At Rel’s suggestion I took a look at what could possibly be happening after my projection level is reached. Perhaps they tend to extend by a preferred amount? This led to me study the specific data more carefully, and I’ve uncovered a lot more about the levels.
First, the extensions, which are based off of the PP base I’m using.
about 50% of moves that go beyond the projection extend only up to 2x (projection+projection). Which is not bad, but the data is showing a steady decease as the bins get larger, indicating that the results are not truly ratio driven.
Next, the actual typical PP levels. Close to 45% of the PP levels are within the 40-60 sweet spot area. There’s also a good chuck that’s 30 pips or under, and knowing that the large majority of days move more than 30 pips in one direction, this causes some issues in the projections. For example, these low thresholds will lead to the direction being hit easily, boosting the HBP/LBP hit count from the earlier post.
the under 30 pip projections make up close to 30% of the data as well.
Skewed direction-hit bias..
Clearly most 30 pip days have at least 1 projection hit, along with an increased both direction hit percentage. Also expectedly, the chance that no pivot is hit decreases as the PP level increases.
All in all, these are the kind of results you would expect if you were to assign a random number as your projection everyday in my opinion. This suggests to me that the levels are not dynamic, or interacting with the metadata, enough. What I’m really interested in is capturing the 40-60 pip days since they are most common, and in combining data I end up with something like 85% 1 direction hit which is good, but a near 24% probability to hit both bounds; close to 1/4 is too high for me. I’d like to see that probability drop down to under 10% if possible. It looks like I won’t be leaving this idea as early as I thought. Looks like I’ll be building my first “panel” to speed up the process and analyze different projections faster.