Monday assorted links

1. Photos from Belarus, interesting in their own right but all the more so now.

2. More comments on the models.

3. The lockdown culture that is Ontario: “19-year-old charged after Mercedes clocked doing 308 km/h.”

4. Two million chickens to be killed because there aren’t enough workers to kill them.

5. Covid-19 has largely spared the baseball world (model this).

6. An argument that all will be well soon enough.  Not my view, but happy to pass along this perspective from Lars Christensen.


Not your view, or not your hope?

2. Who is this person and why should I trust them over Microsoft or John Carmack? A lot of unbacked assertions sent out 140 characters at a time. can read the source code.

What a joke man. We have a golden primary source (the literal source code, or at least a cleaned-up derivative of it) and you're clear that you won't look at it and will only accept interpretation of it if it aligns with your priors.

"Oh it says the model is bad? must be a biased source with no expertise in the matter"

Why bother reading anything when you've decided to stop thinking?

>or at least a cleaned-up derivative of it)

We have a cleaned-up derivative, months after that derivative ceases to matter. The lesson to take from this is everyone who doesn't publish source code and underlying data isn't doing real science, and their work product shouldn't be taken seriously by policy makers* "Science" requires replication and review, and you can't get that by merely studying output.

*whether you want to take that risk in any particular situation is up to you; it's your skin.

This tweet shatters her whole argument, which is an appeal to ignorance.

Eh, an Appeal to authority shatters her whole argument. Maybe if her argument was an appeal to ignorance that would sort of be true. However, she lists multiple specific reasons why she objects to the code.

Just a few of which are:
* It contains a 450 wide parameter base including hotel guests and shop closures and digital contact apps. But does not even consider care homes.

* Yes thats right, the seed numbers don't seem to generate deterministic outputs when running in a multi-threaded mode - but do in a single threaded mode.

* The code itself contains many special rules held in imperative C++. There is no explaination as to why these are present. For example, why are hotel's excluded in the "place sweep"? Who knows, but they are.

* This model does not even produce deterministic outputs for a given seed.

(This is a pretty damning comment, if true. If it's significantly true, then this code isn't even a model. It's just a kludge.)

Those are hardly appeals to ignorance.

The most damning part is the 450 parameters, with no documentation for setting them to any particular value.

I mildly disagree. Yes, 450, ill-defined parameters is bad. But think about it, if the code doesn't generate reliable output for a given input, with even 1 input, it's fundamentally broken.

Adding 450 parameters is just dumping 5 gallons of icing on a bad cake.

As someone else noted below, in multi-threaded situations that can happen due to race conditions and tracking it down is difficult. It also might be only affecting it at a trivial order of precision. I literally spent days one time tracking down a non-determinism at the 9th decimal place of precision. It happens. It might only be affecting one of the smallest and most unimportant parts of your system.

> Recall that many of these matrices of numbers will be derived from real datasets and their derivation will reside in published papers,

Ironically that argument is the real appeal to ignorance, suggesting we assume the values are good and rigorously derived because we don't have evidence to the contrary.

Can you find the published papers that show the derivation?

No because there is little to no documentation.

Where is the burden of proof?

It was a blatant appeal to authority.

And some people listened.

We live in a strange world.

Where is Microsoft volunteer's endorsement? Refactoring is not an endorsement.

Were is Carmack's endorsement? Carmack seemed to be implying career programmers also face tough challenges. That's not the same as an endorsement.

#2 Again, why does anyone expect much else? Most software is awful (this from an experienced software dev), and even good models only point a view at a given point in time with a given set of assumptions.

Someone paid too much attention to the map and drove into the pond that wasn't on it. Map != territory.

2/ "The majority of these are not based on any ground truth data that I can see and are just..."magic numbers" presented in the model. Any one of them can change the outputs of the model in unpredictable ways."

Yes, a basic class in dynamical systems, chaos theory, or partial differential equations would teach you all about that. You are modeling a chaotic system.

"The model is essentially a giant, complex state machine."

All computer programs are complex state machines.

"All this wouldn't matter if it delivered reasonable results. "

That's not how models work.

Overall, a fairly weak biased takedown. No wonder this Twitter handle has no blue checkmark unlike say Carmack. The irony of ironies is this person complains about aggressive responses from others but writes loaded comments like:

"Here are just a few of the free parameters you can fiddle with in order to "make up" a result which suits your narrative."

