How to use machine learning and R to build algorithms to trade and arbitrage Bitcoin and other currencies.

1. Bitcoin Algorithmic
Trading with CoinTrader
and R
Tom Johnson, PhD
Bitcoin SF Dev Group
6-23-2014

2. Bitcoin Algo Trading Development
Objective:
Make Money,
Help Bitcoin,
Have Fun

3. Why focus on Price
rather than Pipes?
Price Volatility
Big deterrent to public acceptance
Smooth rise good for all!
Big Opportunities for:
Products - Derivatives, Transaction info
Services - Price guides, Block execution
Trading/Arb - Hedging, Funds, Day-trade
Coins - Medium of xchg vs Store of value

4. What does R have to contribute?
R is Open Source, Global, Data Science
R script - Interactive, Algo development,
Prototyping, Live control, Easy to migrate
Packages - 5800, 120k functions
Community - Statistics & Machine Learning
Pros - Stable, fast math, answers/resources
Cons - Single threaded, in-memory, silos
R tools well matched to Bitcoin challenges
So are others like Python and ?

5. So how volatile is Bitcoin?
Compared with EURUSD?

6. Compared with Gold?

9. Comparing Bitcoin, EUR & Gold

10. But getting out of, or into
BTC will cost you

11. And what you see isn’t
what you get

12. And nothing’s isolated; it impacts
other Crypto, Home, Cross Pairs

13. So what does Bitstamp’s BTCUSD
Orderbook look like?
Demo with CoinTrader CSV file in AWS

14. With all the turbulence -
What about arbitrage opportunities?

15. And there’s Triangular Arbitrage --
Demo Example
Priceonomics.com --
Find best 2-currency plus Bitcoin arbitrage
among 3 currencies in live price feed:
Find most profitable sequence of transactions
from any Chosen Home Currency, through an
Intermediate Currency, then through Bitcoin
and finally back to the Home Currency.

16. Exchange Rates at a Point in Time
{
USD_JPY: "95.7422091",
USD_USD: "1.0000000",
JPY_EUR: "0.0080872",
BTC_USD: "105.5641218",
...
}

17. R Script
xchgTable <- xchgTable()
foreignCurrs <- allCurrs[!allCurrs %in% c(input$homeCurr, "BTC")]
foreignAmts <- btcAmts <- endHomeAmts <- rep(0, length(foreignCurrs))
names(foreignAmts) <- names(btcAmts) <- names(endHomeAmts) <- foreignCurrs
for (foreignCurr in foreignCurrs) {
foreignAmts[foreignCurr] <- xchgTable[input$homeCurr, foreignCurr] *
as.numeric(input$initHomeAmt)
btcAmts[foreignCurr] <- xchgTable[foreignCurr, "BTC"] * foreignAmts[foreignCurr]
endHomeAmts[foreignCurr] <- xchgTable["BTC", input$homeCurr] *
btcAmts[foreignCurr]
}
maxArbCurr <- foreignCurrs[which.max(endHomeAmts)]
cat(c("Trade", input$initHomeAmt, input$homeCurr, "to", round(foreignAmts[maxArbCurr],2),
maxArbCurr, "to", round(btcAmts[maxArbCurr],6), "BTC", "to",
round(endHomeAmts[maxArbCurr], 2), input$homeCurr, "which =",
round(((endHomeAmts[maxArbCurr]/as.numeric(input$initHomeAmt) - 1) * 100), 2), "% Gross
Gain"), "nnn")

18. Running in Amazon Cloud
AWS - Account, EC2, private key, static IP
Stack -
AMI (Medium), Windows Server, Firefox,
Google Drive, R, Shiny, RStudio, Apps
Set up URL with AWS namespaces
Demo -- www.next1up.com

19. Demo
Run Shiny App in cloud
Access through browser
Basic - Calculate & Display
Next - Monitor, Control & Execution

21. Machine Learning can help
Prediction & Classification
Lots of raw, currrent data (CoinTrader
CSV)
Code up savvy features
Forget the scholarly models
Use many non-linear methods
Don’t overfit or let the future leak in!
Cross-validate, analyze resduals
Performance going forward is what
matters!

22. Steps to build robust Algos & Rules
to Trade, Hedge, Place Limit Orders
Feature primitives -
Full feature variables - Differences, EMAs, etc.
Data - Training, Testing, Reserved data test
Variables - Bagging, Boosting, Selection
Models - Trees, RF, GLM, GA, GP, Ensemble
Use final code or live signal in CoinTrader

23. The Take-Away
Buy when under valued;
Sell when over valued.
Price changes faster and,
for reasons other, than value.
Good information & analysis can help.
The bigger the mispricing,
the bigger the opportunity
Whether trading, products, services or
coins.
Making money helps us all and Bitcoin!

24. Sources
Tom Johnson PhD
Tom@AI-Realtime.com
Arbitrage Demo http://next1up.net
https://www.linkedin.com/in/tom-johnson-62a16
Prices www.cryptocoincharts.info
APIs www.quandl.com/markets/bitcoin
Cool www.bitlisten.com
Analyses www.priceonomics.com
Shiny www.rstudio.com/shiny