Types of financial risks in Higher Education Institutions
The following analysis will answer RQ1 in identifying the main financial risks HEIs face. According to Chen W [23], the financial risk associated with universities may be conceptualized as the probability of these institutions failing to promptly fulfill their obligations of principal and interest repayment to banks, owing to many uncertain elements. In financial management, HEIs are faced with a series of financial risks, which may come from changes in the external environment, such as fluctuations in the global economy [24] and policy uncertainties [25], or from internal factors, such as mistakes in management decisions [26] and unreasonable allocation of resources [27] and income distribution [28]. University financial risks focus on debt repayment, capital raising, investment strategies, and financial stability [29], with Hyland et al. identifying risks in capital financing, investment decisions, and daily operations [30].
The top U.S. universities, such as Harvard and Stanford, face various risks, from academic to insurance, underscoring the inevitability of financial risks in higher education. Wenbin Chen et al. outlined HEIs' financial risks in fundraising, investment, capital recovery, and income distribution, noting their concealment and assessment challenges [23]. The Deloitte Center advocates for a holistic risk management approach in HEIs, considering factors like business models and compliance risks [31].
Financial risk management in HEIs involves macro and micro risks, focusing on capital financing, investment decisions, and operational challenges. Key areas include debt capacity, fundraising, and financial stability, with risks like compliance and strategic and reputational factors also being significant. Comprehensive risk management approaches are advocated to address these multifaceted risks effectively. Table 3 summarizes the main types of financial risks in HEIs, as mentioned in the analysis.
Table 3
The main types of financial risks in HEIs
NO. | Risk Types | Citation |
1 | Risks of global economic volatility | [24] |
2 | Policy uncertainty risk | [25] |
3 | Manage the risk of bad decisions | [26] |
4 | Unreasonable resource allocation risks | [27] |
5 | Risk of unreasonable income distribution | [28] |
6 | Capital financing, investment decisions and daily operational risks | [29, 30] |
7 | Capital raising, investment, capital recovery and income distribution risks | [23] |
The causes of financial risks in Higher Education Institutions
The following analysis supports the RQ2, which identifies the causes of financial risks in HEI. Financial risk management evolved from Fayol's 1916 concepts into a structured discipline after the 1932 crisis, significantly influenced by Altman's 1968 work on predictive models for corporate risk [46]. Angelo (2016) highlighted the dynamic nature of risk assessment based on the variance between outcomes and objectives. In higher education, financial risks stem from infrastructure expansion, relying heavily on loans, and flawed financial systems. Rapid growth and weak governance further exacerbate these risks. Table 4 summarizes the leading causes of financial risks in HEIs, as mentioned in the analysis.
Table 4
The leading causes of financial risks in HEIs
NO. | Risk Cause | Citation |
1 | Infrastructure expansion, reliance on loans | [32] |
2 | Rapid growth and weak governance | [33] |
3 | Insufficient external funding and management deficiencies | [34] |
4 | Capital investments, reduced government funding | [35] |
5 | Management's risk misperception, expansion pressures | [36] |
6 | Investments required by national evaluations | [37] |
7 | Enrollment declines, trust issues | [38] |
8 | Financial system reforms, software updates | [39] |
9 | Funding shortages leading to financial crises | [40] |
10 | Funding gap from new campus constructions | [41, 42] |
11 | Complexity of financial management in HEIs | [43, 44] |
12 | Shortage of student tuition income | [45, 46] |
Financial risk management in HEIs has evolved significantly, influenced by foundational works, and exacerbated by challenges such as rapid expansion, governance weaknesses, and reliance on limited funding sources. Key risks include infrastructure costs, funding deficits, and management inefficiencies, with strategies focusing on student success and cost control. The complex blend of commercial and educational financial practices complicates risk management, highlighting the critical role of tuition income and the potential for financial crises in HEIs.
Financial risk early warning assessment in Higher Education Institutions
Several approaches to financial risk assessment in HEIs have been developed to address RQ3, which is to conduct financial risk early warning assessment. Berens et al. established a framework using 19 financial indicators [47], validated by data from 25 universities. Yaacob et al. crafted a system based on operational, investment, and financing cash flows [48]. Li et al. created a risk model categorizing risks into three levels for an early warning system [49].
