Release Notes

Release 1.0.2

Release:1.0.2
Date:September 8, 2016

Enhancements

  • Adds forward fill checkpoint tables for the blaze core loader. This allow the loader to more efficiently forward fill the data by capping the lower date it must search for when querying data. The checkpoints should have novel deltas applied (#1276).
  • Updated VagrantFile to include all dev requirements and use a newer image (#1310).
  • Allow correlations and regressions to be computed between two 2D factors by doing computations asset-wise (#1307).
  • Filters have been made window_safe by default. Now they can be passed in as arguments to other Filters, Factors and Classifiers (#1338).
  • Added an optional groupby parameter to rank(), top(), and bottom(). (#1349).
  • Added new pipeline filters, All and Any, which takes another filter and returns True if an asset produced a True for any/all days in the previous window_length days (#1358).
  • Added new pipeline filter AtLeastN, which takes another filter and an int N and returns True if an asset produced a True on N or more days in the previous window_length days (#1367).
  • Use external library empyrical for risk calculations. Empyrical unifies risk metric calculations between pyfolio and zipline. Empyrical adds custom annualization options for returns of custom frequencies. (#855)
  • Add Aroon factor. (#1258)
  • Add fast stochastic oscillator factor. (#1255)
  • Add a Dockerfile. (#1254)
  • New trading calendar which supports sessions which span across midnights, e.g. 24 hour 6:01PM-6:00PM sessions for futures trading. zipline.utils.tradingcalendar is now deprecated. (#1138) (#1312)
  • Allow slicing a single column out of a Factor/Filter/Classifier. (#1267)
  • Provide Ichimoku Cloud factor (#1263)
  • Allow default parameters on Pipeline terms. (#1263)
  • Provide rate of change percentage factor. (#1324)
  • Provide linear weighted moving average factor. (#1325)
  • Add NotNullFilter. (#1345)
  • Allow capital changes to be defined by a target value. (#1337)
  • Add TrueRange factor. (#1348)
  • Add point in time lookups to assets.db. (#1361)
  • Make can_trade aware of the asset’s exchange . (#1346)
  • Add downsample method to all computable terms. (#1394)
  • Add QuantopianUSFuturesCalendar. (#1414)
  • Enable publishing of old assets.db versions. (#1430)
  • Enable schedule_function for Futures trading calendar. (#1442)
  • Disallow regressions of length 1. (#1466)

Experimental

  • Add support for comingled Future and Equity history windows, and enable other Future data access via data portal. (#1435) (#1432)

Bug Fixes

  • Changes AverageDollarVolume built-in factor to treat missing close or volume values as 0. Previously, NaNs were simply discarded before averaging, giving the remaining values too much weight (#1309).

  • Remove risk-free rate from sharpe ratio calculation. The ratio is now the average of risk adjusted returns over violatility of adjusted returns. (#853)

  • Sortino ratio will return calculation instead of np.nan when required returns are equal to zero. The ratio now returns the average of risk adjusted returns over downside risk. Fixed mislabeled API by converting mar to downside_risk. (#747)

  • Downside risk now returns the square root of the mean of downside difference squares. (#747)

  • Information ratio updated to return mean of risk adjusted returns over standard deviation of risk adjusted returns. (#1322)

  • Alpha and sharpe ratio are now annualized. (#1322)

  • Fix units during reading and writing of daily bar ``first_trading_day `` attribute. (#1245)

  • Optional dispatch modules, when missing, no longer cause a NameError. (#1246)

  • Treat schedule_function argument as a time rule when a time rule, but no date rule is supplied. (#1221)

  • Protect against boundary conditions at beginning and end trading day in schedule function. (#1226)

  • Apply adjustments to previous day when using history with a frequency of 1d. (#1256)

  • Fail fast on invalid pipeline columns, instead of attempting to access the nonexistent column. (#1280)

  • Fix AverageDollarVolume NaN handling. (#1309)

    Performance


  • Performance improvements to blaze core loader. (#1227)
  • Allow concurrent blaze queries. (#1323)
  • Prevent missing leading bcolz minute data from doing repeated unnecessary lookups. (#1451)
  • Cache future chain lookups. (#1455)

Maintenance and Refactorings

  • Removed remaining mentions of add_history. (#1287)

Documentation

Testing

  • Add test fixture which sources daily pricing data from minute pricing data fixtures. (#1243)

Data Format Changes

  • BcolzDailyBarReader and BcolzDailyBarWriter use trading calendar instance, instead of trading days serialized to JSON. (#1330)
  • Change format of assets.db to support point in time lookups. (#1361)
  • Change BcolzMinuteBarReader``and ``BcolzMinuteBarWriter to support varying tick sizes. (#1428)

Release 1.0.1

Release:1.0.1
Date:May 27, 2016

This is a minor bug-fix release from 1.0.0 and includes a small number of bug fixes and documentation improvements.

Enhancements

  • Added support for user-defined commission models. See the zipline.finance.commission.CommissionModel class for more details on implementing a commision model. (#1213)
  • Added support for non-float columns to Blaze-backed Pipeline datasets (#1201).
  • Added zipline.pipeline.slice.Slice, a new pipeline term designed to extract a single column from another term. Slices can be created by indexing into a term, keyed by asset. (#1267)

Bug Fixes

  • Fixed a bug where Pipeline loaders were not properly initialized by zipline.run_algorithm(). This also affected invocations of zipline run from the CLI.

