COVID-19 statistical forecasts
If the page does not update, press shift + refresh
Last update Wed 14 Oct 17:20:32 CEST 2020

Disclaimer: this is a scientific exercise, do not take it too seriously.

Plots on the left report the cumulated number of deaths, plots on the right the daily number of deaths.
Since April 3rd, a 5 day running mean is applied to all data.

Since April 9th, with Davide Proment we implemented the same exercise for England regions on his GitHub page.

Italian data from GitHub repository of Italian Civil Protection.
World data from GitHub repository of Johns Hopkins University.

Methodology note
Forecasts are estimated with Gompertz functions computed via the nls.lm R function. Ribbons show the 95% confidence interval computed with the confint2 R function.
Since April 3rd, a 5 day running mean is applied to all data
The non-linear model to forecast the evolution is reliable only after one third of its path (Winsor, PNAS, 1932). Hence errors are likely larger than currently estimated in situation where the outbreak is just started.
Saturation values are the alpha parameter of the Gompertz fit. Errors are estimated using the half of the distance between the 2.5% and the 97.5% percentiles. When forecasts error is larger than half of the forecast value, i.e. the forecast is completely unreliable, NA is shown.
Forecasts are updated at least twice a day (18.30 and 05:30).
Some inconsistencies in Italian prediction may arise from slight data differences between the Civil Protection and the Johns Hopkins archives.
The code can be found on my GitHub repository, althought it is really badly written.

Italian Time Evolutions of the Prediction

World Time Evolutions of the Prediction

Italian Forecasts Evolutions

Forecasts from the past 10 days

P. Davini (CNR-ISAC), March 2020