Three Types of Data-Driven Forecasting
Part 2 in the 'How I Forecast' series from Top Forecaster, Joey Shurtleff
Forecast is a new community for discussing the future, organized by topics that matter to people. Whether about news, politics, social issues, sports or pop culture, Forecasters work together to share their knowledge about what's going on today to build a better understanding of what will happen tomorrow. Download the app to join the community.
In this 3 post guest series, we welcome Joey Shurtleff, Forecast's all-time top Forecaster. In addition to being a top Forecaster with over 400 forecasts and 900,000 points in reason supports to his name, Joey is also a serial entrepreneur and world traveler. Last time, Joey discussed his Analysis strategy. In this post, he’ll share some stories about making forecasts by keeping up with world news and other data sources. Next time, he’ll discuss exploits.
Forecast is as much a breaking news app as it is a forecasting app. Forecasters can earn lots of points by moving markets when news breaks, and I’ve often learned about breaking news from Forecast before seeing it elsewhere.In this post, I’ll focus on how I’ve kept a finger on the pulse of the goings on in the world, which is reflected in three forecast types: news, external forecasts, and live data.
News
As you can see from the chart above, my third most lucrative forecast type was what I’m calling “news”: when I spot a piece of highly relevant news earlier than other forecasters and get a good position at a great price. I gained over 125k pts from this, representing over 15% of my total gains.
My biggest winner of this type was this question:
I learned from a prior market that the news about the Billboard chart often breaks before the new chart is added to the website, and I made a mental note to Google for this in the final days before this question was to be resolved. Sure enough, even though other Forecasters thought that Taylor was 90+% likely to keep the top spot for a 7th week, I found an article saying that the data showed that she’d end up behind Big Sean’s “Detroit 2” and three other albums. I bought the “6 weeks” position all the way up from a price of 8 pts to a price of 97 pts, and gained over 12,000 pts when the market was finally resolved.
Another winner worth highlighting is this one about one of the Autumn California wildfires:
There had been a number of fire containment questions on Forecast previously, and I learned that it was challenging to forecast questions like this until the fire department released an estimated containment date. I kept a browser tab open on my phone to a search for “glass fire containment” and several days later this enabled me to notice the containment date announcement before other forecasters and generate a gain of over 6,000 pts.
My only significant loss trading on news was on this question about Uber:
I thought this was highly likely since Uber messaged its users on October 26 saying that it would suspend service at the end of the day, but I was not aware that a judge would grant an emergency stay of injunction that evening and Uber would rescind their threat (this miscue only cost me less than 600 pts).
External Forecasts
My fourth most lucrative forecast type was “external forecasts”, which is when I take market positions based on external forecasts such as Vegas betting lines, FiveThirtyEight models, or something similar. I gained over 120k pts from this, representing about 15% of my total gains.
I’ve done this most aggressively with the various NBA and MLB markets. The most lucrative of these markets was this one:
I based my action primarily on the Vegas odds, and therefore initially took stakes in the Lakers, Clippers, and Bucks. I noticed a couple other forecasters bidding heavily on the Celtics; I suspect that either they were big Celtics fans or they were making their bids based on FiveThirtyEight’s model, which consistently believed the Celtics (and the Heat) were significantly better than the Lakers. I had strong conviction that the Vegas odds were a far better reflection of the true odds than FiveThirtyEight’s model so I continued to take stakes in the Lakers, Clippers, and Bucks at somewhat reduced prices. I continued to track this market and update prices as games were played and concluded, and I ended up taking large lucrative positions in the Heat and Nuggets as those teams made it further than most thought they would. Even though the Clippers and Bucks were eliminated early, my large stakes in the Lakers and Heat were the primary contributors to the over 25,000 pts I gained from this market.
My biggest loser of this forecast type was “Who will win the 2020 Emmy for Outstanding Lead Actress in a Drama Series?”; I had significant stakes in each of the three favorites, and this award ended up being the only major surprise of the night (Zendaya won), costing me over 4,000 pts net.
Live Data
Live data, where I make forecasts based on constantly updating data, was my most volatile forecast type. I gained over 30k pts from this, representing just under 4% of my total gains; on the 6 questions in this category, I registered significant gains (over 1k pts) on 3 and significant losses (over 1k pts) on the other 3.
One of the most lucrative examples of this was the question “How much money will Summer Games Done Quick 2020 raise by August 23, 2020?” A large late contribution of over $250k dramatically changed the trajectory of this outcome and I recognized this and other shifts quickly enough to gain over 8,000 pts on this (very thinly traded) market.
Some of my biggest per-question losses have come from this category. The biggest loss was from the question “How many times will Donald Trump tweet or retweet between October 25 and November 1, 2020?”. Here’s the price chart for this question
I had a large position on between 250 and 399 (I’m pretty sure I drove most of the price movement in the last few days of this market) and the President’s 400th tweet came less than 90 minutes before the end of the question’s window, costing me over 5,000 pts net. Other large losses were on two questions pertaining to the “air quality index” in US west coast cities during this Autumn’s wildfire season; these each cost me over 2,000 pts net.
The analysis above suggests that keeping up with breaking news is often highly rewarding and carries minimal risk, while forecasting based on live data can be exciting but is akin to playing with fire. Next week, I’ll finish this post series by covering the last set of forecast types - exploits - and suggesting some product improvements that could minimize Forecaster gains from non-value-add actions.