Remote Data Access

Warning

In pandas 0.17.0, the sub-package pandas.io.data will be removed in favor of a separately installable pandas-datareader package. This will allow the data modules to be independently updated to your pandas installation. The API for pandas-datareader v0.1.1 is the same as in pandas v0.16.1. (GH8961)

You should replace the imports of the following:

from pandas.io import data, wb

With:

from pandas_datareader import data, wb

Functions from pandas.io.data and pandas.io.ga extract data from various Internet sources into a DataFrame. Currently the following sources are supported:

It should be noted, that various sources support different kinds of data, so not all sources implement the same methods and the data elements returned might also differ.

Yahoo! Finance

In [1]: import pandas.io.data as web

In [2]: import datetime

In [3]: start = datetime.datetime(2010, 1, 1)

In [4]: end = datetime.datetime(2013, 1, 27)

In [5]: f = web.DataReader("F", 'yahoo', start, end)

In [6]: f.ix['2010-01-04']
Out[6]: 
Open               10.17000
High               10.28000
Low                10.05000
Close              10.28000
Volume       60855800.00000
Adj Close           9.33868
Name: 2010-01-04 00:00:00, dtype: float64

Yahoo! Finance Options

*Experimental*

The Options class allows the download of options data from Yahoo! Finance.

The get_all_data method downloads and caches option data for all expiry months and provides a formatted DataFrame with a hierarchical index, so it is easy to get to the specific option you want.

In [7]: from pandas.io.data import Options

In [8]: aapl = Options('aapl', 'yahoo')

In [9]: data = aapl.get_all_data()

In [10]: data.iloc[0:5, 0:5]
Out[10]: 
                                             Last  Bid  Ask  Chg PctChg
Strike Expiry     Type Symbol                                          
34.29  2016-01-15 call AAPL160115C00034290  90.95    0    0    0  0.00%
                  put  AAPL160115P00034290   0.03    0    0    0  0.00%
35.71  2016-01-15 call AAPL160115C00035710  88.30    0    0    0  0.00%
                  put  AAPL160115P00035710   0.04    0    0    0  0.00%
37.14  2016-01-15 call AAPL160115C00037140  90.10    0    0    0  0.00%

# Show the $100 strike puts at all expiry dates:
In [11]: data.loc[(100, slice(None), 'put'),:].iloc[0:5, 0:5]
Out[11]: 
                                            Last  Bid  Ask  Chg PctChg
Strike Expiry     Type Symbol                                         
100    2015-05-15 put  AAPL150515P00100000  0.03    0    0    0  0.00%
       2015-05-22 put  AAPL150522P00100000  0.04    0    0    0  0.00%
       2015-05-29 put  AAPL150529P00100000  0.05    0    0    0  0.00%
       2015-06-05 put  AAPL150605P00100000  0.06    0    0    0  0.00%
       2015-06-12 put  AAPL150612P00100000  0.11    0    0    0  0.00%

# Show the volume traded of $100 strike puts at all expiry dates:
In [12]: data.loc[(100, slice(None), 'put'),'Vol'].head()
Out[12]: 
Strike  Expiry      Type  Symbol             
100     2015-05-15  put   AAPL150515P00100000    1311
        2015-05-22  put   AAPL150522P00100000      49
        2015-05-29  put   AAPL150529P00100000      29
        2015-06-05  put   AAPL150605P00100000    1000
        2015-06-12  put   AAPL150612P00100000       2
Name: Vol, dtype: int64

If you don’t want to download all the data, more specific requests can be made.

In [13]: import datetime

In [14]: expiry = datetime.date(2016, 1, 1)

In [15]: data = aapl.get_call_data(expiry=expiry)

In [16]: data.iloc[0:5:, 0:5]
Out[16]: 
                                             Last  Bid  Ask  Chg PctChg
Strike Expiry     Type Symbol                                          
34.29  2016-01-15 call AAPL160115C00034290  90.95    0    0    0  0.00%
35.71  2016-01-15 call AAPL160115C00035710  88.30    0    0    0  0.00%
37.14  2016-01-15 call AAPL160115C00037140  90.10    0    0    0  0.00%
38.57  2016-01-15 call AAPL160115C00038570  86.45    0    0    0  0.00%
40.00  2016-01-15 call AAPL160115C00040000  87.52    0    0    0  0.00%

Note that if you call get_all_data first, this second call will happen much faster, as the data is cached.

