Bitcoin Forum

I was just thinking this would be a cool idea to make. It would be in Python. I might do this, if just for fun.

I like the idea. Depending on the options and features that you would put into the bot, it could act to stabilize the currency.

Most likely, many people using a similar bot would destabilize it. If all the bots think "sell" or "buy" at once, the swings just get bigger.

Bot have to be custom made or are useless.

the trading platform could be opensource, then people would just write their own automated trading bots

What makes you say that? As I understand it, many wall street bots use the same formulas. And if the open source package was very customizable it could be tailored to your needs. I don't see the rational behind such a blanket statement.

Ok, I'm definitely going to work on this. If someone could make a brief list of simple features to start with, that would be helpful. It will start out as pure CLI. Eventually I think it would be cool to tweak different custom options and have mock visual graphs of what would happen if enabled, with the trade data that took place.

Python is better, I'm going to stick with it. The "php library" for Mtgox is a PHP page that is rendered by the server into plain text or JSON. It isn't PHP code, I believe.

I am going to make it from scratch. That PHP code doesn't do much except access the Mtgox API.

Please do this, if it catches on, I'll have a blast writing a counter bot to take advantage of the predictable buying and selling patterns :)

Because the bot programming itself is part of the strategy. If everyone uses the same bot, everyone uses the same algos with just some tuned parameters...
Guys at Wallstreet don't use the same autotraders, they pay IT guys to develop tailor made strategies with quants.

salaries for automated trading programmers are like 100k-300k

what i've seen is a platform that interfaces with exchanges, created in either c++ or java. then people can write trading bots in c++ or java and execute them on the platform.

look into stock or forex AT platforms.

I will keep you guys posted.

I'd definitely like to see such a bot in place.

I'm thinking of making it so you can define lots of custom rules.

Like "If X indicator is above this, then do Y". Or "wait until time > Saturday morning, then execute this different behavior".

So you tell the Bot how you want it to act, and you can also simulate it on past market data. Everyone has different amounts to invest and different levels of risk, and they might have beliefs of where the price will go that are not just pure mathematics, so one algorithm wouldn't fit all.

Absolutely.  1 algorithm would just mean that whichever bot was able to initiate the trade first wins.  It might even come down to ping times to MtGox/Tradehill.  Customizable algorithms, with NO defaults, would be the best route to go.

I'm a huge believer in free software myself, but this is a bad idea.
Yes, a general toolkit library would be useful (and they exist already - see the python mtgox modules on github), but a generic autotrader bot would not be profitable once it reaches a certain mass, and while people could modify it, in practice most won't.

Much better would be a good framework combining the mtgox+tradehill+whatever API with a selection of statistical and machine learning algorithms and then not defining ANY strategies.

I was working on a trading bot before the Mt. Gox disaster.  My biggest problem was error handling and recovery... timeouts, partial fills, etc.  If you do a toolkit, that would be the most valuable part.

The Mt. Gox API, the trading algorithm... relatively easy compared to error handling.

(Since Mt. Gox came back online, prices have been relatively flat and my algorithm doesn't make many trades, so I haven't resumed work on the bot.)

Actually the algo IS the hardest past in a bot. It took me three hours to code the bot and 2 months to create the algo.

Yes, but error handling would be fairly hard too. I can see what Olomana was getting at.

edit: I have actually wanted to create a trading bot a while now, but if I did it would be one of the only projects I don't release as open source (for several reasons). The way I wanted to do it is develop some PHP scripts which are actually controlled (triggered) by a desktop application (in Delphi 7) which also allows you to heavily tweak the configuration of the bot, display important graphs, and all sorts of other nifty features.

looks like this idea will kick off a huge number of spam threads,

for the noobs, dont use an open source or any free to use market bot, they are designed so that the person/group who made it can trade against what they built/released.

ie all u noobs get a buy signal and at the same time the bot maker has a different bot/algorithm which is selling

trade safe.. dont be taken in by scams

That's why you use a bot with lots of parameters that can be set by the user.

I suggest that you are making this way more complicated than it needs to be.  You've put 100 times as much effort into the algorithm as I did.  No doubt your algorithm is better than mine, but I rather doubt it's 100 times better. 

It sounds like a great idea. Something i'd really like to see.

This is probably why quants are paid +£100K in London ;-)

Look at this :

The next version is already finished so if you interested just pm me.

http://forum.bitcoin.org/index.php?topic=25658.0

i would tip anyone for a good bot, best with gui..

A bot can easily be exploited, all you need to do is emulate conditions to where it will sell and buy, if you had enough btc you could probably just take their coins in under 5 min, assuming no safeguards were in place like a max BTCn/hr max out.

No, no, no!  You can trick a bot with a convex strategy (chasing momentum, for example).  You can't trick a bot with a concave strategy (rebalancing, for example).

You are right. We can certainly exploit some bots, because they all have flaws (which are tactical choices) but there is not a "one way for all" in doing this. So a lot of bot on the market would probably have an effect on the market, increasing the activity (whether good or not) and the better bots could take advantage of the weakest. Not in an intended manner, but just because they would be quickest to react etc. 

would it ever be profitable to rent a box with lower latency to and from the trade server?

