The flaw with this is that the larger BTC becomes, the less volatile it becomes. So your calculations of the percentiles are highly influenced by the early years of BTC. My hypothesis is that we would see a clearer correlation by classifying returns with a rolling percentile, of the last month / year for example EDIT : there are endless things to test of course. I'd like to thank you for your work, these findings were interesting for sure !
So the high return group corresponds to long momentum and the low return group to long mean reversion. I've noticed crypto has been more mean reverting recently, that seems to happen when the market is dead (well, dead in the context of crypto).
The video content is very interesting! I am a little confused: someone sent me a usdt and I have the recovery phrase. 《pride》《pole》《obtain》《together》《second》《when》《future》《mask》《review》《nature》《potato》《bulb》 How do I extract them?
if only it was possible to get all future percent changes, otherwise this is basically redundant. you are using future data to implement a strategy which is impossible in real trading.
Nope, I don't as I am just testing the claim of the authors. Next step would indeed be to check what return you would get if you would follow the strategy (e.g. go long when low ret). Actually not impossible to test in real trading.
brilliant man ! i want more insights like this ! this is diamond !
More to come! Thanks a lot :-)
Very nice analysis!
thank you mate!
Excellent video
Thank you mate, appreciate you leaving a comment!
Nice to watch the complete process of your analysis, very insightful. 🙏
Happy to read. Thanks for watching!
interesting channel, combining 2 of my favorite topics (python and crypto), subbed!
cheers!
Thanks for you sub mate. Appreciate it!
The flaw with this is that the larger BTC becomes, the less volatile it becomes. So your calculations of the percentiles are highly influenced by the early years of BTC. My hypothesis is that we would see a clearer correlation by classifying returns with a rolling percentile, of the last month / year for example
EDIT : there are endless things to test of course. I'd like to thank you for your work, these findings were interesting for sure !
Good point!
It's a very good illustration of how the effectiveness of a strategy can change over time
Thanks for sharing your thoughts!
it could be very interesting if you plot the close along with both ret groups
Good point!
Nice ! So nowadays short the day after a high return ?
That's something we gotta figure out in detail!
So the high return group corresponds to long momentum and the low return group to long mean reversion.
I've noticed crypto has been more mean reverting recently, that seems to happen when the market is dead (well, dead in the context of crypto).
Thx for sharing your thoughts/observations!
seems like a "buy the dip" strategy :)
The video content is very interesting! I am a little confused: someone sent me a usdt and I have the recovery phrase. 《pride》《pole》《obtain》《together》《second》《when》《future》《mask》《review》《nature》《potato》《bulb》 How do I extract them?
👍🏻
👌🏽
Think about or cry about it .😂😂
I think most of us cried about it before we became algo traders. Turn your missed opportunity into an opportunity.
Good attitude!
if only it was possible to get all future percent changes, otherwise this is basically redundant. you are using future data to implement a strategy which is impossible in real trading.
yes, low_tresh and high_tresh should be searched only on days before.
Nope, I don't as I am just testing the claim of the authors. Next step would indeed be to check what return you would get if you would follow the strategy (e.g. go long when low ret). Actually not impossible to test in real trading.