Document Type : Research Paper
Authors
1
Prof., Faculty of Physics and Energy Engineering, Amirkabir University of Technology, Tehran, Iran
2
Ph.D Candidate, Faculty of Management and Accounting, Allameh Tabataba’i University, Tehran, Iran
3
2Prof., Faculty of Management and Accounting, Allameh Tabataba’i University, Tehran, Iran
4
Associate Prof., Faculty of Management and Accounting, Allameh Tabataba’i University, Tehran, Iran
10.22059/frj.2024.367421.1007532
Abstract
Aim: The study is done to present a novel model for predicting the future commitments of insurance companies. Traditionally, insurance companies use the Cahin_ladder approach as a statistical tool to forecast future claims and the trend of claims development. The forecasted amount of money insurers calculate for their futute liabilities are called claim reserves. The Chain_ladder as statistical approach is also favored by regulatory authorities in various countries due to its simplicity in assumptions and clear interpretation. However, some assumptions such as the stability in data development and linear relationships between variables can affect the efficiency of this model when faced with internal policies or external factors like the COVID-19 pandemic. Forecasting future commitments close to reality may lead to the financial stability of Insurance companies. The amount that insurance companies allocate to meet their future technical obligations is known as reserves. Calculating reserves less than the required amounts can pose challenges for insurance companies in fulfilling their commitments while calculating more than necessary amounts, can impact the financial statements of insurance companies negatively.
Method: In this study, a dynamic model based on machine learning algorithms is proposed. The model's output, which is the combination of the number and timing of bodily injury accidents, plays a key role in calculating reserves in nonlife insurance products. This model is trained to predict the frequency of accidents in a Vehicle Third-Party liability insurance. The model can identify hidden layers and non-linear, complex relationships among claims data. A Long Short-Term Memory (LSTM) neural network algorithm is implemented in this study which is known for its high predictive capability in time series data. The model is trained by the historical data of Karafarin Insurance Company from 1396 to 1400.
Findings: The performance of the model is highly related to the hyperparameters chosen for the model. Two of the most common approaches for tuning the hyperparameters are tested in this study. These Two models are grid and random search. The Root Mean Square Error (RMSE) is used as a performance metric, and it indicates that the grid search has a lower RMSE than the random search for the training data with a slight difference (16.33 versus 17.4). However, the results for the test data in the grid search have a sign of overfitting.
Conclusion: The recommendation of this study is to use the random search for tuning the hyperparameters of the model to predict the frequency of daily incidents. The evaluation between the two approaches for tuning hyperparameters reveals that the random search is more suitable for working with unfamiliar data and handling overfitting situations. Overfitting occurs when the model is heavily influenced by the training data and learns not only the actual patterns but also the noise and minor details of the data. This issue may affect the generalizations of the model.
Keywords
Main Subjects
)