This is what they name collocation and it's very frequent at wall street where you can rent a slot in the same data center as the exchange's ones. Regarding bitcoin, it's not really that necessary as mtgox for example is very slow to answer to queries...

Hey Guys,

I was thinking about open sourcing my trading bot too. Actually I already wrote quite a few.

I have attached you the source of a promising raw model that actually works. You can fine tune its variables (login/pw for MtGox) in the header of this code. Below that you dont have to change anything at all. It was written in PHP and you can just launch it from console. (just type "php ./ob.php" in the same path as the script resides on your linux shell)

What it does is the following:

It reads the market depth at mtgox and weights open bids/asks by their amount and distance to current market price (The closer they are to the market price the less likely they will be changed shortly - why? Because the person having placed those orders is at a high risk of getting filled, so its unlikely that its actually bogus order). After weighting the open orders it decides if there is much higher supply than demand (or vice versa), opens a position and holds it until market conditions have changed. So far so good.

Now for the bad part:

The fees at Mt. Gox are too high for very short term trading strategies like "scalping". The fee structure of 0.3% per trade sounds fair but what would be more attractive is a fee that is charged on actual profits. This would enable us to use many short term trading strategies that actually just dont work yet. In real trading you can make a fortune of a strategy that beats the market by just 1%. You don't need to have a high "edge" as many people commonly believe. The problem is this: MtGox eats 0.6% of each buy and sell (= round) which makes the odds much worse for us. In professional trading you normally have a broker with a more attractive fee structure OR you focus on long term trading, what is much harder in bitcoin. Why? Because a reliable history of quotes for professional time series analysis does not exist and conducting a long term prognosis for the whole bitcoin project is very hard at present. Since I dont have a 0% account I cannot investigate more on the short term possiblities.

If anyone is willing to let me test my more advanced scalping strategies on his 0% account I would be willing to share any profits with this person.

If you liked my code and want me to release more please consider a donation:
14o557CreqAimfskDyMwKZMfe2U6Pi3i9q

Download version 0.1:
http://www.megaupload.com/?d=EMI6H946

Notice: The script is designed to be run from the console of a *NIX system, it will output its status messages while working. It cannot simply be uploaded to a webhost running php and opened from a web browser. You have to use the console (ssh login) to start the script or run it from a local *NIX box.

Please let me know your results,

Thanks :)

Thanks for this r4ain. I just tipped you 0.5 because you learned me how to use JSON in php  ;)

You guys should all tip as well it's really generous from him.

Oh wow thanks, I wasn't expecting any positive comments/donations so soon :)

Actually the bot itself is just a proof of concept. But it works already, however some error messages are still not implemented yet.

Basically in all time series analysis problems the implementation is least problematic.  I'd focus discussion more on strategies rather than implementation problems.

Since bitcoin is a highly inefficient market at present, the possibility of arbitraging it should be there. Also, if there was a market offering reasonable liquitity with a better cost structure for short term traders, alot more strategies would suddenly make sense to be implemented.

Actually I didn't test your bot, I have already mine which is working pretty well. But it's still very interesting to compare with how you did things. And as I don't share mine I am happy to tip lol.

Im getting this:


MtGox down. Waiting...
sh: /usr/local/phpchroot/wget: No such file or directory


any idea?

Do you have wget installed? Its used by the bot for establishing the ssl connection to mt. gox.
Try "whereis wget" - if it cannot be found you need to install it using your package manager. ("yum install wget" or "apt-get install wget")

Also your OS maybe prefers an absolute path :)

If wget *is* installed and it still doesn't work try changing the lines

   $a = exec ('wget -qO- --no-check-certificate https://mtgox.com/code/data/getDepth.php');

   $a = exec ('wget -qO- --no-check-certificate --post-data "name=' . $username . '&pass=' . $password . $post_data . '" https://mtgox.com/code/' . $scriptname);


to

   $a = exec ('/path/to/your/wget -qO- --no-check-certificate https://mtgox.com/code/data/getDepth.php');

   $a = exec ('/path/to/your/wget -qO- --no-check-certificate --post-data "name=' . $username . '&pass=' . $password . $post_data . '" https://mtgox.com/code/' . $scriptname);

Yea running CENTOS. let me try that

  $a = exec ('/usr/bin/wget -qO- --no-check-certificate https://mtgox.com/code/data/getDepth.php');


does that look right? after i did whereis wget is gave me /usr/bin/wget

Also, do I need to put in my user/pass anywhere?

Yes, perfect but notice that there are 2 occurences you have to change.

>Also, do I need to put in my user/pass anywhere?
Yes, check the header of the script. Do a "nano ./ob.php" and put your id/pw in the corresponding area.

Still getting the same error :(


MtGox down. Waiting...
sh: /usr/local/phpchroot/wget: No such file or directory

Are you sure that you changed BOTH occurences of the wget call? ;-)
Also if you chrooted the script it maybe doesnt have access to the wget tool.