Not in good faith at all.

"All computer programs are complex state machines."

No, that's certainly not true. State machines explicitly require memory and have an ongoing state. Their output changes over time based upon their current state. However, much (most) software is designed to always initialize at a default state and then respond to input. Such software should always return the same output for a given input.

The criticism of the model is that it should return the same output for a given input, but that multiple people have noticed that under certain, rather broad conditions it doesn't.

" This model does not even produce deterministic outputs for a given seed."

To model a stochastic system, I would expect standard software practices to apply. So you would expect, that given the same seed input, the same output is always recorded. (This is a basic check that the software actually works as expected). Then to use the model stochastically, you input random numbers.

To be fair here, I was really talking about the common definition of state machines, which is machines that when you turn on, have some kind of internal stored memory and always start from that last state, as opposed to a more normal software state of re-intializing every time it is "turned-on" and using the current conditions (sensors, etc) from the real world to proceed from.

For example, a normal piece of code might re-initialize when you start your car and determine the correct fuel air mixture based upon sensor data. Whereas, a state machine would respond to it's last running state and if there was a change to the machine while it was off the state machine would have to be adjusted. The state machine would always assume that the last state it was in was the correct state, but if something had been changed, Cam position for example, it won't operate correctly and must be adjusted.

I agree every program is a state machine in the space of 2 to the power of the number of bits of volatile memory. It's just jargon and nobody cares. What she described I don't think most people would say is a state machine because it's real world parameters like "proportion of hotel attendees who are local". If the table replaces code I would call it a state machine. Parameters do not replace code; you still have to write code.

With four parameters I can fit an elephant, and with five I can make him wiggle his trunk said Mr. Von Neumann
With 450 parameters set by the modeler any resemblance to the real world is purely coincidental.

Agree. You can say what you want about non-determinism. There's a bug in the source, it mgiht be affecting a trivial level of precision. But having 450 parameters to play with and no justification for why any of them are set that way is just an elaborate way of making shit up.

>Most software is awful

Most software isn't the basis of multi-trillion dollar bets.

1. Well NO software should be the basis of trillion dollar bets
2. There IS a lot of software being used to do trillion dollar bets - Wall St runs on software, Im sure academic softwares-based models are used to make economic policy decisions and are just as bad
3. The whole stink coming from this website about this is rank hypocrisy. Tyler has long been a strong advocate of ever more software making ever more important decisions in our lives and our society. He mocks “anti-tech” people as irrelevant. He cheerleads ever ML algorithm and product. Oh a lot of software is actually dog crap? Gee Wiz!!! So this whole thing from Tyler now about “bad software” is insanely cynical and hypocritical and if you fall for this act from him you just have a baby brain. What can I tell you?

Can we appreciate that in essence 2 contradicts 1 here?

Turing completeness means that 1 implies that it's impossible to make 'trillion dollar bets' (I assume we mean decisions) whether by machine or human reasoning. Also what's the cuttoff? $999 billion? Seems a bit arbitrary because it is.

Speaking of baby brains, maybe you should check out what hypocrisy is. Pretty sure you can be a proponent of X, and an opponent of y ∈ X without "pretending [to have] virtues, principles, or beliefs that one in fact does not have"

The stock market is run on software and is full of multi trillion dollar bets. Google, Microsoft, Amazon, and Apple are full of software and are worth a trillion each.

Most software isn't the basis of trillion dollar bets but you can't place a trillion dollar bet without software.

>The stock market is run on software and is full of multi trillion dollar bets.

Sure, and it's written to a higher standard than your random iphone app.

And no software is going to be built to that standard "just because" - it is always going to be built based on the economic stakes of the moment.

This model was built in academia over a decade or two with the goal of publishing papers.

No one came and said, hey we are going to make a several-trillion-dollar bet, here's a few billion and a couple years, built me something!!

Instead, the need to make this bet came up "out of nowhere" and they just grabbed whatever software happened to already exist. Now we're all looking at the software that just happened to be grabbed, and judging it by the standards as if it was actually planned and built to the purpose we've assigned it ex post.

Yeah, very stupid and very cynical of Tyler to try to make this an issue. But it just shows how boneheaded his readership is.

