Sunday, 5 April 2020

Presidential Prediction


While watching an interview with Edward Thorp (mathematician, first to prove the edge of card counting in blackjack and successful fund manager) he mentioned a book called the "The Keys to the White House”. The book’s author (Allan Lichtman) created a prediction tool to for the outcome of the US presidential election, which was back tested to all American presidential elections from 1860. Interestingly, the tool doesn’t rely on any polling data.



It reminded me of the strength that simple algorithms can have in outperforming expert predictions, for example the Bordeaux equation by Orley Ashenfelter, a formula which outperformed expert predictions on the quality and therefore price appreciation of particular vintages of wine ("Grape Expectations" at the bottom of the article)



The tool requires answering of 13 questions either as “true” or “false”, which cover things like the performance of the economy and if there has been a major scandal. Some of the questions do require judgement, and I have grouped the questions below to reflect this. If 8 or more are True then the incumbent is expected to win.



By my estimation 8 were True, 3 False and the 2 questions on the economy are hanging in the balance, with the impact of the coronavirus yet to be fully experienced. If for example we did see a deep recession, causing both social unrest and a significant reduction in per capita incomes then the tool would suggest Biden as the eventual winner. As it stands Trump is the favourite and this is mirrored in the betting market, where he is an implied 53% favourite.

  

Indisputable questions:

Party Mandate: After the midterm elections, the incumbent party holds more seats in the U.S. House of Representatives than after the previous midterm elections. False



Contest: There is no serious contest for the incumbent party nomination. True



Incumbency: The incumbent party candidate is the sitting president. True



Third party: There is no significant third party or independent campaign. True



Short-term economy: The economy is not in recession during the election campaign. False / True



Long-term economy: Real per capita economic growth during the term equals or exceeds mean growth during the previous two terms.  False / True



Policy change: The incumbent administration effects major changes in national policy. True



Social unrest: There is no sustained social unrest during the term. True



Incumbent charisma: The incumbent party candidate is charismatic or a national hero. (Note to achieve this the benchmark is Ronald Reagan, JFK and Theodore Roosevelt) False



Challenger charisma: The challenging party candidate is not charismatic or a national hero. True



Debatable:



Scandal: The incumbent administration is untainted by major scandal. False



Foreign/military success: The incumbent administration achieves a major success in foreign or military affairs. True



Foreign/military failure: The incumbent administration suffers no major failure in foreign or military affairs. True

Sunday, 9 February 2020

US Presidential odds update - Bloomberg odds shortening

It has been a month since I last aggregated the probabilities associated with the US election (linked here) and in that time Peter Buttigieg won the most votes in the Iowa caucus, the result of which was delayed due to technical issues.

Of the updated data below, the most interesting change has been in the betting market's probability for Mike Bloomberg to become President. Bloomberg now sits at an 11% implied probability from 5% a month ago. This is the second highest amongst the Democratic candidates. However, it is in sharp contrast to the 538 primary forecasting model which sees his chances of winning the Democratic nomination first as just 2%.

Bloomberg's primary strategy is untested - he will only begin campaigning on the Super Tuesday states, missing out earlier chances to pickup delegates. He is outspending the rest of the field put together, a strategy only possible due to his personal financial position. This could backfire given the focus in recent years on rising inequality and the behaviour of the very wealthy. The narrative of him trying to "buy" the election is something I would expect to see other Democratic candidates using if he began winning delegates. However, the betting markets clearly give him a lot of credit for being able to bring those resources to bear, and for what he brings as a candidate as a successful businessman and ex mayor of New York.

My personal view is that it is an interesting bet to lay against him winning the presidency, taking the 538 forecasting model as more likely to reflect the true probability of his chances over the betting market, given their data driven approach.

Betting market probabilities

Probabilities across sources

Sunday, 2 February 2020

Harry and Meghan - divorce odds


After Harry and Meghan stepped back from the British royal family, UK bookmakers started offering odds on various other scenarios that might occur to the couple. The most interesting of these to analyse are the odds of them getting divorced, in part because there are good datasets on divorce rates.

How should you think about analysing the probability of this?

3/1 are the bookmakers odds on them being divorced by 2025. For simplicity let's assume this is a 5 year time horizon. For this to be an attractive proposition you would want the probability of them getting divorced to be materially higher than the 25% implied probability the 3/1 bet offers. If the true probability was exactly 25%, out of 4 trials, 3 times you would lose your stake and 1 time you would make 3 times your stake.

In analysing a proposition like this you want to first establish a base rate; an average probability across the population that gives you a starting point from which you can bias your estimate of the "true" probability up or down depending on specific factors to the case in hand. The UK Office of National Statistics has good datasets on divorce, broken down by the time since the couple has been married.

Harry and Meghan were married more than a year and a half ago, the marriage data is grouped by year of marriage, for simplicity we will assume that they have been married for 2 years.

The base rate we would like to know is what is the probability of a British couple who has been married for 2 years getting divorced over the next 5? Table 2 of the ONS link posted below has the relevant data, with the most recent year that this figure is available for as couples married in 2006 - at a 13% probability. The number has been on the decline from peaks of 16% in the late 1980s.