If a given expiry date is not available, data for the next available expiry will be returned (January 15, 2015 in the above example).

Available expiry dates can be accessed from the expiry_dates property.

In [17]: aapl.expiry_dates
Out[17]: 
[datetime.date(2015, 5, 15),
 datetime.date(2015, 5, 22),
 datetime.date(2015, 5, 29),
 datetime.date(2015, 6, 5),
 datetime.date(2015, 6, 12),
 datetime.date(2015, 6, 19),
 datetime.date(2015, 6, 26),
 datetime.date(2015, 7, 17),
 datetime.date(2015, 8, 21),
 datetime.date(2015, 10, 16),
 datetime.date(2016, 1, 15),
 datetime.date(2017, 1, 20)]

In [18]: data = aapl.get_call_data(expiry=aapl.expiry_dates[0])

In [19]: data.iloc[0:5:, 0:5]
Out[19]: 
                                             Last  Bid  Ask  Chg PctChg
Strike Expiry     Type Symbol                                          
70     2015-05-15 call AAPL150515C00070000  53.88    0    0    0  0.00%
75     2015-05-15 call AAPL150515C00075000  52.42    0    0    0  0.00%
80     2015-05-15 call AAPL150515C00080000  45.60    0    0    0  0.00%
85     2015-05-15 call AAPL150515C00085000  39.80    0    0    0  0.00%
90     2015-05-15 call AAPL150515C00090000  37.45    0    0    0  0.00%

A list-like object containing dates can also be passed to the expiry parameter, returning options data for all expiry dates in the list.

In [20]: data = aapl.get_near_stock_price(expiry=aapl.expiry_dates[0:3])

In [21]: data.iloc[0:5:, 0:5]
Out[21]: 
                                            Last  Bid  Ask  Chg PctChg
Strike Expiry     Type Symbol                                         
127    2015-05-22 call AAPL150522C00127000  2.45    0    0    0  0.00%
       2015-05-29 call AAPL150529C00127000  2.93    0    0    0  0.00%
128    2015-05-15 call AAPL150515C00128000  1.21    0    0    0  0.00%
       2015-05-22 call AAPL150522C00128000  1.90    0    0    0  0.00%
       2015-05-29 call AAPL150529C00128000  2.38    0    0    0  0.00%

The month and year parameters can be used to get all options data for a given month.

Google Finance

In [22]: import pandas.io.data as web

In [23]: import datetime

In [24]: start = datetime.datetime(2010, 1, 1)

In [25]: end = datetime.datetime(2013, 1, 27)

In [26]: f = web.DataReader("F", 'google', start, end)

In [27]: f.ix['2010-01-04']
Out[27]: 
Open            10.17
High            10.28
Low             10.05
Close           10.28
Volume    60855796.00
Name: 2010-01-04 00:00:00, dtype: float64

FRED

In [28]: import pandas.io.data as web

In [29]: import datetime

In [30]: start = datetime.datetime(2010, 1, 1)

In [31]: end = datetime.datetime(2013, 1, 27)

In [32]: gdp=web.DataReader("GDP", "fred", start, end)

In [33]: gdp.ix['2013-01-01']
Out[33]: 
GDP    16502.4
Name: 2013-01-01 00:00:00, dtype: float64

# Multiple series:
In [34]: inflation = web.DataReader(["CPIAUCSL", "CPILFESL"], "fred", start, end)

In [35]: inflation.head()
Out[35]: 
            CPIAUCSL  CPILFESL
DATE                          
2010-01-01   217.488   220.633
2010-02-01   217.281   220.731
2010-03-01   217.353   220.783
2010-04-01   217.403   220.822
2010-05-01   217.290   220.962

Fama/French

Dataset names are listed at Fama/French Data Library.

