PyDatastream is a Python interface to the Refinitiv Datastream (former Thomson Reuters Datastream) API via Datastream Web Services (DSWS) (non free), with some extra convenience functions. This package requires valid credentials for this API.
Note: Up until version 0.5.1 the library has been using SOAP API of DataWorksEnterprise (DWE). As of July 1, 2019 this interface was discontinued by Thompson Reuters, and at the moment Datastream content is delivered through Datastream Web Services (DSWS). Starting version 0.6 pydatastream library is using REST API of DSWS. Any earlier versions of pydatastream do not work anymore.
The latest version of PyDatastream is always available at GitHub at the master branch. Last release could be also installed from PyPI using pip:
pip install pydatastreamTwo external dependencies are pandas and requests.
All methods to work with the Datastream is organized as a class, so first you need to create an object with your valid credentials:
from pydatastream import Datastream
DS = Datastream(username="ZXXX000", password="XXX000")If necessary, the proxy server could be specified here via extra proxy parameter (e.g. proxy='proxyLocation:portNumber').
The following command requests daily closing price data for the Apple asset (Datastream mnemonic "@AAPL") in 2008:
data = DS.get_price('@AAPL', date_from='2008', date_to='2009')The data is retrieved as a pandas.DataFrame object, which can be easily plotted (data.plot()), aggregated into monthly data (`data.resample('M', how='last')``) or manipulated in a variety of ways. Due to extreme simplicity of resampling the data in Pandas library (for example, taking into account business calendar), I would recommend to request daily data (unless the requests are huge or daily scale is not applicable) and perform all transformations locally. Also note that thanks to Pandas library format of the date string is extremely flexible.
For fetching Open-High-Low-Close (OHLC) data there exist two methods: get_OHLC to fetch only price data and get_OHLCV to fetch both price and volume data. This separation is required as volume data is not available for financial indices.
Request daily OHLC and Volume data for Apple in 2008:
data = DS.get_OHLCV('@AAPL', date_from='2008', date_to='2009')Request daily OHLC data for S&P 500 Index from May 6, 2010 until present date:
data = DS.get_OHLC('S&PCOMP', date_from='May 6, 2010')If mnemonics for specific fields are known, then more general function can be used. The following request
data = DS.fetch('@AAPL', ['P', 'MV', 'VO'], date_from='2000-01-01')fetches the closing price, daily volume and market valuation for Apple Inc.
Note: several fields should be passed as a list of strings, and not as a comma-separated string (i.e. 'P,MV,VO' as a field name will raise an exception).
If called without any mnemonics for the fields (e.g. DS.fetch('@AAPL')) it will fetch one series for the "default" field, which is usually P. In this case the single column might not have any name due to inconsitency in API.
fetch can be used for requesting data for several tickers at once. In this case a MultiIndex Dataframe will be returned (see Pandas: MultiIndex / Advanced Indexing).
res = DS.fetch(['@AAPL','U:MMM'], fields=['P','MV','VO','PH'], date_from='2000-05-03')The resulting data frame could be sliced, in order to select all fields for a given ticker (res.loc['U:MMM']) or data for the specific field for all tickers ('res['MV'].unstack(level=0)').
Note: in order to force resulting dataframe to have MultiIndex even in case of a single ticker requested, set argument always_multiindex=True:
res = DS.fetch('@AAPL', fields=['P','MV','VO','PH'], date_from='2000-05-03', always_multiindex=True)Starting and ending dates could be provided as strings or pandas.Timestamp objects. One useful option is to fetch the data using BDATE as a starting date. In this case the whole series will be fetched from the first available date will be fetched. For several series - the earliest starting date of all of them should be used:
res = DS.fetch(['@AAPL','U:MMM'], 'P', date_from='BDATE')Note: In some cases, especially for many mis-aligned series, using BDATE, the Datastream server returns mis-formatted response (series for individual tickers are of different lengths and do not match the common date index), which could not be parsed. In this case it is suggested either to use explicit date_from or to split request into several smaller ones.
Important: "Default field" is the most likely reason for "E100, INVALID CODE OR EXPRESSION ENTERED" error message when fetching multiple symbols at once, even if they could be fetched one-by-one. This was observed so far in exchange rates and economic series.
There's a slight ambiguity of what "P" stands for. For cash equities there's a datatype "P" which correspond to adjusted price. However when one does the request to an API, "P" also stands for the default field. And if no fields (datatypes) are supplied, API will assume that the default field "P" is requested.
