S&P 500 E-mini FuturesEducation
Flawed concepts: returns & log-returns
Considering we are almost in 2k23 (Merry Christmas btw xd), obsession with returns & log-returns is hilarious:
1) It implies sampling, which is info loss by definition, I don't even wanna mention that most don't even bother with proper one-point estimates of dem datapoints, dem just use Close prices;
2) Then the differencing process itself, whatever the lag is, implies info loss, higher the lag - higher the info loss.
I can totally get it people doing it 30 years ago when getting data was a problem, but now?
As I said in "there's only one strategy" post and in other previous post, what you see on your chart is aggregated (binned) trading activity, a real-world process that can be later 'modeled' with math or algorithmized. Ultimately, doesn't matter on which one of some correlated assets it's happening, prices move up when there's a consistent wave of buying volume (buy market orders & aggressively placed bids), and prices move down when there's a consistent wave of selling volume (sell market orders & aggressively placed asks). Gas intakes, like nitro bro! The same principle propagate on every resolution, only the bin sizes are different.
Returns disregard all this, they show you prices magically jumping one time step after another.
Not understanding all this is in modern days is what I call a textbook-scientist/library-guy/paycheck quant syndrome. These people know a lot of stuff, read a lot of stuff, remember a lot of stuff that you can google in 2 minutes, they can calculate fast, remember all the primes for some reason, but can't create anything demselves. And usually, their understanding of a complex system stops on a model they've found in a paper, usually a bad one, but they don't get the principles how & why smth works. They are like these people who have huge muscles they like to stare at and show to others, but when it comes to entering The Octagon it ain't really helps. They don't improvise, adapt and overcome, if something is not in a book, they're stunned. It's all easy to fix tho: open your mind, start trusting yourself and start to use your own head for thinking. Understand that it's all about information, and in most cases every1 has the equal access to the most important info, and if not, the info can be inferred. Everything is connected in the Universe.
If not returns/log returns than what?
VD, volume delta/log volume delta, if the volumes are legit, or at least are "ok", like on BABA stock.
IVD, inferred volume delta/inferred log volume delta, if the actual volume data is incomplete, like on ES futures.
You can google how to calculate volume delta yourself, ima tell you how to infer it, it's simply Close minus Open.
Don't believe? Check for yourself, open a nice vehicle where volumes make sense, calculate both VD and IVD, check dem correlation.
Unlike returns & log returns, Close minus Open not loosing any info, ain't no sampling & differencing happening.
Was it really so hard to figure out? It was in front of your eyes for at least 20 years, Open prices been publicly available for while, and electronic trading platforms started to appear around that time. Oh, if you care for gaps some reason, you can take IVD and add the same logic as is used in so called "True Range", these are the bar sizes that take gaps into account. Use the same logic with IVD.
On the chart you see 4 studies, top-down best to worst. These things suppose to quantify changes in direction/slope/gradient, see for yourself which one worked better, pay attention to the highlighted period marked with vertical lines.
1) Lol, simply a weighted mean on Close minus Open datapoints, one line of code essentially, works exceptionally;
2) A dead simple seasonal differencing, for some reason called "Momentum", one line of code, worse, but super easy to calculate;
3) Non-weighted average over log returns, harder to calculate, but the result is the same as number 2 xd, so even worse;
4) Macd, even more funny stuff goes inside, numerous parameters, smoothed for some reason (smoothing is info-loss by definition), much worse.
The easiest one line of code study worked better, because it simply uses more information.
Just wanna mention it again, if you trade manually you need none of these, you simply look at the charts and your embedded organic neural network sees it all instantly. If your embedded organic neural network doesn't know how to do it, read all other posts, they explain levels & waves, it's all you need.
The whole game is about information.
1) It implies sampling, which is info loss by definition, I don't even wanna mention that most don't even bother with proper one-point estimates of dem datapoints, dem just use Close prices;
2) Then the differencing process itself, whatever the lag is, implies info loss, higher the lag - higher the info loss.
I can totally get it people doing it 30 years ago when getting data was a problem, but now?
As I said in "there's only one strategy" post and in other previous post, what you see on your chart is aggregated (binned) trading activity, a real-world process that can be later 'modeled' with math or algorithmized. Ultimately, doesn't matter on which one of some correlated assets it's happening, prices move up when there's a consistent wave of buying volume (buy market orders & aggressively placed bids), and prices move down when there's a consistent wave of selling volume (sell market orders & aggressively placed asks). Gas intakes, like nitro bro! The same principle propagate on every resolution, only the bin sizes are different.
Returns disregard all this, they show you prices magically jumping one time step after another.
Not understanding all this is in modern days is what I call a textbook-scientist/library-guy/paycheck quant syndrome. These people know a lot of stuff, read a lot of stuff, remember a lot of stuff that you can google in 2 minutes, they can calculate fast, remember all the primes for some reason, but can't create anything demselves. And usually, their understanding of a complex system stops on a model they've found in a paper, usually a bad one, but they don't get the principles how & why smth works. They are like these people who have huge muscles they like to stare at and show to others, but when it comes to entering The Octagon it ain't really helps. They don't improvise, adapt and overcome, if something is not in a book, they're stunned. It's all easy to fix tho: open your mind, start trusting yourself and start to use your own head for thinking. Understand that it's all about information, and in most cases every1 has the equal access to the most important info, and if not, the info can be inferred. Everything is connected in the Universe.
If not returns/log returns than what?
VD, volume delta/log volume delta, if the volumes are legit, or at least are "ok", like on BABA stock.
IVD, inferred volume delta/inferred log volume delta, if the actual volume data is incomplete, like on ES futures.
You can google how to calculate volume delta yourself, ima tell you how to infer it, it's simply Close minus Open.
Don't believe? Check for yourself, open a nice vehicle where volumes make sense, calculate both VD and IVD, check dem correlation.
Unlike returns & log returns, Close minus Open not loosing any info, ain't no sampling & differencing happening.
Was it really so hard to figure out? It was in front of your eyes for at least 20 years, Open prices been publicly available for while, and electronic trading platforms started to appear around that time. Oh, if you care for gaps some reason, you can take IVD and add the same logic as is used in so called "True Range", these are the bar sizes that take gaps into account. Use the same logic with IVD.
On the chart you see 4 studies, top-down best to worst. These things suppose to quantify changes in direction/slope/gradient, see for yourself which one worked better, pay attention to the highlighted period marked with vertical lines.
1) Lol, simply a weighted mean on Close minus Open datapoints, one line of code essentially, works exceptionally;
2) A dead simple seasonal differencing, for some reason called "Momentum", one line of code, worse, but super easy to calculate;
3) Non-weighted average over log returns, harder to calculate, but the result is the same as number 2 xd, so even worse;
4) Macd, even more funny stuff goes inside, numerous parameters, smoothed for some reason (smoothing is info-loss by definition), much worse.
The easiest one line of code study worked better, because it simply uses more information.
Just wanna mention it again, if you trade manually you need none of these, you simply look at the charts and your embedded organic neural network sees it all instantly. If your embedded organic neural network doesn't know how to do it, read all other posts, they explain levels & waves, it's all you need.
The whole game is about information.
Related publications
Disclaimer
The information and publications are not meant to be, and do not constitute, financial, investment, trading, or other types of advice or recommendations supplied or endorsed by TradingView. Read more in the Terms of Use.