Is it possible to create AI so it can invest money in stock exchange and other types of investment?

Is it possible to create AI so it can invest money in stock exchange and other types of investment?

Yes, great thinking, you could make a profit out of that idea...... if it was the 80s.

They have already been doing that for decades already.

Yes until the market crashes and the ai didn't account for that

Black rock do it.

There already is and many investors use them. They are not as good as an experienced human investor though and are only useful in some situations.
The main reason is that businesses (and the market in general) are still run and sustained by humans and are not purely "logical" entities.
For example your AI won't actively search and look at a news article about a business owner meeting with a shady looking Chinese investor and make any conclusion or further research about what it means for the business.
It also can't make any judgement about whether the new "Iphone 123gorillion" with its microwave-recharging feature is gonna sell or not.

I spend a lot of free time studying the market. I don't even trade, I just have some autistic interest in it I guess. It's really given me a greater appreciation for what I guess is called the philosophy of science or the scientific method because our human mind is really really bad when it comes to looking at raw data your mind starts to play a lot of tricks on you trying to make you think that there is a pattern and that there are an uncanny number of coincidences surely it must mean something, when there really isn't anything there it's just a ghost in your mind.

How the fuck do you think the economy works? It's just AIs interacting and competing. You'd need to make a superior AI to the huge corporations invested in it. Even an individual can't be successful at all with trading/investing without automating a lot of it and designing AI.

They aren't AIs though, they are dumb statistical models. By dumb I don't mean they are not sophisticated, because they are very sophisticated. I mean dumb in the sense that if garbage goes in then garbage comes out so they have to have a team of phds who's full time job is to be constantly vigilant about the data going into their models and make sure the data is clean and scrubbed of any errors.

>They aren't AIs though, they are dumb statistical models.

what do you think AI is?

A intelligent human can discern the difference between and figure out if that's a a bad tick or if the market really just suddenly spiked 100 ticks and gapped. A model will just blindly follow whatever it's black box is. I would presume an artificial intelligence would simulate the human intelligence's ability.

Pic related.
>A intelligent human can discern the difference between and figure out if that's a a bad tick or if the market really just suddenly spiked 100 ticks and gapped. A model will just blindly follow whatever it's black box is. I would presume an artificial intelligence would simulate the human intelligence's ability.
What you're describing isn't anything fundamentally different from how an artificial neural network program would operate. I can say you're "blindly following what your black box is" too (and in fact we most definitely are doing that since what we're aware of that goes on with our cognition is a deeply flawed child's scribble reproduction of the tiniest fraction of what are brains are actually doing during those occasions when we believe "we're" coming up with answers), that doesn't really say much one way or the other about quality.
These programs aren't following explicit instructions, they're generating output based on weighted connections that have evolved over time to be more and more reliable in producing the "right" answers. Better AI in the future will do the same thing but more of it, much like how we're doing the same thing but more of it. There isn't going to be a magical "understanding" formula that gifts machines with the dignity of human souls. Our brains are still network based, it's just that we've had billions of years for an insanely convoluted mess of 100 trillion connections to evolve into place. If you had billions of years to train an artificial program you'd probably end up making some decent gains too.
And if you're concerned about big mistakes, you can just make sure to more heavily penalize with that sort of error in training, much like how we learn not to touch hot lightbulbs because they not only hurt but hurt a tremendous amount, you can make networks learn not to mistake a small fluke for large trend by making their error increase a tremendous amount.

Markets don't follow a program. They are mostly driven by the impulsive and emotional decisions of many people all at once. Then there are the big whales who manipulate and prey on such people, they also have big market effects.

Despite what Joogle or other AI developers tell you, we are a long way of actual AI. No AI can think let alone make abstract decisions. It's still just complex software following Boolean logic to mimic intelligence.
The only predictable part of any market is the 'dead cat bounce'
All the 'cycle theory' and other crap people peddle are bunk

You're saying AI in finance isn't profitable anymore?

No, they can't. 98% of stockbrokers are ridiculously bad at their job, losing money at 2/3 or more of their trades.

They are extremely profitable. But 'AI' is not exactly the right term. It'd be more correct to call it automated trading or algorithmic trading.

AI is a bit of a loaded CS term.

You have to:
>Learn a lot about how the markets work
>Find an inefficiency in the market
>Implement an extremely high performance realtime trading server
>Move close to an exchange so your latency is lower than that of other companies who have direct microwave links
>Implement an algorithm to trade on that inefficiency
>Hope nobody else finds it
>Hope you aren't yourself introducing a new inefficiency that another trading firm with more PhD statisticians and quants can't exploit
>Lose all of your money because of a bug in your code due to bad software engineering practices

Don't bother, no seriously, don't bother. If you are capable of doing this, you are capable of doing this for a trading firm and then you don't have to use your own capital and then you have access to their hardware and get paid a wage and bonuses

Good luck otherwise. You will fail

what ive actually been wondering about is if its possibly to do extremely high frequency micro trading. my idea is to spread out, say $1000, over as many stocks as possible. stock x price rises even a dollar, sell some of it, stock y price raises even 50 cents, sell some of it. and repeatedly do this over a lot of different stocks and keep investing profits into more and more stocks and keep up the micro trading. is it possible?

No it's not that simple, there isn't enough volatility to do this

Also you're not taking into account fees and large swings that go outside your arbitration window

Isn't it like a couple dollars to make a trade?

It's possible in the cryptocurrency market though. It's actually apparently quite easy and I would be working on one if I wasn't involved in other ideas.

intredasting.... im gonna look into this. currently unemployed but have math degree and learning neural networks now. but why is it possible in cryptocurrency? what makes it so?

Let's take it to the extreme.
AIs are used to design products. AIs are used to manage businesses effectively. AIs are used to market the product.
The stock market doesn't go away. AIs become totally dominate over the stock market, and they run it.

What kind of emergent patterns could we begin to see, if any?

>AI is a bit of a loaded CS term.

it is, but the knowledgeable are converging on a group definition of "AI" with little complaint. It's no longer about being a magic consciousness in a box, but rather any sufficiently advanced artificial neural network and algorithms.

Markets fluctuate very highly, obviously.

A bot made over half a million dollars on the Ethereum wave starting in January or some shit. Just sold and bought.

Efficient market theory pretty much states that you cannot forecast future value from previous values, and if it was somehow possible to do the market would readjust so that whatever it is you did is no longer possible, money in the stock market comes from the fact that the index rises some 8% on average per year

>Efficient market theory
except that every single time they investigate this, they find that the market isnt efficient.

its usually a % of the trade, with some minimum amount.

>apparently quite easy
its not. you CAN make money with it easily, but that's because the average value of coins is increasing so much, its extremely difficult to make more money that you would have made just buying and holding 100 different coins.

>A bot made over half a million dollars on the Ethereum wave starting in January
considering that the value of ETH increased from $7 to $350 in that time, just buying and holding would have turned $10k into half a million. (I had $100 and sold it for $3000 with just buy and hold)


If you want to use IA to do this, you will have to make something novel, you cant just download sk-learn, take the price data, shove it onto a neural network or SVM and think the results will be good. Even after weeks of feature engineering and custom algos the best ive ever managed to predict was 56%. its enough to make money, but factoring in trading costs and slippage you end up just breaking even (or at least even with the average of the market).

Yes and it's been used by Decades.
Invest was one of the 1st places where AI was applied.
Too primitive decades ago. But It's improving.
Bankers use AI to get more Sheckels.

That's what Financial Engineering is