This is a simple Bellman-Ford bot that uses the negative cycle detection feature of the algorithm to find favorable currency trades to make in forex markets (in this case we are targeting bitcoin/fiat/scryptcoin markets on btc-e and other exchanges).
Typically the trades the bot finds are less than 0.5% profit, will take 3 steps, and must be filled quickly to be profitable.
The main entry point into the application is the class BtceBFBot.java which is a reference implementation and runs the BFAlgorithm against Btc-e.com exchange public API in a loop.
The following image demonstrates the principal of jumping through a set of currencies in order to make a profit in pricing discrepancies. In this project the various currency markets on Btc-e are represented as a graph and profitable cycles are detected via the eponymous algorithm. Each currency is represented as a Node and each exchange rate is represented as an Edge. The graph is a directed graph but each node will always have reciprocal edges so it may thought of as cyclic and undirected.
The application consists of two sets of code:
1. The cycle detection system. This creates a set of trades to execute.
2. The trade executor. It takes a set of trades and executes them (Incomplete)
The code currently has the begining of a Btc-e implementation but is meant to provide interfaces to easily build for other exchanges.
Using the Bellman-Ford algorithm to detect so-called "negative" cycles in a graph is a very procedure in forex markets. Applying this to thinly traded and unregulated bitcoin markets seemed like a pretty fun idea to me so I implemented a cycle detector that will take into account corresponding fees that will be taken to make n number of trades in a cycle. Negative cycles are found by taking the log of the bid or ask price, using that as the wieght of the edge between nodes and as you may well be aware the log of a number less than 1 and greater than 0 yields a negative.
The underlying mechanics of the application are as follows:
1. The Bellman-Form algorithm is used to find a negative cycle in the currency graph
2. Build a Cycle of trades from this
3. Calculate profitability
4. Send profitable trade steps to the trade executor as a multi-step trade.
5. Wait for the multi-step trade to fill or fail to fill before we send another.
The application's trading system is based on an existing library that is wrapped to provide multi-step execution and controlled deviation from the initial limit order intended to be executed. I have packaged the libarary I intend to use as a Jar file in the /libs directory.
You can find that library here: https://github.com/alexandersjn/btc_e_assist_api
The trade executor should implemented with something like a Fill-Or-Kill order type with a given number of retries and a trade tolerance which will allow slightly less profitable trades to be placed if the initial optimal order price cannot be filled. There should be a time limit and retry limit for each step so that the executor can decide when to abort and trade back into the base currency. In my case this would be USD.
While initially I was optimistic about the potential profitability of running an arbitrager bot on a bitcoin exchange I have come to believe that due to the percentage based trading fee, high probability of my own oversight in programming a trade executor and general low reward / work ratio I decided to share the cycle detector system and hope that some avaricious sole may want to help me finish the executor system.
Additionally I have become more interested in market-making algorithms after reading Irene Aldrige's High Frequency Trading book, but I think it is almost impossible to implement a decent market maker on percentage based fees.
Anyway, I don't think this little bot will help you end up like this guy:
