I think it would be interesting to analyze the correlation between earnings announcement and the low qpi values- specifically whether, given the low qpi that is the trigger, does excluding the earnings announcement days improve or degrade the results. I know that earnings announce dates with a before open/ after close flag is difficult to source but it would be iworth looking at.
Curious if you are doing more RealTest - this QP indicator I can't figure out how to code in RealTest yet. Would be interested to know if it's possible or if you need to use more sequential/looping like in Amibroker or Python (or a dll)
Here's another idea... the outsized N day drop allows for trade entry. However, trade entry, once allowed, is only done after close > open and close > yesterday's close. This might save you from a dreaded 9-day falling knife. Another interesting one to test is wait for day of unusually low volume (below some average). Typically volume picks up on these multi-day drops, buying after the frantic selling is done seems like a reasonable alternative. Note: in either of the cases above I've seen ones that fail (just continue dumping after a brief bounce) so I'm not expecting "all winners" out of these ideas, but maybe an improvement on win ratio. Another note: the turn-around bar idea pretty well guarantees entry in the right direction. The volume based idea doesn't, super high volume could persist throughout a turn around... but my guess is this is quite rare (that you don't see volume dry up at some favourable point). For what it's worth, I did this sort of thing to "overweight" trend positions. Worked out rather well (generally added to overall performance). It was RSI/BB%, wait for up day, then hold 'til RSI/BB% allowed exit (high enough), wait for RSI/BB% rollover. Initial stop was same as for underlying trend position. I'm sure I could have improved on such a simple thing. Looking forward to future development here. Cheers.
The test takes previous days and calculates the probability of the accumulated drop considering the down days being completely independent, it just takes the accumulated drop. I wonder if it would improve win rate to handle them as a Markov chain so they also provide a "probability of next move" built in
Fun experiment... N is the number of days in the pre-trade distribution, then let P be the number of days in the post-trade distribution. Then you can compare the N vs. P distributions. Not exactly sure how I'd perform the comparison for meaningful analysis, but I think it's a decent starting for some interesting study. Perhaps more fun... let P be the number of days in the post-trade distribution of highs (instead of close) for long trades.
Instead of doing short, can we apply market regime filter (for instance based on ATR or similar)?
Hi Georgiy! Thanks for the idea, I'll try it! Cheers!
I think it would be interesting to analyze the correlation between earnings announcement and the low qpi values- specifically whether, given the low qpi that is the trigger, does excluding the earnings announcement days improve or degrade the results. I know that earnings announce dates with a before open/ after close flag is difficult to source but it would be iworth looking at.
This is a great comment. I believe excluding earnings days will improve the results. Need to check, though...
Hi, this is great - do you have a link to the code for your backtesting infra?
Nope, it's proprietary :)
The code is propietary, but is the backtest software also yours? or do you use any python open library?
Curious if you are doing more RealTest - this QP indicator I can't figure out how to code in RealTest yet. Would be interested to know if it's possible or if you need to use more sequential/looping like in Amibroker or Python (or a dll)
Hi Daniel! I developed it originally using Python... I think the guys at the RealTest forum recreated it for RT:
https://forum.mhptrading.com/t/quantitativo-s-mean-reversion-indicator/4058
thanks mate, onto it :)
Here's another idea... the outsized N day drop allows for trade entry. However, trade entry, once allowed, is only done after close > open and close > yesterday's close. This might save you from a dreaded 9-day falling knife. Another interesting one to test is wait for day of unusually low volume (below some average). Typically volume picks up on these multi-day drops, buying after the frantic selling is done seems like a reasonable alternative. Note: in either of the cases above I've seen ones that fail (just continue dumping after a brief bounce) so I'm not expecting "all winners" out of these ideas, but maybe an improvement on win ratio. Another note: the turn-around bar idea pretty well guarantees entry in the right direction. The volume based idea doesn't, super high volume could persist throughout a turn around... but my guess is this is quite rare (that you don't see volume dry up at some favourable point). For what it's worth, I did this sort of thing to "overweight" trend positions. Worked out rather well (generally added to overall performance). It was RSI/BB%, wait for up day, then hold 'til RSI/BB% allowed exit (high enough), wait for RSI/BB% rollover. Initial stop was same as for underlying trend position. I'm sure I could have improved on such a simple thing. Looking forward to future development here. Cheers.
On your backtest, did you assume limit orders being filled or being filled at the closing price?
The test takes previous days and calculates the probability of the accumulated drop considering the down days being completely independent, it just takes the accumulated drop. I wonder if it would improve win rate to handle them as a Markov chain so they also provide a "probability of next move" built in
Just a niggle... "if we had traded this since 2023" ... think the year is off, just not sure which it should be ??
Fun experiment... N is the number of days in the pre-trade distribution, then let P be the number of days in the post-trade distribution. Then you can compare the N vs. P distributions. Not exactly sure how I'd perform the comparison for meaningful analysis, but I think it's a decent starting for some interesting study. Perhaps more fun... let P be the number of days in the post-trade distribution of highs (instead of close) for long trades.