Forecasting Newsletter: April 2020
A forecasting digest with a focus on experimental forecasting.
The newsletter itself is experimental, but there will be at least five more iterations. Feel free to use this post as a forecasting open thread.
Why is this relevant to EAs?
Some items are immediately relevant (e.g., forecasts of famine).
Others are projects whose success I’m cheering for, and which I think have the potential to do great amounts of good (e.g., Replication Markets).
The remaining are relevant to the extent that cross-polination of ideas is valuable.
Forecasting may become/is becoming a powerful tool for world-optimization, and EAs may want to avail themselves of this tool.
Conflict of interest: With Foretold in general and Jacob Laguerros in particular. This is marked as (c.o.i) throughout the text.
Index
Prediction Markets & Forecasting platforms.
Augur.
PredictIt & Election Betting Odds.
Replication Markets.
Coronavirus Information Markets.
Foretold. (c.o.i).
Metaculus.
Good Judgement and friends.
In the News.
Long Content.
Prediction Markets & Forecasting platforms.
Forecasters may now choose to forecast any of the four horsemen of the Apocalypse: Death, Famine, Pestilence and War.
Augur: augur.net
Augur is a decentralized prediction market. It will be undergoing its first major update.
Predict It & Election Betting Odds: predictIt.org & electionBettingOdds.com
PredictIt is a prediction platform restricted to US citizens or those who bother using a VPN. Anecdotically, it often has free energy, that is, places where one can earn money by having better probabilities, and where this is not too hard. However, due to fees & the hassle of setting it up, these inefficiencies don’t get corrected. In PredictIt, the world politics section...
gives a 17% to a Scottish independence referendum (though read the fine print).
gives 20% to Netanyahu leaving before the end of the year
gives 64% to Maduro remaining President of Venezuela before the end of the year.
The question on which Asian/Pacific leaders will leave office next? also looks like it has a lot of free energy, as it overestimates low probability events.
Election Betting Odds aggregates PredictIt with other such services for the US presidential elections.
Replication Markets: replicationmarkets.com
Replication Markets is a project where volunteer forecasters try to predict whether a given study’s results will be replicated with high power. Rewards are monetary, but only given out to the top N forecasters, and markets suffer from sometimes being dull. They have added two market-maker bots and commenced and conclude their 6th round. They also added a sleek new widget to visualize the price of shares better.
Coronavirus Information Markets: coronainformationmarkets.com
For those who want to put their money where their mouth is, there is now a prediction market for coronavirus related information. The number of questions is small, and the current trading volume started at $8000, but may increase. Another similar platform is waves.exchange/prediction, which seems to be just a wallet to which a prediction market has been grafted on.
Unfortunately, I couldn’t make a transaction in these markets with ~30 mins; the time needed to be included in an ethereum block is longer and I may have been too stingy with my gas fee.
Foretold: foretold.io (c.o.i)
Foretold is an forecasting platform which has experimentation and exploration of forecasting methods in mind. They bring us:
A new distribution builder to visualize and create probability distributions.
Forecasting infrastructure for epidemicforecasting.org.
Metaculus: metaculus.com
Metaculus is a forecasting platform with an active community and lots of interesting questions. They bring us a series of tournaments and question series:
The Lightning Round Tournament: Comparing Metaculus Forecasters to Infectious Disease Experts. “Each week you will have exactly 30 hours to lock in your prediction on a short series of important questions, which will simultaneously be posed to different groups of forecasters. This provides a unique opportunity to directly compare the Metaculus community prediction with other forecasting methods.” Furthermore, Metaculus swag will be given out to the top forecasters.
The Salk Tournament for coronavirus (SARS-CoV-2) Vaccine R&D.
/(Good Judgement?[^]*)|(Superforecast(ing|er))/gi
Good Judgement Inc. is the organization which grew out of Tetlock’s research on forecasting, and out of the Good Judgement Project, which won the IARPA ACE forecasting competition, and resulted in the research covered in the Superforecasting book.
The Open Philantropy Project has funded this covid dashboard by their (Good Judgement Inc.’s) Superforecasting Analytics Service, with predictions solely from superforecasters; see more on this blogpost.
Good Judgement Inc. also organizes the Good Judgement Open (gjopen.com)[https://www.gjopen.com/], a forecasting platform open to all, with a focus on serious geopolitical questions. They structure their questions in challenges, to which they have recently added one on the Coronavirus Outbreak; some of these questions are similar in spirit to the short-fuse Metaculus Tournament.
