Notes: Quantopian

  • The finance industry is very secretive. Not only about what they do (so to avoid strategy/capability leak to competitors) but also in who they are.
    Consequently it is difficult to get hired.
    The best strategy is to become well known first and they will recruit you.
    The industry’s biggest problem is finding and hiring talent.
    It usually takes 1 or 2 years to train a new hire.
    Quantopian solves this problem by providing a platform to develop trading strategies, proving your abilities in the field.
  • Because of existing tools, many quant traders cannot leave their job.
    Even the super intelligent and successful ones simply cannot reconstruct a firms tooling.
  • Tools break down into three components:
    Quantopian holds clean (normalized for splits/dividends) data.
    They have minute trading data for almost all companies over the past 10 years (free).
    Including companies which went bankrupt, or stopped trading, so as not to introduce survivorship bias.
    Comparing an algorithmic performance using a different backtester’s is not apples-to-apples, because each backtester has differences in the way it handles splits, dividends, commissions, slippage, borrowing costs if shorting, in-sample v. out-sample, etc.
    Quantopian allows comparison across algorithms, because the backtester is the same for everyone on the site.
    Implementation Plumbing
    Once an algorithm has been developed and rigorously tested, programmer may still introduce bugs when implementing it for live trading.
    Every trading platform API is different (IB is horrible, but the standard).
  • Quantopian provides in-browser editing and development, removing the largest hassle: setup.
  • Proud to list some of the tools used: Rightscale, Flask, Gevent, Zero MQ, Pandas, SciPy, NumPy, Heroku, MongoHQ, Highcharts, Zipline, and others.
  • Competitor (QuantBot?) allows people to create algo’s by connecting different boxes.
    Much more flexible to let people write python.
    Quantopian targets ease of uses and is trying to grow the community.
    They do not focus on HFT development.
  • Future monetization could come from charging for specialized data as a premium feature (futures, derivatives, commodities, etc); charging for cpu-hr during backtesting.
    They want to avoid commission if/when the support live trading, because that would seriously complicate people’s backtesting.