  • Fixed a bug that caused the %%zipline IPython cell magic to fail (533233fae43c7ff74abfb0044f046978817cb4e4).

  • Fixed a bug in the PerTrade commission model where commissions were incorrectly applied to each partial-fill of an order rather than on the order itself, resulting in algorithms being charged too much in commissions when placing large orders.

    PerTrade now correctly applies commissions on a per-order basis (#1213).

  • Attribute accesses on CustomFactors defining multiple outputs will now correctly return an output slice when the output is also the name of a Factor method (#1214).

  • Replaced deprecated usage of pandas.io.data with pandas_datareader (#1218).

  • Fixed an issue where .pyi stub files for zipline.api were accidentally excluded from the PyPI source distribution. Conda users should be unaffected (#1230).

Documentation

  • Added a new example, zipline.examples.momentum_pipeline, which exercises the Pipeline API (#1230).

Release 1.0.0

Release:1.0.0
Date:May 19, 2016

Highlights

Zipline 1.0 Rewrite (#1105)

We have rewritten a lot of Zipline and its basic concepts in order to improve runtime performance. At the same time, we’ve introduced several new APIs.

At a high level, earlier versions of Zipline simulations pulled from a multiplexed stream of data sources, which were merged via heapq. This stream was fed to the main simulation loop, driving the clock forward. This strong dependency on reading all the data made it difficult to optimize simulation performance because there was no connection between the amount of data we fetched and the amount of data actually used by the algorithm.

Now, we only fetch data when the algorithm needs it. A new class, DataPortal, dispatches data requests to various data sources and returns the requested values. This makes the runtime of a simulation scale much more closely with the complexity of the algorithm, rather than with the number of assets provided by the data sources.

Instead of the data stream driving the clock, now simulations iterate through a pre-calculated set of day or minute timestamps. The timestamps are emitted by MinuteSimulationClock and DailySimulationClock, and consumed by the main loop in transform().

We’ve retired the data[sid(N)] and history APIs, replacing them with several methods on the BarData object: current(), history(), can_trade(), and is_stale(). Old APIs will continue to work for now, but will issue deprecation warnings.

You can now pass in an adjustments source to the DataPortal, and we will apply adjustments to the pricing data when looking backwards at data. Prices and volumes for execution and presented to the algorithm in data.current are the as-traded value of the asset.

New Entry Points (#1173 and #1178)

In order to make it easier to use zipline we have updated the entry points for a backtest. The three supported ways to run a backtest are now:

  1. zipline.run_algo()
  2. $ zipline run
  3. %zipline (IPython magic)

Data Bundles (#1173 and #1178)

1.0.0 introduces data bundles. Data bundles are groups of data that should be preloaded and used to run backtests later. This allows users to not need to to specify which tickers they are interested in each time they run an algorithm. This also allows us to cache the data between runs.

By default, the quantopian-quandl bundle will be used which pulls data from Quantopian’s mirror of the quandl WIKI dataset. New bundles may be registered with zipline.data.bundles.register() like:

@zipline.data.bundles.register('my-new-bundle')
def my_new_bundle_ingest(environ,
                         asset_db_writer,
                         minute_bar_writer,
                         daily_bar_writer,
                         adjustment_writer,
                         calendar,
                         cache,
                         show_progress):
    ...

This function should retrieve the data it needs and then use the writers that have been passed to write that data to disc in a location that zipline can find later.

This data can be used in backtests by passing the name as the -b / --bundle argument to $ zipline run or as the bundle argument to zipline.run_algorithm().

For more information see Data Bundles for more information.

String Support in Pipeline (#1174)

Added support for string data in Pipeline. zipline.pipeline.data.Column now accepts object as a dtype, which signifies that loaders for that column should emit windowed iterators over the experimental new LabelArray class.

Several new Classifier methods have also been added for constructing Filter instances based on string operations. The new methods are:

  • element_of()
  • startswith()
  • endswith()
  • has_substring()
  • matches()

element_of is defined for all classifiers. The remaining methods are only defined for string-dtype classifiers.

Enhancements

  • Made the data loading classes have more consistent interfaces. This includes the equity bar writers, adjustment writer, and asset db writer. The new interface is that the resource to be written to is passed at construction time and the data to write is provided later to the write method as dataframes or some iterator of dataframes. This model allows us to pass these writer objects around as a resource for other classes and functions to consume (#1109 and #1149).

  • Added masking to zipline.pipeline.CustomFactor. Custom factors can now be passed a Filter upon instantiation. This tells the factor to only compute over stocks for which the filter returns True, rather than always computing over the entire universe of stocks. (#1095)

  • Added zipline.utils.cache.ExpiringCache. A cache which wraps entries in a zipline.utils.cache.CachedObject, which manages expiration of entries based on the dt supplied to the get method. (#1130)

  • Implemented zipline.pipeline.factors.RecarrayField, a new pipeline term designed to be the output type of a CustomFactor with multiple outputs. (#1119)

  • Added optional outputs parameter to zipline.pipeline.CustomFactor. Custom factors are now capable of computing and returning multiple outputs, each of which are themselves a Factor. (#1119)

  • Added support for string-dtype pipeline columns. Loaders for thse columns should produce instances of zipline.lib.labelarray.LabelArray when traversed. latest() on string columns produces a string-dtype zipline.pipeline.Classifier. (#1174)

  • Added several methods for converting Classifiers into Filters.