Of the questions which have been added recently to the Good Judgment Open, the crowd doesn’t buy that Tesla will release an autopilot feature to navigate traffic lights, despite announcements to the contrary. Further, the aggregate...
is extremely confident that, before 1 January 2021, the Russian constitution will be amended to allow Vladimir Putin to remain president after his current term.
gives a lagging estimate of 50% on Benjamin Netanyahu ceasing to be the prime minister of Israel before 1 January 2021.
and 10% for Nicolás Maduro leaving before the 1st of June.
forecasts famine (70%).
Of particular interest is that GJOpen didn’t see the upsurge in tests (and thus positives) in the US until until the day before they happened, for this question. Forecasters, including superforecasters, went with a linear extrapolation from the previous n (usually 7) days. However, even though the number of cases looks locally linear, it’s also globally exponential, as this 3Blue1Brown video shows. On the other hand, an enterprising forecaster tried to fit a Gompertz distribution, but then fared pretty badly.
In the News
Forecasts in the time of coronavirus: The Financial times runs into difficulties trying to estimate whether some companies are overvalued, because the stock value/earnings ratio, which is otherwise an useful tool, is going to infinity as earnings go to 0 during the pandemic.
Predictions are hard, especially about the coronavirus: Vox has a short and sweet article on the difficulties of prediction forecasting; of note is that epidemiology experts are not great predictors.
538: Why Forecasting COVID-19 Is Harder Than Forecasting Elections
COVID-19: Forecasting with Slow and Fast Data. A short and crisp overview by the Federal Reserve Bank of St Louis on lagging economic measurement instruments, which have historically been quite accurate, and on the faster instruments which are available right now. Highlight: “As of March 31, the WEI [a faster, weekly economic index] indicated that GDP would decline by 3.04% at an annualized rate in the first quarter, a much more sensible forecast than that which is currently indicated by the ENI (a lagging measure which predicts 2.26% growth on an annualized basis in the first quarter)”.
Decline in aircraft flights clips weather forecasters’ wings: Coronavirus has led to reduction in number of aircraft sending data used in making forecasts.
The World in 2020, as forecast by The Economist. The Brookings institution looks back at forecasts for 2020 by The Economist.
Forbes brings us this terrible, terrible opinion piece which mentions Tetlock, goes on about how humans are terrible forecasters, and then predicts that there will be no social changes because of covid with extreme confidence.
The Challenges of Forecasting the Spread and Mortality of COVID-19. The Heritage foundation brings us a report with takeaways of particular interest to policymakers. It has great illustrations of how the overall mortality changes with different assumptions. Note that criticisms of and suggestions for the current US administration are worded kindly, as the Heritage Foundation is a conservative organization.
Why most COVID-19 forecasts were wrong. Financial review article suffers from hindsight bias.
Banks are forecasting on gut instinct — just like the rest of us. Financial Times article starts with “We all cling to the belief that somebody out there, somewhere, knows what the heck is going on. Someone — well-connected insider, evil mastermind — must hold the details on the coming market crash, the election in November, or when the messiah will return. In moments of crisis, this delusion tightens its grip,” and it only gets better.
‘A fool’s game’: 4 economists break down what it’s like forecasting the worst downturn since the Great Recession. “‘My outlook right now is that I don’t even have an outlook,’ Martha Gimbel, an economist at Schmidt Futures, told Business Insider. ‘This is so bad and so unprecedented that any attempt to forecast what’s going to happen here is just a fool’s game.’”
IMF predicts −3% global depression. “Worst Economic Downturn Since the Great Depression”.
COVID-19 Projections: A really sleek US government coronavirus model. See here for criticism. See also: Epidemic Forecasting (c.o.i).
Some MMA forecasting. The analysis surprised me; it could well have been a comment in a GJOpen challenge.
Self-reported COVID-19 Symptoms Show Promise for Disease Forecasts. “Thus far, CMU is receiving about one million responses per week from Facebook users. Last week, almost 600,000 users of the Google Opinion Rewards and AdMob apps were answering another CMU survey each day.”
Lockdown Policy and Disease Eradication. Researchers in India hypothesize on what the optimal lockdown policy may be.
