[ 3 / biz / cgl / ck / diy / fa / ic / jp / lit / sci / vr / vt ] [ index / top / reports ] [ become a patron ] [ status ]
2023-11: Warosu is now out of extended maintenance.

/biz/ - Business & Finance

View post   
File: 43 KB, 1366x730, upload.png [View same] [iqdb] [saucenao] [google]
18594531 No.18594531 [Reply] [Original]

welcome to my trading desktop

>> No.18594594

>>18594531
>R
>a linear regression
never gonna make it

>> No.18594596

>>18594531
as a freshman maths computer science major, is algorithm trading something worth looking into?

Seems kinda memeish?
Also is it fiat youre trading or boomer stocks?

>> No.18594608

>>18594594
and why is that?

>> No.18594612

>>18594531
No one cares fucking freshman

>> No.18594629 [DELETED] 

>>18594612
be at a certain point in my career that i cannot control. Get insulted by someone who probably has/wont ever get a degree from a worthwhile school

>> No.18594653 [DELETED] 

>>18594629
You are a dime a dozen midwit trying to flex a fucking linear regression in R
At least use matlab/octave so it looks nice

>> No.18594660

>>18594531
OK, and what's the regression you are plotting? Could be anything, how is it trading related?

>> No.18594662 [DELETED] 

>>18594653
im not OP teehee idiot learn how to use 4chan

>> No.18594667

>>18594531
My eyes. At least use r studio.

>> No.18594674

>>18594653
i am considering looking at octave

>> No.18594686

>>18594596
nothing prevents you from trying something and running simulations on historic data. Then you check manually if the bot is not doing anything stupid, and you might have something nice.

I did that some years ago, had some good results but did not finish like the retard I am

>> No.18594693

OP is a faggot

>> No.18594698

>>18594608
Study.

>>18594596
It's a meme, they are all crashing. No matter what they tell you, most of the "techniques" are just trying to prop up momentum or trying to catch mean reverting trends.
There's potential for nice things, but you'd have to be in an elite group like renaissance tech to be a part of those efforts.

>> No.18594699

>>18594662
Shiiiet i jumped the gun there
>>18594674
I find octaves syntax less annoying

>> No.18594715

>>18594699
delete you posts if you want, no hard feelings. We've kinda ruined this thread

>> No.18594738

>>18594698
Also, you have two approaches:
1) an economist has a hypothesis and an algorithm to take advantage of it -> runs backtest and hope for the best
2) you actually know about mathematics and you develop a model based on the probabilistic behaviour you're seeing ---> backtest

Also, you need 10 years of time series data to properly backtest, backtesting is NOT something really simple and clear and it most definitely doesn't guarantee you anything.
Basic premise about fin markets is that they are non-ergodic, which imples that past behaviour tells you nothing about future behaviour, so everyone is kind of playing it blind. You might spot that something is happening but you need to be lucky to be able to exploit it while it's still happening (could be a week or years).

>> No.18594743

>>18594715
>>18594699
Why can't we ever have nice things on /biz? Guys locking antlers all the time, that's why.

>> No.18594832

>>18594743
Now I feel bad

>> No.18594839

>>18594738
>Also, you need 10 years of time series data to properly backtest[..]

>Basic premise about fin markets is that they are non-ergodic, which imples that past behaviour tells you nothing about future behaviour,

I don't get this. If markets tend to morph to eat up any alpha a given approach has, then why do you need 10y of backtesting data? Seems like less data would be an advantage here, as the old data would just lead you to false conclusions?

>> No.18594927

>>18594839
I basically said two contradictory things, and they are, because they are based on two different assumptions:

1st: assume there are consistent behaviours in markets --> then basic statistics tells you that you need 10 years of daily trading data to be able to even try to properly validate your hypothesis

2nd: assume markets are non-ergodic --> there is no way that studying the past will ever give you any insight into the future (no time line, not even last week) so everything is astrology

Reality is probably somewhere in between. But you have a lot of people that work based on the 1st assumption, and a lot of people that just try to quickly find exploits. Honestly, I lean into the 2nd assumption, and that's the reason why I have only trade based on gut instinct (and have done reasonably well so far). I am actually studying some weird maths because I want to try to come up with stuff to see if backtesting yields anything, but it's mainly for fun, because I fundamentally believe in the 2nd hypothesis.

>> No.18594947

>>18594531
Fuck mate, check your residuals, I reckon you can fit something

>> No.18595051

>>18594927
I am just dabbling in algos, but I believe that

1) there are immature markets like crypto where it might work better as the quants with their HTF trading haven't been that prominent

2) On a high-timeframe, even things like mean reversion will work on sufficiently large shocks. E.g. check the latest corona dump. A mean reversion algo pouncing on the drop would have done good. Whether that's better than just buying funds, IDK, as sufficiently large shocks are rare and you'd just sit in half money half stocks most of the time.

