk prediction procedure includes a number of processes that evaluate the possible risks a firm can encounter on its financial path (Peng & Huang, 2020).It 1 Chongqing Vocational College of Finance and Economics, China 2 Department of Finance, University of Sabah, China https://journals.e-palli.com/home/index.php/ajftiAm.J. Financ.Technol.Innov.2(1) 15-24, 2024 is crucial to comprehend how this approach will affect a company's stability and operations (Lee et al., 2022).In this process, & Riyanto (2020), emphasizes that liquidity ratios, such as the quick ratio and current ratio, measure a company's short-term solvency while Nukala& Prasada (2021), emphasizes that leverage ratios, such as debt-to-equity ratios, evaluate its long-term financial structure and have significant importance in the financial risk prediction process.The importance of striking a balance between these ratios has also been emphasized by Dianova & Nahumury, (2019), as high leverage can increase the risk of financial distress and low liquidity can make it difficult for a firm to meet immediate obligations.However, there may also be some shortcomings that financial risk experts need to investigate (Dianova & Nahumury, 2019;Maisharoh & Riyanto, 2020;Nukala & Prasada Rao, 2021).As Stephany et val. (2023) noted, financial indicators can give conflicting signals when assessing risks, hence it is important to carefully examine these signals when predicting financial risk (Stephany et al., 2020).Such as Mohamed, (2022), underlined the necessity for a nuanced interpretation and an investigation of deeper underlying concerns if a business displays a high Return on Assets (ROA) despite bearing a significant debt load (Mohamed, 2022).Multicollinearity among financial indicators, which show strong correlations, is one such significant issue.Lasso regression is one of the statistical methods that Urdes et al. (2022) devised to deal with multicollinearity and increase the resilience of prediction models.These techniques aid in separating the web of connected indications (Urdes et al., 2022).However, regularity in the data is necessary for the implementation of such procedures, which is frequently disrupted by missing values.Winsemius et al. (2018), illustrates how assessing financial risk may be hampered by missing or inadequate data.Data imputation and amputation are two techniques that Washburn et al. (2018) cover in their discussion of viable approaches to this problem.These techniques simplify dataset reconstruction and enable more thorough risk assessments (Washburne et al., 2018;Winsemius et al., 2018).The selection of accounting standards is yet another important issue that needs to be carefully taken into account (Weygandt et al., 2018).The decision between international accounting standards like IFRS and nationspecific elements has relevance in the globalized financial landscape for standardizing indicators in cross-border risk assessment.Swanepoel (2018), has looked at how these decisions may affect how reliable and comparable risk assessments are in different international contexts.Zio (2018), has stated that construction of reliable risk models requires a thorough understanding of various financial risk prediction components and how they interact.The study has further explored the collective knowledge of the field and increase our understanding of financial risk prediction by combining ideas from various academic studies (Chen et al., 2021;Swanepoel, 2018).

Implication of Textual Indicators in Financial Risk Prediction
Textual indicators, such as managerial tone and tone indexes, are becoming more and more important parts of the process of predicting financial risk (Zhang et al., 2022).They have been cited as playing crucial roles in improving risk assessment by several academics.According to.Iqbal & Riaz (2021), the management's tone of a company's communications, including annual reports or press releases, might offer insightful information (Iqbal & Riaz, 2021).Investor confidence and subsequent financial performance can be impacted by positive or negative management attitude (Platonova et al., 2018).Additionally, biases in textual data may be inherent and result from biased reporting or inaccurate sentiment analysis.It is important to note that financial experts are aware that manual evaluation of textual data might be more effective at eliminating these biases (Metaxa et al., 2021) .This practical method enables a more precise analysis of subtle textual clues.Textual indicators also interact with financial measurements like Return on Assets (ROA) and Return on Equity (ROE), therefore they do not exist in a vacuum (Alduais, 2022).These interactions, as explained by Zio (2018), influence the results of risk assessments and give a comprehensive picture of a company's financial health.The study highlights that professionals may collect nuanced information, improve decision-making, and lessen data biases by including textual indicators into the financial risk prediction process (Zio, 2018).This integration acknowledges the importance of textual data in the modern financial sector while reflecting the changing environment of risk assessment.

