Categories
politicalbetting.com » Blog Archive » Unpicking the presidential election forecasts
politicalbetting.com » Blog Archive » Unpicking the presidential election forecasts
A number of groups publish statistical models, usually updated daily, which attempt to assign probabilities to the possible outcomes of the US presidential election. The best-known is Nate Silver’s fivethirtyeight.com. Others include models from The Economist, Decision Desk HQ, and the New Statesman.
0
This discussion has been closed.
Comments
Wasn't it the last election when 538's star fell? Or was it the mid-terms? I know there was one when I decided Nate wasn't such a smart guy after all ...
The 538 odds on the more extreme outcomes do seem intuitively wrong. Is there something in their model which adds in too much randomness to allow for unpredictable events ?
Of course if the polling moves that will move other models more.
[Various sensible-sounding things that reduce uncertainty like the polls never ever changing for months on end]
> The magnitude of the difference between the polling-based national snapshot and the fundamentals forecast. A wider gap means more uncertainty.
> The standard deviation of the component variables used in the FiveThirtyEight economic index. More economic volatility means more overall uncertainty in the forecast.
> The volume of major news, as measured by the number of full-width New York Times headlines in the past 500 days, with more recent days weighted more heavily. More news means more uncertainty.
https://fivethirtyeight.com/features/our-election-forecast-didnt-say-what-i-thought-it-would/
The fundamentals part is quite Trumpy, especially because of the economy being good, allegedly, sort-of. Then economic volatility must be all over the place. And apparently there's a lot of news going on...
The "shy trumper" argument boils down to arguing that nearly all polls are systemically biased to Biden. I don’t see that myself as it would be less in Internet panels etc.
Another interesting model is the Plural Vote one, which adds in state based Internet search data to published polling. The author is a Democrat so not neutral, but worth a look.
https://www.pluralvote.com/
Intuitively, I have more sympathy with their approach because it seems to me that this is an exceptionally volatile election with 2 deeply flawed and unpredictable candidates for either of whom something could go seriously wrong by November. I don't think this is because a lot of people will change their mind, my guess is that something like 80% of the electorate would not vote for the other candidate no matter what, but whether they will decide that their guy is not worthy of support after all and therefore don't vote.
- people in the US much more invested and interested in shares directly
- a strong correlation between the share indices and the election winner
- a lot of volatility and momentum in the stock market with QE and the potential for a vaccine
- Trump good at linking himself with stock market performance
It is a much better focus for Trump than the culture wars. People have made up their minds on the culture wars already, yes Trumps lot will be motivated but this time so will the anti Trumpers. The few votes that can be shifted will be on the economy.
I think the economist model is simply underestimating the chance of Biden's lead narrowing before November. Given Trump's EC advantage Biden effectively has a 4-5% lead. It's like the opposition party having a 4-5 % lead 8 weeks before a general election.
I think around 30%Trump seems about right at this point, even if the shape of 538's set of outcomes might be off.
So what are the chances of the gap narrowing here? Well, there are still significant set piece events like the debates left, with the risk of Biden doing very badly. The economy will also presumably recover further, although how much credit Trump will get when it'll still be below January levels is unknown. And how Covid-19 develops is also likely to have a role to play, most likely way below peaks but still worse than most countries overall. And while there will be other unpredictable events and news cycles between now and then, there can only be so many in eight weeks.
Finally, on the modelling itself I'd observe that the volatility in polling may have been fitted to 2016, where things ebbed and flowed between significant and very small Clinton leads (see here: https://projects.fivethirtyeight.com/2016-election-forecast/), wheres in 2020 Biden's lead has been much more stable (see here: https://projects.fivethirtyeight.com/polls/president-general/national/). If I were modelling this, I'd include the volatility of polling to date as one of the predictors of future polling volatility, rather than relying on the factors described by edmundintokyo above.
https://www.youtube.com/watch?time_continue=2&v=4WdzMVekUBU&feature=emb_title
https://www.ft.com/content/b3297609-e63b-4161-8287-7ab1179d0c40
How are students not getting a discount on their already ridiculous fees when there is no contact time? And the universities couldn't even be bothered to tell the students this until after they had paid for accommodation.
Yet another example of UK2020 failing our young people.
And TBF there is polling volatility in there too, I only listed the last three things. But I do think it's notable that basically all the pro-Biden indicators are simple and objective, whereas the ones that bring it back towards Trump are quite wild and conjectural, especially when applied to a pandemic scenario that's never happened before.
