Re:Microsoft Prediction Lab
by tedsanders on Monday September 29, 2014 @05:18PM
(#8385362)
I'm excited that Microsoft is dedicating some resources toward building a prediction market platform. It joins the space now occupied by Foresight Exchange, SciCast, Inkling Markets, Lumenogic, Predictious, Truthcoin, and similar software platforms. Hopefully Microsoft will build a better product and find more success, though so far it seems like the lack of prediction market uptake seems to be the result of low demand, not low supply.
As a long-time participant in prediction markets, I'll a list of my initial impressions of the Microsoft platform.
(1) I think it's big mistake to mix opinion polls with forecasts. It's quite confusing when one set of questions asks for facts and another set asks for opinions. It's also confusing when one set affects your score and another set does not. I think users would have a better experience if the two categories were totally separated (or if the opinion polls were just axed).
(2) The scoring method they're using is not at all transparent. I suspect it's the logarithmic scoring rule (best for low-volume markets), but the interface doesn't make it clear how the bets relate to the odds. (Look at SciCast.org for one way of doing this.) I understand that the math behind betting can be offputting to new users, but it's also tough to get engagement when new users don't really understand the possible actions or their consequences.
(3) The default bet sizes are awful. 100 points out of a 1,000 point balance means you'll be practically done with the site after 10 bets. Even if your bets are accurate, you can't bet again until they resolve. The SciCast play money market has had significant problems making accurate predictions because of a lack of liquidity for its high-volume users, who also tend to be the most accurate users.
(4) I think I've already run into some bugs. When I vote in polls, it looks like my vote is being assigned to a choice different from the one I voted on.
(5) I thought the design looked nice at first, but there are some usability problems. It's annoying how forecasts with three options require you to click a small button to scroll from side to side. A big part of casual non-algorithmic prediction market participation is quickly scanning over the markets to look for obvious mistakes. Making it hard to scan over the forecasts makes the market harder to use and less accurate.
Anyway, the momentum behind prediction markets has been growing for years. Perhaps in the future more companies will try experimenting with internal markets to predict things like whether projects will finish on time or on budget.
Re:Microsoft Prediction Lab
by freehunter on Monday September 29, 2014 @04:17PM
(#8385000)
I'm not sure I understand. I have 1000 points, but I only have 400 points for this week? When I spend my points on a bet, my available points don't decrease? There are only three polls I can bet on, the rest are just free opinion polls?
Re:Microsoft Prediction Lab
by adam on Monday September 29, 2014 @07:12PM
(#8385873)
MS has experimented with prediction markets internally off and on. This particular market is being run by David Rothschild who is a researcher at Microsoft. He's run the site http://predictwise.com for quite some time and done a really nice job with it. I've met David in person and he's an uber smart guy with some really good ideas. This is a "labs" project for the time being as the name implies, but I wouldn't be surprised to see some of this work make its way in to MS products in one form or another, i.e. embedded prediction market style questions in Sharepoint or MS Project, or even as ad units on Bing to collect a massive pool of forecasts.
As a long-time participant in prediction markets, I'll a list of my initial impressions of the Microsoft platform.
(1) I think it's big mistake to mix opinion polls with forecasts. It's quite confusing when one set of questions asks for facts and another set asks for opinions. It's also confusing when one set affects your score and another set does not. I think users would have a better experience if the two categories were totally separated (or if the opinion polls were just axed).
(2) The scoring method they're using is not at all transparent. I suspect it's the logarithmic scoring rule (best for low-volume markets), but the interface doesn't make it clear how the bets relate to the odds. (Look at SciCast.org for one way of doing this.) I understand that the math behind betting can be offputting to new users, but it's also tough to get engagement when new users don't really understand the possible actions or their consequences.
(3) The default bet sizes are awful. 100 points out of a 1,000 point balance means you'll be practically done with the site after 10 bets. Even if your bets are accurate, you can't bet again until they resolve. The SciCast play money market has had significant problems making accurate predictions because of a lack of liquidity for its high-volume users, who also tend to be the most accurate users.
(4) I think I've already run into some bugs. When I vote in polls, it looks like my vote is being assigned to a choice different from the one I voted on.
(5) I thought the design looked nice at first, but there are some usability problems. It's annoying how forecasts with three options require you to click a small button to scroll from side to side. A big part of casual non-algorithmic prediction market participation is quickly scanning over the markets to look for obvious mistakes. Making it hard to scan over the forecasts makes the market harder to use and less accurate.
Anyway, the momentum behind prediction markets has been growing for years. Perhaps in the future more companies will try experimenting with internal markets to predict things like whether projects will finish on time or on budget.
I wish Microsoft success with this new product.
Not sure if I'm missing something.
Related article: http://www.npr.org/blogs/parallels/2014/04/02/297839429/-so-...
( http://blogs.microsoft.com/next/2014/06/11/how-does-bing-pre... )
http://web.archive.org/web/20080820025756/http://ipredict.ms...
It's from around 1995 or 1996, as you can probably tell from the site design.