Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Ask HN: Has anyone used machine learning to trade?

news.ycombinator.com

Post reply on HN
Print view View original

11–20 of 64 posts

Re: Ask HN: Has anyone used machine learning to trade?

#11

Post by dpapathanasiou »

I have friends who do this, and they told me, anecdotally, that for every 10 trades they do, 8 or 9 lose money.

While they do make it up on the rest, you have to ask yourself if you can afford the costs (brokerage commissions and fees) of executing your strategy in that context (they all work for banks and hedge funds).

If you really want to get into this, you might be better off applying to a quant-based hedge fund.

A firm like that will have the capital and brokerage connections to make the game profitable.

Also, take a look at Ed Thorp's web site (http://edwardothorp.com/index.html).

He was one of the original quants and has many interesting articles on the industry.

Re: Ask HN: Has anyone used machine learning to trade?

#12

Post by rw »

haliax wrote: post #3
cperciva wrote: post #2

Is it a complete waste of time to try without being a big fish (I've only got about 10K to venture on this)? Yes. The intraday movement of stocks is small enough that you'd need to be using a large fraction of your capital on every single trade in order to overcome trading costs; and based on the Kelly criterion you'd need to have unrealistically high odds of betting correctly in order to avoid going broke.

Do you think that interday would be more suitable then?

A strategy: run a LOT of simulations, keep running them every day. When they are too good to ignore, go raise some capital.

Re: Ask HN: Has anyone used machine learning to trade?

#14

Post by haliax »

rw wrote: post #12
haliax wrote: post #3

Earlier quoted context omitted.

Do you think that interday would be more suitable then?

A strategy: run a LOT of simulations, keep running them every day. When they are too good to ignore, go raise some capital.

There's a saying in the industry: your strategy will go south as soon as you put real money on it -- you can't raise capital without a track record of real money success.

Re: Ask HN: Has anyone used machine learning to trade?

#16

Post by agbell »

This is not what you are asking but a market with more inefficiencies and less transaction costs is Betfair. There is a simple api and free historical data.

I spent a bunch of time with Neural Nets and SVM and that data. Sadly I didn't find a profitable strategy. I would love to hear about anyone else who is looking in that area.

Re: Ask HN: Has anyone used machine learning to trade?

#17

Post by haliax »

agbell wrote: post #16

This is not what you are asking but a market with more inefficiencies and less transaction costs is Betfair. There is a simple api and free historical data. I spent a bunch of time with Neural Nets and SVM and that data. Sadly I didn't find a profitable strategy. I would love to hear about anyone else who is looking in that area.

What type of analysis did you look into? i.e. what were your input vectors / output?

Re: Ask HN: Has anyone used machine learning to trade?

#19

Post by copernicus »

In general, machine learning would probably not be a great approach to day trading. There are several problems:

1) Limited market data: As an individual, you have limited access to market data. At best, you probably have real time top of book data but without paying for it, you almost certainly aren't getting a real time depth feed. Going off of only last trade data gives you even less to work with. In most cases free quotes are delayed making them worthless to day traders (although not longer term investors).

2) Limitations of market data: Even if you have a lot of historical data with a full order book, that is still of limited utility for simulations. The problem is that everything you do affects the market so any trade you simulate needs to be doing small enough quantity to not make your simulation unreasonable. How to deal with simulating market-affecting trades is a complex problem and the only way to truly know is to run your trades in the market.

3) For day trading, there is a lot of profit to be captured simply from the microstructure of the market (its moves back and forth, etc.). To capture much of this it helps to be fast but you won't be able to colo boxes at an exchange with only 10K.

Machine learning seems more suited to longer term investments and trading on a slower platform than making a quick profit arbing or scalping. You may want to consider this rather than day trading since not only does day trading require much larger amounts of capital, but its also extremely risky.

Also, if you plan to trade equities, enjoy all of the ridiculous SEC regulation (order marking, etc.).

Re: Ask HN: Has anyone used machine learning to trade?

#20

Post by agbell »

haliax wrote: post #17
agbell wrote: post #16

This is not what you are asking but a market with more inefficiencies and less transaction costs is Betfair. There is a simple api and free historical data. I spent a bunch of time with Neural Nets and SVM and that data. Sadly I didn't find a profitable strategy. I would love to hear about anyone else who is looking in that area.

What type of analysis did you look into? i.e. what were your input vectors / output?

I would generate a back or lay output signal for 'Correct score' bets for soccer matches before the game started. Input would be everything you can pull out of the the data set especially including the odds on other more standard bets like the match odds and the over / under odds.

There are also arb opportunities between the various arcane bets, but that involves speed more than ML.

Post reply on HN