Any of you faggots using neural nets for trading?

so you're basically trying to predict the future lol

which algo did you use?

Sure, it's a series of GRUs with dropouts, i've implemented it with keras.

There are some pretty good examples on kaggle using lstms and grus for market data regression tasks, what i'm interested in, now, is doing classification to generate buy/sell/hold signals on a reinforcement learning context.

get out

Yes.

Don't you need a simulation of the dynamics of the system in order to implement a RL algorithm?
How are you planning to tackle that?

Attached: Selection_010.png (736x357, 35K)

does the NN predicting the trend EXACTLY not stand out to you at all?
have you ever heard of the term "overfitting"?

Please for the love of god do not use this as a trading strategy. Save your money.

Before you become some mastermind that can predict the future, you should first learn how to at least label your graphs with matplotlib.

Attached: 1446493314032.jpg (236x279, 18K)

You'd be better off keeping a close eye on the news and using some intuition about what impact events will have on the market as a whole, and buying ETFs for the S&P. The factors driving the stock market aren't in-band

It only saw the orange part, i've trained the network on a wide range of financial data. Here's another chart.

Attached: Screenshot_2.png (1026x769, 76K)

Give it 2/3 of a dataset, see how well it predicts the final third. Show.

What's the timeframe nigger? Days?

Do you really think the stock market is predictable yet none of the crazy rich brokers/investment firms have paid a team of ml programers, each 100x, smarter than you to do this? That would be their greedy-asses ticket to infinite wealth and they would 100% go for it in a heartbeat if it were possible.

Guarantee you cannot predict the market accurately enough over a long enough period to make more money (after taxes and fees) than just buying and holding some index funds.

>thinking you're going to build a model that beats what teams of math/stat/etc phds on wallstreet have, also while having no where near the same access to data they do, and it being their full time job.
If you think you've found a signal, then that means thousands of others also have and that signal ceases to be useful.

doing it right now, will take some time as i don't have a gpu yet.

yes

I think they all have this and better, but that doesn't mean i can't eat some of the pie.

Cool. And really, if there is any granularity where it reliably predicts the value you can use it. IE, if a good portion of the time it correctly predicts one day in advance but no more than that, it is still useful.

They don’t study after study after study over and over again shows so few brokers (

This. 95% of hedge funds underperform the S&P and probably 99% long term. They're literally scams that take 2% upfront and want 20% of any profits with no downside risk.

>I think they all have this and better
Then why can't they beat index funds?

>and the actual price is in blue
I don't see any blue there user. What kind of algo do you have?

guys im making money on bullish periods

Attached: autist.png (1120x840, 178K)

Ok then, go ahead and use this to trade. No one in this thread knows what they are talking about.

Please update us after you have big gains.

Its always priced in. I wish the markets were predictable

So I should just long term invest in index ETFs?

Only if you want to go bankrupt in the great recession of 2019. Its better to have it all in bullion buried in your backyard

Highjacking the thread.
What's a recommended online broker that allows me to interface my buy / sell order with an API?

>orange and green train and test respectively

Wait, your only input is the stock price, as in, no real world data like news or tweets?

yes, the net only input so far is the last 20 previous prices

Attached: Screenshot_1.png (1071x802, 85K)

how fast is all the training with 2-3 1080ti 's ?

Pretty darn fast.

Attached: bfz.png (1280x927, 81K)

That seems dangerous. Looks to me like it's trying to predict the previous increase due to ... well, the previous increase. At the least, have them analyze twitter feeds of executives and the stock buying/selling behavior of CEOs for these companies. I say at least, knowing full well that thats probaby difficult (I don't know how to code,) but that's what a real person who is successful with stocks would do.
After that, teach them how to short sell

Are these guys using neural nets?

en.m.wikipedia.org/wiki/Renaissance_Technologies

Attached: images.jpg (366x137, 5K)