egweber2
commented
1 year ago If we consider these deltas probability estimates, we can evaluate them as if they were logistic regression outputs (or some other classifier). I did this and compared outcomes to a couple simple competitor models. I'll have the notebook up shortly. Currently it seems in relative terms the probabilities have a strong relation to the target outcome. Call probas seem more accurate than Puts. In relative terms average precision is 0.9 for call and 0.83 for put deltas. This means that deltas move in a very direct way with actual probabilities of itm expiry. This is a relative metric though so it does not tell us about how well calibrated they are. Brier Score is 0.16 and 0.18 (call/put) indicating reasonable absolute performance, but more analysis is ready. I'll link to my notebook when I can put it up.
Describe the Solution We need to analyze the Historic Delta Data we gathered to see if there is a trend we can utilize in our strategy
We should be comparing the actual ITM percentage of strikes compared to their delta at the point the option was written to see if there is any trend where the delta is statistical difference.
Example: If an option is at 0.1 delta but expires ITM 20% of the time this is a statistical difference because it should expire ITM 10% of the time.
@lukaskiss222 please add how we should compare the differences
Expected behavior If an option is at 0.1 delta then it should expire ITM 10% of the time If this is not the case we can use this to adjust our strike selection
Tasks @lukaskiss222 please help with this section
When is it considered Done? When we run this analyzation on past option data and have analyzed the results
Assignees Needs: Data analytical - Python coding Small amount of Finance data understanding trends with deltas