Shuhua Tsao et al. conducted an in-depth analysis using a fuzzy comprehensive assessment of financial data from 2020 and 2021. Lu Liu emphasized a comprehensive framework, integrating key indicators with neuron dynamic models and learning algorithms for nuanced assessment [50], enriching HEI financial risk evaluation methods. Concurrently, various methods have been devised for financial risk evaluation in HEIs, including Berens et al.'s 19-indicator framework (2021), Yaacob et al.'s cash flow analysis (2019), Li et al.'s three-tier risk model (2018), and Shuhua Tsao et al.'s fuzzy comprehensive evaluation applied to financial data (2023), all contributing to the development of effective risk assessment strategies.
HEI financial risk evaluation has evolved with diverse models, from Berens et al.'s 19-indicator framework to Li et al.'s three-tiered risk system. Innovations like Shuhua Tsao et al.'s fuzzy analysis and Lu Liu's neuron dynamic models highlight a shift towards nuanced, data-driven risk management strategies in higher education.
Financial risk early warning model
An Early Warning System (EWS) is an integrated system that reduces the impact of disasters by disseminating early warning information to facilitate preparedness and response mechanisms [51]. The development of the financial risk early warning model has followed a trend from simple to complex. The following analysis supports the RQ4, which identifies the model used in the early warning of financial risk in HEIs.
Univariate model
In 1932, Fitz Patrick pioneered the use of financial ratios in analysis, marking a significant shift in financial research [46]. He identified that ratios like debt-to-equity and net equity interest could distinguish between firms at high bankruptcy risk and stable ones, highlighting their predictive value for financial health.
Expanding on this, Beaver's 1966 study [52] revealed distinct differences in financial metrics between stable and distressed firms, enhancing the understanding of corporate financial health. However, the reliance on univariate models in early research, which focused on single indicators, often resulted in inaccuracies in complex business settings. This led to the development of more advanced multivariate prediction methods to improve the precision and dependability of financial risk assessments.
Multivariable model
Since Altman's groundbreaking bankruptcy prediction model in 1968, the diversity and number of such models have surged, driven by advancements in statistical and information technologies [53]. Altman's analysis of 33 companies using five key indicators resulted in the Z-SCORE model, boasting up to 95% accuracy a year before a financial crisis [54]. Despite its success, Altman's model assumes financial variables follow a normal distribution, a condition often unmet. This leads to adopting logistic regression models for their flexibility in distribution assumptions.
Blum's 1974 refinement of the Z-score with cash flow metrics led to the F-score model, enhancing predictive capabilities [55]. However, Chou et al.'s study indicated the Z-score model's lower accuracy for non-bankrupt firms, prompting further model innovations. Burke's 2015 multiple linear regression model demonstrated adequate financial risk warning, focusing on critical financial ratios [56].
In summary, multivariate models have evolved from Altman's Z-SCORE to more nuanced approaches, improving accuracy but still grappling with assumptions and variable distribution challenges.
Artificial neural network model
Since the 1990s, artificial neural networks (ANNs) have become a vital tool in financial risk early warning, gaining widespread application due to their unique advantages. Odom et al. pioneered using ANNs in this area in 1990, creating a model based on Altman's research with 81.75% and 78.18% accuracy for predicting distressed and healthy firms, respectively [57]. This initial success spurred further advancements [58, 59, 60, 61, 62], significantly impacting enterprise failure prediction research. Pan's work on optimizing regression algorithms for neural networks [63] showcased the method's potential and efficiency in financial risk prediction, laying a theoretical and methodological foundation for future studies.
In the field of bankruptcy prediction, many scholars have begun to adopt various methods from machine learning and artificial intelligence. Examples include rough set theory [64, 65, 66, 67], case-based reasoning[68], and support vector machines[69, 68, 70, 71]. Rough set theory, originally designed to deal with the problem of the identifiability of objects in a set, has been widely used in various types of financial decision analysis, with accuracy reports in bankruptcy prediction ranging from 76–88%.
