Preparing Synthetic Event Data for Alphalens Event Studies
Summary
This notebook demonstrates how to prepare synthetic prices and sparse event signals for Alphalens. It creates a small panel of prices for six securities, then marks selected date-security pairs in an event factor while leaving other entries missing. The factor and prices are passed to Alphalens to calculate forward returns over several horizons and produce an event-study tear sheet.
The central lesson is that the factor should contain values only where events occur; missing values exclude other observations from the analysis. The example also notes that positive event values represent long-side events and negative values represent short-side events, affecting cumulative-return plots. The notebook shows output plots but provides no numerical performance results or comparison against a benchmark. Its synthetic prices and hand-entered events are intended to illustrate input formatting and experimentation, so they do not establish that any real event has predictive value. The code's introductory comment mentions a price threshold that the displayed event matrix does not actually calculate.
Key ideas
- Alphalens uses non-missing factor entries to identify dates and securities for analysis.
- A sparse factor matrix can represent event occurrences by leaving all other cells missing.
- Positive and negative event values indicate long and short directions in cumulative-return plots.
- Forward returns can be calculated across several holding periods for event observations.
- Synthetic examples help validate data formatting but do not demonstrate a tradable edge.
Tags
Full text
# Synthetic data examples
# Synthetic data examples
In this Notebook we will build synthetic data suitable to Alphalens analysis. This is useful to understand how Alphalens expects the input to be formatted and also it is a good testing environment to experiment with Alphalens.
```python
%matplotlib inline
from numpy import nan
from pandas import (DataFrame, date_range)
import matplotlib.pyplot as plt
from alphalens.tears import (create_returns_tear_sheet,
create_information_tear_sheet,
create_turnover_tear_sheet,
create_summary_tear_sheet,
create_full_tear_sheet,
create_event_returns_tear_sheet,
create_event_study_tear_sheet)
from alphalens.utils import get_clean_factor_and_forward_returns
```
```python
#
# build price
#
price_index = date_range(start='2015-1-10', end='2015-2-28')
price_index.name = 'date'
tickers = ['A', 'B', 'C', 'D', 'E', 'F']
data = [[1.0025**i, 1.005**i, 1.00**i, 0.995**i, 1.005**i, 1.00**i]
for i in range(1, 51)]
prices = DataFrame(index=price_index, columns=tickers, data=data)
```
```python
prices.plot()
```
```python
prices.head()
```
Now it's time to build the events DataFrame, the input will give to Alphalens.
Alphalens calculates statistics for those dates where the input DataFrame has values (not NaN). So to compute the performace analysis on specific dates and securities (like an event study) then we have to make sure the input DataFrame contains valid values only on those date/security combinations where the event happens. All the other values in the DataFrame must be NaN or not present.
Also, make sure the event values are positive (it doesn't matter the value but they must be positive) if you intend to go long on the events and use negative values if you intent to go short. This impacts the cumulative returns plots.
Let's create the event DataFrame where we "mark" (any value) each day a security price fall below 30$.
```python
#
# build factor
#
factor_index = date_range(start='2015-1-15', end='2015-2-13')
factor_index.name = 'date'
event = DataFrame(index=factor_index, columns=tickers,
data=[[1, nan, nan, nan, nan, nan],
[4, nan, nan, 7, nan, nan],
[nan, nan, nan, nan, nan, nan],
[nan, 3, nan, 2, nan, nan],
[1, nan, nan, nan, nan, nan],
[nan, nan, 2, nan, nan, nan],
[nan, nan, nan, 2, nan, nan],
[nan, nan, nan, 1, nan, nan],
[2, nan, nan, nan, nan, nan],
[nan, nan, nan, nan, 5, nan],
[nan, nan, nan, 2, nan, nan],
[nan, nan, nan, nan, nan, nan],
[2, nan, nan, nan, nan, nan],
[nan, nan, nan, nan, nan, 5],
[nan, nan, nan, 1, nan, nan],
[nan, nan, nan, nan, 4, nan],
[5, nan, nan, 4, nan, nan],
[nan, nan, nan, 3, nan, nan],
[nan, nan, nan, 4, nan, nan],
[nan, nan, 2, nan, nan, nan],
[5, nan, nan, nan, nan, nan],
[nan, 1, nan, nan, nan, nan],
[nan, nan, nan, nan, 4, nan],
[0, nan, nan, nan, nan, nan],
[nan, 5, nan, nan, nan, 4],
[nan, nan, nan, nan, nan, nan],
[nan, nan, 5, nan, nan, 3],
[nan, nan, 1, 2, 3, nan],
[nan, nan, nan, 5, nan, nan],
[nan, nan, 1, nan, 3, nan]]).stack()
factor_groups = {'A': 'Group1', 'B': 'Group2', 'C': 'Group1', 'D': 'Group2', 'E': 'Group1', 'F': 'Group2'}
```
```python
event.head(10)
```
```python
event_data = get_clean_factor_and_forward_returns(event, prices,
quantiles=None, bins=1,
periods=(1, 2, 3, 4, 5, 10, 15), filter_zscore=None)
```
```python
event_data.head(10)
```
```python
create_event_study_tear_sheet(event_data, prices, avgretplot=(5, 10))
```
Shown in full with attribution under the source's licence. Licence: Apache-2.0
This summary was written by Stratmill's research agent from the original; it is not a copy of the source.