An empirical study on machine learning-based future loss prediction for insurance firms: evidence from China’s property and casualty insurance sector

Accounting estimates are crucial in financial reports, especially in the insurance industry, where actuaries’ estimates of future claims reserves directly affect corporate financial transparency and operational decisions. Existing research has revealed the uncertainty and subjectivity of traditional accounting estimation methods, and has recognized the positive role of machine learning, but pure machine learning faces the challenge of insufficient interpretability. Based on financial data of Chinese property insurance companies from 2014 to 2022, this paper selects five major insurance types, adopts three machine learning algorithms: random forest, gradient boosting machine, and NGBoost to predict future losses, and compares them with actuary predictions. The results show that the machine learning models are superior to actuary predictions in terms of indicators such as mean absolute error (MAE) and root mean square error (RMSE). After incorporating actuary prediction data into the model, accuracy and reliability are significantly improved. On this basis, this paper analyzes the interpretability of the optimal machine learning model through feature importance ranking, verifies the value of actuary predictions, and provides a feasible path for actuaries to transform into algorithm supervisors and auditors.