pandas 0.8.1 documentation

Computational tools

Statistical functions

Percent Change

Both Series and DataFrame has a method pct_change to compute the percent change over a given number of periods (using fill_method to fill NA/null values).

In [187]: ser = Series(randn(8))

In [188]: ser.pct_change()
Out[188]: 
0         NaN
1   -1.602976
2    4.334938
3   -0.247456
4   -2.067345
5   -1.142903
6   -1.688214
7   -9.759729
In [189]: df = DataFrame(randn(10, 4))

In [190]: df.pct_change(periods=3)
Out[190]: 
          0         1         2         3
0       NaN       NaN       NaN       NaN
1       NaN       NaN       NaN       NaN
2       NaN       NaN       NaN       NaN
3 -0.218320 -1.054001  1.987147 -0.510183
4 -0.439121 -1.816454  0.649715 -4.822809
5 -0.127833 -3.042065 -5.866604 -1.776977
6 -2.596833 -1.959538 -2.111697 -3.798900
7 -0.117826 -2.169058  0.036094 -0.067696
8  2.492606 -1.357320 -1.205802 -1.558697
9 -1.012977  2.324558 -1.003744 -0.371806

Covariance

The Series object has a method cov to compute covariance between series (excluding NA/null values).

In [191]: s1 = Series(randn(1000))

In [192]: s2 = Series(randn(1000))

In [193]: s1.cov(s2)
Out[193]: 0.00068010881743109321

Analogously, DataFrame has a method cov to compute pairwise covariances among the series in the DataFrame, also excluding NA/null values.

In [194]: frame = DataFrame(randn(1000, 5), columns=['a', 'b', 'c', 'd', 'e'])

In [195]: frame.cov()
Out[195]: 
          a         b         c         d         e
a  1.000882 -0.003177 -0.002698 -0.006889  0.031912
b -0.003177  1.024721  0.000191  0.009212  0.000857
c -0.002698  0.000191  0.950735 -0.031743 -0.005087
d -0.006889  0.009212 -0.031743  1.002983 -0.047952
e  0.031912  0.000857 -0.005087 -0.047952  1.042487

Correlation

Several methods for computing correlations are provided. Several kinds of correlation methods are provided:

Method name Description
pearson (default) Standard correlation coefficient
kendall Kendall Tau correlation coefficient
spearman Spearman rank correlation coefficient

All of these are currently computed using pairwise complete observations.

In [196]: frame = DataFrame(randn(1000, 5), columns=['a', 'b', 'c', 'd', 'e'])

In [197]: frame.ix[::2] = np.nan

# Series with Series
In [198]: frame['a'].corr(frame['b'])
Out[198]: 0.010052135416653445

In [199]: frame['a'].corr(frame['b'], method='spearman')
Out[199]: -0.0097383749534998149

# Pairwise correlation of DataFrame columns
In [200]: frame.corr()
Out[200]: 
          a         b         c         d         e
a  1.000000  0.010052 -0.047750 -0.031461 -0.025285
b  0.010052  1.000000 -0.014172 -0.020590 -0.001930
c -0.047750 -0.014172  1.000000  0.006373 -0.049479
d -0.031461 -0.020590  0.006373  1.000000 -0.012379
e -0.025285 -0.001930 -0.049479 -0.012379  1.000000

Note that non-numeric columns will be automatically excluded from the correlation calculation.

A related method corrwith is implemented on DataFrame to compute the correlation between like-labeled Series contained in different DataFrame objects.

In [201]: index = ['a', 'b', 'c', 'd', 'e']

In [202]: columns = ['one', 'two', 'three', 'four']

In [203]: df1 = DataFrame(randn(5, 4), index=index, columns=columns)

In [204]: df2 = DataFrame(randn(4, 4), index=index[:4], columns=columns)

In [205]: df1.corrwith(df2)
Out[205]: 
one      0.803464
two      0.142469
three   -0.498774
four     0.806420

In [206]: df2.corrwith(df1, axis=1)
Out[206]: 
a    0.011572
b    0.388066
c   -0.335819
d    0.232412
e         NaN

Data ranking

The rank method produces a data ranking with ties being assigned the mean of the ranks (by default) for the group:

In [207]: s = Series(np.random.randn(5), index=list('abcde'))

In [208]: s['d'] = s['b'] # so there's a tie

In [209]: s.rank()
Out[209]: 
a    2.0
b    4.5
c    3.0
d    4.5
e    1.0

rank is also a DataFrame method and can rank either the rows (axis=0) or the columns (axis=1). NaN values are excluded from the ranking.