The argument seems to be that:
- 450 parameters are too many [449 would be OK?]
- the parameter inputs are not well documented [surprise, surprise]
- the random number generator is not deterministic in multi-thread mode [actually a serious issue, but less important in Monte Carlo use]

I have read code for determining gross takeoff and landing weight of commercial jet aircraft. For a given class of aircraft, there were from 50-200 parameters, some straightforward, many based on empirical testing and negotiations with the manufacturer, the manufacturer's re-insurance company, the pilot's union, the airline's re-insurer and the FAA. There was a good supply of ad hoc code that was inserted for fudging things here and there. It was in an ancient FORTRAN dialect, and I was lucky in that I had access to a coding old timer who remembered many of the crashes or close calls that prompted the additions. The flight planning system was much uglier and had its own stack of parameters of similar parentage.

Just counting the parameters and calling out a few for ridicule is what they usually call "a cheap shot". The lack of any parameters explicitly for old age care homes might be a problem, or might be because the model uses some other proxy parameter. It would be nice to know. Yes, 450 parameters is a lot of parameters, but just counting them isn't a useful critique.

Usually, I'm a big one for being able to recreate the random number sequence in simulations. It can be critical for debugging, but that's what single thread mode is for, and doing it right multi-thread can be a pain. I gather this model is used for Monte Carlo analysis, so it isn't clear this is a problem in production.

It is quite possible this code is a steaming heap, but that designation requires more than innuendo.

She explicitly says "The majority of these are not based on any ground truth data that I can see "

Also I doubt that a commercial jet aircraft has quite as much chaos involved as an epidemiological model. The vast majority of those parameters are pretty fixed, and pretty well documented. They aren't just left free for the modeler to play with. And of course, we're not dealing with a well understood 50 year old architecture, but a novel virus for which non of the input parameters are well established. At thins point , nobody has the kind of detailed knowledge for most of those 450 parameters to do anything other than guess. You should be using models that detailed until you understand the system you are working on in that much detail that you can fix the values of those parameters, and only be changing a few.

Should not.

Or well, maybe you could get crazy and have a genetic algorithm try to fit the parameters against real-world empirical data. That would be fun.

This ain’t no Monte Carlo.

They were running with given seed inputs, getting deaths off by several tens of thousands with the same seed, averaging the garbage and publishing it as the Gospel.

Ever wonder why we don’t have a probability density function?

Because it’s pure bullshit and apparently 90% of our overlords are innumerate

Japan continues to do very well. Claims about how poorly it was really doing look steadily more inaccurate. Not only does it have a much lower case and fatality rate than Germany, it even has a per capita fatality rate down with South Korea.

A lot of people got way too sold into the idea of under-reporting for the Olympics, but even if you look at mortality data, Japan's pretty well off. (And I believe that masks are indeed a big deal.)

Masks probably help and a survey found 67% of Japanese were wearing them in February. Still, I think almost never shaking hands and never kissing strangers when greeting is probably a greater factor in low deaths. With only 600 deaths out of 126 million people it seems that there's also something else going on.

Again, something happened with the Swine Flu , H1N1, in 2009/10 in Japan. There were about 100 deaths in Japan and 12,500 deaths in the U.S. On a per capita basis, the U.S. had 48 times as many H1N1 deaths as Japan and now there are 50 times as many H1N1 deaths in the U.S. as Japan.

I can't recall the last time I heard Tyler say anything nice about Japan, maybe due to their stubborn unwillingness to go along with the international playbook.

Tyler promotes "free" (ok really highly managed ) trade, broad diversity and open borders. Japan explicitly rejects all of these and still does very well.

Japan was in the WTO and on board with TPP whereas Krugman was against it. Japan is also as open as the U.S. in terms of (imports + exports)/GDP. Diversity has increased a lot in Japan since 1992 when I first lived there even though you wouldn't mistake it for California. Borders...

Chickens: they have historically been the political victim, in trade agreements and sanctions and now pandemics, because they are so vulnerable and yet so delicious! “We were concerned that the chickens were still being talked about as commodities...They are the victims.” Poultry is not paltry. Save the chickens.

+1 for “Poultry is not paltry.”

1 dead chicken is a tragedy, a million dead chickens is a successful restaurant franchise!