LITERATURE REVIEW
Financial risk prediction is a crucial field of research in finance and economics, having important consequences for organizations, investors, and decision-makers (Win et al. 2018).Arnold et al. (2018), explains that financial risk forecasting heavily relies on historical financial data.Risk assessment is based on historical financial performance, which includes income statements, balance sheets, and cash flow statements (Goh et al. 2022) 2018), highlights that he use of sentiment analysis and news sentiment data as fresh indicators of financial risk has also grown in popularity.News and social media attitude can influence the state of the market and the value of assets (Masuda et al. 2022).
In essence, components that can anticipate financial risk include sentiment analysis, credit risk indicators, market and macroeconomic conditions, and historical financial data.By combining these elements with cutting-edge analytical methods, risk assessments can be improved, assisting decision-makers (Terzi et al. 2019).

Integrating Financial and Textual Indicators for Financial Risk Prediction
The effects of combining financial and textual data in financial risk prediction are extensive.For a thorough risk assessment, certain businesses need specialised financial criteria (Nyman et al. 2021).Risk prediction is greatly influenced by historical financial performance, including debt ratios and earnings stability (Sathyamoorthi, 2022).When projecting financial risk, objective financial health metrics like liquidity and solvency ratios frequently outperform market sentiment, especially during economic downturns (Nazareth & Reddy, 2023).When assessing financial stability, short-term financial measures like liquidity ratios take center stage (Zorn et al. 2018).Sector-wide statistics, however, may outperform firm-specific indicators in high-risk circumstances (FLACHENECKER l et al., 2020).The accuracy and reliability of risk prediction are improved by text indicators, such as sentiment analysis and tone indices (Zhang et al., 2022).According to Mushtaq et al. (2022), risk evaluations are influenced by how management tone and language sentiment interact with financial measures like ROA and ROE.The study further highlights that the efficiency of textual indicators is impacted by legislative changes and current affairs.Furthermore, Manual data review can be used to address possible biases in textual data, such as sentiment analysis errors and reporting biases (Díaz et al. 2018).With consequences that vary among industries, historical settings, and market dynamics, the combination of financial and textual indicators enhances the forecast of financial risk (Cavalcante et al. 2016).Additionally, there is a huge area of study that will improve this integration, increasing the accuracy of risk assessment models.((2018).highlights that a process both unstructured textual data and structured financial data is required.The above approach takes into account the dynamic relationships between FI and TI to assist fast risk assessment.Integration of these technologies offers more precise and responsive financial risk models as Python and big data continue to develop (Fu et al. 2021).

Literature Gap and Hypothesis Development
The observed gap in the literature and the theoretical groundwork extracted to the literary analysis serve as a strong foundation for the hypotheses developed in this study.The analysis of the literature found a paucity of thorough studies integrating both financial and textual indicators for improved financial risk prediction using deep learning techniques.The work uses well-established financial risk prediction theories and models to close this gap while also recognizing the growing importance of textual indicators.The foundation for the assumptions comes from theoretical frameworks including Altman's Z-score model, Beaver's financial ratios, and contemporary deep learning methods.To fill the current research gap, these hypotheses reflect an original strategy that blends conventional financial analysis by employing the financial and textual indicators through state-of-theart deep learning techniques.Based on these observations the following hypothesis are formed: Hypothesis 01: Financial risk prediction factors like financial and textual indicators has significant positive trends over the years.
H2: Financial indicators has significant positive impact on financial risk prediction H3: Textual indicators has significant positive impact on financial risk prediction H4: Both Financial and Textual indicators has significant positive correlation and have significant positive impact on financial risk assessment or organizations.The presented study used a quantitative research methodology to look at the elements that financial risk experts find most useful in predicting financial risk.The cross-sectional approach was used to examine and comprehend interrelationships.The choice of quantitative research depends on its capacity for efficient and objective data collecting and processing.The positivist viewpoint places a strong emphasis on employing unbiased, verifiable data to support the progression of the process.According to Lombardo et al. (2019), more generalizable results are associated https://journals.e-palli.com/home/index.php/ajftiAm.J. Financ.Technol.Innov.2(1) 15-24, 2024 with bigger sample sizes.Quantitative research thrives when using standardized and methodical data gathering approaches, as Creswell and Hirose (2019) explained.

