Today, we have the privilege of speaking with Ms. Nigar Karimova, a distinguished expert in financial risk management who has made remarkable strides in the industry. Ms. Karimova's career has been marked by her exceptional performance at Citibank, where she garnered national and international success. Her profound expertise and innovative thinking set her apart as a leader in her field.
As Co-Founder and Partner of a leading venture capital fund, Ms. Karimova has demonstrated unparalleled leadership in strategic investments. Her leadership was instrumental in securing a significant $3.5 million investment in NUCS AI, a testament to her dedication to promoting technical innovation and supporting transformative initiatives within the financial sector.
She is also actively working on an exciting proposal aimed at introducing a Debt Repayment Capacity Calculation for Wholesale Credit Risk, which will be presented to the Central Bank of Azerbaijan. This initiative highlights her critical role in advancing financial risk management practices in the country and showcases her commitment to fostering innovation in the industry. Currently, Azerbaijan’s top banks are eager to benefit from her insights and expertise, recognizing her as a valuable asset in the financial sector.
In this interview, we will explore Ms. Karimova’s insights, experiences, and the significant impact she continues to make in the field of financial risk management. Specifically, we will discuss her most notable achievement: pioneering the integration of artificial intelligence into risk analysis, which has transformed how financial institutions assess and manage risk.
– Dear Nigar, it is a great honor for us to discuss your upcoming research work on the application of artificial intelligence for credit risk assessment of small and medium enterprises. What inspired this research, and what was the reason for referring to this topic as the main focus?
– Thank you for having me here. The inspiration for this work, or selection as you might call it, lies in the implications of applying AI models in credit risk modeling. By doing so, financial institutions can enhance the accuracy of identifying potential defaulters, thereby reducing their credit risk. Additionally, this approach can decrease the unfair rejection of credit access for SMEs, significantly contributing to economic growth.
– In your work, you compared an AI-based random forest model with traditional credit risk assessment methods. What were your key findings, and how did the AI model show advantages over traditional approaches?
– There are many findings illustrating the benefits of applying the AI model. Time management is at the forefront of these reasons. The application of AI in various fields has gained momentum in recent years, which has bolstered efficiency and effectiveness of organizational outcomes. The financial system is not an exception and has benefited from multiple contributions of AI and machine learning. Credit risk modeling is one of the areas prone to human errors. Additionally, the manual work involved in calculating default scores through assessing each entry in the database is a cumbersome and time-consuming process. AI is, however, not only powerful for completing long processes in a relatively short period but also can capture complex relationships better compared to traditional models. Traditional models rely on the static nature of variables in the financial world, which is not the case. A dynamic nature of variables must be incorporated into the analysis to make more informed decisions.
– Were any external macroeconomic factors, such as amendments or changes in interest rates or currency fluctuations, taken into account when building the model?
– Certainly! Learning from historical data, machine learning algorithms can provide a much more effective outcome in terms of accuracy. AI models investigate internal factors such as performance measures of companies and analyze macroeconomic and political external factors, which can have a consequential impact on the performance of firms.
– How did you evaluate the importance of different features in the model, and which features proved to be most significant for predicting default?
– By using historical data to build simulations and learn from thousands of iterations, AI models predict the variations in external variables and their impact on the performance of SMEs instead of focusing on limited historical data of the performance of SMEs themselves. The research aims to see a practical improvement in the accuracy and prediction of the model used traditionally for credit scoring after the application of the random forest model.
– How does the model handle non-standard situations, such as crises or sharp changes in industry conditions?
– Talking about non-standard situations of the firms, AI in credit risk assessment has been used on multiple fronts. Firstly, the capability to analyze a vast amount of data in a short period of time has been one of the most attractive features of AI application. Additionally, the complexity of analyzed data is particularly impressive with AI as it can effectively divulge relationships between complex variables, capture non-linearity, and provide accurate outcomes for predictions. AI models also reduce noise in the data, indicating that these models select the most important variables in the dataset.
– How do you envision the process of integrating the random forest model into an existing credit scoring system of a financial institution? What factors need to be considered when assessing the economic efficiency of implementing the model?
– The value of integrating random forest models is high in the case of SME default projections because these models can handle non-linear data and relationships. Oftentimes, relationships between variables cannot be accurately analyzed by assuming linear relationships. Therefore, random forest models are a powerful tool in this regard. Regarding the economic efficiency of implementing the model, the application of AI in various fields has gained momentum in recent years, which has bolstered efficiency and effectiveness of organizational outcomes. The financial system is no exception and has benefited from multiple contributions of AI and machine learning.
– How do you assess the potential impact of your research on the development of the financial market in Azerbaijan and other countries with developing economies?
– The power of the model lies in the fact that it incorporates multiple decision trees and captures non-linear relationships between variables, which is one of the shortcomings of traditional models. Considering that there are multiple dynamic risk factors impacting SMEs in Azerbaijan and in general, the model provides useful insights into the complex nature of relationships between variables.