>> No.18595111

>>18594927
HFT works so there must be a way to scalp gains over a large number of trades
Retail investors don't have access to the tools/connections that are probably necessary for this however

>> No.18595130

>>18595051
Man, I meant the quant's high frequency trading algos, not high time frame, of course. Sorry

>> No.18595133

>>18595051
>>18595111
HTF is mostly just making money of the bid/ask spread, or small momentum pushes. for things to go at that speed there cannot be anything smart going on.

your intuition about those things might say that they anecdotally might work, but scientifically you cannot get a validation for that, so it's still only your intuition in the end telling you "it will work".

Also, when you compare most of those strategies with buy and hold the sp500, most of the times you find out they underperform it, so it was only an illusion that they were making money. Also, hindsight is 20/20, and you cannot magically guess when to activate and deactivate your strategy so that it only trades when there's going to be profit

>> No.18595248

>>18595133
to add to this, the most common mistake of designing trading strategies is to see what happened in the past and assume it's going to happen again. This is basically over-fitting, as you are designing your strategy based on the training data itself, it's the most amateurish thing possible. You need at least to cross validate it. Overfitted strategies will backtest amazing profits, but as soon as run them in real life you will get nothing and everything you get will be pure luck.
The analogy is this: you check that the lottery winners for the past two months all started with either 3 or 5, so you bet that it will happen again. Truthfully, you will get fucked.

>> No.18595262
File: 26 KB, 1536x746, eur-usd.png [View same] [iqdb] [saucenao] [google]
18595262

>>18594531
You are like a little baby, watch this. Buy when lower window indicator goes up, sell when it goes down.

>> No.18595289

>>18595262
did you seriously just fit a sine wave?
that's really stupid and ingenious at the same time

>> No.18595320

>>18595262
Why the amplitude modulation if you are just fucking with OP by showing a sine wave?

>> No.18595330

>>18595289
>>18595320
Anon please, it is not a sine, I applied custom band pass filter to the time series to reduce noise and to reveal hidden market cycles.

>> No.18595335

>>18595051
> I believe that
> there are immature markets like crypto where it might work better as the quants with their HTF trading haven't been that prominent
You believe wrong. Unless you're talking about micro mini scp shitcoins on scam exchanges

>> No.18595343

>>18595330
Ok, that was really creative. But why a band pass and not just a low pass?

>> No.18595360

>>18595330
Also, some of the wave cycles do not match, so I'm not 100% convinced it was a filter and not and amplitude modulation fit.

>> No.18595438

>>18595343
You need to get rid of low frequency trend component. You can use it too, but I've found out that there are a big delay.

>>18595360
Of course it's not 100%, that would be a holy grail. Markets are not ideal. Still you can make profit and predict movement with good probability.

>> No.18595488

For all of you intersted in quntitative trading I recommend this guys blog https://imetricablog.com/.. He also has github with some juicy apps, I personally use dumbed down version of his mdfa method for R.

>> No.18595503

>>18595438
The reason I was complaining about wave cycles not matching wasn't because of model being god or bad, but because a filter doesn't work like that. A filter shifts the phase but it doesn't change the frequency that's leftover, so there should not be big peaks that do not match. But this is just visually, so it's gut feeling saying that it looks fishy.

>> No.18595540

>>18595330
>hidden market cycles
Is this the new fibonacci

>> No.18595583

>>18594738
If I knew how to computer program, I would want a program that tells me which stock/company has the highest positive publicity and overall publicity. That would factor in with heavy weight.

>> No.18595602
File: 49 KB, 584x450, grug.jpg [View same] [iqdb] [saucenao] [google]
18595602

tfw no idea what anyone is talking about

>> No.18595652

>>18595602
Buy dwsh. The market is trending down.

>> No.18595659

>>18595583
NLP is still very dumb but there already funds trying to use it. Basically doing webscraping for news and taking out the text to get a general overlook score. What those methods forget is that news =/= traders and news are biased as well.

>> No.18595660

>>18595602
It's just a bunch of people trying to use math to make good trading decisions. Giant institutions can't do it, but these tards think they can. Only retards think modeling can be applied directly and it will magically work.

>> No.18595801

>>18595660
giant institutions can't market sell like a retard can with a 100 lignes python script trading bot managing 1000$ of btc

>> No.18595820

>>18595660
these threads always make me a bit sad since I can code but know almost nothing about statistics

>> No.18595849

>>18595660
the bigger your positions, the more likely it is that every trade you make fucks up your assumptions
it's easier for an individual trader with 100k to manage to make 40% than for a giant hedge fund

also, some of us actually argued that there's no way to prove that you can have consistent profit

>> No.18595920

>not using superior MATLAB
kek

>> No.18595926

>>18595801
thousands of guys using the same data and the same statistical models to place a single trade is the same thing as one guy placing thousands of trades.