Delicious, deep fried, juicy, cordon-bleu-baked, misfortune indeed.

8. The lockdown culture that is suburban Detroit includes drag racing:

The culture of drag racing in suburban Detroit has always been there - the lockdown has just provided more open streets on which to exercise it.

5. Outdoors

3. Why is this lockdown based? Idiot teens have always liked to speed with their parent’s cars.

The lockdown means fewer other drivers to get in the way.
308 is fast.

Mercedes should consider it Free Advertising. 191 MPH is pretty darn fast for a mass production vehicle.

308 KPH in a Mercedes C63 AMG. I was skeptical at first but a quick search shows examples of 300++ top speed in this model. This was not street racing and I'm surprised that the young man pulled over. I'm a little concerned about a police officer taking to social media to moralize about high speeds on a controlled access highway. It is a controversial topic but let's not forget that a great deal of data exists from Germany and elsewhere that shows that speed deviation kills, not speed itself. Your mileage may vary.

I know of no roads which have been designed or maintained so that 300 kph is a speed at which control of the vehicle should be expected. I leave it to the engineers at Mercedes to opine about whether that car could be 'in control' at 300 kph under track conditions.

> shows that speed deviation kills, not speed itself

I'd say a speed deviation of 160-200 km/h is pretty deadly. These highways are *less* occupied, not deserted.

Speed deviation increases risk but I’m not sure how “deadly” it is compared to other risky behaviors such as driving while sleep deprived or texting while driving. Germany has taken a different approach to mitigate the risk but each measure taken, such as minimum speed limits, is based on speed deviation rather than absolute speed. Decreasing the speed limit saves lives. So does increasing the speed limit. In both cases the underlying mechanism is speed deviation and the effects are temporary until behavior naturally adjusts to the new norm.

Have you been on the QEW in Toronto? Unless he was driving at 3am, lockdown provided the absence of traffic necessary to hit those speeds.

I looked at the live traffic cameras and the QEW around Burlington looks empty at 5pm rush hour on a Monday. The CBC article shows the car on a tow truck at night.

#6 - I hope this guy doesn't manage money for a living. He seems to be living in an alternate universe. We will be luck if schools including colleges and universities open on time in the fall. I wonder how the GMU econ department are modeling this? Will classes resume? Will they be in person or by distance learning? Will there be a faculty cutback if enrollment drops? Inquiring minds want to know.

Schools and colleges are opening in Europe right now. Why would we be so far behind, especially if we are talking about some states?

Colleges open in May for the regular academic year? University of Chicago's Autumn quarter is/was scheduled to start September 29th. (Yes, quarter system, but still.)

I've only seen notices of grade school openings. Certainly no college openings in the major Euro countries. Do you have a citation?

6. I want to believe, but...

> "The markets is telling us so" [sic]

The stock market right now is like a general fighting the previous war. It's anticipating a splendid 2009-style boom but there's no guarantee. Back then QE and near-zero interest rates were a revolutionary shock to the system; this time around you can try cranking the firehose volume up to eleven but it was already at ten. The impact won't be the same.

> "Most layoffs are temporary"

Jobs with pre-existing comorbidities aren't coming back. A lot of brick-and-mortar retail was on its last legs anyway and Covid has now pushed it over the edge. Costco and Amazon have made permanent gains.

Restaurants also have a high infant mortality rate, by the nature of the industry. They bleed money even in good times. They're born with congenital comorbidity. How many will reopen? Maybe eventually, but not right now. Not with the very real short-term risk that you might be forced to shut down all over again.

So what would have been a gradual adjustment that played out over several years in those sectors is now happening all at once. That alone suggests that six months until November won't be nearly enough to bring unemployment back to normal levels.

> Consumption will rebound sharply – the money is there

People who never dreamed that their world could collapse overnight will now be extra cautious. All the previous recessions took maybe half a year or more to set in, not sudden onset like this. A lot of people will now find religion about saving up a three-month emergency fund, and others who were forced to downsize their consumption may find that it actually suits them.

History tells us that the American consumer just won't quit, and memories are short. So consumption will indeed rebound in a few years, but not in the next six months.

TL;DR: good things will hopefully happen, but not by November. His time frame is way too short.

