TA and algos

[Anonymous No.10841538]

I'm tinkering with a genetic algorithm that evolves buy and sell indicators over time. The indicators are comprised of 'genes' each of which is a basic data check against backtest candles. Highly profitable indicators breed with other highly profitable indicators and spread their genes to the next generation, then some of the genes are mutated, then the cycle commences again.

Most complex TA instruments really just boil down to a series of very basic data checks stacked on top of each other, like "is the moving average of the last 14 periods higher than than the moving average of the past 7 periods". Or etc. The thing is, I am writing this algo precisely I suck dick at manual TA, so I am not necessarily well versed in popular indicators that people use.

For successful TAfags out there, what are your most useful indicators that you employ. I want to take the instrument and look at the kinds of data relationships they are checking on, and incorporate those relationships into my algorithm.

[гeluctant kneepad girl No.10841560]

me most successful algo am trading around Ichti Clouds
Double Cloulds, Tripple Clouds, 2x Time Clouds, ect

these algos most consistently make 1000+ profit but the catch is they are pretty impossible for humans to trade because so many variables

[Anonymous No.10841594]

>>10841538
But how do you deal with the fact that many things that will impact tomorrow's price will only happen tomorrow?

[Anonymous No.10841641]

>>10841538

Most of TA amounts to magical thinking and ex post facto explanations -- it's of limited predictive power and that's the thing which matters most. Moving averages, price momentum and 52-week highs and lows are all useful, however. Basic indicators.

Just setup OTO orders (one triggers the other). Set a trailing stop sell order of 5% or more and combine that with a trailing stop buy order of 5% or more. Adjust your trailing percentage higher or lower depending upon your aims and acceptable window in terms of trading frequency.

[Anonymous No.10841658]

stop larping

[Anonymous No.10841672]

>>10841594
It is operating on the premise that all TA operates on, which is to ignore events/news and to presume that certain patterns in historical data can predict future price movement. It doesn't concern itself with things it can't anticipate. And of course it only needs to be right 51% of the time to make money.

Right now I have it playing with fun money with 5x leverage on bitmex, it is strictly looking at bitcoin patterns and long/short dominance. It's basically break even right now after two weeks and I'd like to hope that some new ways of parsing the data might be the thing it needs to get consistently in the green.

>>10841560
I keep looking at ichi and I'm pretty sure all the components of it are emulable by my bot. It's probably the perfect example of something that looks hopelessly complication but under the hood it is really a series of very simple data relationships being tested. They're just stacked on top of each other until the chart looks like a mess to the uninitiated.

[Anonymous No.10841745]

>>10841658
Not a larp. eg, pic attached is one of the indicators it has shat out that I use live on bitmex at the moment. It consistently evolves towards VERY tight trailing stops, even with large slippage estimated in. It doesn't afraid to stop out 9 times out of 10, if it can ride an upward bart that 10th time it makes all its money back. Unfortunately that's all it's doing at the moment, just breaking even.

[Anonymous No.10841817]

>>10841745
buddy, if youre using a computer algo, you don't need to manually look for good indicators.
just plug in every indicator which ever library package youre using into the pool and use apply GA. it should find the best indicators for you, let the computer do the work.
for parameter tuning/optimization, use something like grad descent or maybe go pull blown NN right outta the package.

[Anonymous No.10841857]

>>10841672
Interesting, could you be bothered to give us a zero-knowledge proof it actually works?

Pick a time interval you are comfortable with (like 90min etc), make a predictions a put them in a pure text file. take md5sum or other hash of the file and use it as a passprhase and encrypt the file with gpg (or similar). Post encrypted trades together with the time frame. Then at specified time post the md5 to decrypt the file so we can see how the trades went. That way there will be no more people accusing you of larping...

As a side: would using popular off-the-shelf prediction algos as an input signal to represent what other tbot-traders are thinking will happen would add anything?

[Anonymous No.10841895]

>>10841745
>if it can ride an upward bart that 10th time it makes all its money back

What do you do, if your stoploss lets you crash down 50% or even more? It happened last year with ETH and could happen every time.

And even more: if you need to rely on abnormal Barts (which you rarely see in non-manipulated markets), why do you try to make money in between?

[Anonymous No.10841909]

>>10841538
No indicator is the best indicator. Maybe Fib retracement and extension. But don't use shit like MACD, RSI and Fisher transform.

[Anonymous No.10841932]

>>10841672
>And of course it only needs to be right 51% of the time to make money.

Less than that if you trade in such a way that your winning trades make more money than your losing trades lose. And you should be trading that way from the start as your strategy is poor if this isn't the case.

[Anonymous No.10841956]

the word you're looking for is 'strategy' not indicator
also GP is better suited for what you're 'doing'.

[Anonymous No.10842253]

>>10841538

Been there, done that. Doesn't work anymore, the barts killed all my algo overnight. They're a clever trick by the whales, they don't consistently occur on some set of indicators. I was doing a consistent 1 - 2% per day previously. Machine learning and such isn't magic, there's no pattern to trade any more, not for automation anyway.

[Anonymous No.10842281]

>>10841817

Tensorflow combined with Scikit-learn is amazing. Recommend it for all bot authors.