Table Of Contents

  1. Docs
  2. Get Started

Get Started

Through two examples, we will discover how to perform an event study analysis on a single event or on an aggregate of events.

Note

You can use the interactive version of this tutorial to play with the functions yourself.

Preliminary work

  1. Load the eventstudy module and its dependencies: numpy and matplotlib:

    import eventstudy as es
import numpy as np
import matplotlib.pyplot as plt

  2. Set the parameters needed for your events: the returns and Fama-French factors (using es.Single.import_returns()and es.Single.import_FamaFrench()):

    es.Single.import_returns('returns.csv')
es.Single.import_FamaFrench('famafrench.csv')

Example 1: A single event

As an introductory example, we will compute the event study analysis of the announcement of the first iphone, made by Steve Jobs during MacWorld exhibition, on January 7, 2007.

  1. Run the event study, here using the Fama-French 3-factor model:

    event = es.Single.FamaFrench_3factor(
    security_ticker = 'AAPL',
    event_date = np.datetime64('2013-03-04'),
    event_window = (-2,+10), 
    estimation_size = 300,
    buffer_size = 30
)

    Note

    You can easily play with the parameters and adjust the event study analysis to your needs.

    For more details, see the documentation on FamaFrench_3factor function.

    See also other models’ function. You can even set your own modelisation functions

  2. Display results:

    • In a plot:

    event.plot(AR=True)
plt.show() # use standard matplotlib function to display the plot

    Single event plot

    grey area: confidence interval (here at 90%); blue line: CAAR; black bars: AR

    Note

    You can remove the confidence interval (set CI = False) or change its level of confidence (set confidence = .95 for a confidence interval at 95%).

    By default AR are note displayed (set AR = True to add them to the plot).

    For more details, see the documentation on plot function .

    • Or in a table:

    event.results(decimals=[3,5,3,5,2,2])
AR Variance AR CAR Variance CAR T-stat P-value
-5 -0.053 0.00048 -0.053 ** 0.00048 -2.42 0.01
-4 0.012 0.00048 -0.041 * 0.00096 -1.33 0.09
-3 -0.013 0.00048 -0.055 * 0.00144 -1.43 0.08
-2 0.004 0.00048 -0.051 0.00192 -1.15 0.13
-1 0 0.00048 -0.051 0.00241 -1.03 0.15
0 -0.077 0.00048 -0.128 ** 0.00289 -2.37 0.01
1 -0.039 0.00048 -0.167 *** 0.00337 -2.88 0
2 0.027 0.00048 -0.14 ** 0.00385 -2.26 0.01
3 0.024 0.00048 -0.116 ** 0.00433 -1.77 0.04
4 -0.024 0.00048 -0.14 ** 0.00481 -2.02 0.02
5 0.023 0.00048 -0.117 * 0.00529 -1.61 0.05
6 0.057 0.00048 -0.06 0.00577 -0.79 0.21
7 0.011 0.00048 -0.049 0.00626 -0.62 0.27
8 0.011 0.00048 -0.038 0.00674 -0.46 0.32
9 0.017 0.00048 -0.021 0.00722 -0.24 0.4
10 0.003 0.00048 -0.018 0.0077 -0.2 0.42
11 -0.01 0.00048 -0.028 0.00818 -0.31 0.38
12 0.008 0.00048 -0.02 0.00866 -0.21 0.42
13 -0.009 0.00048 -0.028 0.00914 -0.3 0.38
14 0.012 0.00048 -0.016 0.00962 -0.16 0.43
15 0.011 0.00048 -0.005 0.01011 -0.05 0.48
16 0.024 0.00048 0.018 0.01059 0.18 0.43
17 0.001 0.00048 0.019 0.01107 0.18 0.43
18 0.009 0.00048 0.028 0.01155 0.26 0.4
19 -0.002 0.00048 0.026 0.01203 0.24 0.41
20 -0.02 0.00048 0.006 0.01251 0.06 0.48

Note

Asterisks are added automatically to highlight the level of significance (Significance level: *** at 99%, ** at 95%, * at 90%). You can remove asterisks by setting asterisks parameter at False.

decimals is a list of integer setting for each column (except index) the rounding decimal. You can also set one integer (e.g. decimals = 3) if you want all columns to be rounded the same.

See the documentation on this results function for more details.

Example 2: An aggregate of events

The eventstudy package offer three different entry points. The list of events can be provided using a csv file, a plain python text or a python list.

In this second example, based on the same data and preliminary work, we will compute an event study on GAFA annual report (10-K form) releases.

