COVID-19 FORECAST: COMPARISON AND COMBINATION OF AU- TOREGRESSIVE AND COMPARTMENTAL MODELS

Resumo

The Imperial College was responsible for the firsts forecasts for covid-19 propagation, in the early
stages of the pandemic. Their study had a positive effect in the British Lockdown and it may have saved many
lives. Besides the Imperial College study, many other studies were proposed based on different approaches. Some
models have projected much more deaths than we have observed, while others have projected much less. There
are plenty reasons to explain why the model can fail, it can be caused by bad data, virus mutation, health care
system collapse, and others. One strategy to reduce the error in the propagation forecast is to combine different
approaches. In this work, we are going to present error analysis for the forecast of covid-19 propagation based on
autoregressive models (such as ARIMA and SARIMA models) and compartimental models (such as SIR model
and variations) and how the forecast can be improved by the combination of these two approaches.