One related issue to this though is how peaky it looks, particularly tech stocks. It is arguable that the Amazon rise is justified by a permanent shift in the market, but much less so for Apple for example, yet the rise in Apple shares since Feb is astonishing, and looks very bubbly. There was a correction at the end of last week, but a major crash back to values of just six months ago is very possible. Early autumn is stock market crash season, and that could be very bad for Trump.
Supposedly the US economy was healthily growing last year but then the US Budget Deficit was nearly 5% of GDP and a trillion dollars. During so-called growth. That is madness, absolutely fiscally incontinent madness.
Of course the whole market could crash and that would be a nail in the coffin for Trump, but from a presidential betting angle the stock market is the most likely thing that can change the result of the election.
As always with Trump, you can never quite tell if there's some brilliant cunning strategy at work or if it's just a confused old man reacting to what he sees on the telly.
Yes without QE asset values would be a lot less, but there is loads of QE and govts are addicted to it, so asset values are high.
Trump is the most disastrous President for the economy of my lifetime. He is Gordon Brown on steroids. He has pumped primed the economy to the nth degree - with it running a nearly 5% deficit last year before the recession even hit.
We made a big deal recently about the fact that during the peak of the recession our total national debt reached the £2 trillion threshold. That was including furlough and an all-time cumulative total figure - including all borrowing before, during and after the financial crisis. In contrast the US deficit (not debt) last year alone was $1 trillion.
If anyone at all thinks Trump has been good for the economy then not a single lesson has been learnt from the financial crisis. Trump is creating another crisis with his totally irresponsible mismanagement of the economy.
In the last full economic year before the recession hit the the UK's deficit was 1.2% and had fallen every year for a decade. Debt to GDP was falling.
In the last year before the recession hit the USA's deficit was 4.8% and had increased every single year of Trump's term. Debt to GDP was rising.
Increasing the deficit and increasing debt to GDP during a so-called growth period is absolutely unforgivable and irresponsible.
Yes, QE makes a difference but the rise in US exchanges has been much more bubbly than our own, other European and Asian markets, even allowing for the tech heavy nature of the US one. Apple is near double its pre covid valuation, and that looks very peaky to me, as one example among many.
Fair enough about the contact time though. Less teaching should mean lower fees.
Anyone who supports this economic insanity should be thoroughly ashamed of themselves. Deficits are a tax on the future so Trump is one of the highest tax Presidents ever.
Its one thing running a deficit during a recession, or reducing it gradually after one, but he's increased it to 5% before the recession. I shudder to think what it will be this year.
Its probably as safe as Government bond right now.
"In the last normal year for which figures are available – 2018 – Britain spent 0.34 per cent of our GDP on state aid, compared to 1.45 per cent for Germany and 0.79 per cent for France."
There's a pretty solid PB consensus here that Biden is a buy at current odds and I concur. In fact I've gone in quite heavily for me. The down side seems so low.
I don't believe the shy-Trumper theory. Most of those I've met are pretty noisy. Perhaps they've got quieter but it's a non-evidence based theory so file in the round cabinet.
The economy is a different matter. If he can keep it going until election day he does at least have one achievement to crow about, even if it is down to reckless mismanagement.
BLM may be helping him too, but the evidence is equivocal and Biden appears to be playing this one right. Trump's approach seems designed to fire up a base that is already fired up. It doesn't extend it. I can't see it making much difference in November.
That leaves 'the debates'. Experience suggests they are rarely game changers and there is no particularly reason to think they will play well for him. They may even play badly. Biden did ok in the Dem debates. Trump is out of practice and some of his recent live performances have been a bit odd.
Seven points behind in the National polls Trump really needs a game changer. Biden can pretty much run down the clock. That is what I think will happen. A polarised electorate seems to have made its mind up. The game isn't going to change.
Therefore, the historical precedents used to calibrate the model might exaggerate the potential for a change in opinion during the campaign. Normally with this sort of event the paucity of historical data would produce the opposite effect, of a model without sufficient dispersion.
And, anyway, all the models implicitly assume that all the votes will be counted. How confident of that can we be?
My guess is that this mostly affects safe states either way. If you live in Wisconsin, it won't be hard to find people who agree with your preference, whatever it is.
And in theory it could be cheaper but that only comes by increasing numbers to reflect the fact you are no longer constrained by room sizes. But no university can do that at the moment those changes are probably 3-5 years away.
