diff --git a/docs/index.rst b/docs/index.rst index fa37478e..9df0fb79 100644 --- a/docs/index.rst +++ b/docs/index.rst @@ -23,7 +23,7 @@ matplotlib, scipy, statsmodels, and sklearn support development, analysis and visualization of state-of-the-art trading systems. Zipline is currently used in production as the backtesting engine powering -_ free, community-centered platform that allows +``_, a free, community-centered platform that allows development and real-time backtesting of trading algorithms in the web browser. Zipline was released at PyData NYC'12. @@ -33,6 +33,7 @@ Contents .. toctree:: :maxdepth: 4 + manifesto.rst installation.rst overview.rst modules.rst diff --git a/docs/manifesto.rst b/docs/manifesto.rst new file mode 100644 index 00000000..dbcf190b --- /dev/null +++ b/docs/manifesto.rst @@ -0,0 +1,140 @@ +******************** +Quantopian Manifesto +******************** + +Wall Street's culture was born in an age of information scarcity. +Hoarding information and keeping secrets were the norm. The world has changed. +Today's world is defined by information that wants to be free. The new +scarcity is people: people with the talent and drive to wring insight from all +of that data. + +Quantopian's mission is to attract the world's algorithmic and +financial talent. We want to attract today's quants, and we want to +attract talent that hasn't yet had the opportunity to be a quant. We +want to bring this talent together, provide them with the tools that they +require, and help them build a community. First and foremost, our community is +rooted in openness and sharing. Members share code, know-how, and data. +Quantopian sets the tone by providing open-sourced code, discussing our +techniques, and supplying the historical data needed for algorithmic investing. + +By educating more people about statistical arbitrage and data mining for +finance, we aim to dispense with the secrecy and raise the state of the art. +Rather than hoard data, we relentlessly push data to our community. We want to +diversify the data that can be mined, and permit our members to explore as much +as they like. Our members' success in analyzing and investing will help +us draw more data and more members to our community. Every individual's +success will also help other Quantopians. + +The Evolution of Algorithmic Finance +==================================== + +Charting +-------- + +Algorithmic finance originated as chart reading. Chartists would look for +certain patterns in price history charts. The patterns were always graced with +artfully chosen names like 'head and shoulders,' 'spinning top', or 'morning +star'. Chart reading looks a lot like palm reading, and for the skeptics among +us the similarities don't end with appearances. Still, chart reading is an +attempt to infer the balance of buying and selling appetites in the markets +from a stock's history. Viewed that way, chart reading pursues the noble goal +of prediction. Charting is so common that certain events can trigger market +responses, possibly because so many participants infer the same meaning from a +stock's price chart. + +Technical Analysis +------------------ + +Analysis grew more sophisticated as chartists gave way to +computer scientists writing algorithms. These algorithms have more scientific +sounding names like Moving Averages, Volume Weighted Moving Averages, Bollinger +Bands, Relative Strength Indicators, and Pearson's Correlation +Coefficient. Building technical analysis algorithms looks a lot like modern +statistics, and the optimists among us would say the similarities run deep. +Technical analysts take algorithmic approaches to the same concept: inferring +future behavior from trailing data. In addition to greater sophistication, +technical analysts can also test their algorithms over historic data. Imperfect +to be sure, but a giant leap from staring at a chart. + +Reasonable people can disagree about the 'correctness' of +inferring future events from past behavior. Rather than dwell on that question, +we choose to point out a different limitation of both charting and modern +technical analysis: **both interpret the movement of a single stock in +isolation**. This limitation is both a blessing and a curse. + +On the one hand, there is little room for sophisticated statistics or machine +learning when you have just a single time series for both your signal and your +prediction target. + +On the other, technical analysis can still be intuitive, which makes it easier +to get acquainted with the idea of automated trading. Often there is a mental +leap for people to make from understanding the interpretation of a price series +to issuing orders. Because the signals are easy to understand, technical +analysis makes for a good initial learning experience to explore risk and +performance evaluation as well as order management: the price going above its +30 day moving average is something you can visualize. So, you can focus your +attention on the financial and trading aspects of the problem. + +Statistical Arbitrage +--------------------- + +Statistical Arbitrage is the grandchild of chart reading. +Like technical analysis it relies on algorithms and statistics, but it departs +in one very significant way: 'stat arb' looks for relationships +among many stocks. The challenge with stat arb is twofold: + +* visualizing the relationships can be quite difficult, since the relationships + can have high dimensionality +* the data processing load is quite high - a simple linear regression for all + stocks results in 32 million individual regressions. Assuming a 10-day + window, that can be 320 million individual calculations. To prepare, + backtest, and trade a stat arb strategy required both familiarity with the + mechanics of trading, knowledge of statistics, and a strong computer science + background. + +As stat arb matured, the competition to find stat arb strategies that work +became a two part race: + +1. execute the trades faster +2. find new ways to identify relationships within + market data + +We think the pursuit of faster trades reached diminishing returns when the +market hit sub-millisecond trade execution. We think that the resulting high +level of liquidity is a good thing, but we agree with Thomas Petterffy that +`pursuing even faster trades +`_ +"has absolutely no social value". + +Finding new relationships in the market data is possible and more important now +than ever. In the summer of 2007, there was a sudden meltdown in quantitative +trading firms. Subsequent analysis points to quants crowding into the same +arbitrage bets, and an unforeseen fund liquidation driving all the quants to +unwind those bets concurrently. We believe finding new relationships should +permit investments with lower correlation and lower risks. + +Algorithmic Investing and the Future +==================================== + +A revolution in market understanding happens next. We want Quantopian to enable +more quants than all of Wall Street combined. We want quants, new and old, to +explore and share new ways to view the market. We want to clear away the +obstacles that have so far kept all but a few from doing algorithmic investing +by: + +* simulating with clean, high-quality market data for free +* easy access to markets through trusted brokers +* providing a robust, flexible open-source backtester to permit evaluation and + iteration of algorithms +* supporting a community that fosters the exchange of knowledge, ideas, code + solutions, and data sources + +The community will find new ways to identify market opportunities. It may take +the form of new, non-market data sources, like news feeds or Twitter. It may be +new algorithmic techniques. Most likely, it will be something we +haven't heard of yet: your idea. The one you keep coming back to. The +idea you couldn't test without data. The idea that needs backtesting, +and iteration, and encouragement from other quants. + +Do you want to unleash your idea? This is your chance. `Come hack Wall Street +`_.