It looks like, that when one requests several symbols, API will treat the "P" (even if it is implied, i.e. no datatypes are specified) strictly as a datatype. So if the datatype "P" does not exist (e.g. for exchange rates: "EUDOLLR" or equity indices: "S&PCOMP") the request will be resulting in an error: e.g. here $$"ER", E100, INVALID CODE OR EXPRESSION ENTERED, USDOLLR(P).
So in order retrieve several symbols at once, proper datatypes should be specified. For example:
res = DS.fetch(['EUDOLLR','USDOLLR'], fields=['EB','EO'])As it will be discussed below, it may be convenient to set up property raise_on_error to False when fetching data of several tickers. In this case if one or several tickers are misspecified, error will no be raised, but they will not be present in the output data frame. Further, if some field does not exist for some of tickers, but exists for others, the missing data will be replaced with NaNs:
DS.raise_on_error = False
res = DS.fetch(['@AAPL','U:MMM','xxxxxx','S&PCOMP'], fields=['P','MV','VO','PH'], date_from='2000-05-03')
print(res['MV'].unstack(level=0).head())
print(res['P'].unstack(level=0).head())Please note, that in the last example the closing price (P) for S&PCOMP ticker was not retrieved. Due to Thomson Reuters Datastream mnemonics, field P is not available for indexes and field PI should be used instead. For the same reason, calling method get_OHLC() for getting the open, high, low, close values of the index would not work. This could be instead fetched using the direct call to fetch:
data = DS.fetch('S&PCOMP', ['PO', 'PH', 'PL', 'PI'], date_from='May 6, 2010')fetch method also collects the currencies for the given mnemonics and fields. This information is not returned on fetch by default, and kept in the property last_metadata of the main class:
res = DS.fetch(['@AAPL','U:MMM'], fields=['P','MV','VO','PH'])
print(DS.last_metadata['Currencies'])Note: Currency is not the same as the unit of the given field (data type). For example, the field VO has currency of U$ here, however the volume is reported in the number of shares.
It can be also explicitly returned by the fetch method, if the property return_metadata is set to True:
res, meta = DS.fetch(['@AAPL','U:MMM'], fields=['P','MV','VO','PH'], return_metadata=True)
print(meta['Currencies'])In this case fetch also collects the names of the symbols and fields (data types):
print(meta['SymbolNames'])
print(meta['DataTypeNames'])PyDatastream also has an interface for retrieving list of constituents of indices:
DS.get_constituents('S&PCOMP')Note: In contrast to retired DWE interface, DSWS does not support fetching historical lists of constituents.
By default the method retrieves many various mnemonics and codes for the constituents, which might pose a problem for large indices like Russel-3000 (sometimes the request might be killed on timeout). In this case one can request only symbols and company names using:
DS.get_constituents('FRUSS3L', only_list=True)Another useful method is get_all_listings(), that retrieves mnemonics (columns MJ and CD), countries (column GL) and exchanges (column MK) for different listings of the same stocks, as well as indicator for primary listing (columns PQ). For example for IBM and Vodafone:
DS.get_all_listings(['VOD', 'U:IBM'])Finally, various symbols, codes and mnemonics could be fetched using the following method:
DS.get_codes(['U:IBM', '@MSFT'])Futures contracts have mnemonics of the form XXXMMYY, where XXX is a code for a futures market and MM and YY encode date and year of the contract. For example, LLC0118 would denote market LLC (Brent Crude Oil) expiring in January 2018.
For a given market code it is possible to get a list of all active contracts (together with some extra information, such as exchange traded, tick size, contract size, last trading date etc.):
DS.get_futures_contracts('LLC')By default only active contracts are retrieved. In order to get the full list of contracts, use include_dead=True argument:
DS.get_futures_contracts('LLC', include_dead=True)Similar to the index constituents, it is possible to set up an argument only_list=True in order to limit output to only mnemonics and names. This argument is suggested if server returns an error "Index was outside the bounds of the array." on the full request, such as in the following case:
DS.get_futures_contracts('NCT', include_dead=True, only_list=True)Note: Futures contracts do not operate with the "Price" field (P) and instead they use "Settlement Price" (PS):
DS.fetch('LLC0118', 'PS', date_from='BDATE')List of constituents of indices, that were considered above, is an example of static request, i.e. a request that does not retrieve a time-series, but a single snapshot with the data.
Within the PyDatastream library static requests could be called using the same fetch() method by providing an argument static=True.