In [210]: df = DataFrame(np.random.randn(10, 6))

In [211]: df[4] = df[2][:5] # some ties

In [212]: df
Out[212]: 
          0         1         2         3         4         5
0  0.085011 -0.459422 -1.660917 -1.913019 -1.660917  0.833479
1 -0.557052  0.775425  0.003794  0.555351  0.003794 -1.169977
2  0.815695 -0.295737 -0.534290  0.068917 -0.534290 -0.513855
3  1.465947  0.021757  0.523224 -0.439297  0.523224 -0.959568
4 -0.678378  0.091855  1.337956  0.792551  1.337956  0.711776
5 -0.190285  0.187520 -0.355562  1.730964       NaN -1.362312
6 -0.776678 -2.082637 -0.165877  0.357163       NaN  0.631662
7 -1.295037  0.367656 -1.886797 -0.531790       NaN  1.270408
8  1.106052  0.848312 -0.613544  1.338296       NaN -1.150652
9  0.309979  1.088439  0.920366 -0.750322       NaN  1.563956

In [213]: df.rank(1)
Out[213]: 
   0  1    2  3    4  5
0  5  4  2.5  1  2.5  6
1  2  6  3.5  5  3.5  1
2  6  4  1.5  5  1.5  3
3  6  3  4.5  2  4.5  1
4  1  2  5.5  4  5.5  3
5  3  4  2.0  5  NaN  1
6  2  1  3.0  4  NaN  5
7  2  4  1.0  3  NaN  5
8  4  3  2.0  5  NaN  1
9  2  4  3.0  1  NaN  5

rank optionally takes a parameter ascending which by default is true; when false, data is reverse-ranked, with larger values assigned a smaller rank.

rank supports different tie-breaking methods, specified with the method parameter:

  • average : average rank of tied group
  • min : lowest rank in the group
  • max : highest rank in the group
  • first : ranks assigned in the order they appear in the array

Note

These methods are significantly faster (around 10-20x) than scipy.stats.rankdata.

Moving (rolling) statistics / moments

For working with time series data, a number of functions are provided for computing common moving or rolling statistics. Among these are count, sum, mean, median, correlation, variance, covariance, standard deviation, skewness, and kurtosis. All of these methods are in the pandas namespace, but otherwise they can be found in pandas.stats.moments.

Function Description
rolling_count Number of non-null observations
rolling_sum Sum of values
rolling_mean Mean of values
rolling_median Arithmetic median of values
rolling_min Minimum
rolling_max Maximum
rolling_std Unbiased standard deviation
rolling_var Unbiased variance
rolling_skew Unbiased skewness (3rd moment)
rolling_kurt Unbiased kurtosis (4th moment)
rolling_quantile Sample quantile (value at %)
rolling_apply Generic apply
rolling_cov Unbiased covariance (binary)
rolling_corr Correlation (binary)
rolling_corr_pairwise Pairwise correlation of DataFrame columns

Generally these methods all have the same interface. The binary operators (e.g. rolling_corr) take two Series or DataFrames. Otherwise, they all accept the following arguments:

  • window: size of moving window
  • min_periods: threshold of non-null data points to require (otherwise result is NA)
  • freq: optionally specify a :ref: frequency string <timeseries.alias> or DateOffset to pre-conform the data to. Note that prior to pandas v0.8.0, a keyword argument time_rule was used instead of freq that referred to the legacy time rule constants

These functions can be applied to ndarrays or Series objects:

In [214]: ts = Series(randn(1000), index=date_range('1/1/2000', periods=1000))

In [215]: ts = ts.cumsum()

In [216]: ts.plot(style='k--')
Out[216]: <matplotlib.axes.AxesSubplot at 0x11043e110>

In [217]: rolling_mean(ts, 60).plot(style='k')
Out[217]: <matplotlib.axes.AxesSubplot at 0x11043e110>
_images/rolling_mean_ex.png

