Photo Credit: Lars Plougmann || Indeed, this seems like a race, and the S-Curve is a major challenge to drive through
This will be brief, because I am still working on it, but it is my weak conviction that as far as the markets are concerned, the COVID crisis will largely be over by next Friday. How certain am I? Not very — I give it a probability value of around 30%.
If my thesis is correct, reported new cases of COVID-19 in the US will peak by Friday of this week, and will be 90% complete by next Friday. I will be watching how many new cases are reported. New cases tend to peak when total cases increase at a mid-teens percentage rate over the prior day. Because reporting is noisy, you don’t see that so easily, but the inverse logistic curves I am estimating are consistent on that figure for all the countries I have modeled so far.
I’ve run models for South Korea and Italy as well, and I’ll run them for a few more countries tomorrow. They are all pretty consistent with each other. Italy’s new cases should peak tomorrow, if they haven’t already.
I know everything is dark and gloomy now. Even if my modeling is wrong, which is a significant likelihood (I am extrapolating), I find it difficult to believe that we will still be in crisis mode by tax day.
So, cheer up. The number of COVID-19 cases is unlikely to be overwhelming, and we are all likely to survive this. The markets will revive, though maybe not energy stocks for six months. Those are a separate issue.
And if new cases track my estimates, I will put more money into the market. That’s all for now.
I’d love to know why you think the USA will peak then when it isn’t really testing people :). Italy I could believe, Spain maybe in 12 days, but very surprised by your USA call. Can you break it down?
The model is a simple one, and it fits the data for the countries modeled pretty well. Extrapolation is always a risk, though. This model says new cases for Italy and Spain peaked on 3/15. It fits an inverse logistic curve to the total cases data for each nation. It assumes that the processes for discovering new cases are constant for each nation, though different nations discover them at different speeds. China = slowest. Spain = fastest. The US is in-between. The percentage increase in total cases is slowing down in the US — now it could have a surprise like China did back on 2/12/2020. The real question is whether additional testing will show that much… there are a lot of false positives. What does say is that with what testing is going on, the new case levels are slowing down their growth.
If testing becomes more common, and there aren’t a lot of false positives, this analysis could very well be wrong. That’s why I am not acting on this now, but am going to wait about one week as more data comes in, and act then.
Does that help?
It does, but is it taking into account the differences in testing rates between the various countries?
Implicitly the model does take into account those testing differences. More important are the distancing measures, and willingness to quarantine those who have been exposed… the model assumes that all of those policy measures are constant with a given nation.
As an aside, nations that got to it slow had to be more drastic to catch up. At present the incremental policies of the US government and states are merely action to look like we are doing something, and nothing effective that will help us more.
I hope you are right, but I’m not sure that sufficient effective measures to slow transmission were really implemented until late last week, and my reading suggests that new cases won’t peak until almost two weeks after that.
The other issue is that even the current measures are sufficient to slow transmission, these measures are not going to be relaxed anytime soon. I suspect most restaurant server and flight attendants will still be off work come Independence Day, let alone Tax Day.
But as you say, you are only giving this a 30% probability and it is good to have a plan …
I started running a similar model and got the results you mentioned above. However, when I performed the back testing (on South Korea where there is enough data), the regression didn’t predict well until near the start of the slow down. Taking out the last two weeks of South Korea gave a growth peak a couple days early but a prediction of total cases roughly 55% of the actual.
Appreciate you trying to model this.
As of Wednesday morning, 2,382 people had tested positive in New York State, an increase of more than 800 since Tuesday. In New York City, 1,339 people have tested positive, up from 814 yesterday.
Would this be consistent with your model? To be clear, this is a sincere question. I haven’t modeled it so I don’t know, but that seems on a gut level to be inconsistent.
I also wonder if the size of the U.S. will affect the curve compared to other countries.
Thank you for sharing your analysis. I think we all look forward to the rational, fact-based analyses and the commentary you post. Helpful, as always. Thanks.
David! I’m trying to avoid confirmation bias, but I’m essentially in agreement with you. I have not modeled explicitly as you did, but looking at all of the separate country data indicates a peaking in this manner you describe. I have been generally more pessimistic than those around me during this crisis, but it seems that ALL I hear now is the horrific scenarios of many millions infected and hospitals swamped all over the country, and the markets are reacting as if we don’t get a handle on this for several months.
That just doesn’t track with what has happened in most other countries, as you’re suggesting. So, I’m hoping we’re both right.