    The new methods are: - element_of() - startswith() - endswith() - has_substring() - matches()

    element_of is defined for all classifiers. The remaining methods are only defined for strings. (#1174)

  • Added BollingerBands factor. This factor implements the Bollinger Bands technical indicator: https://en.wikipedia.org/wiki/Bollinger_Bands (#1199).

  • Fetcher has been moved from Quantopian internal code into Zipline (#1105).

  • Added new built-in factors, RollingPearsonOfReturns, RollingSpearmanOfReturns and RollingLinearRegressionOfReturns (#1154)

Experimental Features

Warning

Experimental features are subject to change.

  • Added a new zipline.lib.labelarray.LabelArray class for efficiently representing and computing on string data with numpy. This class is conceptually similar to pandas.Categorical, in that it represents string arrays as arrays of indices into a (smaller) array of unique string values. (#1174)

Bug Fixes

None

Performance

None

Maintenance and Refactorings

None

Build

None

Documentation

  • Updated documentation for the API methods (#1188).
  • Updated release process to mention that docs should be built with python 3 (#1188).

Miscellaneous

  • Zipline now provides a stub file for the zipline.api module. This module is normally dynamically created so the stub file provides some static information for utilities that can consume it, for example PyCharm (#1208).

Release 0.9.0

Release:0.9.0
Date:March 29, 2016

Highlights

  • Added classifiers and normalization methods to pipeline, along with new datasets and factors.
  • Added support for Windows with continuous integration on AppVeyor.

Enhancements

  • Added new datasets CashBuybackAuthorizations and ShareBuybackAuthorizations for use in the Pipeline API. These datasets provide an abstract interface for adding cash and share buyback authorizations data, respectively, to a new algorithm. pandas-based reference implementations for these datasets can be found in zipline.pipeline.loaders.buyback_auth, and experimental blaze-based implementations can be found in zipline.pipeline.loaders.blaze.buyback_auth. (#1022).
  • Added new datasets DividendsByExDate, DividendsByPayDate, and DividendsByAnnouncementDate for use in the Pipeline API. These datasets provide an abstract interface for adding dividends data organized by ex date, pay date, and announcement date, respectively, to a new algorithm. pandas-based reference implementations for these datasets can be found in zipline.pipeline.loaders.dividends, and experimental blaze-based implementations can be found in zipline.pipeline.loaders.blaze.dividends. (#1093).
  • Added new built-in factors, zipline.pipeline.factors.BusinessDaysSinceCashBuybackAuth and zipline.pipeline.factors.BusinessDaysSinceShareBuybackAuth. These factors use the new CashBuybackAuthorizations and ShareBuybackAuthorizations datasets, respectively. (#1022).
  • Added new built-in factors, zipline.pipeline.factors.BusinessDaysSinceDividendAnnouncement, zipline.pipeline.factors.BusinessDaysUntilNextExDate, and zipline.pipeline.factors.BusinessDaysSincePreviousExDate. These factors use the new DividendsByAnnouncementDate` and ``DividendsByExDate datasets, respectively. (#1093).
  • Implemented zipline.pipeline.Classifier, a new core pipeline API term representing grouping keys. Classifiers are primarily used by passing them as the groupby parameter to factor normalization methods. (#1046)
  • Added factor normalization methods: zipline.pipeline.Factor.demean() and zipline.pipeline.Factor.zscore(). (#1046)
  • Added zipline.pipeline.Factor.quantiles(), a method for computing a Classifier from a Factor by partitioning into equally-sized buckets. Also added helpers for common quantile sizes (zipline.pipeline.Factor.quartiles(), zipline.pipeline.Factor.quartiles(), and zipline.pipeline.Factor.deciles()) (#1075).

Experimental Features

Warning

Experimental features are subject to change.

None

Bug Fixes

  • Fixed a bug where merging two numerical expressions failed given too many inputs. This caused running a pipeline to fail when combining more than ten factors or filters. (#1072)

Performance

None

Maintenance and Refactorings

None

Build

  • Added AppVeyor for continuous integration on Windows. Added conda build of zipline and its dependencies to AppVeyor and Travis builds, which upload their results to anaconda.org labeled with “ci”. (#981)

Documentation

None

Miscellaneous

  • Adds ZiplineTestCase which provides hooks to consume test fixtures. Fixtures are things like: WithAssetFinder which will make self.asset_finder available to your test with some mock data (#1042).

Release 0.8.4

Release:0.8.4
Date:February 24, 2016

Highlights

  • Added a new EarningsCalendar dataset for use in the Pipeline API. (#905).
  • AssetFinder speedups (#830 and #817).
  • Improved support for non-float dtypes in Pipeline. Most notably, we now support datetime64 and int64 dtypes for Factor, and BoundColumn.latest now returns a proper Filter object when the column is of dtype bool.
  • Zipline now supports numpy 1.10, pandas 0.17, and scipy 0.16 (#969).
  • Batch transforms have been deprecated and will be removed in a future release. Using history is recommended as an alternative.