They can also be applied to DataFrame objects. This is really just syntactic sugar for applying the moving window operator to all of the DataFrame’s columns:

In [218]: df = DataFrame(randn(1000, 4), index=ts.index,
   .....:                columns=['A', 'B', 'C', 'D'])
   .....:

In [219]: df = df.cumsum()

In [220]: rolling_sum(df, 60).plot(subplots=True)
Out[220]: 
array([Axes(0.125,0.747826;0.775x0.152174),
       Axes(0.125,0.565217;0.775x0.152174),
       Axes(0.125,0.382609;0.775x0.152174), Axes(0.125,0.2;0.775x0.152174)], dtype=object)
_images/rolling_mean_frame.png

The rolling_apply function takes an extra func argument and performs generic rolling computations. The func argument should be a single function that produces a single value from an ndarray input. Suppose we wanted to compute the mean absolute deviation on a rolling basis:

In [221]: mad = lambda x: np.fabs(x - x.mean()).mean()

In [222]: rolling_apply(ts, 60, mad).plot(style='k')
Out[222]: <matplotlib.axes.AxesSubplot at 0x11030d190>
_images/rolling_apply_ex.png

Binary rolling moments

rolling_cov and rolling_corr can compute moving window statistics about two Series or any combination of DataFrame/Series or DataFrame/DataFrame. Here is the behavior in each case:

  • two Series: compute the statistic for the pairing
  • DataFrame/Series: compute the statistics for each column of the DataFrame with the passed Series, thus returning a DataFrame
  • DataFrame/DataFrame: compute statistic for matching column names, returning a DataFrame

For example:

In [223]: df2 = df[:20]

In [224]: rolling_corr(df2, df2['B'], window=5)
Out[224]: 
                   A   B         C         D
2000-01-01       NaN NaN       NaN       NaN
2000-01-02       NaN NaN       NaN       NaN
2000-01-03       NaN NaN       NaN       NaN
2000-01-04       NaN NaN       NaN       NaN
2000-01-05  0.703188   1 -0.746130  0.714265
2000-01-06  0.065322   1 -0.209789  0.635360
2000-01-07 -0.429914   1 -0.100807  0.266005
2000-01-08 -0.387498   1  0.512321  0.592033
2000-01-09  0.442207   1  0.570186 -0.653242
2000-01-10  0.572983   1  0.713876 -0.366806
2000-01-11  0.325889   1  0.899489 -0.337436
2000-01-12 -0.389584   1  0.482351  0.246871
2000-01-13 -0.714206   1 -0.593838  0.090279
2000-01-14 -0.933238   1 -0.936087  0.471866
2000-01-15 -0.991959   1 -0.943218  0.637434
2000-01-16 -0.645081   1 -0.520788  0.322264
2000-01-17 -0.348338   1 -0.183528  0.385915
2000-01-18  0.193914   1 -0.308346 -0.157765
2000-01-19  0.465424   1 -0.072219 -0.714273
2000-01-20  0.645630   1  0.211302 -0.651308

Computing rolling pairwise correlations

In financial data analysis and other fields it’s common to compute correlation matrices for a collection of time series. More difficult is to compute a moving-window correlation matrix. This can be done using the rolling_corr_pairwise function, which yields a Panel whose items are the dates in question:

In [225]: correls = rolling_corr_pairwise(df, 50)

In [226]: correls[df.index[-50]]
Out[226]: 
          A         B         C         D
A  1.000000  0.289597  0.673828 -0.589002
B  0.289597  1.000000 -0.041244  0.204692
C  0.673828 -0.041244  1.000000 -0.848632
D -0.589002  0.204692 -0.848632  1.000000

You can efficiently retrieve the time series of correlations between two columns using ix indexing:

In [227]: correls.ix[:, 'A', 'C'].plot()
Out[227]: <matplotlib.axes.AxesSubplot at 0x110330b10>
_images/rolling_corr_pairwise_ex.png

Exponentially weighted moment functions

A related set of functions are exponentially weighted versions of many of the above statistics. A number of EW (exponentially weighted) functions are provided using the blending method. For example, where y_t is the result and x_t the input, we compute an exponentially weighted moving average as

y_t = \alpha y_{t-1} + (1 - \alpha) x_t

One must have 0 < \alpha \leq 1, but rather than pass \alpha directly, it’s easier to think about either the span or center of mass (com) of an EW moment:

\alpha =
 \begin{cases}
     \frac{2}{s + 1}, s = \text{span}\\
     \frac{1}{c + 1}, c = \text{center of mass}
 \end{cases}

You can pass one or the other to these functions but not both. Span corresponds to what is commonly called a “20-day EW moving average” for example. Center of mass has a more physical interpretation. For example, span = 20 corresponds to com = 9.5. Here is the list of functions available:

Function Description
ewma EW moving average
ewvar EW moving variance
ewstd EW moving standard deviation
ewmcorr EW moving correlation
ewmcov EW moving covariance

Here are an example for a univariate time series:

In [228]: plt.close('all')

In [229]: ts.plot(style='k--')
Out[229]: <matplotlib.axes.AxesSubplot at 0x10fd04d10>

In [230]: ewma(ts, span=20).plot(style='k')
Out[230]: <matplotlib.axes.AxesSubplot at 0x10fd04d10>
_images/ewma_ex.png

Note

The EW functions perform a standard adjustment to the initial observations whereby if there are fewer observations than called for in the span, those observations are reweighted accordingly.

Linear and panel regression

Note

We plan to move this functionality to statsmodels for the next release. Some of the result attributes may change names in order to foster naming consistency with the rest of statsmodels. We will provide every effort to provide compatibility with older versions of pandas, however.

We have implemented a very fast set of moving-window linear regression classes in pandas. Two different types of regressions are supported:

  • Standard ordinary least squares (OLS) multiple regression
  • Multiple regression (OLS-based) on panel data including with fixed-effects (also known as entity or individual effects) or time-effects.

Both kinds of linear models are accessed through the ols function in the pandas namespace. They all take the following arguments to specify either a static (full sample) or dynamic (moving window) regression:

  • window_type: 'full sample' (default), 'expanding', or rolling
  • window: size of the moving window in the window_type='rolling' case. If window is specified, window_type will be automatically set to 'rolling'
  • min_periods: minimum number of time periods to require to compute the regression coefficients

Generally speaking, the ols works by being given a y (response) object and an x (predictors) object. These can take many forms:

  • y: a Series, ndarray, or DataFrame (panel model)
  • x: Series, DataFrame, dict of Series, dict of DataFrame or Panel

Based on the types of y and x, the model will be inferred to either a panel model or a regular linear model. If the y variable is a DataFrame, the result will be a panel model. In this case, the x variable must either be a Panel, or a dict of DataFrame (which will be coerced into a Panel).

Standard OLS regression

Let’s pull in some sample data:

In [231]: from pandas.io.data import DataReader

In [232]: symbols = ['MSFT', 'GOOG', 'AAPL']

In [233]: data = dict((sym, DataReader(sym, "yahoo"))
   .....:             for sym in symbols)
   .....:

In [234]: panel = Panel(data).swapaxes('items', 'minor')

In [235]: close_px = panel['Close']

# convert closing prices to returns
In [236]: rets = close_px / close_px.shift(1) - 1

In [237]: rets.info()
<class 'pandas.core.frame.DataFrame'>
DatetimeIndex: 250 entries, 2011-11-09 00:00:00 to 2012-11-07 00:00:00
Data columns:
AAPL    249  non-null values
GOOG    249  non-null values
MSFT    249  non-null values
dtypes: float64(3)

Let’s do a static regression of AAPL returns on GOOG returns:

In [238]: model = ols(y=rets['AAPL'], x=rets.ix[:, ['GOOG']])

In [239]: model
Out[239]: 
-------------------------Summary of Regression Analysis-------------------------
Formula: Y ~ <GOOG> + <intercept>
Number of Observations:         249
Number of Degrees of Freedom:   2
R-squared:         0.1691
Adj R-squared:     0.1658
Rmse:              0.0157
F-stat (1, 247):    50.2798, p-value:     0.0000
Degrees of Freedom: model 1, resid 247
-----------------------Summary of Estimated Coefficients------------------------
      Variable       Coef    Std Err     t-stat    p-value    CI 2.5%   CI 97.5%
--------------------------------------------------------------------------------
          GOOG     0.4791     0.0676       7.09     0.0000     0.3466     0.6115
     intercept     0.0013     0.0010       1.28     0.2012    -0.0007     0.0032
---------------------------------End of Summary---------------------------------

In [240]: model.beta
Out[240]: 
GOOG         0.479054
intercept    0.001278

If we had passed a Series instead of a DataFrame with the single GOOG column, the model would have assigned the generic name x to the sole right-hand side variable.