Data preparation

The first step is to gather the data about the events and to format it. For the purpose of this example we will use the csv file methods. The csv file must contains all information needed to compute the model.

The event study will be computed using the market model. According to the (/api/eventstudy.Single.market_model.html#eventstudy.Single.market_model)[API documentation] aside from event study parameters which will be set globally for all events (event_window, estimation_size and buffer_size), the market model needs the security_ticker, the market_ticker and the event_date parameters.

Thus, the csv file should contain three columns with the exact same names as the parameter names.

Note

The csv file must use comma (‘,’) as separator.

security_ticker market_ticker event_date
AAPL SPY 31/10/2019
AAPL SPY 05/11/2018
AAPL SPY 03/11/2017
AAPL SPY 26/10/2016
AAPL SPY 28/10/2015
AAPL SPY 27/10/2014
AAPL SPY 30/10/2013
AAPL SPY 31/10/2012
AAPL SPY 26/10/2011
AAPL SPY 27/10/2010
GOOG SPY 04/02/2020
GOOG SPY 05/02/2019
GOOG SPY 06/02/2018
GOOG SPY 03/02/2017
GOOG SPY 11/02/2016
AMZN SPY 31/01/2020
AMZN SPY 01/02/2019
AMZN SPY 02/02/2018
AMZN SPY 10/02/2017
AMZN SPY 29/01/2016
AMZN SPY 30/01/2015
AMZN SPY 30/01/2013
AMZN SPY 01/02/2012
AMZN SPY 28/01/2011
AMZN SPY 29/01/2010
MSFT SPY 01/08/2019
MSFT SPY 03/08/2018
MSFT SPY 02/08/2017
MSFT SPY 28/07/2016
MSFT SPY 31/07/2015
MSFT SPY 31/07/2014
MSFT SPY 30/07/2013
MSFT SPY 26/07/2012
MSFT SPY 28/07/2011
MSFT SPY 30/07/2010
FB SPY 30/01/2020
FB SPY 31/01/2019
FB SPY 01/02/2018
FB SPY 28/01/2016
FB SPY 29/01/2015
FB SPY 31/01/2014

Event study computation

To create an event study analysis of an aggregate of events, we will use the Multiple class. This class is designed following the same spirit and structure of the Single class.

Run the event study using the csv importer method:

release_10K = es.Multiple.from_csv(
    path = '10K.csv', # the path to the csv file created
    event_study_model = es.Single.market_model,
    event_window = (-5,+10),
    estimation_size = 200,
    buffer_size = 30,
    date_format = '%d/%m/%Y',
    ignore_errors = True
)

Note

The event_study_model parameter must point to a function. This function can either be a model function provided by the Single class or a user-defined function.

When the ignore_errors parameter is set to True (which is its default value), the event study analysis will be computed regardless of any error encounter. Event triggering an error will be remove of the sample. Use error_report method to get the full list of error and their explanation. Errors are often coming from data unavailability.

Set the ignore_errors to False if you prefer the analysis to stop at the first error.

Display results:

  • In a plot:

release_10K.plot(confidence=.95)
plt.show() # use standard matplotlib function to display the plot

Mutliple events plot

grey area: confidence interval (here defined by the user at 95%); blue line: CAAR

  • Or in a table:

release_10K.results(decimals=[3,5,3,5,2,2])
AAR Variance AAR CAAR Variance CAAR T-stat P-value
-5 -0.004 1e-05 -0.004 ** 1e-05 -1.7 0.04
-4 0.001 1e-05 -0.004 1e-05 -1 0.16
-3 0.001 1e-05 -0.003 2e-05 -0.64 0.26
-2 -0.002 1e-05 -0.005 2e-05 -1.04 0.15
-1 0.004 1e-05 -0.001 3e-05 -0.26 0.4
0 0.009 1e-05 0.008 4e-05 1.29 0.1
1 -0.004 1e-05 0.004 4e-05 0.54 0.29
2 -0.002 1e-05 0.002 5e-05 0.27 0.39
3 -0.001 1e-05 0.001 6e-05 0.14 0.45
4 -0.002 1e-05 -0.001 6e-05 -0.1 0.46
5 -0.004 1e-05 -0.004 7e-05 -0.53 0.3
6 -0.004 1e-05 -0.009 7e-05 -1.01 0.16
7 0 1e-05 -0.009 8e-05 -0.97 0.17
8 -0.001 1e-05 -0.01 9e-05 -1.09 0.14
9 0.002 1e-05 -0.008 9e-05 -0.81 0.21
10 -0.003 1e-05 -0.011 0.0001 -1.13 0.13

Note

Asterisks are added automatically to highlight the level of significance (Significance level: *** at 99%, ** at 95%, * at 90%). You can remove asterisks by setting asterisks parameter at False.

decimals is a list of integer setting for each column (except index) the rounding decimal. You can also set one integer (e.g. decimals = 3) if you want all columns to be rounded the same.