Enhancements

  • Adds a way for users to provide a context manager to use when executing the scheduled functions (including handle_data). This context manager will be passed the BarData object for the bar and will be used for the duration of all of the functions scheduled to run. This can be passed to TradingAlgorithm by the keyword argument create_event_context (#828).
  • Added support for zipline.pipeline.factors.Factor instances with datetime64[ns] dtypes. (#905)
  • Added a new EarningsCalendar dataset for use in the Pipeline API. This dataset provides an abstract interface for adding earnings announcement data to a new algorithm. A pandas-based reference implementation for this dataset can be found in zipline.pipeline.loaders.earnings, and an experimental blaze-based implementation can be found in zipline.pipeline.loaders.blaze.earnings. (#905).
  • Added new built-in factors, zipline.pipeline.factors.BusinessDaysUntilNextEarnings and zipline.pipeline.factors.BusinessDaysSincePreviousEarnings. These factors use the new EarningsCalendar dataset. (#905).
  • Added isnan(), notnan() and isfinite() methods to zipline.pipeline.factors.Factor (#861).
  • Added zipline.pipeline.factors.Returns, a built-in factor which calculates the percent change in close price over the given window_length. (#884).
  • Added a new built-in factor: AverageDollarVolume. (#927).
  • Added ExponentialWeightedMovingAverage and ExponentialWeightedMovingStdDev factors. (#910).
  • Allow DataSet classes to be subclassed where subclasses inherit all of the columns from the parent. These columns will be new sentinels so you can register them a custom loader (#924).
  • Added coerce() to coerce inputs from one type into another before passing them to the function (#948).
  • Added optionally() to wrap other preprocessor functions to explicitly allow None (#947).
  • Added ensure_timezone() to allow string arguments to get converted into datetime.tzinfo objects. This also allows tzinfo objects to be passed directly (#947).
  • Added two optional arguments, data_query_time and data_query_tz to BlazeLoader and BlazeEarningsCalendarLoader. These arguments allow the user to specify some cutoff time for data when loading from the resource. For example, if I want to simulate executing my before_trading_start function at 8:45 US/Eastern then I could pass datetime.time(8, 45) and 'US/Eastern' to the loader. This means that data that is timestamped on or after 8:45 will not seen on that day in the simulation. The data will be made available on the next day (#947).
  • BoundColumn.latest now returns a Filter for columns of dtype bool (#962).
  • Added support for Factor instances with int64 dtype. Column now requires a missing_value when dtype is integral. (#962)
  • It is also now possible to specify custom missing_value values for float, datetime, and bool Pipeline terms. (#962)
  • Added auto-close support for equities. Any positions held in an equity that reaches its auto_close_date will be liquidated for cash according to the equity’s last sale price. Furthermore, any open orders for that equity will be canceled. Both futures and equities are now auto-closed on the morning of their auto_close_date, immediately prior to before_trading_start. (#982)

Experimental Features

Warning

Experimental features are subject to change.

  • Added support for parameterized Factor subclasses. Factors may specify params as a class-level attribute containing a tuple of parameter names. These values are then accepted by the constructor and forwarded by name to the factor’s compute function. This API is experimental, and may change in future releases.

Bug Fixes

  • Fixes an issue that would cause the daily/minutely method caching to change the len of a SIDData object. This would cause us to think that the object was not empty even when it was (#826).
  • Fixes an error raised in calculating beta when benchmark data were sparse. Instead numpy.nan is returned (#859).
  • Fixed an issue pickling sentinel() objects (#872).
  • Fixed spurious warnings on first download of treasury data (:issue 922).
  • Corrected the error messages for set_commission() and set_slippage() when used outside of the initialize function. These errors called the functions override_* instead of set_*. This also renamed the exception types raised from OverrideSlippagePostInit and OverrideCommissionPostInit to SetSlippagePostInit and SetCommissionPostInit (#923).
  • Fixed an issue in the CLI that would cause assets to be added twice. This would map the same symbol to two different sids (#942).
  • Fixed an issue where the PerformancePeriod incorrectly reported the total_positions_value when creating a Account (#950).
  • Fixed issues around KeyErrors coming from history and BarData on 32-bit python, where Assets did not compare properly with int64s (#959).
  • Fixed a bug where boolean operators were not properly implemented on Filter (#991).
  • Installation of zipline no longer downgrades numpy to 1.9.2 silently and unconditionally (#969).

Performance

  • Speeds up lookup_symbol() by adding an extension, AssetFinderCachedEquities, that loads equities into dictionaries and then directs lookup_symbol() to these dictionaries to find matching equities (#830).
  • Improved performance of lookup_symbol() by performing batched queries. (#817).

Maintenance and Refactorings

  • Asset databases now contain version information to ensure compatibility with current Zipline version (#815).
  • Upgrade requests version to 2.9.1 (2ee40db)
  • Upgrade logbook version to 0.12.5 (11465d9).
  • Upgrade Cython version to 0.23.4 (5f49fa2).

Build

  • Makes zipline install requirements more flexible (#825).
  • Use versioneer to manage the project __version__ and setup.py version (#829).
  • Fixed coveralls integration on travis build (#840).
  • Fixed conda build, which now uses git source as its source and reads requirements using setup.py, instead of copying them and letting them get out of sync (#937).
  • Require setuptools > 18.0 (#951).