This figure is much lower than the level at which the bet might have value, however I would suggest it should be biased upwards for one key reason - celebrity couples are more likely to get divorced for reasons I won't speculate on here. A study done by the Marriage Foundation found that celebrity marriages are twice as likely to end in divorce as the general population for a 14 year period they studied. If we conservatively bias our base rate higher, based on this, by say 10ppts, we are still short of the proposition being attractive (23% calculated vs. 25% implied) and even if we double our estimate it is still not enough margin for error to percieve the bet as attractive (26% calculated vs. 25% implied).

This isn't an unexpected outcome, given bookies are good price makers, but I hope it was useful in showing how one might think about similar propositions.

Links:

https://www.dailymail.co.uk/news/article-7883397/Prince-Harry-Meghan-Markle-likely-divorce-rejoin-Royals-time-bookmakers-say.html
https://www.forbes.com/sites/guymartin/2020/01/15/the-megxit-fallout-day-7-what-the-british-bookmakers-are-telling-us-with-their-odds-on-a-divorce-for-prince-harry/#39ca9a030185
https://www.ons.gov.uk/peoplepopulationandcommunity/birthsdeathsandmarriages/divorce/datasets/divorcesinenglandandwalesageatmarriagedurationofmarriageandcohortanalyses
https://marriagefoundation.org.uk/celebrity-marriages-2/

Sunday, 12 January 2020

US Presidential odds 08/01/20

Below is a snapshot of current odds for the US Presidential election and Democratic nomination across various prediction sources taken as of the 8th of Jan. This will be updated periodically as we approach the election date.

Betting odds are taken from a website which aggregates bookies odds with the question being, "who will be the winner of the 2020 presidential election".The longest odds across the bookmakers for each outcome is shown, with the implied probability of the event calculated.

The "Good Judgement" number is taken from the crowd sourcing prediction website - The Good Judgement Project with the question slightly different - "When will Donald Trump cease to be President?". The odds for "After Inauguration day 2021" are shown. This is slightly different from whether he will win the election given there are some unlikely incidents where he wins the election but has to step down for whatever reason before inauguration.

Finally the predictions of the 538 statistics blog website are shown of the potential outcome of the Democratic nomination.

It is interesting to see Biden and Sanders priced closely by the betting markets (4pts difference), while 538 puts Biden's democratic nomination probability 14pts higher than Sanders.

Also interesting to see Bloomberg priced the same as Buttigieg and Warren.







 Links:

https://www.oddschecker.com/politics/us-politics/us-presidential-election-2020/winner
https://www.gjopen.com/questions/1290-when-will-donald-trump-cease-to-be-president-of-the-united-states
https://projects.fivethirtyeight.com/2020-primary-forecast/

Saturday, 4 January 2020

A quick take: How did Boris do it?

Background
The UK general election held on the 12th of December 2019 saw Boris Johnson and the Conservative party win its largest parliamentary majority since 1987 (365 out of 650 seats) and Jeremy Corbyn's Labour party experience their worst result since 1935 (202 out of 650 seats). For both parties it was a sizeable swing from the June 2017 election (+48 seats for the Conservatives and -60 for Labour), what were some of the factors that drove this?

Poll data takeaways
Detailed polling data from Lord Ashcroft (https://lordashcroftpolls.com/2019/12/how-britain-voted-and-why-my-2019-general-election-post-vote-poll/) highlighted the key themes of this election. "Asked to choose their top three broad reasons for their decision, Conservative voters were most likely to say their party or leader “was the most likely to get the Brexit outcome I wanted” (68%), “would do a better job of running the economy” (64%), and that the leader “would make a better Prime Minister” (58%)." While "The top reasons Labour voters chose were that they “trusted the motives of the party I voted for more than those of other parties” (65%), that they “preferred the promises made by the party I voted for more than the promises of other parties” (59%), and that they thought Labour would do a better job of running the economy (though only 39% chose this as a reason). Only 19% of Labour voters said that believing the party would get the Brexit outcome they wanted was among their top three reasons for doing so."

While Brexit was clearly key, worth highlighting is the difference in trust on the economy and its relatively higher importance for Conservative voters.

Winning on these issues allowed the Conservatives to gain the largest share (43%) of the working and non-working social class vote (https://en.wikipedia.org/wiki/NRS_social_grade) something which Theresa May failed to do in 2017 (41% to Labour's 43%). Labour's 37% share of this demographic will perhaps be their biggest dissappointment given the focus of their campaign and is another datapoint in the breakdown of the traditional voter coalitions which we came to expect in Western democracies.

Campaign comparison
"Get Brexit done" was the memorable and clear slogan of the Conservative campaign, just three words which, as the poll data showed, reflected the frustration of many in the UK over the lack of progress on this issue, and signposted the Conservative's focus in this parliament. Strong campaigns have memorable slogans, a factor which I think has increased in importance in the information age, given the number of messages, emails and adverts we recieve each day. Can you recall the Labour slogan, given the election was only a few weeks ago? "It's time for real change", I had to look it up.