– What steps need to be taken to accelerate the implementation of innovative solutions in credit scoring in financial institutions?
– Several steps are taken in the model to arrive at a final outcome. Firstly, the model takes sample data from the original data and creates training datasets. The randomness of the selection of data points from the original data and replacement contributes to the robustness of the model. This step is known as bootstrap sampling.
– What challenges may arise when scaling the model to a large volume of data and for many clients?
– With regards to limitations, there are certain shortcomings of the model that need to be discussed. Firstly, the accuracy of the model is as good as its input. The inputs for key variables of the model were provided by the financial institution (a local bank in Azerbaijan) in agreement with SMEs in the data. As there is no publicly audited financial data on these SMEs, it is possible that there are inaccuracies in the data, as SMEs could have been interested in manipulating the data for a better credit score outcome. Additionally, the decision-making rationale and the process by AI models cannot be easily explained. Machine learning models are challenging to interpret, and financial institutions could find it difficult and sometimes unethical to explain to SMEs why their application for a loan has been rejected. Therefore, simply referring to a higher accuracy of a particular model in justifying a loan application decision can be difficult to substantiate. The limitations of the model rely on historical data provided by SMEs to a financial institution, which can be inherently biased to present the financial conditions and performance of firms favorably. Also, a random forest is a complex model based on machine learning, and its application to sift through potential defaulters might be challenging to explain by banks and can carry potential ethical concerns as the work principles of machine learning algorithms are not clear enough to substantiate decisions based on them.
– What research directions do you consider most promising for further development of AI-based credit risk models?
–AI models also predict the variations in external variables and their impact on the performance of SMEs instead of focusing on limited historical data of the performance of SMEs themselves. The research aims to see a practical improvement in the accuracy and prediction of the model used traditionally for credit scoring after the application of the random forest model.
– How do you plan to expand your research to include other types of loans or other emerging markets?
– According to my research, financial institutions usually reject loan applications from SMEs or charge them a higher interest rate to cover for potential risks. However, as a random forest model is more dynamic and can integrate the future potential of the enterprises better into the analysis, it is also more likely that these models with the help of machine learning can provide a fairer assessment of the creditworthiness of SMEs. This will lead to an allocation of funds to SMEs which merit it with their historical information.
– What role do you think regulators should play in the development and application of AI in the financial sector? What regulatory frameworks are needed to ensure responsible use of AI?
– AI has revolutionized finance and has been applied in credit score modeling extensively. AI in credit risk assessment has been used on multiple fronts. Firstly, the capability to analyze a vast amount of data in a short period of time has been one of the most attractive features of AI application. Additionally, the complexity of analyzed data is particularly impressive with AI as it can effectively divulge relationships between complex variables, capture non-linearity, and provide accurate outcomes for predictions. AI models also reduce noise in the data, indicating that these models select the most important variables in the dataset. These variables are external factors such as overall economic conditions in the economy, trends in the market, consumer behavior, and even political risks. By capturing real-time data and disclosing its impact on risk factors, the prominence of AI in financial markets has increased.
– Based on your significant experience in risk management, how does this research relate to your career development and professional trajectory?
– Based on my significant experience in risk management, this research directly aligns with my career development and professional trajectory. By delving into the application of AI in credit risk modeling, I can leverage cutting-edge technology to enhance traditional risk assessment methods. This not only broadens my expertise in innovative risk management strategies but also positions me at the forefront of industry advancements. The insights gained from this research allow me to contribute more effectively to my organization’s risk management practices and drive impactful, data-driven decisions that support sustainable growth.
– The study focuses on small and medium enterprises in Azerbaijan, which face unique challenges. Why did you choose this context, and how do you think the research results can be applicable on a broader scale?
– The findings also have broader implications for the use of AI in finance along with the context of Azerbaijan. The results lend support to the use of AI in finance, including credit risk assessment, by highlighting the fact that the accuracy and precision of the model rise significantly when a machine learning model is used instead of a traditional logistic regression. It is also evident that machine learning models need a considerable dataset to be trained to achieve a reliable outcome. The practical implications from this fact are that relying on a limited dataset for an accurate outcome with the use of machine learning can backfire. Finally, ethical issues in the use of machine learning, such as content for the inclusion of data in training the model and transparency, should be considered before interpreting and applying the model outcomes in practice. However, the solution to some of these issues can be beyond the control of banks, such as transparency in machine learning algorithms, as banks, particularly in developing countries like Azerbaijan, rely on ready algorithms instead of determining the work principles of complex machine learning models themselves.
– Nigar, we thank you for your insights and engaging conversation, and wish you great success with your research!
– Thank you for the opportunity to engage in this great conversation. I look forward to our work contributing positively to our country.