See the documentation on this results function for more details.

  • Get CAR distribution:

release_10K.get_CAR_dist(4)
Mean Variance Kurtosis Min Quantile 25% Quantile 50% Quantile 75% Max
-5 -0.0042 0.0003 0.7127 -0.0463 -0.0098 -0.005 0.0045 0.0373
-4 -0.0035 0.0006 -0.1992 -0.0595 -0.0151 0.0006 0.0091 0.0515
-3 -0.0028 0.0009 0.0228 -0.0839 -0.0231 -0.0052 0.0185 0.0627
-2 -0.0052 0.0007 -0.0799 -0.0631 -0.0209 -0.0078 0.007 0.0574
-1 -0.0015 0.0012 0.2771 -0.0705 -0.0221 -0.008 0.0158 0.0868
0 0.0078 0.0036 1.5911 -0.1059 -0.0271 0.0022 0.0224 0.2023
1 0.0036 0.0043 1.1048 -0.103 -0.0378 -0.012 0.0212 0.2108
2 0.0019 0.0048 0.2683 -0.0974 -0.0383 -0.0163 0.0312 0.1854
3 0.001 0.0051 0.5488 -0.1184 -0.0415 -0.0181 0.0289 0.196
4 -0.0007 0.005 0.6936 -0.1127 -0.0413 -0.0199 0.0238 0.2054
5 -0.0044 0.0055 0.7422 -0.1586 -0.0412 -0.0178 0.0253 0.2121
6 -0.0087 0.0057 0.7992 -0.1747 -0.0439 -0.0215 0.0234 0.2106
7 -0.0087 0.0056 0.8942 -0.1896 -0.047 -0.0175 0.027 0.2005
8 -0.0101 0.0054 0.8968 -0.1736 -0.0481 -0.0172 0.025 0.2064
9 -0.0078 0.0052 0.5841 -0.1534 -0.0481 -0.0122 0.03 0.1969
10 -0.0113 0.0057 0.7229 -0.175 -0.0522 -0.0153 0.0227 0.2034

Bonus: Loop in event

The true strengh of this package is to use python. You can create complex algorithms to compute programatically event stuy analyses. From the same data as above, we will compute an event study of 10-K form releases for each company.

Data preparation

This time, we will use the eventstudy.Multiple.from_list method which set event list using a python list. This let the user manipulate the list of event before computing event studies.

This list must contains all parameters needed to compute the selected model (here we will use the Fama-French 3-factor model). The Fama-French factors’ data has to be set using the eventstudy.Single.import_FamaFrench method (see preliminary step).