We can do a moving window regression to see how the relationship changes over time:

In [241]: model = ols(y=rets['AAPL'], x=rets.ix[:, ['GOOG']],
   .....:             window=250)
   .....:

# just plot the coefficient for GOOG
In [242]: model.beta['GOOG'].plot()
Out[242]: <matplotlib.axes.AxesSubplot at 0x10fdb2890>
_images/moving_lm_ex.png

It looks like there are some outliers rolling in and out of the window in the above regression, influencing the results. We could perform a simple winsorization at the 3 STD level to trim the impact of outliers:

In [243]: winz = rets.copy()

In [244]: std_1year = rolling_std(rets, 250, min_periods=20)

# cap at 3 * 1 year standard deviation
In [245]: cap_level = 3 * np.sign(winz) * std_1year

In [246]: winz[np.abs(winz) > 3 * std_1year] = cap_level

In [247]: winz_model = ols(y=winz['AAPL'], x=winz.ix[:, ['GOOG']],
   .....:             window=250)
   .....:

In [248]: model.beta['GOOG'].plot(label="With outliers")
Out[248]: <matplotlib.axes.AxesSubplot at 0x10fd9e390>

In [249]: winz_model.beta['GOOG'].plot(label="Winsorized"); plt.legend(loc='best')
Out[249]: <matplotlib.legend.Legend at 0x111d7f590>
_images/moving_lm_winz.png

So in this simple example we see the impact of winsorization is actually quite significant. Note the correlation after winsorization remains high:

In [250]: winz.corrwith(rets)
Out[250]: 
AAPL    0.991873
GOOG    0.984418
MSFT    0.998519

Multiple regressions can be run by passing a DataFrame with multiple columns for the predictors x:

In [251]: ols(y=winz['AAPL'], x=winz.drop(['AAPL'], axis=1))
Out[251]: 
-------------------------Summary of Regression Analysis-------------------------
Formula: Y ~ <GOOG> + <MSFT> + <intercept>
Number of Observations:         249
Number of Degrees of Freedom:   3
R-squared:         0.2347
Adj R-squared:     0.2285
Rmse:              0.0144
F-stat (2, 246):    37.7281, p-value:     0.0000
Degrees of Freedom: model 2, resid 246
-----------------------Summary of Estimated Coefficients------------------------
      Variable       Coef    Std Err     t-stat    p-value    CI 2.5%   CI 97.5%
--------------------------------------------------------------------------------
          GOOG     0.4502     0.0712       6.32     0.0000     0.3107     0.5897
          MSFT     0.2654     0.0748       3.55     0.0005     0.1188     0.4120
     intercept     0.0009     0.0009       0.94     0.3482    -0.0009     0.0027
---------------------------------End of Summary---------------------------------

Panel regression

We’ve implemented moving window panel regression on potentially unbalanced panel data (see this article if this means nothing to you). Suppose we wanted to model the relationship between the magnitude of the daily return and trading volume among a group of stocks, and we want to pool all the data together to run one big regression. This is actually quite easy:

# make the units somewhat comparable
In [252]: volume = panel['Volume'] / 1e8

In [253]: model = ols(y=volume, x={'return' : np.abs(rets)})

In [254]: model
Out[254]: 
-------------------------Summary of Regression Analysis-------------------------
Formula: Y ~ <return> + <intercept>
Number of Observations:         747
Number of Degrees of Freedom:   2
R-squared:         0.0244
Adj R-squared:     0.0231
Rmse:              0.2105
F-stat (1, 745):    18.6418, p-value:     0.0000
Degrees of Freedom: model 1, resid 745
-----------------------Summary of Estimated Coefficients------------------------
      Variable       Coef    Std Err     t-stat    p-value    CI 2.5%   CI 97.5%
--------------------------------------------------------------------------------
        return     3.1731     0.7349       4.32     0.0000     1.7327     4.6136
     intercept     0.1888     0.0111      16.95     0.0000     0.1670     0.2107
---------------------------------End of Summary---------------------------------

