I've been posting here every couple of months for about a year to get input on a stock prediction algorithm I've been...

I've been posting here every couple of months for about a year to get input on a stock prediction algorithm I've been building. Based on its predictions over the last couple of months, and on simulated data sets, I'm fairly convinced that it is working.

I want to know what I need to do to capitalize the algorithm. As far as I can tell there are two options: I can invest my own money and see where that takes me, or prove that it works and sell the algorithm. I have 200k student loan debt from medical school right now, and I can't even come close to affording the 25k investment floor for day trading, nor do I understand capital gains tax or any of that shit.

So I want to sell it. How do I prove it works without giving away the algorithm, and also without any real money to invest? Would a company be interesting in buying an algorithm if it indeed works? What kind of yearly return does a prediction algorithm need to be achieve to be valuable?

>nor do I understand capital gains tax or any of that shit.
Capital gains isn't really that complex. If you can do med school I'm sure you can figure it out.

Give me some background. What does it trade? Blue Chips? Pennies?
What kind of returns are we looking at?
What's back testing look like?

Trades ~3000 stocks listed on yahoo API. It's not limited to that, its just that I didn't want to build my own price scraper so I used Yahoo. To be considered the list price has to be greater that 5$, and it must have been publicly traded for more than 3 years, though those are changeable parameters. It makes a prediction for every stock listed, and for the 10-20 stocks with the highest certainty of prediction it makes an 'investment'.

Back testing works by initializing to a year ago, so it'd be 3/13/16, makes a prediction for 3/14, then checks how much the indicated stocks changed. It then increments the day, and tracks accuracy over the entire year. Based on 2015-2016 I got a 213% return, and 2016-2017 189%. This, tho, only considers the percentage change in the price, so it assumes equally distributed funds across indicated stocks, doesn't consider that an investment may occupy more than 5% stake in the company, nor does it consider that investing in a stock may actually effect its price. Even still, a 200% return/year is high enough that even if these assumptions are really bad, I'd bet the algorithm is still valuable.

>it worked the last two years
>it will work the next two years

nice meme

It's generalized, so reason it would stop working

>fairly convinced

The point of the thread was to ask how I could convince myself further :)

That's impressive.

Why haven't you borrowed 1mil from some nigga at 30% apr yet?

Because even then they will realize quickly not trading paper that there is a flaw in the system, something they forgot to account for that will fuck them up, have friends that do the same thing and sometimes it's a boom, sometimes they are just breaking even, it depends on the market and other crappy human emotion factors.

Trading is pretty much just gambling with what you assume will happen.

Get in contact with people who have a major in mathematics and physic. Develop additional models together and see how it works out for some more time. Put your own money in it. Work on it together. If you maintain this track record for another two years or so it won't be hard to get your first 10 mil investment capital...

You surely have not found the holy grail though. But you might be on to something. Don't think small but be realistic.

He's right:Can't be convinced it works without putting in funds, but don't wanna put in funds unless I know it works.

I'm not sure I have enough ambition to really pursue that much. I just put this thing together as a hobby.

Backtesting only filters out terrible algorithms. It doesn't actually tell you if the algorithm is good, especially when your backtest environment doesn't account for slippage.
Paper trade that shit. No one in their right mind would give you money for a bot that performs well in a barebones backtest with no other trading. Did you even simulate trading fees?

>backtest environment doesn't account for slippage
At n=250 trading days, slippage should be accounted for on average correct, assuming I'm using daily averages?
>Did you even simulate trading fees?
Why bother if the goal is to sell the algorithm to a brokerage anyway?
>Paper trade that shit
I've started doing that, it's quite slow tho.

>>Did you even simulate trading fees?
>Why bother if the goal is to sell the algorithm to a brokerage anyway?

kek

>Why bother if the goal is to sell the algorithm to a brokerage anyway?
Because if it isn't profitable after trading fees, it isn't profitable period.

Fair point. I'll backtest again with some simulated fees.

This thread confirms me that algorithms will cause more distortions on a financial market and therefor give humans more chances to profit.

People really believe the finance industry will be interested in their "automated stock picking software".

The most successful hedge fund in the world is entirely algorithmic.

dont listen to some of these undewhelming, past low peformance loser. if your algo works, then it works. keep testing it.

some places like quantiacs may interest you, but im sure theyll rip your algo. then, use it themselves elseehere outside of their quantiacs setup.

Every unnecessary complicated system fails on the long run.
I'm not saying algorithm do not work but I'm saying they will never replace humans on the market. As long as there are only a few algorithm they will probably make a profit. But profit is made with a market anomaly and if there are too many algorithms outcompeting each other there will be no profit.

I wish all the best to OP but there are a few things I want to point out.
I'm sure OPs algorithm is based on some fundemental analysis or methods already known to the market player. Methods already researched by others. Correct me if I'm wrong OP.
You tested/backtested your algorithm in a bull market. How is the performance with the opposite? Trading sideways?
There is a huge difference if you trade with 1 million or 1 billion or 100 billions. One trade of your algorithm could trigger a market reaction if enough money is in play and how the market reacts is a big question mark.
What are your exit strategies?
How does your algorithm perform in crazy market situations? Black Monday. Lehman Brother bankruptcy.

>People really believe the finance industry will be interested in their "automated stock picking software"
I don't believe anything, I just put this together as a hobby, and wanted to know how to tell if it is useful. It's also still a work in progress.

>Methods already researched by others
Sort of true, but not on stocks. I worked in a machine learning lab in undergrad, and spliced together 2 algorithms used for genome assembly, and modified to handle a different data type. haven't seen anything else like it in finance.

>There is a huge difference if you trade with 1 million or 1 billion or 100 billions
this is true, but no real way to test that kind of thing without the actual cash

>What are your exit strategies?
>How does your algorithm perform in crazy market situations? Black Monday. Lehman Brother bankruptcy.
No clue :^)

I want to emphasize that I don't really know what I'm doing if that wasn't already obvious enough

Most people working in finance have no clue what they are doing.

My advice. Keep working on it as a hobby. Maybe make your trades/performance public.
If your algorithm continues being profitable I can assure you people will contact you.

Algos have been evolving since they were first applied to markets. The money to be had back then is all but gone. The one's working today (at the big guys and at smaller established firms) will rip you to shreds like piranhas, even if you manage build a "very good system".

If it is a hobby then it's a hobby and you may learn a lot. Big data is a good field to get into.

Bruh sometimes you gotta just wing it. I had an algo that did remarkably well in backtests so I gambled my savings on it in live trading with a hardstop of losing 25% and its been providing me handsome returns so far.

>Maybe make your trades/performance public
That probably the route to go.

I'll shore up the back testing to include transaction fees, and penalize to account for any market reaction to the investment. Then test in some bear markets

Thanks hombres

> I have a model that picks winning stocks!
> Isn't rich

hahaha.

Okay man.

I'm a great gambler too. I can pick the winner of every pro football and pro basketball game. IDK why I'm not rich though. So strange.