Actuarial Learning for Pension Fund Mortality Forecasting (2504.05881v1)

Abstract: For the assessment of the financial soundness of a pension fund, it is necessary to take into account mortality forecasting so that longevity risk is consistently incorporated into future cash flows. In this article, we employ machine learning models applied to actuarial science ({\it actuarial learning}) to make mortality predictions for a relevant sample of pension funds' participants. Actuarial learning represents an emerging field that involves the application of ML and AI techniques in actuarial science. This encompasses the use of algorithms and computational models to analyze large sets of actuarial data, such as regression trees, random forest, boosting, XGBoost, CatBoost, and neural networks (eg. FNN, LSTM, and MHA). Our results indicate that some ML/AI algorithms present competitive out-of-sample performance when compared to the classical Lee-Carter model. This may indicate interesting alternatives for consistent liability evaluation and effective pension fund risk management.

Actuarial Learning for Pension Fund Mortality Forecasting

This paper presents a comprehensive paper on the use of machine learning techniques in actuarial science, specifically for forecasting mortality rates in pension fund participants. This emerging domain, termed "actuarial learning," leverages advanced algorithms and computational models to analyze substantial volumes of actuarial data, which traditionally rely on the classical Lee-Carter model for mortality prediction.

Methodology and Models

The authors employ a suite of models including regression trees, random forests (RF), boosting, XGBoost, CatBoost, and neural network architectures such as Feedforward Neural Networks (FNN), Long-Short Term Memory (LSTM), and Multi-Head Attention (MHA). These models are evaluated against the traditional Lee-Carter model widely used in the literature.

The paper utilizes data from Brazilian pension funds, focusing on participants aged over 30 years. The machine learning models are implemented primarily in R, with CatBoost executed in Python, and neural networks in Keras. The purpose is to predict one-year-ahead mortality rates, assessing out-of-sample performance through metrics such as Mean Absolute Error (MAE) and Root Mean Squared Error (RMSE).

Numerical Results and Comparative Analysis

The results demonstrate that some machine learning algorithms showcase competitive out-of-sample performance relative to the Lee-Carter model. CatBoost and FNN, in particular, are highlighted for achieving the best MAE and RMSE scores, respectively, indicating their potential in providing accurate mortality forecasts. The FNN stands out for producing smooth and increasing out-of-sample predicted mortality curves, a desirable quality in mortality modeling.

The analysis reveals that actuarial learning can effectively address the mortality heterogeneity found in countries like Brazil, by providing predictions that are more aligned with the specific population segment of pension fund participants rather than relying solely on national statistics.

Implications and Applications

This paper has significant practical implications for pension fund management and risk evaluation. Accurate mortality forecasts are essential for assessing financial liabilities and managing longevity risk, which are crucial components of pension fund solvency and sustainability.

Applications conducted with the FNN model include forecasting life expectancy changes, evaluating impacts from unforeseen events such as the COVID-19 pandemic, and calculating expected cash flows for pension plans. These applications are integral for tasks like asset and liability management and market risk analysis in pension schemes.

The integration of machine learning into actuarial practices showcases a shift towards more data-driven methodologies, offering deeper insights into risk evaluation and enabling more tailored actuarial strategies. Future research could further explore the scalability of these approaches in different actuarial contexts and populations.

Conclusion

The advancement of actuarial learning through machine learning techniques promises enhanced capabilities in mortality forecasting, thereby enriching the toolset available to actuaries for comprehensive risk assessment and decision-making in pension systems. This paper underscores the vitality of adopting innovative methodologies in actuarial science to tackle the complexities of modern financial landscapes.