Using a delay-adjusted case fatality ratio to estimate under-reporting.
The first modern pandemic. In which Bill Gates names covid-SARS “Pandemic I” and offers an informed overview of what is yet to come.
36,000 Missing Deaths: Tracking the True Toll of the Coronavirus Crisis.
There is a shadow industry which makes what look to be really detailed reports on topics of niche interest: Here is, for example, a $3,500 report on market trends for the Bonsai
An active hurricane season will strain emergency response amid pandemic, forecasters warn. “Schlegelmilch stresses that humanity must get better at prioritizing long-term strategic planning.”
Long Content
Atari, early. “Deepmind announced that their Agent57 beats the ‘human baseline’ at all 57 Atari games usually used as a benchmark.”
A failure, but not of prediction; a SlateStarCodex Essay.
Philip E. Tetlock on Forecasting and Foraging as a Fox; an interview with Tyler Cowen. Some highly valuable excerpts on counterfactual reasoning. Mentions this program and this study, on the forefront of knowledge.
Assessing Kurzweil’s 1999 predictions for 2019. Kurzweil made on the order of 100 predictions for 2019 in his 1999 book The Age of Spiritual Machines. How did they fare? We’ll find out, next month.
Zvi on Evaluating Predictions in Hindsight. A fun read. Of course, the dissing of Scott Alexander’s prediction is fun to read, but I really want to see how a list of Zvi’s predictions fares.
An oldie related to the upcoming US elections: Which Economic Indicators Best Predict Presidential Elections?, from 2011′s Nate Silver.
A rad comment exchange at GJOpen in which cool superforecaster @Anneinak shares some pointers.
As the efficient markets hypothesis turns 50, it is time to bin it for a Financial Times article, from Jan 1st and thus untainted by coronavirus discussion. Related: This LW comment by Wei Dai and this tweet from Eliezer Yudkowsky. See also a very rambly article by an Australian neswpaper: Pandemic highlights problems with efficient-market hypothesis.
Nice to see a newsletter on this topic!
Clarification: The GJO coronavirus questions are not funded by Open Phil. The thing funded by Open Phil is this dashboard (linked from our blog post) put together by Good Judgment Inc. (GJI), which runs both GJO (where anyone can sign up and make forecasts) and their Superforecaster Analytics service (where only superforecasters can make forecasts). The dashboard Open Phil funded uses the Superforecaster Analytics service, not GJO. Also, I don’t think Tetlock is involved in GJO (or GJI in general) much at all these days, but GJI is indeed the commercial spinoff from the Good Judgment Project (GJP) that Tetlock & Mellers led and which won the IARPA ACE forecasting competition and resulted in the research covered in Tetlock’s book Superforecasting.
Thanks for the correction; edited.
Note that the headline (“Good Judgement Project: gjopen.com”) is still confusing, since it seems to be saying GJP = GJO. The thing that ties the items under that headline is that they are all projects of GJI. Also, “Of the questions which have been added recently” is misleading since it seems to be about the previous paragraph (the superforecasters-only questions), but in fact all the links go to GJO.
Edited again. If you want, throw me a bone: what’s the last explicit probabilistic prediction you’ve made? Also, I liked your review on How to Measure Anything, which feels relevant to the topic at hand. NNTR.
The headline looks broken in my browser. It looks like this:
/(Good Judgement?[^]*)|(Superforecast(ing|er))/gi
The last explicit probabilistic prediction I made was probably a series of forecasts on my most recent internal Open Phil grant writeup, since it’s part of our internal writeup template to prompt the grant investigator for explicit probabilistic forecasts about the grant. But it could’ve easily been elsewhere; I do somewhat-often make probabilistic forecasts just in conversation, or in GDoc/Slack comments, though for those I usually spend less time pinning down a totally precise formulation of the forecasting statement, since it’s more about quickly indicating to others roughly what my views are rather than about establishing my calibration across a large number of precisely stated forecasts.
I’ve written this interactive notebook in Foretold prediction platform. It is meant to be completely beginner friendly and takes about 2 hours to go through. I’ve used it as the basis for a workshop, and the accompanying slides can be found at the bottom of the notebook.
From the notebook:
The time horizon for this is “before 1 June 2020.” That seems reasonable.
Oh hey, I’ve seen you around on GJOPen. Thanks for the correction; edited.