Documentation

  • Document the release process for developers (#835).
  • Added reference docs for the Pipeline API. (#864).
  • Added reference docs for Asset Metadata APIs. (#864).
  • Generated documentation now includes links to source code for many classes and functions. (#864).
  • Added platform-specific documentation describing how to find binary dependencies. (#883).

Miscellaneous

  • Added a show_graph() method to render a Pipeline as an image (#836).
  • Adds subtest() decorator for creating subtests without nose_parameterized.expand() which bloats the test output (#833).
  • Limits timer report in test output to 15 longest tests (#838).
  • Treasury and benchmark downloads will now wait up to an hour to download again if data returned from a remote source does not extend to the date expected. (#841).
  • Added a tool to downgrade the assets db to previous versions (#941).

Release 0.8.3

Release:0.8.3
Date:November 6, 2015

Note

We advanced the version to 0.8.3 to fix a source distribution issue with pypi. There are no code changes in this version.

Release 0.8.0

Release:0.8.0
Date:November 6, 2015

Highlights

Enhancements

  • Account object: Adds an account object to context to track information about the trading account. Example:

    context.account.settled_cash
    

    Returns the settled cash value that is stored on the account object. This value is updated accordingly as the algorithm is run (#396).

  • HistoryContainer can now grow dynamically. Calls to history() will now be able to increase the size or change the shape of the history container to be able to service the call. add_history() now acts as a preformance hint to pre-allocate sufficient space in the container. This change is backwards compatible with history, all existing algorithms should continue to work as intended (#412).

  • Simple transforms ported from quantopian and use history. SIDData now has methods for:

    • stddev
    • mavg
    • vwap
    • returns

    These methods, except for returns, accept a number of days. If you are running with minute data, then this will calculate the number of minutes in those days, accounting for early closes and the current time and apply the transform over the set of minutes. returns takes no parameters and will return the daily returns of the given asset. Example:

    data[security].stddev(3)
    

    (#429).

  • New fields in Performance Period. Performance Period has new fields accessible in return value of to_dict: - gross leverage - net leverage - short exposure - long exposure - shorts count - longs count (#464).

  • Allow order_percent() to work with various market values (by Jeremiah Lowin).

    Currently, order_percent() and order_target_percent() both operate as a percentage of self.portfolio.portfolio_value. This PR lets them operate as percentages of other important MVs. Also adds context.get_market_value(), which enables this functionality. For example:

    # this is how it works today (and this still works)
    # put 50% of my portfolio in AAPL
    order_percent('AAPL', 0.5)
    # note that if this were a fully invested portfolio, it would become 150% levered.
    
    # take half of my available cash and buy AAPL
    order_percent('AAPL', 0.5, percent_of='cash')
    
    # rebalance my short position, as a percentage of my current short
    book_target_percent('MSFT', 0.1, percent_of='shorts')
    
    # rebalance within a custom group of stocks
    tech_stocks = ('AAPL', 'MSFT', 'GOOGL')
    tech_filter = lambda p: p.sid in tech_stocks
    for stock in tech_stocks:
        order_target_percent(stock, 1/3, percent_of_fn=tech_filter)
    

    (#477).

  • Command line option to for printing algo to stdout (by Andrea D’Amore) (#545).

  • New user defined function before_trading_start. This function can be overridden by the user to be called once before the market opens every day (#389).

  • New api function schedule_function(). This function allows the user to schedule a function to be called based on more complicated rules about the date and time. For example, call the function 15 minutes before market close respecting early closes (#411).

  • New api function set_do_not_order_list(). This function accepts a list of assets and adds a trading guard that prevents the algorithm from trading them. Adds a list point in time list of leveraged ETFs that people may want to mark as ‘do not trade’ (#478).

  • Adds a class for representing securities. order() and other order functions now require an instance of Security instead of an int or string (#520).

  • Generalize the Security class to Asset. This is in preperation of adding support for other asset types (#535).

  • New api function get_environment(). This function by default returns the string 'zipline'. This is used so that algorithms can have different behavior on Quantopian and local zipline (#384).

  • Extends get_environment() to expose more of the environment to the algorithm. The function now accepts an argument that is the field to return. By default, this is 'platform' which returns the old value of 'zipline' but the following new fields can be requested:

    • ''arena': Is this live trading or backtesting?
    • 'data_frequency': Is this minute mode or daily mode?
    • 'start': Simulation start date.
    • 'end': Simulation end date.
    • 'capital_base': The starting capital for the simulation.
    • 'platform': The platform that the algorithm is running on.
    • '*': A dictionary containing all of these fields.

    (#449).

  • New api function set_max_leveraged(). This method adds a trading guard that prevents your algorithm from over leveraging itself (#552).

Experimental Features

Warning

Experimental features are subject to change.

  • Adds new Pipeline API. The pipeline API is a high-level declarative API for representing trailing window computations on large datasets (#630).
  • Adds support for futures trading (#637).
  • Adds Pipeline loader for blaze expressions. This allows users to pull data from any format blaze understands and use it in the Pipeline API. (#775).