In a panel model, we can insert dummy (0-1) variables for the “entities” involved (here, each of the stocks) to account the a entity-specific effect (intercept):

In [255]: fe_model = ols(y=volume, x={'return' : np.abs(rets)},
   .....:                entity_effects=True)
   .....:

In [256]: fe_model
Out[256]: 
-------------------------Summary of Regression Analysis-------------------------
Formula: Y ~ <return> + <FE_GOOG> + <FE_MSFT> + <intercept>
Number of Observations:         747
Number of Degrees of Freedom:   4
R-squared:         0.7843
Adj R-squared:     0.7834
Rmse:              0.0991
F-stat (3, 743):   900.6175, p-value:     0.0000
Degrees of Freedom: model 3, resid 743
-----------------------Summary of Estimated Coefficients------------------------
      Variable       Coef    Std Err     t-stat    p-value    CI 2.5%   CI 97.5%
--------------------------------------------------------------------------------
        return     4.0218     0.3483      11.55     0.0000     3.3392     4.7044
       FE_GOOG    -0.1396     0.0089     -15.66     0.0000    -0.1571    -0.1221
       FE_MSFT     0.3052     0.0089      34.16     0.0000     0.2877     0.3227
     intercept     0.1244     0.0077      16.24     0.0000     0.1093     0.1394
---------------------------------End of Summary---------------------------------

Because we ran the regression with an intercept, one of the dummy variables must be dropped or the design matrix will not be full rank. If we do not use an intercept, all of the dummy variables will be included:

In [257]: fe_model = ols(y=volume, x={'return' : np.abs(rets)},
   .....:                entity_effects=True, intercept=False)
   .....:

In [258]: fe_model
Out[258]: 
-------------------------Summary of Regression Analysis-------------------------
Formula: Y ~ <return> + <FE_AAPL> + <FE_GOOG> + <FE_MSFT>
Number of Observations:         747
Number of Degrees of Freedom:   4
R-squared:         0.7843
Adj R-squared:     0.7834
Rmse:              0.0991
F-stat (4, 743):   900.6175, p-value:     0.0000
Degrees of Freedom: model 3, resid 743
-----------------------Summary of Estimated Coefficients------------------------
      Variable       Coef    Std Err     t-stat    p-value    CI 2.5%   CI 97.5%
--------------------------------------------------------------------------------
        return     4.0218     0.3483      11.55     0.0000     3.3392     4.7044
       FE_AAPL     0.1244     0.0077      16.24     0.0000     0.1093     0.1394
       FE_GOOG    -0.0153     0.0073      -2.10     0.0358    -0.0295    -0.0010
       FE_MSFT     0.4295     0.0071      60.08     0.0000     0.4155     0.4436
---------------------------------End of Summary---------------------------------

We can also include time effects, which demeans the data cross-sectionally at each point in time (equivalent to including dummy variables for each date). More mathematical care must be taken to properly compute the standard errors in this case:

In [259]: te_model = ols(y=volume, x={'return' : np.abs(rets)},
   .....:                time_effects=True, entity_effects=True)
   .....:

In [260]: te_model
Out[260]: 
-------------------------Summary of Regression Analysis-------------------------
Formula: Y ~ <return> + <FE_GOOG> + <FE_MSFT>
Number of Observations:         747
Number of Degrees of Freedom:   252
R-squared:         0.8445
Adj R-squared:     0.7656
Rmse:              0.0979
F-stat (3, 495):    10.7076, p-value:     0.0000
Degrees of Freedom: model 251, resid 495
-----------------------Summary of Estimated Coefficients------------------------
      Variable       Coef    Std Err     t-stat    p-value    CI 2.5%   CI 97.5%
--------------------------------------------------------------------------------
        return     3.4851        nan        nan        nan        nan        nan
       FE_GOOG    -0.1408        nan        nan        nan        nan        nan
       FE_MSFT     0.3037        nan        nan        nan        nan        nan
---------------------------------End of Summary---------------------------------

Here the intercept (the mean term) is dropped by default because it will be 0 according to the model assumptions, having subtracted off the group means.

Result fields and tests

We’ll leave it to the user to explore the docstrings and source, especially as we’ll be moving this code into statsmodels in the near future.