covid-19 forecasting

I've been spending a lot of time this past week on some COVID-19 forecasting (collaboration with my friend Tom). Our spreadsheet tracks trends at a global macro level, state-level, and also includes a US forecasting model to see what happens when we enact different lockdown procedures inspired by different countries (severe / moderate / etc) and the impact of early enforcement.

It's been a particularly sobering experience. I wanted to share 2 things I learned about the disease outcomes from the numbers I am seeing: (1) the pandemic will become much worse than imagined if we don't take necessary precautions fast, BUT ALSO (2) if we follow lockdown precedents from other countries like S. Korea and Italy, early action could lead to a dramatic containment of the disease. We're currently about 10 days behind Italy, and if we can implement what they are doing earlier, many deaths could be avoided.

We have all the raw data and sources cited that we use in the spreadsheet, which is still continuing to grow in our analysis on a day-by-day basis. Please look at the numbers, and feel free to reach out any time if you need help, support, or even if you have any further ideas and insight on what we've done so far. We're in this together. #COVID19

Link here:

- - - - -

Some other data-driven takeaways (supplemented by some charts that are in the link above):

Screen Shot 2020-03-18 at 6.51.42 PM

(1) Growth Rate by Country - really encouraging news from Italy that the growth rate is declining so rapidly from last week, especially if we're able to implement the same rigidity of lockdown style. Acting earlier, even by a couple days, could reduce the amount of infections by an order of magnitude.
>> The growth rate for the US, on the other hand, is really alarming. This could also be due to the huge lack of testing that is happening (limited tests = people who "qualify" for tests are the ones exhibiting the most symptoms = higher % of positive cases), as well as the number of people starting to exhibit these symptoms from the incubation period.
>> We are currently also digging into how much of growth is due to testing (in)availability / testing catchup

Screen Shot 2020-03-18 at 6.54.23 PM

(2) S. Korea and Italy since Lockdown Date - this chart visually summarizes some information from the daily decrease in growth rates from previous lockdowns. A huge amount of mitigation can occur if we have the right policies; the shelter-in-place order that is happening currently in the Bay Area gives me a lot of hope. You can track the restrictions that each state has implemented here:

Screen Shot 2020-03-18 at 6.56.48 PM

(3) State-level Cohort Analysis: Walk Score - the least safest places are cities with high walkscores. This intuitively makes sense as harder to get places = less population clusters / density / people walking on the streets to get from place to place.

>> Implementing measures to "artificially" decrease walk score will have huge forward impacts in terms of limiting disease spread. Social distancing works!