6. One-time supply shocks shouldn’t affect unemployment in the long-term, but this virus could easily last for a long while and affect people’s permanent incomes. I don’t think we’re all going to go back to living off the land, but it’s pretty reasonable to expect that a large number of people are not going to go out as much as they used or that some legal restrictions such as on some forms of international travel will continue until there is a vaccine. This sets the virus apart from something like September 11 which was a one-time shock, and therefore raises the possibility that the economy will remain depressed for quite some time even accepting that this is an example of a Real Business Cycle and not a typical recession.

5. Interesting juxtaposition: baseball players haven't been much affected by covid 19, so baseball players are susceptible to covid 19. Well, this reminds me of the many comments at this blog: covid 19 has not affected most (red) areas in America, so those areas should be allowed to ignore covid 19 and go back to work. What about the baseball players?

I think he is basically correct. Restaurants will reopen with some restrictions, perhaps plexiglas partitions, travelers will fly again ( with masks). After 9/11 we got a lot of security theater , here we will get a lot of pandemic health theater ( temperature checks, etc..) Some sectors will be challenged. If people work more from home, commercial real estate might be in over supply. Demand is still there just pent up.
I notice it myself. I wanted to buy a Nintendo Switch, but it’s $599 currently instead of $299 because of supply shortage, so I’ll wait until the supply improves.
The stock market overreacts short term but generally makes the correct call longer term.
Investors are not fleeing to safety.

Re: 1. All socialists miss Stalin.

Name one.

Actually, I can think of a few socialists who missed Stalin over the years. One of them did manage to hit Lenin in the neck and shoulder, but he survived.

Anyway, we are about to get a lot more data: economies are reopening (Georgia), some never really closed (Florida), lots of people who were anti-social-distancing-and masks went and put their money where their mouths are (Michigan, Colorado).

Either a lot of them are going to get covid right-soon, and everything is going to close up again; or they are not, and more is going to reopen over the summer.

My priors are that it is too early to open up and there are about to be a lot of red-county outbreaks, but there should be a lot of revision through May and into June, which will help me decide if I am going Out West for summer vacation, or buying a freezer full of beef and installing an above-ground pool.

I disagree with your second paragraph. A lot of people will get Covid-19, in place where Covid 19 has not been so harsh yet (red states, China, etc.) and hopefully, nothing is going to close again.

Another prior to watch. Many of those red states have larger, rural dispersed populations.

On #2: they lost me at 450 parameters. Any predictive model of with 450 parameters, especially of what primarily seem to be aggregates, is pretty much automatically BS.

It's like an integrated circuit simulator with 30 types of multi-layer transistors, each with 15 parameters.

It's like a protein folding simulator with 25 parameters per amino acid.

Just try simulating automobile assembly with ONLY 450 parameters.

450 is not a daunting number. This is a simulator, not a curve fitter. It may be 100% bogus, but 450 parameters is not a good argument. For all I've gathered, it might be overly simplified, and a good simulator would have 4,500 parameters, many with even weird names and produced by a broad variety of analyses.

Back in the days of the four year business cycle I had a forecasting rule of thumb that worked amazingly well. The recovery phase of the cycle--from the bottom back to the previous peak -- took the same amount of time as the decline. So if the economy fell for six months it took six months to get back to the previous peak --12 months=12months,etc.. Even in the depression when it took 3 years to fall from the peak to the 1933 bottom it also took 3 years to return to the previous peak just before the 1937 recession. Of course this gave FDR a normal recovery that a lot of people are not willing to concede.

Right now we are still waiting for the bottom, so it is still too early to try and apply this rule to the current recession. but it could make Lars Christensen look very good.

Of course back in the 1950s-1960s most of the people laid off were employed in manufacturing durable goods -- like autos -- and fully expected to go back to their old job in a few months. So they did not go hunting for new jobs. Because of this the BLS changed the question the asked to include the part about looking for work to be counted as unemployed. So these people fueled the early stages of the recovery. So maybe the key point is do the waiters, retail clerks, etc.,
feel this way and will they be recalled to their old jobs.

zizeck could be a hoax/ist

2. An okay criticism, but perhaps not recognizing its own real argument.

What it is really saying is that this really complex pandemic simulator cannot be suddenly used as a single answer policy tool.

You actually use it if you want to sift through 450 variables to see what it's most important. Not to suddenly capture the regulatory and social environment of say Borneo, and predict their death toll next week.