Even then it will fail as university courses are validated by other universities (who to some extent are competitors). And no competitor is going to approve of a course moving fully online at 1/3 -1/2 the cost...
That is why people perceive it should be cheaper.
Doesn't look to be enough right now, but....
* Make it hard to register. Exclude anyone with a criminal record (which is a lot of people in the US). Make the procedure complicated. Require that you do it in person during working hours at a particular location.
* Spread targeted rumours. Target strong Dem communities with rumours that postal votes won't be counted, then near polling day more rumours that the pandemic is rampant, there are gunmen at polling stations. This is exactly the sort of thing Russian bot accounts excel at.
* Have a really short space between when you get your postal vote and when you stop counting them (ideally election day).
* In swing states with a GOP governor, limit the number of polling stations in strong Dem areas. Spread more rumours about the risks of standing in line.
It's these factors that have put me off lumping on Biden, rather than the debate (he'll be fine, Trump is relatively incoherent) or the stock market (must be already factored in).
It's also all very depressing - the sort of behaviour associated with dodgy quasi-democracies run by autocrats. A key point is that with the exception of the rumours, nobody is bothering to deny that the effort is being made.
Anyone who's filled out one of the online forms can tell they're designed for track and trace type activity, not quarantine policing. The assumption behind track and trace is that people should WANT to know if they're at risk as a result of coming into close contact with an infected person.
I suspect part of the reason some countries do better than others at tracking and tracing is that in some countries people are motivated to work with the authorities for the sake of their and public health, and in others all anybody is focussed on is how they can beat the system.
Plus there should be significant in-trade opportunities on the night.
We should note that most of these tactics were being done in 2016 also, so I guess we are focusing on the marginal voter suppression being done this time. Trump probably is in more of a position to do so, so I'd expect that additional suppression to be considerable.
They were, IIRC, told by various regulators (US, EU etc) - no, you can't. Because if they had gone ahead, the regulators assumed there would be a stampede from the old banks to Apple Bank. Leaving them with an entire dead financial system....
I really don't understand what's going on now, it's never been this bad before.
Waste of time.
Ok
Meanwhile, the woman in charge of this mess is being put in charge of the whole public health response across England and Wales.
What could possibly go wrong?
World-beating.
My other half works in an NHS lab, and she says that the hassle of anti-Covid measures has pushed productivity through the floor. She would normally be signing off reports; instead she has spent the day trying to work out where the hell put everybody.
You don't have to do things at a national level, just at the state level.
"And if the 538 model is over-estimating the extremes, flattening the distribution so much, it must be under-estimating the central bands, pulling down Clinton’s quoted chances of victory because it takes a large chunk of the probability out of Clinton’s central winning zone of 300 to 370 ECVs into Trump-victory territory of Clinton getting fewer than 270 (and also some into Clinton landslide territory, but that doesn’t alter the headline odds of him winning)."
The point is that 538 were panned in 2016 for coming up with a model that in the final days gave Trump a possibility of winning of around 30% or so, compared the the consensus in most other models that gave Trump a chance in the range of 10%. I think 538 called Trump's chances correctly based on the evidence then available.
So I'll stick with 538, thanks, and their current assessment that Trump has a 28% chance. Biden should be the favourite by some margin, but at this point it is not nailed on for him yet, just as it wasn't nailed on for Clinton two months before polling day in 2016. Of course, if the polls in 7 weeks time are where they are now, it would be correct to rate Trump's chances as much lower than 28%, but these are not at the moment eve of poll probability ranges that are being assessed.
It's not impossible Trump performs a miracle and wins (somehow) but on the face of the overwhelming evidence before us Biden is currently a clear buy, and great value at current market prices.
Why Trump has to lose.
I am fearful that he won’t, though.
Postal votes are already going out, and the Dems have already established that Trump's nefarious Postmaster General is going to try to delay them, so you need to get them in really early.
Once you've created a situation where Dem votes have overwhelmingly gone in by post already, it gets much harder to run the polling station attacks. Firstly, there will be fewer Dems who need to vote on the day, so it will be harder to create congestion at polling stations. But secondly, since the Dem tribe is saying you should vote by post, and the GOP tribe is saying that postal voting is an evil plot, you risk catching a lot of your own supporters in your attack. You can *partly* mitigate this by focusing very hard on - for example - inner city black precincts, but even there Trump is getting non-zero support. It's easy to spread rumours on social media, but it's much harder to spread *pinpoint* rumours.