METHODOLOGY Research Design
The current study leveraged a deep learning approach implemented using Python through Jupyter Notebook for the analysis of

Analysis and Modeling
To examine the association between study sections and the mean scores given to particular factors, this study used linear regression analysis.Three crucial columns made up the dataset: "Section," "Variable," and "Mean."'Section' stood for several sections, 'Variable' stood for research variables, and 'Mean' included the mean scores for each variable inside each section.

Preparation of Data
The dataset was put into a Panda DataFrame called "df_means," and the "Section" variable underwent label encoding to convert its values into numbers appropriate for regression analysis.

Model for Linear Regression
For the linear regression analysis, Scikit-learn's 'LinearRegression' class was used.Section_encoded served as the independent variable, reflecting encoded section values, while 'Mean' served as the dependent variable, including mean scores related to each variable (Galioulline, et al. 2023).

Model Fitting
Using the 'fit' procedure, the linear regression model was adjusted to the data.The goal of this fitting procedure was to find the regression line that suited the data the best and minimized the gap between anticipated values and actual mean scores.

RESULTS AND DISCUSSIONS Results of Regression
The following important factors were shown to assess model performance:

Intercept
Depicting the y-intercept of the regression line.

Coefficient (Slope)
Identifies the slope of the regression line, indicating how the mean scores vary when the units in the encoded section change.

R-squared
A measure of the coefficient of determination that expresses the amount of variance in mean scores that the model is able to account for.

Data Visualization
The findings were shown as a scatter plot, with blue data points representing the actual mean scores, supported by McDermaid et al. (2019).The regression line was shown by a red line to show how well it suited the data.In order to accomplish the goals of the study, this linear regression analysis provided insights into the link between study sections and mean scores for particular variables (Grotzinger et al. 2019).

Analysis and Conclusions
The results of the deep learning models were analyzed in the study to acquire understanding of how the combination of textual and financial variables improves financial risk prediction.To improve understanding and encourage practical decision-making, qualitative analysis and visualization methods were used (Martins, et al. 2022).This method is an innovative approach for predicting financial risk since, as Kang et al. highlights that it combines deep learning and NLP to glean insightful information from both organized and unstructured data sources.The results of this ground-breaking study will be presented and discussed in the following parts, with an emphasis on their consequences and potential applications in the field of financial risk assessment (Babich et al. 2018).

Participants Information
The frequency analysis demonstrated in Fig. 1 shows that majority of truth financial risk experts participated in the study are familiar with financial risk prediction through deep learning approaches, whereas an equal amount of participants has opted for less familiarity to slightly familiar in the study.The fig 2 illustrates that majority of the participants had 2-4 years of experience whereas considerable number of participants have 4-5 years of experience, but also a good amount of participants were observed to does not have much familiarity with financial risk prediction integrating the textual and financial indicators.

Section 02: Financial Risk Predictors
The mean values derived from financial risk experts' responses, collected on a 5-point Likert scale, provide insights into their views on identified financial risk prediction factors:

Preferences of Liquidity
The observed mean on the preferences of liquidity ratios over leverage ratios as a financial risk predictor is 1.996.It reflects that experts generally agree that liquidity plays a crucial role in financial risk prediction as compared to leverage ratios.

Financial Indicators and Conflict Signals
The observed mean for the creation of conflicting signals such as high ROA but high debt was 2.029.This illustrates that experts tend to agree that managing financial risk effectively involves dealing with conflicting signals from financial indicators.

Multicollinear Issues
The observed mean for on the multicollinear issues in the financial risk prediction can be resolved by stepwise regresses ion and leads model instability as compared to Lasso regression Mean is 1.933.This shows that there is an agreement that multicollinearity among financial indicators can lead to model instability.

Missing Data
The observed mean on the employment on amputation techniques to resolve missing values is: 1.750.It reflects that Experts agree that missing data problems can be resolved using amputation techniques.

Standardizing Indicators
The mean of the collected responses on the preference of IFRS while dealing with financial indicators in international contexts as compared to country-specific factors for standardizing indicators for cross-border risk assessment was observed in the analysis is 1.667.This illustrates that the lowest mean value indicates strong agreement that international accounting standards are more efficient for standardizing indicators in cross-border risk assessment.

Critical for Specific Industries
The observed mean on the preference of financial indicators over textual indicators is 1.65.Experts generally agree that certain financial indicators hold industry-specific significance in financial risk assessment.This underscores the recognition of tailored risk evaluation approaches.

