Auquan’s SEIR Model: Prediction Error Analysis
June 22, 2020 by Vishal
Our disease model is a modified SEIR model, adapted with additional compartments to account for special characteristics of COVID 19, specifically the asymptomatics.
More details here: https://blog.auquan.com/page/cvdmdl2\ or see our our other document: ‘How we're predicting the Coronavirus Pandemic - our SEIR Model’
The model is fit to each country’ s past data to infer parameters, with thought bounds. This allows us to capture effects like different testing capacities, hospital capacities and mortality rates in different countries. We also weigh recent data heavily compared to older data to capture effects of the most recent lockdown policies. Finally, these models tend to be sensitive to initial starting points, so we use a metaheuristic based approach to fit the model.
Below is the RMSE (root mean square error) of actual deaths to predicted mortalities for 1d, 2d, 3d, 1w and 2w ahead for select countries.
Note that due to the changing nature of the pandemic, govts change policies frequently, often within 1-2 w with immediate implementations, hence the higher error on 2w forecasts. Even then our model is able to estimate mortality within ~10% error range.
Share this Article:
Auquan is a data science solutions provider for asset managers and hedge funds. Our state of the art technology empowers Portfolio Managers to stay ahead of the trend and achieve better returns.