So I've done a really poor and simple algo. The idea was that people on the internet influence trades in the market. so if I troll certain data sources I would get some top stock picks. I decided to attempt doing this will penny stocks and see if peoples influence before the start of the day could potentially predict stocks with a high enough yield per day.
Where the journey started. I decided to troll the data from various penny stock subreddits using my raspberry pi and generate a list of stock picks from users and hold on to their context ( I now have data going back about 5-6 months, context being their whole comment). I then would try and see if there was any users that would generate a stock pick from my 6:00am snapshot that would yield over 2%. Lessons learned.
1. online users are not consistent
2. online users are not frequent repeat posters
So the above did not work. Ok….. on to the next test. Does the hive mind generate a positive result based on their popular vote…..ie most penny stock mentions by 6am. This did not work….. at least looking at opening vs close in a day. As I look through the graphs I notice there is a dip every morning at the start of the day. so I adjust my buy/sell time to part way in the morning and before the end of the day. the results are positive hurrah, I'm going to get rich…..(fat chance). As time went out this strategy was not consistent or profitable in the long term. So that leads me to this morning where I just spent a few hours trying to get some ML subject matter analysis happening on comments from users who result in a stock with a high yield vs one that doesn't. 3 hours later I get it all working and find out it has an accuracy rating of 0.26….. and then start laughing like a crazy person. I've built one hell of a machine here that doesn't work and I loved every minute of doing it. Below is my total % gain/loss since the start of this per day.
Submitted November 11, 2020 at 11:24AM by spikyness27