A few things to note:
1. This is a free service!
2. Some of you may remember this site from a previous post of mine. I got a lot of great feedback that helped shape tons of dev work since then, and am looking for more feedback now.
3. I just finished out the spam report system as well, which will be greatly useful in protecting the integrity of the data as the system moves forward. Essentially, as users interact with the system and report irrelevant data, the system will learn and get better at determining what it worth taking into account for each security. My hope is that some beta users can help me make the system smarter.
Website: Quikfo.com
Here's a poster for my thesis on this subject which sort of fuels a lot of the backend:
https://i.imgur.com/RVZxDTP.png
Basically, I made a system which tracks article headlines & social media posts for the Russell 3000 and some top market-cap cryptos, then keeps track of word association vs next-day performance, and finally uses Bayesian classification to come up with a score for each asset or security, based upon its current headlines, which indicates the likelihood that the price will rise or fall over the next day.
You can also follow the system on Twitter @myQuikfo, it posts long and short picks every morning at ~8:30am. Please keep in mind this whole thing is in beta though, so when an aspect of my system crashes shit can hit the fan and the twitter may post nonsense.
The radius of the bubble is the frequency of occurrence of a word, the hue of the bubble is the positive or negative severity (green being good, red being bad).
The system works quite well so far. Looking back on the SP500, for example, the optimal trading strategy has been to buy at market open and sell at market close anything which is scored >=85. This strategy would have returned ~13.3% trading from 12/1/17 to 2/28/18.
I'll be building out a full API for developers who want to use this data themselves. An early version is available on the site now with some brief documentation. Please feel free to make use of it and let me know what more you would like to see added to it!
And yes, I'm aware that some of the positive/negative word associations seem crazy. How can a good word be bad or a bad word be good, one might ask. Well all this does is indicate that when a word appears, it more often than not leads to a specific market movement. One could notice that 'Fraud' is particularly green...perhaps this is because when everyone is talking about fraud, then it's already priced in and may have found a bottom. The question of why can't really be answered by the system, that's more up to the inferences of the user.
Looking for feedback from the community, please let me know what you think. There is tons of data being collected that I haven't even gotten around to building a front-end interface for. The possibilities are endless and I would love to tailor the project based upon the direction that this, and other communities, would like to see.
Thank you!
Submitted April 05, 2018 at 10:59AM by TheLoneDonut https://ift.tt/2H1m0IB
No comments:
Post a Comment