yepher
359 posts
msg #123872
- Ignore yepher |
5/25/2015 12:54:22 PM
Here is a Numberphile that seemed like it could be applied to guessing the direction of the next trading day with 68% chance instead of 50%. Although it maybe possible I can't find a good solution to find a good way to generate K with the appropriate random range..
Numberphile video:
https://www.youtube.com/watch?v=ud_frfkt1t0
My test (in progress):
https://docs.google.com/spreadsheets/d/1K2NBgZOxG0Y66OqT1sVqAaX1vCZ_aUu1AqKWz22xT60/edit?usp=sharing
-- Yepher
|
fotchstecker
304 posts
msg #126387
- Ignore fotchstecker |
12/5/2015 5:39:58 PM
This is very interesting. Are you still working on it?
Also, is there a reason for not using Google Finance functions to pull the last close in?
https://support.google.com/docs/table/25273?hl=en
Thank you.
|
fotchstecker
304 posts
msg #126394
- Ignore fotchstecker |
12/7/2015 8:48:27 AM
You might be interested in this, too:
http://traders.com/files/ADCC-Tables.xlsx
From this month's issue.
|
fotchstecker
304 posts
msg #129348
- Ignore fotchstecker |
6/22/2016 11:11:38 PM
I'd love to revisit this. A few questions:
How are you generating new rows for new dates?
Is there a reason the most recent trading session isn't inserted on that day (after close)?
I tried messing around with the range coefficient but came up empty.
How would you use this to project an idea of the coming session's probabilities?
|
Downed
1 posts
msg #130016
- Ignore Downed
modified |
7/24/2016 12:33:44 AM
Wrote up a script for this in python comparing this vs a few other methods of close-to-close-direction-prediction. This does indeed have an edge with a simple normal distribution using high-to-low mean of the dataset... But it's nothing close to 66.66% (at least I haven't been able to replicate it).
E.g., going long on AAPL every close with 1 day holds till next close with a 6 year backtest has a 51.795% success rate. Using this method going long/short (whatever it signals) has a 52.582% success rate (with 1000 meta-trials over the data to account for the randomness of K). Not insignificant, but probably not worth it either.
Also, more volatile stocks have much higher rates, like 50 long vs 55 this method. Probably because of the larger gaps between data points that was talked about in the video.
Fotchstecker, if you are interested in passing ideas on to me to improve the rates, I'd be happy to implement them and share the script with you.
|
fotchstecker
304 posts
msg #149829
- Ignore fotchstecker |
12/7/2019 11:14:31 AM
Downed, you have only one post ever and I'm replying to our 3-year-old exchange. :/
Any chance you're a lurker and seeing this now?
|