Modeling and Forecasting Mortality Rates During the Covid-19 Pandemic Using the Second Adapted Nolfi Model and Auto Arima

Modeling and forecasting mortality rates have been widely performed using various approaches. One such approach is the Second Adapted Nolfi model, which is one of three adaptations derived from the Nolfi and Generalized Nolfi models. Unfortunately, its application remains limited compared to widely used models like Lee-Carter and Cairns-Blake-Dowd. Previous studies on this model have shown satisfactory performance, particularly in residual analysis. However, those studies were conducted before the COVID-19 pandemic, and no study has yet applied it in the pandemic or post-pandemic periods. Although the pandemic may appear less relevant in 2025, the absence of such studies highlights the importance of further investigation into the model’s performance under extreme demographic conditions. This study addresses that gap by evaluating the Second Adapted Nolfi model using data from the Human Mortality Database (HMD) for the United States, the United Kingdom, and Italy. The model was applied to data up to 2019, and Auto-ARIMA was used to forecast from 2020 onward. The modeling results indicate that the logarithmic mortality curves align with established patterns, such as high rates at age 0, a decline through childhood, a sharp increase in early adulthood, and a continued rise into old age. The results also show that HMD mortality rates exceed the forecasted values for individuals aged 80 and above, suggesting increased elderly mortality during the pandemic. Three error metrics were used, yielding RMSE values from 0.01 to 0.18, MAE from 0.004 to 0.07, and MAPE from 28 to 286. Although Italy had the highest MAPE, the United States and the United Kingdom also showed notable errors. These findings reveal both the pandemic’s demographic impact and limitations of the model in capturing sudden shocks. Future studies may enhance this model through new adaptations, further comparison with other models, or alternative smoothing techniques to develop more robust mortality forecasts.