Historical Performance
The observed mean on the influence of historical performance in te financial risk prediction process is 1.48).The mean indicates a strong agreement that historical financial indicator performance influences financial risk predictions.This reflects the experts' belief in the predictive power of past financial data.

Financial Health and Risk Indicators
The observed mean on the preference of financial health indicates for the risk assessment over market and investors perception indicators is 1.45.This reflects that financial risk experts strongly agree that financial health and risk indicators outweigh market and investor perception indicators in financial risk predictions.This highlights the priority placed on fundamental financial metrics.

Short-term Financial Indicators
The observed mean on the preference of short-term financial indicators over long term indicators is 1.43.The mean suggests a consensus that short-term financial indicators, like liquidity ratios, hold higher importance when assessing financial stability compared to long-term indicators.

Historical Sector
The observed mean on the preference of historical sector wide is preferred over firm specific data in highrisk scenario is 1.67. it illustrated that there is strong agreement that, in cases of high-risk indications, historical sector-wide data is preferred over firm-specific data.This emphasizes the importance of broader industry context in risk assessment.
In the context of the study, these means signify a shared belief among experts regarding the significance of industry-specific considerations, historical data, and fundamental financial health metrics in the financial risk prediction process.It underscores the value of these factors in developing comprehensive risk assessment models.

Section 4: Implication of Textual Indicators in Financial Risk Prediction
The mean values for Textual Indicators and Their Implication variables, gathered on a 5-point Likert scale, provide valuable insights:

Textual Indicators Reliability
The observed mean of the reliability of textual indicators in the financial risk prediction process is 1.95.This reflects that experts tend to agree that textual indicators, such as sentiment analysis, are accurate and reliable for financial risk prediction.This suggests their confidence in the utility of textual data in risk assessment.

Management Tone and Tone Indexes
The observed mean on the importance of.textual indicators, such as management tone and tone indexes in financial risk prediction is 1.6.The mean indicates agreement that management tone and tone indexes play a significant role in financial risk prediction, underlining the relevance of management communication in risk assessment.

Regulatory Changes or News Events
The observe mean on the employment of manual review of textual data more efficiently in the financial risk assessment is 1.59.This illustrates that experts generally agree that regulatory changes and news events influence the use of textual indicators in financial risk prediction.This highlights the timeliness of textual data.

Potential Biases
The observe mean on interaction of with specific financial indicators (e.g., ROA, ROE) in shaping risk assessment outcomes is 1.95.This reflects that there is a consensus that potential biases in textual data, like sentiment analysis inaccuracies or reporting biases, can be more efficiently resolved through manual review.This reflects a practical approach to mitigating biases.

Language Sentiment
The observed mean on the language sentiment analysis in the financial risk prediction is 2.05): The highest mean suggests that textual indicators, like management tone and language sentiment, interact significantly with specific financial indicators, impacting risk assessment outcomes.In the study context, these means underscore the experts' acknowledgment of the reliability of textual indicators, the influence of management tone and external events, and the importance of addressing potential biases.They emphasize the intricate relationship between textual and financial data in enhancing financial risk prediction models.

CONCLUSION
Employing quantitative techniques, this research investigates the fusion of financial and textual indicators for predicting financial risk.Tackling multicollinearity problems and handling liquidity are significant risk prediction findings.Focusing on industry-specific metrics, experts prioritize short-term data when evaluating highrisk sectors, while historical trends hold less weight.According to textual indicators, sentiment analysis is just one of the reliable predictors, with the need to address biases included.While textual indicators exhibit a weaker bond, linear regression reveals a substantial relationship between financial prediction factors and indicators.Enhanced methodologies are achieved through the correlation between their interdependence in risk prediction, resulting in improved decision-making.

RECOMMENDATIONS
Based on the findings of the current study that, financial risk assessment professionals emphasize the significance of liquidity indicators in risk evaluation, address the management of conflicting signals from financial data, adopt strategies for mitigating multicollinearity problems, consider the use of amputation techniques for missing data challenges, and investigate the application of International Financial Reporting Standards (IFRS) for standardizing cross-border reporting.The should also employ past sector-wide data for high-risk scenarios while giving previous financial performance data and short-term financial indicators priority.Continue to place your faith in textual indications like sentiment analysis and managerial tone, but aggressively address any biases through manual inspection and look into deep learning approaches to improve the integration of financial and textual data even more.These initiatives will support more thorough and accurate financial risk assessments, eventually enhancing risk analysis decision-making procedures.

