Modelización Estadística Para La Estimación Y Predicción De La Incidencia De COVID-19 En España Statistical Modelling for Estimating and Forecasting COVID-19 Incidence in Spain

Introduction: Basing decision-making processes on data containing errors and inaccuracies is unavoidable in many situations. The COVID-19 pandemic related data is a clear example, where the information provided by official sources was often unreliable due to data collection mechanisms and the amount of asymptomatic cases. Objectives: To estimate the amount of misreported data in a time series and reconstructing the most probable evolution of the process and provides a discussion on the more appropriate statistical methods able to yield reliable forecasts in this context. Methods: The usage of a model based on autoregressive conditional heteroskedastic time series is proposed, estimating the parameters by Bayesian synthetic likelihood. Results: Only around 51% of the cases of COVID-19 in the period from February 23rd, 2020 to February 27th, 2022 were observed in Spain, also detecting remarkable differences in the reporting issues between Autonomous communities. Conclusion: The presented method allows generating realistic predictions under different possible scenarios, and therefore it represents a valuable tool for policy makers in order to improve the evaluation of the evolution of a situation.