Otherwise, financial institutions could simply divide their stack over thousands of traders, right? (which they do btw)

>> No.18595990

>>18595262
This is only like 50% accurate bro. Just count. Not even better than guessing lmao.

>> No.18596016

>>18595920
>using a proprietary language

>> No.18596041

>>18595926
I may not know what a low pass is
but I can tell you with a 100% certainty that any algorithm posted here is gonna be less than useless, so there's no "thousands of guys using the same data and the same statistical models", and if there are, they're not making profit. An institution will have to evaluate its models after every significant trade it makes, or create a chain of trades that rely on the previous trades having the predicted outcome. The former limits their profits, the latter is beyond the comprehension of a single person.

>> No.18596046

>>18595820
Same here. I can code day in day out but I know shit about statistics or signal theory. And as a 9-5 wagecuck with family I don't have the time it feels.

>> No.18596063

>>18596046
>And as a 9-5 wagecuck with family I don't have the time it feels.
I can relate
I play video games and shitpost on biz all day, don't have time to study numbers n shiet

>> No.18596076

>>18596041
>but I can tell you with a 100% certainty that any algorithm posted here is gonna be less than useless
And yet there are people out there who have coded up algos that make them profit over the long run. Those "I can with the authority of laziness tell you with 100% certainty it can't work" smells like sour grapes.

>> No.18596107

>>18596076
>And yet there are people out there who have coded up algos that make them profit over the long run
Did I say there aren't?
Those people know that they shouldn't post their algos publicly
are you projecting your insecurities friend?

>> No.18596114

>>18596016
Just use octave or get a license from your school/job bro

>> No.18596148

>>18596114
I used mathlab in Uni for a few courses, but I use python whenever I have the ability to choose.

By the way, Matlab has mistake sin some formulas, and you cannot edit them. I've detected the same in python libs but you can correct those.

>> No.18596169

>>18596107
They don't post their algos, obviously. Which is just more maddening.

It's not my insecurity, but the talking down on others who might or might not have what it takes to get there, always riles me up. Yes, 99% will end up losing interest or money or both. But some will come out with something that works, is all.

>> No.18596180

>>18596148
>Matlab has mistake sin some formulas,
wait, what?

>> No.18596194
File: 34 KB, 438x421, 1587407748262.jpg [View same] [iqdb] [saucenao] [google]
18596194

>>18596148
>use python
>By the way, Matlab has mistake sin some formulas, and you cannot edit them. I've detected the same in python libs but you can correct those.

yeah ok pajeet

>> No.18596195

>>18596169
>maddening
you're upset that people want to protect their source of income?
Are you some kind of communist?

>> No.18596199

>>18596169
I don't have algos that work, I only trade on gut instinct, but I am studying some stuff to develop a framework for trading.; not sure if it will work or not.
However, I assume that if you manage to find something that works you should disclose it, that's 101 of trading, if you want to make money you don't disclose your method unless you are getting something equivalent in return.

>>18596180
Yes, like all products, sometimes there are bugs. The finmath libraries had mistakes, I remember that clearly.

>> No.18596209

>>18596195
No, I am upset when people tell other people to not even try something because "they know 100% it doesn't work"

>> No.18596215

>>18596194
I am not a pajeet, I just come from a scientific background so python is good because it's dimwit tier in terms of programming, but it's still more flexible that fucking matlab (which is retard tier).

>> No.18596229

>>18596209
Personally, I just shoot down the people that proudly proclaim that they are SURE this will make money. I think it's definitely possible to make money, but it's 100% impossible to know beforehand.

>> No.18596249

>>18596209
I didn't say that
I suggested that taking any algo or method that is posted here as-is, will NOT make anyone money

>> No.18596272

>>18596229
Which is fine and actually the right mindset. But then I wasn't talking about you. You are more the "tough kind of love" guy stripping people of illusions. What I found unnerving was >>18596041, who is the same kind of cocky but in the negative sense. But w/e

>> No.18596289
File: 278 KB, 2453x1309, resultado XMR ETH.png [View same] [iqdb] [saucenao] [google]
18596289

>>18594531
yeah welcome to mine
I use matlab

>> No.18596322

>>18594531
R is gay
use python3 and stop wasting your time

>> No.18596506
File: 140 KB, 1391x757, chrome_a6jKYoWskO.png [View same] [iqdb] [saucenao] [google]
18596506

my volatility algo on HA candles

>> No.18596724

>>18596506
heikin-ashi? i have no idea what that is, i'm going to read a bit about it

>> No.18597199

>>18596506
Looks like it's a bit late in selling? Buying seems mostly OK, but apart from the huge drop, it mostly shorted too late.