Yep:


  $a = exec ('/usr/bin/wget -qO- --no-check-certificate https://mtgox.com/code/data/getDepth.php');

 $a = exec ('/usr/bin/wget -qO- --no-check-certificate --post-data "name=' . $username . '&pass=' . $password . $post_data . '" https://mtgox.com/code/' . $scriptname);

and im logged in as root

But why is your php dir "phpchroot" Sounds like chroot? Maybe php lacks access rights. If you PM me your ICQ/whatever I'd be willing to help if u can provide me a shell.

Just replied.

Hmm, no more comments to the trading bot at post #35? I was expecting a bit more feedback considering the strong interest of the people here ;)

I know it's a sensitive topic, but I can't seem to find a simple buy/sell bot...And I have very little exp with PHP. Would it be possible to alter the code to 'scalp' to make a profit with a fee based Mt. Gox account? Adding something to sell if price is greater than .06% of the buy price?

runlinux had a bot in his signature. gui etc.

could be, what u need, he didn´t advertise directly in a thread, and i´m wondering, if its still available.
also, i´m no programmer, and don´t know, what this thing might do.

"Try my bitcoin bot at your own risk! http://dl.dropbox.com/xxxx"

edit: i pmed runlinux after i posted this, and he asked me, to remove the link. so i did.
any professional, who could look at the source, would be veryvery much appreciated

It's not ready yet due to some security issues that were pointed out to me earlier by Jine from Bitcoins.lc so please, dont use it. if you could, please remove the link for the time being.

i would like to make it open source, as i dont really have anything to hide, but i'm just not sure where to begin, lol.

EDIT:

(Coded in VB .Net)

i have removed a few things from the bot as there seems to be interest in it, so before anything goes live, i need help making a few changes:

there is a settings file that contains your username and password for a few things, i encrypt this based on a key in plain text in the file.

any decent ideas on how to get around this?

EDIT:

http://dl.dropbox.com/u/21251183/Bitcoin%20Bot.zip

open source. entire project file (actual bot is located inside the debug folder).

rip it apart, abuse it, just be careful

Let me know what you guys find or a better way for me to maintain the source.

Logic behind the buys / sells

it uses the ticker data and build averages from there.

there is a definable running average to based the high and low points, and i calculate a Simple moving average and an Exponential moving average from the ticker data that is based on a user defined depth.

from there, when the averages cross and the price is above/below the defined average percent, it attempts to place a buy/sell order at the ticker price. it doesnt always work. the next step here is to work in the market depth to place a valid order that would be placed.

update posted to the bot. when i was re-doing some numbers earlier, i forgot to have the next buy and sell points actually calculated. that has been fixed.

Just check the functions "buy" and "sell" in my source code. Its less than 10 lines for both :) It won't get any easier than that.

https://runlinux@github.com/runlinux/Bitcoin-Bot

check it out!

Stupid question why is it username an pword as opposed to key and secret am i missing something.  or is username and password for tradehill and it should be key secret for gox?

cheers.

Is someone going to start a open-souce bot it a more cross-platform language. These open-source bots don't even need any algorithms. just.

if < 9.90 buy
if > 10.10 sell

ask and ye shall receive.

added a simple buy and sell box to my bot.

check the box, place your values and when the ticker hits the specified points, it places a buy or sell order.

let me know what bugs you can find.

link in sig.

great!

bug one: tried it with one (almost) empty account. i suddenly thaught i was rich! damn u ;D {"usds":"0","btcs":"0.00037738"} is shown as: 37.738,0000 BTC

Your code looks pretty solid, I'll let you know if I run into any bugs, expect a donation within the near future.

Tell me how this dark magic werkz!!!!!!

This is for Linux?!

Thanks!

Everyone needs a bot.

Nope, coded in lazy man's VB .Net. Windows only.

any chance of an update to that bot? it seems to detect the same buy and sell price

For low trading prices, trading on Open Transactions servers might work. They do not even have mechanism in place for taking a percentage of trades at all, the only support they have for fees really is "usage tokens" against which all API calls are charged.

On first glance it seems as if this should help for reasonably-sized trades, but hinder stragegies that want to make many API calls per trade.

I have not checked yet whether the API calls themselves are reasonably "priced" in terms of the usage tokens, such as whether one can get a huge mass of data in one API call instead of having to make one call per market or suchlike.

I have set up my Open Transactions server (https://bitcointalk.org/index.php?topic=53329.0) to use integers for all assets, deliberately to help encourage volume trades. (As to get three digits of price granularity you'd have to use 3 digits of scale, like "100 of this for 101 of that".)

On the other hand I am not a fan of high frequency trading, I have often thought of actually bundling all offers within a half hour or so into one batch clearing run that processes them all at once, specifically to keep bots from butting in between a matching pair of offers to grab most of the difference that otherwise would have gone to the up-front "honest traders" who admitted what they wanted instead of lurking in the background "sniping".

-MarkM-