No one is going to capture the environment of a country or even a city when it really does take hundreds of parameters to represent it.

How much does the average person in Huntington Beach fear the beach? How much does the average person in Huntington Beach fear the market? Etc. Etc.

Huntington Beach chosen for their place in the news.

> On #2: they lost me at 450 parameters. Any predictive model of with 450 parameters, especially of what primarily seem to be aggregates, is pretty much automatically BS.

Yes indeed. Give any model enough knobs and the biases become an integral part of the model. How likely are you to catch covid when ordering from a menu? I dunno, if the guy that previously used the menu had covid, then I'd guess 0.1%. But that could readily be 1% or 10%. Nobody knows. And the modeler is just guessing.

Researchers at Utrecht University, Erasmus Medical Center and Harbour BioMed (HBM) today reported that they have identified a fully human monoclonal antibody that prevents the SARS-CoV-2 (COVID-19) virus from infecting cultured cells. The discovery, published online today in Nature Communications, is an initial step towards developing a fully human antibody to treat or prevent the respiratory disease COVID-19 caused by the novel coronavirus SARS-CoV-2.

Not a vaccine, more of a seasonal antibody booster. Monoclonal means it is a heavy chain single B cell configuration. A polyclonal would require two or more antibody B cell to work in tandem ( my short cut in thinking about this.)

That could be the best news we've had in a week.

Let's hope it:

a.) Works well in real life.
b.) Production can be scaled up rapidly.

Assuming it can protect and treat, how many times a prophylactic dose is a person dying from GOVID-19 going to require? The sick could suck up a lot of this.

"Sue Denim" has followed up their original criticism of the ICL/Ferguson model with more information. (Apologies for reposting the link - it makes more sense to attach it to today's assorted links!)

Taking the average of garbage still yields garbage.

Hilarious that they assumed the errors would average out. There’s a lot of assumptions going into that, none of which would be defensible

That's every engineering design program ever. Be afraid. Be very afraid.

It's also how voting works.

1) I love Victory Day in the former Soviet countries. It's a bit like our 4th of July but a much bigger deal. The insane sacrifice of the Soviet populace during WWII is hard to fathom--over 25 million deaths.

I lived over there, close to 15 years ago now. Veterans would wear their medals on their jackets year round; it was quite normal and I assume still is for the remaining guys. Strangers would thank them for their service on the street, they'd be introduced as veterans of the Great Patriotic War at weddings and pretty much any big event. The guys would tell you about how the Soviets saved the world from Nazis, and who can argue? The U.S. saved the world as well, but with not nearly the level of sacrifice. It's all a bit of nostalgia mixed with patriotism searching for something to be proud of 70+ years on.

I love the big hats. Iranian and N. Korean militaries also go for the big hats. Meaningful trend?

6. An argument that all will be well soon enough.

However, where most commentators are wrong is assuming that this has to be seen as a normal recession. I on the other hand would argue that this has little to do with a normal recession. In fact I am increasingly thinking that the use of the term ‘recession’ is a misnomer in relation to this crisis.

"This time is different', a common mistake.
Not so, even the Fed regime change will not be much different than the last two times. We have the same revenue sharing problems as the last two times, and the same pricing of intergenerational change. The same informational technology shock. It is just our time, and we get better at it on each trial.

6. Enough Americans are dead and going to die from COVID-19 that economic forecasters should be predicting the effects of the reduced demand for housing stock and the effect of a very large number of people receiving inheritances within a relatively short period.


Wot? You don't think the perhaps 8% decrease in expected population growth and 4% increase in mortality 100,000 US dead represents isn't worth paying attention to? Dunno how it works in the United States, but inheritances here tend to be treated like windfalls and increase spending immediately and for several years afterwards.

Note the particularly high death toll in New York State which is the state with the second highest per capita GDP and currently has maybe 1 in 700 dead from COVID-19.

You are right about US population stagnation, but for the wrong reasons. CV-19 is accelerating some deaths for people who were very likely to die in the next decade anyway, so over a 10 year period it isn’t doing anything to increase mortality.