Bug Fixes

  • Fix a bug where the reported returns could sharply dip for random periods of time (#378).
  • Fix a bug that prevented debuggers from resolving the algorithm file (#431).
  • Properly forward arguments to user defined initialize function (#687).
  • Fix a bug that would cause treasury data to be redownloaded every backtest between midnight EST and the time when the treasury data was available (#793).
  • Fix a bug that would cause the user defined analyze function to not be called if it was passed as a keyword argument to TradingAlgorithm (#819).

Performance

  • Major performance enhancements to history (by Dale Jung) (#488).

Maintenance and Refactorings

  • Remove simple transform code. These are available as methods of SIDData (#550).

Build

None

Documentation

  • Switched to sphinx for the documentation (#816).

Release 0.7.0

Release:0.7.0
Date:July 25, 2014

Highlights

  • Command line interface to run algorithms directly.
  • IPython Magic %%zipline that runs algorithm defined in an IPython notebook cell.
  • API methods for building safeguards against runaway ordering and undesired short positions.
  • New history() function to get a moving DataFrame of past market data (replaces BatchTransform).
  • A new beginner tutorial.

Enhancements

  • CLI: Adds a CLI and IPython magic for zipline. Example:

    python run_algo.py -f dual_moving_avg.py --symbols AAPL --start 2011-1-1 --end 2012-1-1 -o dma.pickle
    

    Grabs the data from yahoo finance, runs the file dual_moving_avg.py (and looks for dual_moving_avg_analyze.py which, if found, will be executed after the algorithm has been run), and outputs the perf DataFrame to dma.pickle (#325).

  • IPython magic command (at the top of an IPython notebook cell). Example:

    %%zipline --symbols AAPL --start 2011-1-1 --end 2012-1-1 -o perf
    

    Does the same as above except instead of executing the file looks for the algorithm in the cell and instead of outputting the perf df to a file, creates a variable in the namespace called perf (#325).

  • Adds Trading Controls to the algorithm API.

    The following functions are now available on TradingAlgorithm and for algo scripts:

    set_max_order_size(self, sid=None, max_shares=None, max_notional=None) Set a limit on the absolute magnitude, in shares and/or total dollar value, of any single order placed by this algorithm for a given sid. If sid is None, then the rule is applied to any order placed by the algorithm. Example:

    def initialize(context):
        # Algorithm will raise an exception if we attempt to place an
        # order which would cause us to hold more than 10 shares
        # or 1000 dollars worth of sid(24).
        set_max_order_size(sid(24), max_shares=10, max_notional=1000.0)
    

    set_max_position_size(self, sid=None, max_shares=None, max_notional=None) -Set a limit on the absolute magnitude, in either shares or dollar value, of any position held by the algorithm for a given sid. If sid is None, then the rule is applied to any position held by the algorithm. Example:

    def initialize(context):
        # Algorithm will raise an exception if we attempt to order more than
        # 10 shares or 1000 dollars worth of sid(24) in a single order.
        set_max_order_size(sid(24), max_shares=10, max_notional=1000.0)
    
    ``set_max_order_count(self, max_count)``
    Set a limit on the number of orders that can be placed by the algorithm in
    a single trading day.
    Example:
    
    def initialize(context):
        # Algorithm will raise an exception if more than 50 orders are placed in a day.
        set_max_order_count(50)
    

    set_long_only(self) Set a rule specifying that the algorithm may not hold short positions. Example:

    def initialize(context):
        # Algorithm will raise an exception if it attempts to place
        # an order that would cause it to hold a short position.
        set_long_only()
    

    (#329).

  • Adds an all_api_methods classmethod on TradingAlgorithm that returns a list of all TradingAlgorithm API methods (#333).

  • Expanded record() functionality for dynamic naming. The record() function can now take positional args before the kwargs. All original usage and functionality is the same, but now these extra usages will work:

    name = 'Dynamically_Generated_String'
    record( name, value, ... )
    record( name, value1, 'name2', value2, name3=value3, name4=value4 )
    

    The requirements are simply that the poritional args occur only before the kwargs (#355).

  • history() has been ported from Quantopian to Zipline and provides moving window of market data. history() replaces BatchTransform. It is faster, works for minute level data and has a superior interface. To use it, call add_history() inside of initialize() and then receive a pandas DataFrame by calling history() from inside handle_data(). Check out the tutorial and an example. (#345 and #357).

  • history() now supports 1m window lengths (#345).

Bug Fixes

  • Fix alignment of trading days and open and closes in trading environment (#331).
  • RollingPanel fix when adding/dropping new fields (#349).

Performance

None

Maintenance and Refactorings

  • Removed undocumented and untested HDF5 and CSV data sources (#267).
  • Refactor sim_params (#352).
  • Refactoring of history (#340).

