Predicted market correct last 13 days in row(twitter.com)
twitter.com
Predicted market correct last 13 days in row
https://twitter.com/dynofuz?src=hnws
3 comments
In this case they have only one Twitter account but they delete old posts and start over.
Predicted market correct last 5 days in row https://news.ycombinator.com/item?id=6535057
Predicted market correct last 5 days in row https://news.ycombinator.com/item?id=6535057
Correct, I only have that one account. You can look at the dates on the tweets. They are indeed from the same time as my previous post for 5 days in a row. The only way i could dupe users is by posting both directions and deleting the incorrect one after the end of the trading day. Of course, I'm not doing that.
One way to put things in perspective is to compare against predictor that always chooses the most frequent label (higher in this case).
dynofuz made 14 predictions, 13 correct (at least they claim so) and their predictor agrees with majority label predictor 10/14 = 71% of the time. The difference is just 3 correct predictions.
correction: 13 predictions, 10/13 = 77% same as majority label predictor.
dynofuz made 14 predictions, 13 correct (at least they claim so) and their predictor agrees with majority label predictor 10/14 = 71% of the time. The difference is just 3 correct predictions.
correction: 13 predictions, 10/13 = 77% same as majority label predictor.
the 14th prediction is for today. I didnt want to jump the gun and say it was correct. I agree with the majority label issue you brought up. You'll just have to keep an eye on it for when the market begins turning down.
You have graph from last 3 months in the site and you claim that "We've been correct about 60% of the time." Getting it 10% better than coin toss in upwards trending market is not that hard.
Yeah the 60%+ is over the last 2+yrs. 60% is really not as easy as you'd think as it gives you a 20% edge 60% right - 40% wrong. Also my model's learning everyday and over the last few weeks has been doing pretty well as you can see. I'm working on an account system right now and will soon update the site with some more disclosure about my predictions and historical accuracy.
What's the probability of correctly predicting HIGHER/LOWER 13 days in a row?
(1/2)^13 = 1/8192
(1/2)^13 = 1/8192
Delete the incorrect accounts the next day and repeat for the next 7 days... You'll have one account left at the end, but it'll be exactly right.