Linked to this was the discpline shown in each party's manifesto (https://vote.conservatives.com/our-plan and https://labour.org.uk/manifesto/). The Conservative manifesto website has the main promises front and centre, allowing you to quickly read what they plan. Contrast this to the Labour manifesto website, which requires clicking through various sections and subsections on the website before you get to the meat.

Finally, another aspect of any political campaign is winning the "Beer Test". When making complicated decisions under uncertainty we often substitute the question for a more simple one subconsciously. In the case of a general election, politicians are keen to present themselves as the candidate you would most like to go for a pint with. Boris's unconventional and semi-serious demeanor, history of hijinks and regular appearances on comedic panel shows like "Have I Got News for You", means I think this contest was over before it started.

While not trying to being comprehensive, I hope this shone a light on some of the important aspects of this election.Comments and constructive criticism welcome.

Sunday, 1 December 2019

Bring back the experts

A common observation of platforms where virtually anyone can write a review (Google places, Yelp, Amazon etc.) is; given reviews can and are being faked, their usefulness decreases. This decrease in usefulness in some cases can be enough for the reviews to be as useful as not having any reviews at all. If all products in a category achieve 4 or 5 star average review ratings what is the point of having them.

Taking the above to its natural conclusion results in the increased importance of trusted experts in a particular field for consumers to ascertain the best product. It is a reversion to the pre everyone review era and companies like Which are cheering / playing on this development.

Assuming that those platforms want people to trust the reviews there, they need to somehow root out the fake reviews from the real which is very tricky. Another solution might be trying to integrate expert opinions into their platform.

A company doing badly at the above I would argue is Deliveroo. Looking today at range of cuisine categories I saw very few restaurants averaging below 4 stars (they have a 5 star scale).

As an example looking at Thai restaurants only there 26 that deliver to me. The range of average star ratings is from 3.7 to 4.8. 15 out of 26 have ratings between 4.5 and 4.8. 9 have ratings between 4 and 4.4 and 2 have ratings less than 4. Given these stats I would suggest most people would avoid restaurants with less than 4 stars - therefore it becomes an imperative for restaurants on the app to achieve greater than 4 stars. This creates a strong incentive for reviews to be faked, from a restaurant's point of view, to bring their average rating up to an 4 star level. As more restaurants do this fewer restaurants exist in the 4 star or less category and a feedback mechanism is created - it becomes more important not to be in the 4 star or less category and so more reviews are faked until there are no restaurants at this level. At some point the cut off for what will be percieved as a weak restaurant average rating might be 4.5 at which point the game starts again...

It is a difficult problem to solve, rooting out fake reviews for Deliveroo is tricky / impossible given that a review may just be a star level without any written text where fakes can be easier to spot. My personal view is that if Deliveroo would like people to trust the reviews on their site they need to integrate expert opinion. This isn't without problems as you can see the potential for this to cause issues with your restaurant partners.

Thursday, 2 February 2017

Peak Irony

It seems as if we have reached a situation where the most visible parts of both the left and right of politics (in the US) could not become more ironic. Ironic may be the wrong word, completely lacking any common sense might be a better description.

Today saw students at Berkeley, supposedly progressive, violently riot in reaction to the speaker Milo Yiannopoulos http://www.bbc.co.uk/news/world-us-canada-38837142 . A defining characteristic of the most abhorrent regimes in history has been the limiting of free speech. It is not a coincidence. By limiting free speech you gain control by limiting opposing ideas. If by publicly voicing ideas contrary to the status quo you risk violence or death effective opposition becomes very difficult to build. In this way would be groups that could overthrow a regime never have the chance to form.

Berkeley students' response of violence to a speaker who's views they dislike is an attempt to limit discourse and free speech. I have little doubt that they see Milo as tantamount to a fascist but in behaving this way they are emulating many of the great fascists. They are fighting something they view as bad (rhetoric) with something worse (violence).

Protest organiser Yvette Felarca defended the demonstration, describing the actions of students as "self-defence".
"We have the right to defend ourselves," she said, adding: "This shutting down Milo Yiannopoulos, and doing whatever's necessary to do that, is our right to self-defence."

The event organiser, while she may or may not condone the violence, displays this obviously warped view that it is ok to try and silence people you disagree with. The phrase “doing whatever’s necessary” suggests she does in fact agree with the violence that occurred.

Moving to the inconsistency on the right of US politics. Trump has decided to ban refugees from 7 Muslim countries https://www.bloomberg.com/politics/articles/2017-01-30/trump-s-refugee-ban-and-the-legal-landscape-ahead-quicktake-q-a . He sights protecting US citizens as the reason for his ban. As the many shared graphic shows there has been not a single life taken by people entering from those countries while the US has one of the toughest immigration processes in the world. If protecting US lives was the priority then there are a great many things he could do better. A focus on mental health the 10th largest killer in the US is an obvious one but there are many more things which would do a lot more to improve the safety of its citizens - http://www.cbsnews.com/news/the-leading-causes-of-death-in-the-us/