Novelty
The study uses advanced deep learning techniques to integrate financial and textual indicators for financial risk prediction.It uses recurrent neural networks and transformers to analyze the synergy between these sources, improving the precision of financial risk models.The study uses a quantitative and qualitative approach, incorporating real-world insights from financial risk experts in China.The research identifies a research gap in the literature and contributes to the field of financial risk prediction by combining financial and textual indicators.

Contribution to Knowledge
The study enhances financial risk prediction by integrating

Research Gap
The study identifies a gap in literature regarding the integration of financial and textual indicators for improved financial risk prediction using deep learning techniques.It emphasizes the need for a holistic approach, focusing on each type of data individually.The study also highlights the lack of deep learning applications in financial risk prediction and the potential benefits of advanced methods.It also calls for more in-depth investigation into the reliability of textual indicators and the integration of quantitative and qualitative research methodologies.
Understanding financial risk has been transformed by the convergence of deep learning, big data, and analysis of financial and textual indicators (FI and TI) (Kim et al. 2022).The prediction of risk has now expanded in new directions with the introduction of Deep leaning methods including python and big data technologies (Abkenar et al. 2021).According to Zhou et al. (2021), Python's machine learning packages make it easier to build deep learning models for integrating FI and TI.Massive datasets, such as real-time financial reports and textual data from news and social media sources, may be collected and stored using big data systems (Hariri et al. 2019).Recurrent neural networks (RNNs) and transformers are examples of deep learning approaches that improve FI-TI synergy by automatically discovering complex correlations (Lienhard et al. 2022).For the purpose of capturing complex market emotions and financial health indicators, Taleb et al.
financial and textual data (Tatsat et al. 2020).The data collection involved the extraction of financial indicators and textual information from diverse sources, including financial reports, news articles, and publicly available data (Pejić et al. 2019).The existing research has then helped in formulating a structured openended survey on the integration of different financial and textual indicators in financial risk passement processed as highlighted by Azizi et al. (2021).Furthermore, The survey responses were gathered qualified financial risk experts of China.The official qualified financial reliable sources to collect accurate and effective insights (Wu et al. 2020).Furthermore, Data preprocessing was a crucial step, encompassing the cleansing and standardization of financial data and natural language processing (NLP) techniques applied to textual data (Aldunate et al. 2022).Based on this the NLP facilitated the extraction of meaningful textual indicators, ensuring the integration of unstructured textual information with structured financial data.

Figure 1 :Figure 3 :
Figure 1: Participants familiarity with financial risk prediction Figure 2: Experience of participants

Figure 4 :
Figure 4: Financial indicators and Their Implication

Figure 5 :Figure 6 :
Figure 5: Regression Between Financial Predictors and Indicators

Figure 7 :
Figure 7: Pearson Correlation Matrix A complete positive linear link between both variables is shown by the positive correlation coefficient of 1 in all four quartiles of the correlation matrix between textual indicators and financial indicators.This suggests a significant positive correlation between textual and financial characteristics in the context of predicting financial risk, such that when one set of indicators rises, the other set rises in lockstep.DISCUSSION Using a deep learning approach, the study sought to examine the fusion of financial and textual indicators in financial risk forecasting.Through analysis, we gained insight into how various factors relate to one another and their potential influence on risk assessment.Constructing the foundation on West et al. (2022) work, analyzing the dataset consisting of "Section," "Variable," and "Mean," financial and textual indicators and exploring deep learning techniques like recurrent neural networks and transformers.It provides insights into risk assessment dynamics and bridges the gap between quantitative and qualitative research.The study identifies a research gap in existing literature and offers practical recommendations for financial risk professionals, emphasizing liquidity indicators and considering IFRS for cross-border reporting.It lays the groundwork for future research in deep learning techniques and financial risk prediction.
assessment is credit risk prediction.In assessing loan defaults, variables including credit ratings, debt ratios, and default probability are crucial.Support vector machines and neural networks are two examples of machine learning techniques that are increasingly being used in credit risk analysis (Teles et Am.J. Financ.Technol.Innov.2(1) 15-24, 2024 al. 2021).Chang & Wang (