But immigration was already cratering, and the pandemic has given Trump an excuse to even further restrict it. Not to mention that anyone with any choice would rather choose Canada or Australia now anyway. The pandemic will also lead a lot of people to reconsider having kids, or having another. The high unemployment rate will also nudge people not to procreate or get married. The high unemployment rate also means that Democrats will pay too high of a political cost to try to get more immigrants into the country.

So you have way fewer immigrants and way fewer babies, when you already had historically low levels of both before the pandemic. And even if they don’t die now from CV-19, boomers just keep getting older so the number of them that die in a given year will just keep going up and up, and they are the second largest demographic cohort in the USA. The opioid epidemic will likely be exacerbated by the pandemic as well, so I expect to see declining life expectancy as well.

My prediction is that the Canadians will now be viewed as the leading light of the Anglophone world.

"My prediction is that the Canadians will now be viewed as the leading light of the Anglophone world."

Why is that? Australia and NZ are far away, but they have done much better than Canada with the virus. They are also showing some backbone against China (NZ still needs to do better). Canada is utterly supine to China and does not even defend its citizens. Meanwhile the UK is a mess.

Yeah, but what is called "showing some backbone to China" is actually Australia caving to the US. I would have preferred staying out of it. It's obvious China is no longer going to be the zoonosis threat it was and there is no point getting jiggy with it at a time when emotions are raw when China seems certain to make all the necessary hygiene changes required on their own accord.

New Zealand and Australia have very small populations, and are geographically isolated. The US will still be a huge market to sell things to, as will be Europe, Latin America, and the Pacific rim. Canada is much better placed to sell things to nations in those regions. Plus, they are much more likely to see the pandemic as an opportunity to grab more and better immigrants. New Zealand and Australia seem much more nativist.

Australia imports more people as a percentage of population than Canada. (Numbers do go up and down a bit though.) Australia and NZ are well ahead of Canada on percentage of population born overseas at 30% vs Canada's 21%.

Sure, but how many of those folks are from the British Isles? My thinking is that Canadians are much more durably committed to immigration than Australia or New Zealand. The impression I have gotten is that Australians are quite eager to cut way back on immigration, but maybe I am wrong about this.

Over the last 20 years, China and India have been the main source of immigrants to NZ. Lots of Filipinos and Koreans too. Although overall the UK still leads due to migration over the preceding 50 years. There are significant populations of other Europeans too, especially Dutch and Germans.

Interestingly, the Indians mostly seem to be Sikhs. Singh is the most common name on North Island.

About three and a half times the number in Canada. It's so bad the whole state of South Australia gave up on having it's own accent and just uses comprehensible British.

3. The most devastating fatal crash ever? Um, dead is dead... not seeing why a fatal crash at 191 mph is more devastating than one at a mere 70.

3. I have to update my priors on Canadians. Didn’t think any of them had it in them to do something so stupid, reckless, and awesome. Perhaps the individuals in question have a future in Silicon Valley or Wall Street.

Friend of mine in LA says that driving on the highways during lockdown is like playing a level of Grand Theft Auto. I can totally understand car affectionados going crazy when the highways are eerily cleared of almost all other traffic.

I'm somewhat sympathetic to Kayla's code analysis, but looking through the rest of her feed ... yeah I think I'm gonna trust microsoft over some lady with an agenda. And I'm sympathetic to her worldview too! It's just that you can't expect reasoned analysis from cranks, either to the left of the right.

As a counterpoint, I think John Cochrane's behavioral Covid-19 models are well thought out and reasoned, because although he has a certain worldview, we all do, and I trust that and reading his model it appears that he approached it objectively. This lady didn't.

Couple of points: Non-determinism is inevitable for a model this big on a multi-threaded system. Race conditions are notoriously difficult to track down. This is an academic model, not developed by insurance companies with a profit incentive attached.

It probably should not have reached the prominence it did, but seeing as insurance companies didnt create models with a profit motive assignments ... well it will do.

Regarding the number of parameters ... that is somewhat problematic. But I'll cite Milton Friedman here ... it doesn't really matter, so long as the model makes accurate predictions. It doesn't matter if it's just a long stupid exercise in curve fitting if future curves fit.

If this were a modern machine learning algorithm with real time data, as would probably be better and she appears to suggest, you would have the exact same problem. But worse, you wouldn't even necessarily know what the parameters even refer to. They just exist, and ML fit the curve. Those algorithms run everything nowadays, and are predicated on the idea that you shouldn't necessarily know or need to know everything about a system to model it.