Build

  • The following dependencies have been updated (zipline might work with other versions too):

    -pytz==2013.9
    +pytz==2014.4
    +numpy==1.8.1
    -numpy==1.8.0
    +scipy==0.12.0
    +patsy==0.2.1
    +statsmodels==0.5.0
    -six==1.5.2
    +six==1.6.1
    -Cython==0.20
    +Cython==0.20.1
    -TA-Lib==0.4.8
    +--allow-external TA-Lib --allow-unverified TA-Lib TA-Lib==0.4.8
    -requests==2.2.0
    +requests==2.3.0
    -nose==1.3.0
    +nose==1.3.3
    -xlrd==0.9.2
    +xlrd==0.9.3
    -pep8==1.4.6
    +pep8==1.5.7
    -pyflakes==0.7.3
    -pip-tools==0.3.4
    +pyflakes==0.8.1`
    -scipy==0.13.2
    -tornado==3.2
    -pyparsing==2.0.1
    -patsy==0.2.1
    -statsmodels==0.4.3
    +tornado==3.2.1
    +pyparsing==2.0.2
    -Markdown==2.3.1
    +Markdown==2.4.1
    

Contributors

The following people have contributed to this release, ordered by numbers of commit:

38  Scott Sanderson
29  Thomas Wiecki
26  Eddie Hebert
 6  Delaney Granizo-Mackenzie
 3  David Edwards
 3  Richard Frank
 2  Jonathan Kamens
 1  Pankaj Garg
 1  Tony Lambiris
 1  fawce

Release 0.6.1

Release:0.6.1
Date:April 23, 2014

Highlights

  • Major fixes to risk calculations, see Bug Fixes section.
  • Port of history() function, see Enhancements section
  • Start of support for Quantopian algorithm script-syntax, see ENH section.
  • conda package manager support, see Build section.

Enhancements

  • Always process new orders i.e. on bars where handle_data isn’t called, but there is ‘clock’ data e.g. a consistent benchmark, process orders.

  • Empty positions are now filtered from the portfolio container. To help prevent algorithms from operating on positions that are not in the existing universe of stocks. Formerly, iterating over positions would return positions for stocks which had zero shares held. (Where an explicit check in algorithm code for pos.amount != 0 could prevent from using a non-existent position.)

  • Add trading calendar for BMF&Bovespa.

  • Add beginning of algo script support.

  • Starts on the path of parity with the script syntax in Quantopian’s IDE on https://quantopian.com Example:

    from datetime import datetime import pytz
    from zipline import TradingAlgorithm
    from zipline.utils.factory import load_from_yahoo
    
    from zipline.api import order
    
    def initialize(context):
        context.test = 10
    
    def handle_date(context, data):
        order('AAPL', 10)
        print(context.test)
    
    if __name__ == '__main__':
        import pylab as pl
        start = datetime(2008, 1, 1, 0, 0, 0, 0, pytz.utc)
        end = datetime(2010, 1, 1, 0, 0, 0, 0, pytz.utc)
        data = load_from_yahoo(
            stocks=['AAPL'],
            indexes={},
            start=start,
            end=end)
        data = data.dropna()
        algo = TradingAlgorithm(
            initialize=initialize,
            handle_data=handle_date)
        results = algo.run(data)
        results.portfolio_value.plot()
        pl.show()
    
  • Add HDF5 and CSV sources.

  • Limit handle_data to times with market data. To prevent cases where custom data types had unaligned timestamps, only call handle_data when market data passes through. Custom data that comes before market data will still update the data bar. But the handling of that data will only be done when there is actionable market data.

  • Extended commission PerShare method to allow a minimum cost per trade.

  • Add symbol api function A symbol() lookup feature was added to Quantopian. By adding the same API function to zipline we can make copy&pasting of a Zipline algo to Quantopian easier.

  • Add simulated random trade source. Added a new data source that emits events with certain user-specified frequency (minute or daily). This allows users to backtest and debug an algorithm in minute mode to provide a cleaner path towards Quantopian.

  • Remove dependency on benchmark for trading day calendar. Instead of the benchmarks’ index, the trading calendar is now used to populate the environment’s trading days. Remove extra_date field, since unlike the benchmarks list, the trading calendar can generate future dates, so dates for current day trading do not need to be appended. Motivations:

    • The source for the open and close/early close calendar and the trading day calendar is now the same, which should help prevent potential issues due to misalignment.
    • Allows configurations where the benchmark is provided as a generator based data source to need to supply a second benchmark list just to populate dates.
  • Port history() API method from Quantopian. Opens the core of the history() function that was previously only available on the Quantopian platform.

    The history method is analoguous to the batch_transform function/decorator, but with a hopefully more precise specification of the frequency and period of the previous bar data that is captured. Example usage:

    from zipline.api import history, add_history
    
    def initialize(context):
        add_history(bar_count=2, frequency='1d', field='price')
    
    def handle_data(context, data):
        prices = history(bar_count=2, frequency='1d', field='price')
        context.last_prices = prices
    

    N.B. this version of history lacks the backfilling capability that allows the return a full DataFrame on the first bar.

Bug Fixes

  • Adjust benchmark events to match market hours (#241). Previously benchmark events were emitted at 0:00 on the day the benchmark related to: in ‘minute’ emission mode this meant that the benchmarks were emitted before any intra-day trades were processed.

  • Ensure perf stats are generated for all days When running with minutely emissions the simulator would report to the user that it simulated ‘n - 1’ days (where n is the number of days specified in the simulation params). Now the correct number of trading days are reported as being simulated.

  • Fix repr for cumulative risk metrics. The __repr__ for RiskMetricsCumulative was referring to an older structure of the class, causing an exception when printed. Also, now prints the last values in the metrics DataFrame.

  • Prevent minute emission from crashing at end of available data. The next day calculation was causing an error when a minute emission algorithm reached the end of available data. Instead of a generic exception when available data is reached, raise and catch a named exception so that the tradesimulation loop can skip over, since the next market close is not needed at the end.

