by bonemaster5000
TLDR: The IHME model is trash. Media will soon notice. Long volatility via calendars on SPX for low risk high reward.
All right boys and girls, listen up. Professor BoneMaster’s class is now in session. I’m a researcher at a university you’ve heard of, and I’ve been following the ‘rona closely since January. I told my students to buy SPY puts in mid-February and that the global economy was going to shut down well into the summer only a couple weeks later. I’m up 300% so far by longing volatility, have been mostly cash for the past couple weeks, but am once again confident enough to be back in. As I’ve now convinced you of my predictive ability, let’s get to it.
Thesis: The IHME model touted by Trump is bad. It is considerably under-shooting projected death counts. This becoming apparent is a very soon eventuality, which will be the catalyst for the next leg down.
I’m going to try and keep this as simple as possible for the retards. The IHME model works as follows:
- Assumes cumulative deaths over time follows a curve like this, but normalized for only positive x,y.
- Fits parameters that control the shape of the curve for a given location using existing data. Specifically, it fits for the time where rate of change is maximum and the maximum rate itself more-or-less. It then predicts a maximum value based off of that.
So why is this approach retarded?
- Assuming a Gaussian-like curve for daily deaths and therefore a sigmoid-like curve for cumulative deaths makes sense in the context of a lot of basic epidemiological (SIR-like) models. We also saw curves like this in Wuhan. We are NOT seeing this in Italy or Spain. See daily infections and deaths for Spain here and Italy here. What we are seeing is a leveling off of to a plateau/low angle slope. More later.
- The US models are fitted using data from Wuhan, Italy, and Spain. We are NOT following social distancing as closely as these countries. As others have posted in this sub, see NY vs Italy vs Spain. Americans are stupid and selfish. We aren’t social distancing nearly as well as other countries.
What does this mean?
- For Italy and Spain, the IHME estimates are getting more and more desperate. I’ve been watching the realized daily deaths blow outside their confidence bounds for the last few updates. For the most recent model from Wednesday, Italy just blew up the model’s UPPER-BOUND today (570 realized vs. 522 max). Spain is likely going to do it tomorrow or Sunday at the latest. Needless to say, a predictive model that consistently fails to predict two days out within the 95% confidence interval IS NOT A GOOD MODEL.
- The prediction for the U.S. is going to get double f*cked. It suffers from the invalid assumption of a nice Gaussian-like curve, plus it’s essentially fit using data that paints a much rosier picture than reality.
Deeper dive:
The model inherently assumes a notable inflection point. This occurred in Wuhan as the strict lockdown essentially killed almost all spread. The looser restrictions within Italy and Spain have kept the disease reproductive number sufficiently high enough to maintain something along the line of a plateau of daily new infections. Neither country is at the point where herd immunity has any effect, so this might be maintained for some amount of time. This could actually be considered the desired effect of flattening the curve, meaning that the restrictions are enough that the disease is under control but not so strict that you’re welding people in their houses.
Given sufficient time, this will result in herd immunity. However, at the current rate, that would take about 25 years for Italy (2.5 years if rate is 10x under-reported, as thought). If this is in fact something of a steady-state, maintaining it also requires that restrictions are also essentially maintained as-is until vaccinations are possible. If the rate is actually plateaued, then even a slight loosening of restrictions will result in an immediate second wave.
What I think is actually happening is something along the lines of a self-limiting population-wide response. With bad news people cower in their homes, but with good news (we passed the peak!) people start going back out again. Good/bad news creates a feedback cycle resulting in a steady-state rate of infections that’s just low enough to not overwhelm the hospitals. I call this response the “BoneMaster Effect”.
I think Fauci et al. is well aware of all of this but waiting to break the news. At least, I would certainly hope so. The IHME models for the US have been relatively accurate, as the curve is absolutely a good fit for initial->peak infections. But as we’ve only just started to hit the peak in certain areas, the BoneMaster Effect hasn’t come into play yet. We have another week or so until IHME says NY infections will reduce by half (covid19.healthdata.org/united-states-of-america/new-york), though their confidence interval is absurd. If by next Friday NY is still getting O(600-700) deaths per day, then people are going to take notice.
Summary of other models:
The CovidActNow model is equally retarded, but in the other direction. It assumes way too high of a reproductive number with social distancing. The only good model with a fancy website I’ve found is the Los Alamos one. I’ve been keeping an eye on it, and they’ve generally been spot on. Basically, they don’t make any explicit assumptions about disease growth rate except that it will trend down, then randomly samples over all possibilities of how the growth rate can change over time. This inherently captures that notion of plateau/low angle decreasing slope. More importantly, take a look at the ‘team’ leading that model and compare them to the original ‘team’ for CovidActNow vs. IHME. Notice anything? CovidActNow was started by a bunch of business, IHME is like 90% attractive bio girls, but, by far, the Los Alamos team is the biggest group of nerds out there. Always bet on the nerds.
Uppies or downies:
Neither. Long volatility. Calendar spreads below present value are the way to go along with a debit call spread for capping “brrrr” risk. Wait until you see the first article out hinting that Trump’s model is completely f*cked.