Over the past several weeks, JPMorgan’s Marko Kolanovic has demonstrated the utility of using a big data approach to modeling the proliferation of COVID-19 in the US.
Generally speaking, he projected an apex in New York sooner than broadly-cited models and suggested a peak in hospitalizations was likely imminent. He also spent a considerable amount of time outlining the extent to which big data and machine learning should be marshaled in the fight not just against the current epidemic, but future pandemics, in order to help humanity stay one step ahead as opposed to one (or more) steps behind.
Kolanovic’s forecasts, as laid out in a series of notes dating from March 24 and running through last week, proved prescient. “The predictions of our models were drastically different from the consensus”, Kolanovic writes, in a new note out Monday. “Current New York State official data confirmed our predictions”, he adds.
In his latest, Marko asks (and attempts to answer) the only question that matters both for market participants and, more importantly, for everyday people, millions of whom are currently jobless and face an uncertain economic future.
“The critical question to direct the prevention of further spread and reopening of the economy is: what percentage of the region’s population has developed immunity?”, Marko writes, noting that “this is currently unknown due to the large number of asymptomatic and recovered cases that were never recorded”.
For Kolanovic, this question can be addressed short of testing the entire population using a combination of geographical and demographic data which give clues to “the true mortality rate”. Here’s Marko:
If we know the true mortality rate of all COVID-19 cases (including not tested, recovered and asymptomatic), we can compare the number of tested positive cases, the number of fatalities, and back out (i.e., estimate) the % of population that currently has or had COVID-19, i.e., level of herd immunity.
This is obviously crucial. In a note making the rounds on Monday, Morgan Stanley cautioned that many workers won’t be able to go back to their jobs “until a vaccine is abundantly available as social distancing cannot be fully relaxed until we have herd immunity”.
Michael Burry (of Big Short fame) last week delivered a series of somewhat inflammatory tweets which, if you could get past the rather abrasive delivery (no small feat) contained a simple message. “Denominator deflation infests COVID-19 stats”, Burry wrote. “In countries with more comprehensive testing, there are better estimates of the true case fatality rate”.
In his Monday missive, Kolanovic notes that the reason Italy, Spain and the UK (for example) have very high fatality rates isn’t because the virus is 20x more efficient at killing people in those countries. Rather, the percentage of the population being tested is smaller, and thus doesn’t count large numbers of positive cases which are mild, asymptomatic or already recovered.
“The analogy is with an iceberg — what is being measured in COVID-19 hot zones is only the tip of an iceberg, hence the more one measures, the more cases that are uncovered”, Marko writes, adding that “theoretically, the closer one is to testing the full population, the better the estimate of the true mortality rate”.
Marko looks at cross-sectional data and points to the following straightforward visual, which is the reported mortality rate versus the percentage of the population tested.
The locales in green represent an interesting series of case studies, to say the least.
In Gangelt, Germany, for example, a super-spreading event led to a large outbreak. Subsequently, a large percentage of the municipality’s population was tested. Here, via the MIT Technology Review, is what researchers found:
The survey in Germany was carried out by virologist Hendrik Streeck and several others at the University Hospital in Bonn, who say they approached about 1,000 residents of Gangelt to give blood, have their throats swabbed, and fill out a survey.
They found that 2% of residents were actively infected by the coronavirus and a total of 14% had antibodies, indicating a prior infection. This group of people, they say, “can no longer be infected with SARS-CoV-2,” as the virus is known to scientists.
From the result of their blood survey, the German team estimated the death rate in the municipality at 0.37% overall, a figure significantly lower than what’s shown on a dashboard maintained by Johns Hopkins, where the death rate in Germany among reported cases is 2%.
Kolanovic also cites the Faroe Islands example. “Given that salmon farming on the islands requires test equipment for Salmon isavirus, Faroe had the resources and knowledge to quickly repurpose equipment for COVID-19”, Marko explains, noting that more than 10% of the population was expeditiously tested and the outbreak was quickly brought under control.
There were just 184 cases, apparently. The mortality was 0%, as was the severe/ICU case rate. Here’s The Guardian with a bit more:
Veterinary scientist Debes Christiansen who adapted his veterinary lab to test for disease among humans rather than salmon is being celebrated for helping the Faroe Islands avoid coronavirus deaths, where a larger proportion of the population has been tested than anywhere in the world.
The north Atlantic archipelago currently has only one person in hospital with Covid-19 and it is one of three European countries, along with Georgia and Liechtenstein to so far not have any deaths from the virus.
His laboratory, which was primarily geared to test salmon for viral infection, was adapted and he purchased the extra ingredients required to test humans. There has been mass testing for virus among salmon farmed in Faroese waters since an outbreak of salmon anaemia virus in 2001 ravaged the species. Of the archipelago’s total export value, 90% is accounted for by fish and half of that is salmon.
According to official records, 10% of the population of 61,000 people have now been tested for the coronavirus. Faroese doctors have been able to track and quarantine everyone who has had contact with the 184 people who had tested positive for the virus. Of those who have fallen ill, 131 have fully recovered.
That’s quite something. Of course, these are small sample sizes, but Kolanovic notes that “Iceland [also] represents an example of successful prevention of the outbreak [and its] data size is larger and hence significant in statistical analysis”.
He then takes things further. For example, he uses an estimate of the true mortality rate to project “which part of the population was affected by the COVID-19 virus in hotspots”, a figure which can then indicate the level of herd immunity. Importantly, he accounts for the average age of the population across various regions. Here’s Kolanovic one more time:
The estimated implied COVID-19 prevalence levels in hotspots are consistent with the Gangelt study. In particular, most affected regions (e.g., Lombardy, Madrid, NYC) all suggest approximately ~15%-20% prevalence of COVID-19 (current and past cases). Given that the epidemic in these regions is showing signs of slowdown and decline, this could indicate that a combination of distancing measures and limited herd immunity could be capping the COVID-19 spread rates already at indicated prevalence rates (during the first wave of epidemic).
Although he notes that the results, taken in conjunction with effective social distancing measures, may cap fatality outcomes in some regions, the implied COVID-19 incidence level (again, based on observed fatalities and the estimated age-adjusted mortality rate) for many regions is still very low. The read-through from that, Kolanovic says, is that “the risk of spread is high and governments should focus on controls and travel restrictions from COVID-19 hot zones to COVID-19 unaffected regions”.
His analysis underscores the notion that the true mortality rate is likely somewhere in the neighborhood of o.4%, and that the percentage of the population affected by the virus is likely much larger than known in hot spots.
Ultimately, Kolanovic’s latest effort is aimed at getting a handle on the size of the immune population. His latest adds to what is now a considerable contribution from Marko to the debate around when and how to safely restart the economy.
More importantly, his recent work could help inform public policy more generally during future epidemics, assuming it gets to the right people.
Perhaps Jamie Dimon could assist when it comes to making sure Kolanovic’s work is seen by lawmakers – just a thought.