A basic disclaimer is warranted before making forecasts or even presenting analysis. We are not medical professionals and have limited access to data (likely worse than those in government agencies and universities making forecasts). Our models, because we are not medical professionals and do not have as good of data, are more simple. Please refer further to our disclaimer on this site.
In our previous report Covid-19 is Temperature Dependent, we concluded that the coronavirus is most effective in mild temperature conditions while doing worse in terms of rates of infection and mortality in both freezing and hot temperatures. In comparison to the consensus understanding that the virus is not temperature dependent or that it is seasonal like the flu (weakness to only warm weather), our conclusion is an outlier.
In this report, we make forecasts for the total number of deaths in the US incorporating our temperature dependent model.
Forecasts
For each of our scenarios, our forecast window goes until June 1, 2020. Our temperature dependent premise assumes that the virus will greatly diminish its strength by June 1st as much of the US will already be experiencing daily temperature highs in the 80s, weakening the virus. We also assume, based on data from other countries, that the natural growth rate of the number of confirmed cases will mostly run its course by this time. There is a real likelihood that the US could experience another surge starting around October as temperatures decline, but we will not go into that here.
Our forecasts through June 1, 2020 are as follows:
Low Estimate ~ 31,000
Base Case ~ 46,000
High Estimate ~ 80,000
Base Case
Here, we assume the US’s trajectory of confirmed cases will likely follow the path of Italy, one of the worst hit countries. The choice of Italy is important and deserves some explanation.
According to our temperature dependent model, Covid-19 is most effective in mild weather, roughly between around freezing to around high-70s. Italy has received the brunt of the coronavirus in Europe, for instance, as it experienced the optimal temperature conditions before others like Germany (which was recently experiencing below freezing temperatures, with Berlin hitting a nightly low of 25 on 3/30).
The US, in general, has begun to pass through its mild weather period. Specifically, New York City, the outbreak’s epicenter in the US, has been experiencing optimal weather conditions for the virus. From a temperature perspective, infection growth should look similar to Italy (or Spain) while these conditions last.
We took the growth rate of Italy and used it to forecast the US. As Europe got hit harder first, we have, depending on your reading of the situation, about 2 weeks lead time that we can use to forecast the US. Afterwards, we assume these growth rates slowly decline to about 0.5% by the end of April, remaining at this level throughout May.
For Mortality Rate, we assume a level of 4%, which is significantly higher than the level the US has experienced thus far. The average since the US went above 100 deaths is 1.78%, but has been moving slowly higher. This upward trend is in-line with other countries that experience an increase in Mortality Rate going into their respective peak number of new cases. The US appears to be following a similar pattern.
Additionally, as we noted in Covid-19 is Temperature Dependent, rates of infection and mortality vary depending on temperature conditions. Mild temperatures tend to push up both infection and mortality rates. The US, therefore, is receiving a push from both of these factors.
The 4% level is also based on shifts in Mortality Rates that we have seen in a variety of countries, not just Italy. The UK, France, Germany, and Spain experience similar patterns.
As you can infer, our model is based on guesstimates around important variables. We are not trying to diminish the work involved with putting together this or any model, but let’s face it while conditions are changing as quickly as they are guesstimates are required.
Low Estimate
Our low estimate uses a model focused around hospitalization rates.
Initially, we had very high hopes for this model as we saw similarities between hospitalization and mortality rates from Covid-19 in NYC and historical US data from flu / pneumonia. With such similarities, we hoped to be able to improve our forecasts.
Additionally, focusing on hospitalization data helps us to get around problems associated with testing. Countries have used different strategies when it comes to testing, thus likely distorting the number of confirmed cases. Hospitalizations, however, seem to be more standardized in the sense that presumably a patient will need to reach a certain level of danger before being admitted to the hospital and then ICU. In short, using this data should provide more standardized data which should be easier to use for forecasting.
However, like many models, this one has a serious flaw — sporadic data. Unlike other data such as the number of confirmed cases or deaths, detailed information on hospitalizations due to Covid-19 are not updated daily or provided from many countries, making forecasting more difficult especially in an environment that changes so rapidly.
Such a model could be worth exploring for those with better access to data, such as forecasters working for the government or even for a large university who have access to such non-public data.
Using the hospitalization data, we estimated a growth rate based on our forecasts of confirmed cases. Then, a ratio of deaths / hospitalizations was applied to our forecasted hospitalizations using historical Covid-19 data to forecast the number of deaths.
High Estimate
For the US, the worst case scenario is that it matches the country hardest hit by Covid-19.