For example, this request retrieves names, ISINs and identifiers for primary exchange (ISINID) for various tickers of BASF corporation (part of the get_codes() method):
DS.fetch(['D:BAS','D:BASX','HN:BAS','I:BAF','BFA','@BFFAF','S:BAS'], ['ISIN', 'ISINID', 'NAME'], static=True)Another example of use of static requests is a cross-section of time-series. The following example retrieves an actual price, market capitalization and daily volume for same companies:
DS.fetch(['D:BAS','D:BASX','HN:BAS','I:BAF','BFA','@BFFAF','S:BAS'], ['P', 'MV', 'VO'], static=True)Datastream allows to apply a number of functions to the series, which are calculated on the server side. Given the flexibility of pandas library in all types of data transformation, this functionality is not really needed for python users. However I will briefly describe it for the sake of completeness.
Functions have a format FUNC#(mnemonic,parameter). For example, calculating moving average on 20 days on the prices of IBM:
DS.fetch('MAV#(U:IBM,20D)', date_from='2013-09-01')Functions could be combined, e.g. calculating moving 3 day percentage change on 20 days moving average:
DS.fetch('PCH#(MAV#(U:IBM,20D),3D)', date_from='2013-09-01')Calculate percentage quarter-on-quarter change for the US real GDP (constant prices, seasonally adjusted):
DS.fetch('PCH#(USGDP...D,1Q)', date_from='1990-01-01')Calculate year-on-year difference (actual change) for the UK real GDP (constant prices, seasonally adjusted):
DS.fetch('ACH#(UKGDP...D,1Y)', date_from='1990-01-01')Documentation on the available functions are available on the Thompson Reuters webhelp.
By default Datastream does padding of the prices on the holidays (i.e. on a vacation days it returns the price of the previous day). In order to remove the padding one need to know which days were holidays in the past.
The history of trading and non-trading days for each country could be returned using the function (get_trading_days):
DS.get_trading_days(['US', 'UK', 'RS'], date_from='2010-01-01')This function will return a dataframe that contains values 1 and NaN, where 1 identifies the business day. So by multiplying this list with the price time series it will remove padded values on non-trading days:
DS.fetch('@AAPL', 'P', date_from='2010').mul(DS.get_trading_days('US', date_from='2010')['US'], axis=0)The full list of available calendars (countries) is contained in the property DS.vacations_list.
Thomson Dataworks Enterprise User Guide suggests to optimize requests: very often it is quicker to make one bigger request than several smaller requests because of the relatively high transport overhead with web services.
PyDatastream allows to fetch several requests in a single API call (so-called "bundle request"):
r1 = DS.construct_request('@AAPL', ['PO', 'PH', 'PL', 'P'], date_from='2013-11-26')
r2 = DS.construct_request('U:MMM', ['P', 'MV', 'PO'], date_from='2013-11-26')
res = DS.request_many([r1, r2])
dfs = DS.parse_response(res)
print(dfs[0].head())
print(dfs[1].head())If request contains errors then normally DatastreamException will be raised and the error message from the DSWS will be printed. To alter this behavior, one can use raise_on_error property of Datastream class. Being set to False it will force parser to ignore error messages and return empty pandas.Dataframe. For instance, since the field "P" does not exist for indices the following request would have raised an error, but instead it will only fill first 3 fields:
DS.raise_on_error = False
DS.fetch('S&PCOMP', ['PO', 'PH', 'PL', 'P'])raise_on_error can be useful for requests that contain several tickers. In this case data fields and/or tickers that can not be fetched will not be present in the output. However please be aware, that this is not a silver bullet against all types of errors. In all complicated cases it is suggested to go with one symbol at a time and check the raw response from the server to spot invalid combinations of ticker-field.
For the debugging purposes, Datastream class has last_request property, contains URL, raw JSON request and raw JSON response from the last call to the API.
print(DS.last_request['url'])
print(DS.last_request['request'])
print(DS.last_request['response'])In many cases the errors occur at the stage of parsing the raw response from the server. For example, this could happen if some fields requested are not supported for all symbols in the request. Because the data returned from the server in the unstructured form and the parser tries to map this to the structured pandas.DataFrame, the exceptions could be relatively uninformative. In this case it is highly suggested to check the raw response from the server to spot the problematic field.
Many data types (fields) in Datastream are asset type-specific. In order to check what exactly types do your symbols have call get_asset_types:
DS.get_asset_types(['MMM', '@AAPL', 'S&PCOMP', 'FRUSS3L', 'EUDOLLR', 'USGDP...D', 'FRFEDFD'])Statistics about the number of requests, datatypes and returned data points for a given account could be requested using the usage_statistics method. By default it returns stats for the present month, but it could also return the usage statistics for some specified month in the past, or even the range of months. For instance: the usage statistics (monthly) for the last year could be fetched using:
DS.usage_statistics(months=12)PyDatastream has two useful methods to work with the Thomson Reuters Economic Point-in-Time (EPiT) concept (usually a separate subscription is required for this content). Most of the economic time series, such as GDP, employment or inflation figures are undergoing many revisions on their way. So that US GDP value undertakes two consecutive revisions after the initial release, after which the number is revised periodically when e.g. either the base year of calculation or seasonal adjustment are changed. For predictive exercises it is important to obtain the actual values as they were known at the time of releases (otherwise the data will be contaminated by look-ahead bias).