I would argue the main problem is a matter of complexity. You don't need separate parameters for time people stay at hotel, etc. a simple population density will do and add a separate parameter for interactions among a population. Doesn't really matter how those interactions occur.

What probably happened was over some, some academic said, hey why don't we account for this, added a parameter, and wrote a paper. Repeat. Then you get hundreds of parameters. But again, most of them are irrelevant, created by some academic who decided to look into it. Their existence is not a sign of malicious intent or a dastardly plan to be incorrect, it's just a lot of excess that can be safely ignored without affecting the overall picture.

While it is pretty clear that the model was inappropriate and it was certainly inappropriate to use it as significantly in this crisis, I'm not taking marching orders from some random lady with an ace to grind.

The problem with the intercept model is not the code. The problem is, as has been pointed out by numerous economists in this, standard SIR models dont capture how people behave in response to this. They aren't dynamic. That's the issue. The model here is asking you to presume a fixed behavioral pattern at a high specificity (that's what all the parameters are for) and then run. You simply cant do that. That's the issue. Arguments over the number or usage of those parameters fundamentally misses the point, the point is that behavior is treated as a static phenomena when it shouldn't be.

+1. Good overall take.
I doubt there was any malicious intent, the problem is that if you have too many parameters to play with, you can play with them until your model does what you think it is supposed to do, and thereby confirm your priors. It's no longer a model at that point, it's an exercise in confirmation bias.

The emergence of influenza virus strains resistant to approved neuraminidase inhibitors and the time constrains after infection when these drugs can be effective constitute major drawbacks for this class of drugs. This highlights a critical need to discover new therapeutic agents that can be used for the treatment of influenza virus-infected patients. The use of broadly neutralizing anti-influenza monoclonal antibodies (MAbs) has been sought as an alternative immunotherapy against influenza infection. Here, we tested in mice previously characterized broadly neutralizing anti-hemagglutinin (HA) stalk MAbs prophylactically and therapeutically using different routes of administration. The efficacy of treatment against an influenza H1N1 pandemic virus challenge was compared between two systemic routes of administration, intraperitoneal (i.p.) and intravenous (i.v.), and two local routes, intranasal (i.n.) and aerosol (a.e.). The dose of MAb required for prophylactic protection was reduced by 10-fold in animals treated locally (i.n. or a.e.) compared with those treated systemically (i.p. or i.v.).


From 2015. Having the thing here, the lungs is where the macrophage go wild, it is the same symptom, different causes. The macrophage are the killer, if we can get a much better anti body in here at the right moments, we can stop, like an inhaler, handy to have around. Good for flu, colds, severe hangover, and everyday aches and pains.

#6: I'm skeptical too. He claims that economies recover quickly from supply shocks, but neglects to give any examples of this actually happening. The three examples of supply shocks that he does have -- the rapid increase in oil prices or disruptions in oil supply -- all created conventional recessions (or stagflations) that as usual took many months to recover from.

If the businesses can magically re-open and re-hire all their laid-off workers, great. In a few weeks the economy picks up where it left off.

But I think most of us rightfully expect that a good chunk of the shut-down businesses are either not going to re-open at all, or will re-open with much reduced demand and workforces.

This all relates to #4 as well, the labor shortage. The economy has taken a huge hit, but the lockdowns are only to blame for a fraction of that. With or without a lockdown, the virus was going to cause a worldwide recession: workers don't want to work, consumers don't want to buy, travelers don't want to travel. Oregon has permitted most manufacturers to stay open but several closed anyway, either because they couldn't re-arrange their plants to provide enough physical distancing, or simply because their workers started getting sick. The NBA shut down voluntarily.

To be sure, government orders put further restrictions on the economy but my raw guess is that more than half of the lost GDP and unemployment would've happened anyway.

#6 Inflation expectations have NOT stabilized. TIPS breakeven rate for 5 years is less than 0.75% pa. Even year 6-10 is only 1.5%. Tell me that signals markets expecting pre-pandemic NGDP growth. While it is possible that employment will be rising and real GDP recovering by November, that will depend very much on whether the Fed gets NGDP expectations back on their pre-pandemic track.

Comments for this post are closed