We chose as a metric, deaths from Covid-19 / population of the country. For medium/large countries, the worst hit is actually Spain (many likely believe it was Italy, they are comparable on this metric but Spain leads).
Taking Spain’s metric and multiplying it by the population of the US, we can infer what the loss of life would look like in a worst case scenario.
There are many problems with this analysis. For instance, it is basically a back-of-the-envelope calculation (which doesn’t necessarily make it wrong). Also, judging from the current state of the US, it is in much better shape than Spain making such a forecast highly improbable.
It would be extremely difficult for the US at this time to become the worst hit country in the world on a per capita basis. It is possible, but highly unlikely.
Comparing to Consensus
As stated, this pandemic is rapidly developing, as must be the consensus. However, we can get some basis from recent forecasts announced through reliable channels.
The Institute for Health Metrics and Evaluation (IHME), an independent global health research center at the University of Washington, which is at least partially funded by the Bill & Melinda Gates Foundation, made a forecast for the ‘next four months’ on March 25, 2020. Their headline forecast was for approximately 81,000 deaths in the US. This is roughly equivalent to our worst case forecast. Our base case is almost half this figure.
The forecast of the White House’s coronavirus task force was announced at a recent press briefing. For CY2020, they provided an expected range between 100,000 and 240,000. In comparison to our forecasts, these figures seem high. If we assume Covid-19 comes back and produces the same impact during the October to December time period (once temperatures cool but stay above freezing in the US), we could essentially double our short-term forecast to create a CY2020 forecast. Such a scenario would see our base case approach the lower band of their forecasts.
For a variety of reasons, we see the scenario of the coronavirus returning with the same impact in the autumn as unlikely. The US will likely be much better prepared assuming there is an outbreak at that time. Additionally, we should assume that improved treatments will be available as well, which might not decrease the number of confirmed cases but would decrease the number of deaths.
An outlier forecast was made by Dr. James Lawler, a professor at the University of Nebraska Medical Center. When making a presentation to the American Hospital Association (AHA), he forecast the number of US deaths at 480,000.
Then, there have been numerous forecasts of deaths in the millions both in the US and around the world. But at this stage, it seems that most believe these forecasts were overly pessimistic and have mostly been disregarded.
Our forecasts appear to be well below consensus figures.
Implications if our Forecasts are Correct
For the US, our base case of around 46,000 makes the impact in terms of deaths equivalent to the flu / pneumonia. Assuming this is the case the markets will come back. It might take time for the unusual economic circumstances to unwind but this will not turn into another great depression.
In order to put our base case forecast in context, we can look to see how it fits with trends in flu and pneumonia. The CDC provides annual death estimates by cause going back decades. By adjusting the data to be relative to the population (number of deaths per 100,000 people) we can see the impact over time. The following chart compares the number of deaths per 100,000 for flu / pneumonia to our estimate for Covid-19.
Source: ZettaCap and CDC
The impact on the US, judging from the prior chart, is significant but more or less in-line with that of the flu / pneumonia. The headline numbers are terrible, there is no getting around that. And, any loss of life is a tragedy. But, in relative terms, adding the loss of life from Covid-19 to the loss of life from flu / pneumonia in 2020 looks to resemble figures of just flu / pneumonia from maybe the 1980s.
Looking Forward
The main problem is that we are not entirely sure what will occur in the autumn. Will the virus come back, will it return changed in some important way, will medical treatments improve, will a vaccine be effective, will we end up fighting the virus or strains of it for years, will future fights also involve quarantines?
There are just too many questions to declare victory, even as we point towards forecasts below consensus. The situation could still worsen over the coming years.
We feel comfortable with our short term forecasts. Over the coming months, we believe that the death toll will be shown to be on the lowest band of what people believed possible. But, again, even under such circumstances, we are not ‘out of the woods’.
Limited death toll will certainly be welcomed by extremely nervous financial markets. Our scenario looking forward is that as this marginally more positive scenario becomes apparent, markets will stabilize and try to tick higher but will be likely capped until we pass through the beginning of flu season (when we expect the conditions for Covid-19 to become optimal once again), sometime in the October to December time frame.
For the next six or so months, equity markets will likely be range-bound — not making significant new lows (due to death totals being lower than expected) while having upside capped by uncertainty surrounding the seasonality of the virus.
Though our forecasts could change, the current situation is feeling more like 1987 (a panic short term equity market crash followed by a new high in the equity markets within two years) than 1929 (a panic short term equity market crash followed by multiples years of bear markets and an economic depression, while only hitting a new equity market high a generation later).