events_db = [
    {'security_ticker': 'AAPL', 'market_ticker': 'SPY', 'event_date': np.datetime64('2019-10-31')},
    {'security_ticker': 'AAPL', 'market_ticker': 'SPY', 'event_date': np.datetime64('2018-11-05')},
    {'security_ticker': 'AAPL', 'market_ticker': 'SPY', 'event_date': np.datetime64('2017-11-03')},
    {'security_ticker': 'AAPL', 'market_ticker': 'SPY', 'event_date': np.datetime64('2016-10-26')},
    {'security_ticker': 'AAPL', 'market_ticker': 'SPY', 'event_date': np.datetime64('2015-10-28')},
    {'security_ticker': 'AAPL', 'market_ticker': 'SPY', 'event_date': np.datetime64('2014-10-27')},
    {'security_ticker': 'AAPL', 'market_ticker': 'SPY', 'event_date': np.datetime64('2013-10-30')},
    {'security_ticker': 'AAPL', 'market_ticker': 'SPY', 'event_date': np.datetime64('2012-10-31')},
    {'security_ticker': 'AAPL', 'market_ticker': 'SPY', 'event_date': np.datetime64('2011-10-26')},
    {'security_ticker': 'AAPL', 'market_ticker': 'SPY', 'event_date': np.datetime64('2010-10-27')},
    {'security_ticker': 'GOOG', 'market_ticker': 'SPY', 'event_date': np.datetime64('2020-02-04')},
    {'security_ticker': 'GOOG', 'market_ticker': 'SPY', 'event_date': np.datetime64('2019-02-05')},
    {'security_ticker': 'GOOG', 'market_ticker': 'SPY', 'event_date': np.datetime64('2018-02-06')},
    {'security_ticker': 'GOOG', 'market_ticker': 'SPY', 'event_date': np.datetime64('2017-02-03')},
    {'security_ticker': 'GOOG', 'market_ticker': 'SPY', 'event_date': np.datetime64('2016-02-11')},
    {'security_ticker': 'FB', 'market_ticker': 'SPY', 'event_date': np.datetime64('2020-01-30')},
    {'security_ticker': 'FB', 'market_ticker': 'SPY', 'event_date': np.datetime64('2019-01-31')},
    {'security_ticker': 'FB', 'market_ticker': 'SPY', 'event_date': np.datetime64('2018-02-01')},
    {'security_ticker': 'FB', 'market_ticker': 'SPY', 'event_date': np.datetime64('2016-01-28')},
    {'security_ticker': 'FB', 'market_ticker': 'SPY', 'event_date': np.datetime64('2015-01-29')},
    {'security_ticker': 'FB', 'market_ticker': 'SPY', 'event_date': np.datetime64('2014-01-31')},
    {'security_ticker': 'AMZN', 'market_ticker': 'SPY', 'event_date': np.datetime64('2020-01-31')},
    {'security_ticker': 'AMZN', 'market_ticker': 'SPY', 'event_date': np.datetime64('2019-02-01')},
    {'security_ticker': 'AMZN', 'market_ticker': 'SPY', 'event_date': np.datetime64('2018-02-02')},
    {'security_ticker': 'AMZN', 'market_ticker': 'SPY', 'event_date': np.datetime64('2017-02-10')},
    {'security_ticker': 'AMZN', 'market_ticker': 'SPY', 'event_date': np.datetime64('2016-01-29')},
    {'security_ticker': 'AMZN', 'market_ticker': 'SPY', 'event_date': np.datetime64('2015-01-30')},
    {'security_ticker': 'AMZN', 'market_ticker': 'SPY', 'event_date': np.datetime64('2013-01-30')},
    {'security_ticker': 'AMZN', 'market_ticker': 'SPY', 'event_date': np.datetime64('2012-02-01')},
    {'security_ticker': 'AMZN', 'market_ticker': 'SPY', 'event_date': np.datetime64('2011-01-28')},
    {'security_ticker': 'AMZN', 'market_ticker': 'SPY', 'event_date': np.datetime64('2010-01-29')},
    {'security_ticker': 'MSFT', 'market_ticker': 'SPY', 'event_date': np.datetime64('2019-08-01')},
    {'security_ticker': 'MSFT', 'market_ticker': 'SPY', 'event_date': np.datetime64('2018-08-03')},
    {'security_ticker': 'MSFT', 'market_ticker': 'SPY', 'event_date': np.datetime64('2017-08-02')},
    {'security_ticker': 'MSFT', 'market_ticker': 'SPY', 'event_date': np.datetime64('2016-07-28')},
    {'security_ticker': 'MSFT', 'market_ticker': 'SPY', 'event_date': np.datetime64('2015-07-31')},
    {'security_ticker': 'MSFT', 'market_ticker': 'SPY', 'event_date': np.datetime64('2014-07-31')},
    {'security_ticker': 'MSFT', 'market_ticker': 'SPY', 'event_date': np.datetime64('2013-07-30')},
    {'security_ticker': 'MSFT', 'market_ticker': 'SPY', 'event_date': np.datetime64('2012-07-26')},
    {'security_ticker': 'MSFT', 'market_ticker': 'SPY', 'event_date': np.datetime64('2011-07-28')},
    {'security_ticker': 'MSFT', 'market_ticker': 'SPY', 'event_date': np.datetime64('2010-07-30')}
]

Event studies computation

Using a list comprehension, the full list of events can be filtered to create one list per company’s ticker. Then we can compute for each company, the event study using the Fama-French 3 factor model and store each event study in a releases dictionary.

tickers = ['GOOG', 'AAPL', 'FB', 'AMZN', 'MSFT']
releases = dict()

for ticker in tickers:
    events = [event for event in events_db if event['security_ticker']==ticker]
    
    releases[ticker] = es.Multiple.from_list(
        events,
        es.Single.FamaFrench_3factor,
        event_window= (-5,+10)
    )

Display results

for ticker, event in releases.items():
    print(ticker)
    event.plot()
    plt.show()

GOOG GOOG events plot

AAPL AAPL events plot

FB FB events plot

AMZN AMZN events plot

MSFT MSFT events plot

Event Study package
API
© Copyright 2019, Jean-Baptiste Lemaire, GNU General Public License v3 (GPLv3). Created using Sphinx.