Apr-22-2024, 01:33 PM
See the below image. So that's an example of the edge case that I have highlighted. So as you see the price for AMD at this point is in time is hovering around $181.80 - then you see these three prints around $180.80 - so they are late prints that need to be removed.
I actually do just use a shift to compare columns, and that works great where the late print is a singular print. For each stock, I define a certain value that if the price moves greater than X in a single tick, that's a possible late print. So in this case let's use a greater than $0.18 movement in a single tick is a red flag.
I need to check:
1. If the abnormal movement is positive or negative - because if the initial movement is positive, for example, the corresponding movement after the abnormality should be negative. If we have a positive flag followed by a positive flag... that could indicate a very aggressive, but legit, price move so we don't want to filter out that. In the screenshot I provided the initial movement is negative, so the return was positive.
2. The price of the late prints could be the same, or it could be different.
3. If I start using shift, well by how many rows do I shift? This is an example of 3 ticks of late prints, but the actual count is unknown. I've seen as many as 13 late ticks in a row, but that doesn't mean that's 13 is an upper limit, that's just what I have seen.
It's easy to create a filter that filters out just a specific case, but I need to filter out all cases while keeping the false positives as low as possible.
I actually do just use a shift to compare columns, and that works great where the late print is a singular print. For each stock, I define a certain value that if the price moves greater than X in a single tick, that's a possible late print. So in this case let's use a greater than $0.18 movement in a single tick is a red flag.
I need to check:
1. If the abnormal movement is positive or negative - because if the initial movement is positive, for example, the corresponding movement after the abnormality should be negative. If we have a positive flag followed by a positive flag... that could indicate a very aggressive, but legit, price move so we don't want to filter out that. In the screenshot I provided the initial movement is negative, so the return was positive.
2. The price of the late prints could be the same, or it could be different.
3. If I start using shift, well by how many rows do I shift? This is an example of 3 ticks of late prints, but the actual count is unknown. I've seen as many as 13 late ticks in a row, but that doesn't mean that's 13 is an upper limit, that's just what I have seen.
It's easy to create a filter that filters out just a specific case, but I need to filter out all cases while keeping the false positives as low as possible.