EPiT allows user to request such historical data. For example, the following method retrieves the initial estimate and first revisions of the 2010-Q1 US GDP, as well as the dates when the data was published:
DS.get_epit_revisions('USGDP...D', period='2010-02-15')The period here should contain a date which falls within a time period of interest (so any date from '2010-01-01' to '2010-03-31' will result in the same output).
Another useful concept is a concept of "vintage" of a data, which defines when the particular series was released. This concept is widely used in economics, see for example ALFRED database of St. Louis Fed and their discussions about capturing data as it happens.
All vintages of the economic indicator could be summarized in a vintage matrix, that represents a DataFrame where columns correspond to a particular period (quarter or month) for the reported statistic and index represents timestamps at which these values were released by the respective official agency. I.e. every line corresponds to all available reported values by the given date.
For example for the US GDP:
DS.get_epit_vintage_matrix('USGDP...D', date_from='2015-01-01')The response is:
2015-02-15 2015-05-15 2015-08-15 2015-11-15 \
2015-04-29 16304.80 NaN NaN NaN
2015-05-29 16264.10 NaN NaN NaN
2015-06-24 16287.70 NaN NaN NaN
2015-07-30 16177.30 16270.400 NaN NaN
2015-08-27 16177.30 16324.300 NaN NaN
2015-09-25 16177.30 16333.600 NaN NaN
2015-10-29 16177.30 16333.600 16394.200 NaN
2015-11-24 16177.30 16333.600 16417.800 NaN
...From the matrix it is seen for example, that the advance GDP estimate for 2015-Q1 (corresponding to 2015-02-15) was released on 2015-04-29 and was equal to 16304.80 (B USD). The first revision (16264.10) has happened on 2015-05-29 and the second (16287.70) - on 2015-06-24. On 2015-07-30 the advance GDP figure for 2015-Q2 was released (16270.400) together with update on the 2015-Q1 value (16177.30) and so on.
Finally, when using economic data in any real-life analysis, it is important to know when the next value will be released by the reporting body. For example the following method returns dates of next 4 releases for US GDP and US Nonfarm Payroll figures:
DS.get_next_release_dates(['USGDP...D', 'USEMPALLO'], n_releases=4)In the response release number is counted from the present day to the future. "DATE_FLAG" indicates whether the dates are given by the official agency or estimated by Thomson Reuters; and "TYPE" field indicates whether the new value is going to be provided or an old value is going to be updated.
Datastream Navigator could be used to look up codes and data types.
All these links could be printed in your terminal or iPython notebook by calling
DS.info()Finally, for quick start with pandas have a look on tutorial notebook and 10-minutes introduction.
Datastream Navigator is the best place to search for mnemonics for a particular security or a list of available datatypes/fields.
Once a necessary security is located (either by search or via "Explore" menu item), a short list of most frequently used mnemonics is located right under the chart with the series. Further the small button ">>" next to them open a longer list.
The complete list of datatypes (including static) for a particular asset class is located in the "Datatype search" menu item. Here the proper asset class should be selected.
Help for Datastream Navigator is available here.
get_constituents() and other useful methodsfetch() for many tickers returns MultiIndex Dataframe instead of former Panel. This follows the development of the Pandas library where the Panels are deprecated starting version 0.20.0 (see here).system_info() or sources()) have been removed as they're not supported by a new API.get_futures_contracts() and get_next_release_dates().always_multiindex argument to fetch()Note 1: any versions of pydatastream prior to 0.6 will not work anymore.
Note 2: Starting version 0.6 Python 2 is no longer supported - in line with many critical projects (including numpy, pandas and matplotlib) dropping support of Python 2 by 2020.
A special thanks to:
PyDatastream library is released under the MIT license.
The license for the library is not extended in any sense to any of the content of the Refinitiv (former Thomson Reuters) Dataworks Enterprise, Datastream, Datastream Web Services, Datastream Navigator or related services. Appropriate contract with the data vendor and valid credentials are required in order to use the API.
Author of the library (@vfilimonov) is not affiliated, associated, authorized, sponsored, endorsed by, or in any way officially connected with Thomson Reuters, or any of its subsidiaries or its affiliates. The names "Refinitiv", "Thomson Reuters" as well as related names are registered trademarks of respective owners.