As a practical and intuitive method for solving real-world problems, case-based reasoning has become essential in business failure prediction due to its simplicity, competitive performance compared to modern methods, and ease of pattern maintenance (Lin et al., 2011). Support vector machines, derived from statistical learning theory, were first applied to business failure prediction in 2005 [72, 73] and are superior to artificial neural networks in some cases (Kim, 2011). These methods enrich the theory and practice of bankruptcy forecasting and provide new perspectives and tools for future research.
In recent years, some studies have shown that ML's risk prediction effectiveness is as high as 98.8%. Many scholars use a machine learning model to conduct research in financial risk early warning [74], for example (Zhu Weidong et al.2022; Tan Benyan et al.2023; Yang He et al.2021; Guachamin et al.2020; Clintworth et al.2023; Liu Mingjin et al.2021; Li Huwei2022; Ma Yong et al.2021; Zhang Jianhui2022; Wen Chunhui et al.2021; Zhang Limin et al.2023; Xu Lu et al.2023; Zhu Xinji et al.2021; Gao Jieying et al.2022; Wei Jingya2022; Tang Xiaobo et al.2020; Yang Bo2020; Ge Lin et al.2022; Chen Jianyuet al.2023; Zou Yao et al.2022; Hou Xuechen2022; Polak Petr et al.2020; Liu Wanan et al.2022; Y. Song et al.,2022). Table 5 clarifies explicitly the classification and related information of machine learning models.
Table 5
No. | Model Name | Advantages | Disadvantages |
1 | Linear Regression[75] | - Easy to implement and interpret, ideal for beginners - High computational efficiency, suitable for small to medium datasets - Strong interpretability, results easily understood by business experts - Suitable for predictive modeling with linear relationships, clear assumptions about relationships | - Sensitive to outliers, affected by anomalous points - Can only handle linear relationships, incapable of fitting non-linear data - Prone to multicollinearity, requires detection and removal of correlated features - Predictive power is limited by linear assumptions |
2 | Decision Trees[76] | - Easy to understand and interpret, can be visualized - Handles both numerical and categorical data - Requires minimal data preprocessing, like normalization - Capable of handling complex non-linear relationships - Naturally handles multi-class classification problems | - Prone to overfitting, especially with deep trees - Sensitive to small changes in training data - Single decision trees often underperform compared to other models - Decision boundaries can sometimes be too complex and unrealistic |
3 | Random Forest[77] | - Performs well with high-dimensional data and large training samples - Reduces overfitting, enhancing model generalization - Handles missing values and feature selection automatically - Effective with non-linear data - Suitable for both classification and regression tasks | - Longer training times, especially with large datasets - Poor model interpretability, difficult to understand decision paths of individual trees - Requires substantial computing resources and memory - Many hyperparameters to tune, complex and time-consuming tuning process |
4 | Support Vector Machines (SVM)[78] | - Excellent performance in high-dimensional spaces, especially when the number of features exceeds the number of samples - Low generalization error, minimal risk - Suitable for complex classification boundaries - Effectively handles non-linear relationships with kernel trick | - Computationally intensive for large datasets - Complex parameter tuning, especially choosing the kernel - Sensitive to missing data - Requires thorough feature scaling |
5 | Neural Networks[79] | - Powerful non-linear modeling capabilities, can simulate complex functions - Suitable for large-scale datasets - Outstanding performance in areas like image and speech recognition - Highly flexible, customizable architectures | - Long training times, high resource consumption - Requires a large amount of data for good performance - Black box model, poor interpretability - Prone to overfitting, requires regularization techniques for control |
6 | K-Nearest Neighbors (KNN)[80] | - Simple algorithm, easy to understand and implement - No training phase required, suitable for dynamically changing data - Applicable for both classification and regression tasks - Few assumptions about data, no parameter estimation required | - High computational cost for large datasets - Sensitive to imbalanced data, influenced by majority classes - Large storage space needed, stores all data - Feature scaling is necessary, sensitive to irrelevant features |
The evolution of risk warning models, from univariate to multivariable models, and then extended to neural network models, shows the continuous progress of technology and improves prediction accuracy. The multi-domain application and literature analysis of these models is crucial to the construction of an efficient university risk early warning system, which helps to stabilize the financial security of colleges and universities and provides strong decision support for school administrators to optimize management strategies and enhance the overall robustness of HEIs.