In [36]: import pandas.io.data as web

In [37]: ip = web.DataReader("5_Industry_Portfolios", "famafrench")

In [38]: ip[4].ix[192607]
Out[38]: 
1 Cnsmr    5.43
2 Manuf    2.73
3 HiTec    1.83
4 Hlth     1.77
5 Other    2.16
Name: 192607, dtype: float64

World Bank

pandas users can easily access thousands of panel data series from the World Bank’s World Development Indicators by using the wb I/O functions.

Indicators

Either from exploring the World Bank site, or using the search function included, every world bank indicator is accessible.

For example, if you wanted to compare the Gross Domestic Products per capita in constant dollars in North America, you would use the search function:

In [1]: from pandas.io import wb

In [2]: wb.search('gdp.*capita.*const').iloc[:,:2]
Out[2]:
                     id                                               name
3242            GDPPCKD             GDP per Capita, constant US$, millions
5143     NY.GDP.PCAP.KD                 GDP per capita (constant 2005 US$)
5145     NY.GDP.PCAP.KN                      GDP per capita (constant LCU)
5147  NY.GDP.PCAP.PP.KD  GDP per capita, PPP (constant 2005 internation...

Then you would use the download function to acquire the data from the World Bank’s servers:

In [3]: dat = wb.download(indicator='NY.GDP.PCAP.KD', country=['US', 'CA', 'MX'], start=2005, end=2008)

In [4]: print(dat)
                      NY.GDP.PCAP.KD
country       year
Canada        2008  36005.5004978584
              2007  36182.9138439757
              2006  35785.9698172849
              2005  35087.8925933298
Mexico        2008  8113.10219480083
              2007  8119.21298908649
              2006  7961.96818458178
              2005  7666.69796097264
United States 2008  43069.5819857208
              2007  43635.5852068142
              2006   43228.111147107
              2005  42516.3934699993

The resulting dataset is a properly formatted DataFrame with a hierarchical index, so it is easy to apply .groupby transformations to it:

In [6]: dat['NY.GDP.PCAP.KD'].groupby(level=0).mean()
Out[6]:
country
Canada           35765.569188
Mexico            7965.245332
United States    43112.417952
dtype: float64

Now imagine you want to compare GDP to the share of people with cellphone contracts around the world.

In [7]: wb.search('cell.*%').iloc[:,:2]
Out[7]:
                     id                                               name
3990  IT.CEL.SETS.FE.ZS  Mobile cellular telephone users, female (% of ...
3991  IT.CEL.SETS.MA.ZS  Mobile cellular telephone users, male (% of po...
4027      IT.MOB.COV.ZS  Population coverage of mobile cellular telepho...

Notice that this second search was much faster than the first one because pandas now has a cached list of available data series.

In [13]: ind = ['NY.GDP.PCAP.KD', 'IT.MOB.COV.ZS']
In [14]: dat = wb.download(indicator=ind, country='all', start=2011, end=2011).dropna()
In [15]: dat.columns = ['gdp', 'cellphone']
In [16]: print(dat.tail())
                        gdp  cellphone
country   year
Swaziland 2011  2413.952853       94.9
Tunisia   2011  3687.340170      100.0
Uganda    2011   405.332501      100.0
Zambia    2011   767.911290       62.0
Zimbabwe  2011   419.236086       72.4

Finally, we use the statsmodels package to assess the relationship between our two variables using ordinary least squares regression. Unsurprisingly, populations in rich countries tend to use cellphones at a higher rate:

In [17]: import numpy as np
In [18]: import statsmodels.formula.api as smf
In [19]: mod = smf.ols("cellphone ~ np.log(gdp)", dat).fit()
In [20]: print(mod.summary())
                            OLS Regression Results
==============================================================================
Dep. Variable:              cellphone   R-squared:                       0.297
Model:                            OLS   Adj. R-squared:                  0.274
Method:                 Least Squares   F-statistic:                     13.08
Date:                Thu, 25 Jul 2013   Prob (F-statistic):            0.00105
Time:                        15:24:42   Log-Likelihood:                -139.16
No. Observations:                  33   AIC:                             282.3
Df Residuals:                      31   BIC:                             285.3
Df Model:                           1
===============================================================================
                  coef    std err          t      P>|t|      [95.0% Conf. Int.]
-------------------------------------------------------------------------------
Intercept      16.5110     19.071      0.866      0.393       -22.384    55.406
np.log(gdp)     9.9333      2.747      3.616      0.001         4.331    15.535
==============================================================================
Omnibus:                       36.054   Durbin-Watson:                   2.071
Prob(Omnibus):                  0.000   Jarque-Bera (JB):              119.133
Skew:                          -2.314   Prob(JB):                     1.35e-26
Kurtosis:                      11.077   Cond. No.                         45.8
==============================================================================

Country Codes

New in version 0.15.1.

The country argument accepts a string or list of mixed two or three character ISO country codes, as well as dynamic World Bank exceptions to the ISO standards.

For a list of the the hard-coded country codes (used solely for error handling logic) see pandas.io.wb.country_codes.

Problematic Country Codes & Indicators

Note

The World Bank’s country list and indicators are dynamic. As of 0.15.1, wb.download() is more flexible. To achieve this, the warning and exception logic changed.

The world bank converts some country codes in their response, which makes error checking by pandas difficult. Retired indicators still persist in the search.

Given the new flexibility of 0.15.1, improved error handling by the user may be necessary for fringe cases.

To help identify issues:

There are at least 4 kinds of country codes:

  1. Standard (2/3 digit ISO) - returns data, will warn and error properly.
  2. Non-standard (WB Exceptions) - returns data, but will falsely warn.
  3. Blank - silently missing from the response.
  4. Bad - causes the entire response from WB to fail, always exception inducing.

There are at least 3 kinds of indicators:

  1. Current - Returns data.
  2. Retired - Appears in search results, yet won’t return data.
  3. Bad - Will not return data.

Use the errors argument to control warnings and exceptions. Setting errors to ignore or warn, won’t stop failed responses. (ie, 100% bad indicators, or a single “bad” (#4 above) country code).

See docstrings for more info.

Google Analytics

The ga module provides a wrapper for Google Analytics API to simplify retrieving traffic data. Result sets are parsed into a pandas DataFrame with a shape and data types derived from the source table.

Configuring Access to Google Analytics

The first thing you need to do is to setup accesses to Google Analytics API. Follow the steps below:

  1. In the Google Developers Console
    1. enable the Analytics API
    2. create a new project
    3. create a new Client ID for an “Installed Application” (in the “APIs & auth / Credentials section” of the newly created project)
    4. download it (JSON file)
  2. On your machine
    1. rename it to client_secrets.json
    2. move it to the pandas/io module directory

The first time you use the read_ga() funtion, a browser window will open to ask you to authentify to the Google API. Do proceed.

Using the Google Analytics API

The following will fetch users and pageviews (metrics) data per day of the week, for the first semester of 2014, from a particular property.

import pandas.io.ga as ga
ga.read_ga(
    account_id  = "2360420",
    profile_id  = "19462946",
    property_id = "UA-2360420-5",
    metrics     = ['users', 'pageviews'],
    dimensions  = ['dayOfWeek'],
    start_date  = "2014-01-01",
    end_date    = "2014-08-01",
    index_col   = 0,
    filters     = "pagePath=~aboutus;ga:country==France",
)

The only mandatory arguments are metrics, dimensions and start_date. We strongly recommend that you always specify the account_id, profile_id and property_id to avoid accessing the wrong data bucket in Google Analytics.

The index_col argument indicates which dimension(s) has to be taken as index.

The filters argument indicates the filtering to apply to the query. In the above example, the page URL has to contain aboutus AND the visitors country has to be France.

Detailed information in the following: