The world is changing fast and to keep up you need local knowledge with global context.
A study by the South African Covid-19 Modelling Consortium has predicted that the Covid-19 pandemic may peak between mid-July and late August with around 1 million cases. It also said that between June and November this year, between 40-000 and 45,000 people could die in South Africa from Covid-19. Throughout the pandemic countries have relied on mathematical models like these to make decisions on whether they should lock their countries down to prevent the worst case scenarios that these models presented. But how accurate are they and what are the consequences for countries that act upon them? In the United Kingdom, the government relied on the model by the virus modeller from Imperial College, Professor Neil Ferguson who predicted on the 16th of March this year that 500,000 people could die from the pandemic in the UK without significant action. It prompted a lockdown that has only been lifted partially in the last two weeks. But experts from all over the world have derided the coding used by Prof. Ferguson with one of them describing it as a “buggy mess that looks more like a bowl of angel hair pasta than a finely tuned piece of programming.” A British data technology company said in the commercial world they would fire anyone for developing code like this. Critical care expert, Dr Guy Richards told Biznews that the predictions of 1 million cases would mean that South Africa would have a fifth of the current world population infections and he thinks that is very unlikely. He said he did not want to discount the model as “they are very clever medical people” but he does not think they have taken into account whether wearing masks will decrease the transmission rate. Dwaine van Vuuren from RecessionALERT has taken a close look at the model of the researchers and has come to the conclusion that their forecast is likely to be overblown and he suspects and hopes that the model is “as way off as the models have proven to be in the rest of the world”. – Linda van Tilburg
SA Covid-19 local forecasts likely overblown
By Dwaine van Vuuren*
I have looked at the pre-lockdown and post-lockdown forecasts painted in the media for SA Coronavirus cases and deaths, that the government is basing important life changing decisions on.
The latest figures touted by the South African Covid-19 Modelling Consortium are over one million cases and 40,000 deaths by the time South Africa is at the peak of the country’s infection curve. Peak infection is expected to occur between early-July and mid-August (1-1.2 million cases)
Now epidemiological forecasting is an art and not an exact science, and while I can appreciate the math behind the sophisticated models and the integrity of those building them, the only thing we know for an absolute fact is that all of these models that were used by governments to make Coronavirus lockdown decisions were all way off the mark when the numbers finally came in. In the UK, severe flaws were found in the coding of the model used and in the US, the forecasts that jolted the country into panic were really way off as shown below:
The problem I have with these latest forecasts is that they would truly put South Africa in outlier territory regarding cases and deaths per million of population. Like two to three standard deviations. That worries me, as there is no reason to believe why South Africa would be that far out, above or below, the rest of the world when it comes to infection rates.
With this in mind I decided to examine the statistics of those countries with a material count of infections, where a daily peak in infections had already been convincingly achieved. We could then look at the cases and deaths per million of population of those countries to get a feel for what the upper and lower bounds for post-peak South Africa are likely to be. The theory is that South Africa is still going into its peak, with estimates that we will peak around September 2020, so it made sense to me to look at the per million statistics of those counties that had already peaked. In this way, instead of making assumptions about the future in our forecasts, we rather examine past history of a decent statistical sample of countries that are “ahead in time” with their infection curves.
So here is the sample of post-peak countries we are using to make the comparison. We have excluded China as their numbers are definitely dodgy, as well as Australia, New Zealand, South Korea and Japan which are massive outliers with respect to their populations (in some cases we have already exceeded their figures, so it is pointless using them in any forecasting attempts.)
Now lets use a scatter plot of these countries showing their cases versus deaths per million population:
You can see SA is stuck on the bottom left as we obviously are early on in the infection cycle. SA is going to migrate toward the top right between now and September, with the dotted line (a linear regression of the sample) showing the most likely path. Now our cases could get as bad or worse than Spain or Luxembourg at 6,000-6,500 per million and our deaths could get as bad or worse than Belgium or Spain at 600-800 deaths per million but the reality is that this is unlikely, unless we are happy to think SA is going to be an outlier.
In reality, by September, we are more likely to end up somewhere in that green shaded box which is the surface area within 1 standard deviation of the case and death means respectively. (Think of the shaded area as a 90% confidence boundary.)
Using a SA population count of 59 million as per the last census, we can use the upper and lower bounds of that green box to arrive at the following post-peak estimates:
Even if we land up at the upper-bound, which is assuming we land up at the top-right of that green shaded area, just worse than Ireland, the cases are certainly nowhere near the millions nor the deaths near 40,000. To reach over a million cases, we would have to be positioned at 20,000 on the x-axis and to reach over 40,000 deaths, we would have to be positioned at over 1,475 on the y-axis. Something like this:
Now a lot of variables could come into play to make our situation worst than most, like our prevalence of HIV and TB, poverty levels, poor finances of government, low-grade logistics capability for proper reporting, testing and tracing and ineffectual lockdown effectiveness in townships, but will this really propel us so far ahead of the rest of the world that we become a six-sigma outlier? And given that the other countries have peaked, could they really migrate to a post-peak cluster of dots around the red dot in the above chart, making SA less of an outlier? I don’t think so.
I suspect, and certainly hope, that in the end when we look back at this, we find our models are just as way off as the models have proven to be in the rest of the world.
At any rate, in the coming months, it will certainly be interesting to see how this scatter plot evolves with time.
- Dwaine van Vuuren has a Bachelor of Science Honours degree majoring in mathematics, computer science and statistics and is a full-time trader, investor and quantitative analyst. His passion for numbers and keen research and analytic ability has helped grow RecessionALERT into a company used by hundreds of hedge funds, brokerage firms and financial advisers around the world.
Cyril Ramaphosa: The Audio Biography
Listen to the story of Cyril Ramaphosa's rise to presidential power, narrated by our very own Alec Hogg.