  • Fix pandas indexing in trading calendar. This could alternatively be filed under Performance. Index using loc instead of the inefficient index-ing of day, then time.

  • Prevent crash in vwap transform due to non-existent member. The WrongDataForTransform was referencing a self.fields member, which did not exist. Add a self.fields member set to price and volume and use it to iterate over during the check.

  • Fix max drawdown calculation. The input into max drawdown was incorrect, causing the bad results. i.e. the compounded_log_returns were not values representative of the algorithms total return at a given time, though calculate_max_drawdown was treating the values as if they were. Instead, the algorithm_period_returns series is now used, which does provide the total return.

  • Fix cost basis calculation. Cost basis calculation now takes direction of txn into account. Closing a long position or covering a short shouldn’t affect the cost basis.

  • Fix floating point error in order(). Where order amounts that were near an integer could accidentally be floored or ceilinged (depending on being postive or negative) to the wrong integer. e.g. an amount stored internally as -27.99999 was converted to -27 instead of -28.

  • Update perf period state when positions are changed by splits. Otherwise, self._position_amounts will be out of sync with position.amount, etc.

  • Fix misalignment of downside series calc when using exact dates. An oddity that was exposed while working on making the return series passed to the risk module more exact, the series comparison between the returns and mean returns was unbalanced, because the mean returns were not masked down to the downside data points; however, in most, if not all cases this was papered over by the call to .valid() which was removed in this change set.

  • Check that self.logger exists before using it. self.logger is initialized as None and there is no guarantee that users have set it, so check that it exists before trying to pass messages to it.

  • Prevent out of sync market closes in performance tracker. In situations where the performance tracker has been reset or patched to handle state juggling with warming up live data, the market_close member of the performance tracker could end up out of sync with the current algo time as determined by the performance tracker. The symptom was dividends never triggering, because the end of day checks would not match the current time. Fix by having the tradesimulation loop be responsible, in minute/minute mode, for advancing the market close and passing that value to the performance tracker, instead of having the market close advanced by the performance tracker as well.

  • Fix numerous cumulative and period risk calculations. The calculations that are expected to change are:

    • cumulative.beta
    • cumulative.alpha
    • cumulative.information
    • cumulative.sharpe
    • period.sortino

    How Risk Calculations Are Changing Risk Fixes for Both Period and Cumulative

    Downside Risk

    Use sample instead of population for standard deviation.

    Add a rounding factor, so that if the two values are close for a given dt, that they do not count as a downside value, which would throw off the denominator of the standard deviation of the downside diffs.

    Standard Deviation Type

    Across the board the standard deviation has been standardized to using a ‘sample’ calculation, whereas before cumulative risk was mostly using ‘population’. Using ddof=1 with np.std calculates as if the values are a sample.

    Cumulative Risk Fixes

    Beta

    Use the daily algorithm returns and benchmarks instead of annualized mean returns.

    Volatility

    Use sample instead of population with standard deviation.

    The volatility is an input to other calculations so this change affects Sharpe and Information ratio calculations.

    Information Ratio

    The benchmark returns input is changed from annualized benchmark returns to the annualized mean returns.

    Alpha

    The benchmark returns input is changed from annualized benchmark returns to the annualized mean returns.

    Period Risk Fixes

    Sortino

    Now uses the downside risk of the daily return vs. the mean algorithm returns for the minimum acceptable return instead of the treasury return.

    The above required adding the calculation of the mean algorithm returns for period risk.

    Also, uses algorithm_period_returns and tresaury_period_return as the cumulative Sortino does, instead of using algorithm returns for both inputs into the Sortino calculation.

Performance

  • Removed alias_dt transform in favor of property on SIDData. Adding a copy of the Event’s dt field as datetime via the alias_dt generator, so that the API was forgiving and allowed both datetime and dt on a SIDData object, was creating noticeable overhead, even on an noop algorithms. Instead of incurring the cost of copying the datetime value and assigning it to the Event object on every event that is passed through the system, add a property to SIDData which acts as an alias datetime to dt. Eventually support for data['foo'].datetime may be removed, and could be considered deprecated.
  • Remove the drop of ‘null return’ from cumulative returns. The check of existence of the null return key, and the drop of said return on every single bar was adding unneeded CPU time when an algorithm was run with minute emissions. Instead, add the 0.0 return with an index of the trading day before the start date. The removal of the null return was mainly in place so that the period calculation was not crashing on a non-date index value; with the index as a date, the period return can also approximate volatility (even though the that volatility has high noise-to-signal strength because it uses only two values as an input.)

Maintenance and Refactorings

  • Allow sim_params to provide data frequency for the algorithm. In the case that data_frequency of the algorithm is None, allow the sim_params to provide the data_frequency.

    Also, defer to the algorithms data frequency, if provided.

Build

  • Added support for building and releasing via conda For those who prefer building with http://conda.pydata.org/ to compiling locally with pip. The following should install Zipline on many systems.

    conda install -c quantopian zipline
    

Contributors

The following people have contributed to this release, ordered by numbers of commit:

49  Eddie Hebert
28  Thomas Wiecki
11  Richard Frank
 2  Jamie Kirkpatrick
 2  Jeremiah Lowin
 1  Colin Alexander
 1  Michael Schatzow
 1  Moises Trovo
 1  Suminda Dharmasena