What is true is that the voting situation creates a significant chaos factor, and this could work in either direction. This is something that the 538 model has actually decided not to count, so maybe it will end up offsetting the slightly eccentric things that 538 *is* treating as signs of uncertainty. It could be that Dems will have postal votes rejected en mass because they're missing signatures. But it could also be that differential postal voting in a pandemic provides a huge organizational benefit to the Democrats.
This chaos factor potentially applies to House and Senate races as well as the presidential race, so maybe they're better places to look for weird upsets.
Do COVID protocols actually vary by hospitals, or is there just standardised national guidance which must be followed, regardless of the prevalence of COVID in the local area?
The claim only has to last until the votes are cast, after all. Give Trump believers a reason to turn out and rationalise their vote. And afterwards (or, if he did it a bit too soon and it was unmistakeably not going to happen):
”The DEMOCRAT-controlled scientists are blocking your SAFETY. It could be all over but they’re INSISTING on POINTLESS extra tests! SLEEPY JOE and PHONY KAMALA are playing politics with your Lives! And they enjoy MUZZLING you and taking away your AMERICAN FREEDOMS! Sad!”
At around 7pm on Monday, the M.E.N tried to book a test using a number of postcodes from the remaining six boroughs in Greater Manchester.
For a variety of different postcodes in Oldham, Rochdale, Trafford, Tameside, Salford and Stockport the website would not load when trying to book a test.
No matter which options we selected, it was not possible to book a test using any postcode in these boroughs.
We tried refreshing the page, but no tests were showing as available.
In a statement, a spokesperson from the Department of Health and Social Care spokesperson said that there is currently 'a high demand for tests'.
They said areas with outbreaks are being made a priority.
The statement read: “There is a high demand for tests and to help stop the spread of the virus we are targeting testing capacity at the areas that need it most, including those where there is an outbreak, as well as prioritising at-risk groups.
“We have the capacity to test for coronavirus at an unprecedented scale.
Lucky Greater Manchester isn't a relative hotspot.
https://www.manchestereveningnews.co.uk/news/greater-manchester-news/how-easy-book-coronavirus-test-18892002
https://twitter.com/Peston/status/1303241687603589120
There are two other accepted approaches in statistics. One is called Bayesian which does not require repeatability, but now assumes that this probability number is itself random, so the 60% might be the mid-point of a distribution from 50-70%. The Nate Silver models are based on a Bayesian foundation.
The approach Richard takes is a different but well documented philosophy:
De Finetti's approach. The argument here is that there is a fair price which determines the probability. A person who is unemotional about the result is prepared to back or lay when the price is not in line with the percieved probability and hold off from betting when the price matches the probability. Of course people who gamble seriously find this approach very natural. My problem with it is that it means that probabilites move from being an external concept (the probabilities remain the same regardless of who is trying to estimate it), to being a belief which can vary from individual to individual.
I particularly like the comment that "A philosopher might ask whether a probability for a one-off unrepeatable event means anything at all, since it is inherently impossible to falsify " which makes a point that is so often missed by many people (@Wulfrun_Phil I'm looking at you!)
Pub closures
I combined it with visiting friends and family. My mother has managed to see her GP once since March (she has ongoing health conditions) and my aunt, who lives next door to a hospital, has been told to drive 50 miles, expect to have an invasive procedure with anaesthesia and then drive back. She is 75 and recovering from an ileostomy. The hospital next door simply refuses to see her.
Every year the government needs to borrow to roll over old debts as well as borrow for the deficit. If you can borrow at negative interest rates then the government can and should be using that functionality to bring down old interest rates. Osborne did this while he was Chancellor, the Treasury quietly whenever interest rates went low or negative borrowed and used that money to repay old debts that were running at a higher interest rate, thus the burden of debt on UK interest payment made out. As a result the UK government is paying a remarkably low share of interest as a proportion of GDP all things considered.
Anything that is borrowed, even at a low interest rate, will need to be repaid. Unless that's going to be repaid by a budget surplus (not going to happen) its going to be repaid by rolling over the debt which could be at a much higher interest rate down the line.
I had no idea we were allowed 30 people in a house and doubt many other people did either so changing the limit probably wont have that much impact.
I'm sure anyone with first-hand experience of running a business which involves a lot of human interaction will tell you how much extra hassle anti-Covid measures add to their life.