S&P 500 E-mini FuturesEducation
Editors' picks
The unknown obvious: there's only one strategy
There's only one trading strategy, one way to trade and +inf number of ways (most of them are senseless) to model it & play around it.
How & why prices move is not a mathematical principle that can be explained with a set of logic.
It's the set of logical principles that can be modeled with mathematics.
These mathematical aka quantitative ways are numerous and generally offer a tradeoff between the computational needs and the resulting quality.
Take a look at my chart, you's see the weighted box plot that includes 80% of the data and weighted mean & standard deviations that also include 80% of the data. They're almost the same! As they should be, since the're modelling the same stuff. Box plot is a lil better since it's non-parametric (works well for all the distributions). WMA & WSTDEV are less on point, but cmon, easier to compute. And then you have the 1st degree model - weighted regression (WLSMA), that for some "unknown" reason sometimes matches the deviations that include 80% of the data?! Hard to compute tho, matrixes, vectorized ops..
Different ways, different tradeoffs, the same end, pick your poison & go.
Thing is it's all modelling, but the real underlying principles are much easier, and strangely, hard to automate 4 real, at least business wise. These things are very hard to algorithmize, and probably impossible to just calculate at all. The principles themselves are easy tho and are the same on any resolution:
1) levels are the places where it was/it is/it will be potentially or proved with evidence as cheap/expensive;
2) everything else in between these levels are buying/selling waves aka directional order flows;
Minding all that, there's only one "strategy" that will let you make the market better & earn money by doing so: buy @ potentially cheap & proved cheap, sell @ potentially expensive & proved expensive. All your momentums & mean reversions etc are ideologically the same things that allow you to do it. Everything else is a question of position sizing and gradual risk loading/offloading.
Now coming back to quantification of all this stuff & automation.
In terms of algorithmization, levels are the nightmare. their origins, positioning, clearing. You'll need to run numerous nested cycles on wide data and query databases nonstop in order to process it all, and you'll need to do it on all the resolutions you use. Waves are even more complicated, they start & end in particular places and levels affect it, they get exhausted & overridden. Wave starts/wave ends are based on levels, sometimes on higher resolutions, recursions are involved. Now imagine you're doing it on multiple assets in business environment. I don't even mentioned many absolutely deterministic & well defined judgmental calls that are made during borderline cases.
You can instead try to approximate it all using mathematics. Since the real original principles will not be reached anyways, the best we can do is to include all the information in our models and pick the formulas & methods that are as much coherent with the source as possible.
Formulas & methods are secondary. Regardless the methods, fancy formulas, what you call ML & AI these days, omg DSP adepts, bloody wavelets and ftts, etc etc etc, you can gain as much information from the data it as it is there.
Information is the main thing. The whole game if about information. Features are inherited from the fundamental particle of the market: a tick. Tho, we more interested in the 'tuple' of the 2 last ticks: current tick and the previous tick (wassup Markov).
1) Price. The actual sampled prices, calculated volume modes of every bar, HLC3, HL2, but never a Close lol;
2) Time. Can estimate the most prominent cycle and divide it by 2 / leave it alone;
3) Sequence matters. May be achieved via linear weighting of the data points;
4) Volume. Weighting by volume/ inferred volume;
5) Direction. Plus or minus? Then multiplied by volume? We might have overshoots due to negative weights tho. Another way?
You'll surely end up with something working.
^^ Funny thing tho, it's all extremely easy to do as an organic life form, you just scroll through different resolutions and see it all on the charts in a matter of seconds w/o any brain damage, without any approximations, without any data loss.
I think at this point you understand that there's absolute zero sense in using any chart studies if you trade 100% manually. If you don't I'm spamming the F button
How & why prices move is not a mathematical principle that can be explained with a set of logic.
It's the set of logical principles that can be modeled with mathematics.
These mathematical aka quantitative ways are numerous and generally offer a tradeoff between the computational needs and the resulting quality.
Take a look at my chart, you's see the weighted box plot that includes 80% of the data and weighted mean & standard deviations that also include 80% of the data. They're almost the same! As they should be, since the're modelling the same stuff. Box plot is a lil better since it's non-parametric (works well for all the distributions). WMA & WSTDEV are less on point, but cmon, easier to compute. And then you have the 1st degree model - weighted regression (WLSMA), that for some "unknown" reason sometimes matches the deviations that include 80% of the data?! Hard to compute tho, matrixes, vectorized ops..
Different ways, different tradeoffs, the same end, pick your poison & go.
Thing is it's all modelling, but the real underlying principles are much easier, and strangely, hard to automate 4 real, at least business wise. These things are very hard to algorithmize, and probably impossible to just calculate at all. The principles themselves are easy tho and are the same on any resolution:
1) levels are the places where it was/it is/it will be potentially or proved with evidence as cheap/expensive;
2) everything else in between these levels are buying/selling waves aka directional order flows;
Minding all that, there's only one "strategy" that will let you make the market better & earn money by doing so: buy @ potentially cheap & proved cheap, sell @ potentially expensive & proved expensive. All your momentums & mean reversions etc are ideologically the same things that allow you to do it. Everything else is a question of position sizing and gradual risk loading/offloading.
Now coming back to quantification of all this stuff & automation.
In terms of algorithmization, levels are the nightmare. their origins, positioning, clearing. You'll need to run numerous nested cycles on wide data and query databases nonstop in order to process it all, and you'll need to do it on all the resolutions you use. Waves are even more complicated, they start & end in particular places and levels affect it, they get exhausted & overridden. Wave starts/wave ends are based on levels, sometimes on higher resolutions, recursions are involved. Now imagine you're doing it on multiple assets in business environment. I don't even mentioned many absolutely deterministic & well defined judgmental calls that are made during borderline cases.
You can instead try to approximate it all using mathematics. Since the real original principles will not be reached anyways, the best we can do is to include all the information in our models and pick the formulas & methods that are as much coherent with the source as possible.
Formulas & methods are secondary. Regardless the methods, fancy formulas, what you call ML & AI these days, omg DSP adepts, bloody wavelets and ftts, etc etc etc, you can gain as much information from the data it as it is there.
Information is the main thing. The whole game if about information. Features are inherited from the fundamental particle of the market: a tick. Tho, we more interested in the 'tuple' of the 2 last ticks: current tick and the previous tick (wassup Markov).
1) Price. The actual sampled prices, calculated volume modes of every bar, HLC3, HL2, but never a Close lol;
2) Time. Can estimate the most prominent cycle and divide it by 2 / leave it alone;
3) Sequence matters. May be achieved via linear weighting of the data points;
4) Volume. Weighting by volume/ inferred volume;
5) Direction. Plus or minus? Then multiplied by volume? We might have overshoots due to negative weights tho. Another way?
You'll surely end up with something working.
^^ Funny thing tho, it's all extremely easy to do as an organic life form, you just scroll through different resolutions and see it all on the charts in a matter of seconds w/o any brain damage, without any approximations, without any data loss.
I think at this point you understand that there's absolute zero sense in using any chart studies if you trade 100% manually. If you don't I'm spamming the F button
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.