Lecture 8-2022

Август 9, 2022

Содержание

2. Lecture 8: Event Study Analysis 821L1: Financial and Time Series Econometrics Slides created by Dr. C.
3. Lecture outline Examples of the relationship between breaking news and share prices. Overview of event study
4. March 16: News breaks that Cambridge Analytica harvested facebook user data to help Trump win.
5. March 15: Rihanna retorts at snapchat’s ‘who would you rather slap…’
6. Feb 21: Kylie Jenner tweets asks “does anyone else not open snapchat anymore?”
7. Motivation Unanticipated events affect stock prices, and other economic time series. These graphs are suggestive, but
8. Overview of an event study Event definition What is the event? Examples: Earnings announcement, oil spill,
9. Overview (Cont’d)
10. Overview (Cont’d) Testing procedure Determine technique of aggregating abnormal returns across firms and over time. Design
11. Timeline T0-T1: Pre-event, or estimation window T1-T2: Event window. Contains the date of the event 0.
12. The Constant Mean Return Model
13. The Market Model
14. Estimation
15. Aggregation over time for one security
16. Hypothesis testing aggregating over time
17. Aggregating CAR and SCAR over multiple securities In general we may be interested in aggregating over
18. Aggregating over securities (cont’d)
19. Sensitivity: Normal Returns Model Using a different model to specify normal returns can affect the point
20. Sensitivity: Clustering We have assumed that events do not overlap in calendar time. This may be
21. Further issues Sampling interval: If we are sure that the market internalizes all new information quickly,
22. Example: The FASB and the SEC regulations necessitate that firms report earnings announcements periodically. Event studies
23. Example (cont’d): Use Institutional Brokers Estimate System to proxy for market expectations. Classify announcements into ‘good
24. Abnormal returns: MM and CMRM (Source: CLM)
25. Discussion The information content of earnings announcements do indeed appear to drive abnormal returns in these
27. Скачать презентацию

Слайд 2

Lecture 8: Event Study Analysis
821L1: Financial and Time Series Econometrics
Slides created

by Dr. C. Rashaad Shabab
Edited and updated by Dr. Gabriella Cagliesi
2022

Слайд 3

Lecture outline
Examples of the relationship between breaking news and share prices.
Overview

of event study analysis
The Constant Mean Return Model and the Market Model
Estimation
Aggregating over time and across securities
Sensitivity
Example
Conclusions.

Слайд 4

March 16: News breaks that Cambridge Analytica harvested facebook user data

to help Trump win.

Слайд 5

March 15: Rihanna retorts at snapchat’s ‘who would you rather slap…’

Слайд 6

Feb 21: Kylie Jenner tweets asks “does anyone else not open

snapchat anymore?”

Слайд 7

Motivation
Unanticipated events affect stock prices, and other economic time series.
These graphs

are suggestive, but as trained econometricians you may have a whole host of other concerns.
Are these differences statistically significant?
How did other, similar shares do in the mean-time?
Could some unobserved process be driving this?
In other words, what is the appropriate counterfactual?
The formal econometric methodology that addresses these concerns is called ‘Event Study Analysis’.

Слайд 8

Overview of an event study
Event definition
What is the event?
Examples: Earnings

announcement, oil spill, CEOs health.
What is the window of time this event will affect the stock price in?
Theory says instantaneous. Usually we take 1-2 days after the event.
Sometimes it can be longer. Train crash investigation shows negligence, then prolonged effect.
Selection criteria
We rarely cover all firms. Usually only data on publicly traded firms are available.
Sometimes we focus on largest firms, by say market capitalization.
Important to be explicit and to think through the potential for bias sample selection may introduce
Internal vs. external validity

Слайд 9

Overview (Cont’d)

Слайд 10

Overview (Cont’d)
Testing procedure
Determine technique of aggregating abnormal returns across firms and

over time.
Design testing framework.
Empirical results
Present results
include diagnostic tests and sensitivity analysis
Interpretation and conclusions
Shed insight on interesting economic phenomena.
….or the power of celebrities/ the demise of liberal democracy

Слайд 11

Timeline
T0-T1: Pre-event, or estimation window
T1-T2: Event window. Contains the date of

the event 0.
T2-T3: Post-event Window.

Слайд 12

The Constant Mean Return Model

Слайд 13

The Market Model

Слайд 14

Estimation

Слайд 15

Aggregation over time for one security

Слайд 16

Hypothesis testing aggregating over time

Слайд 17

Aggregating CAR and SCAR over multiple securities
In general we may be

interested in aggregating over many different securities.
Statistically, this is very straightforward as the event study methodology is fairly flexible.
We define time relative to the event date for each security and define the estimation and event windows for each security (NB: not calendar time).
We then compute the CAR for each security and can take averages.
An important distributional assumption is that the event windows do not overlap (where they do overlap, additional steps need to be taken, see CLM textbook)

Слайд 18

Aggregating over securities (cont’d)

Слайд 19

Sensitivity: Normal Returns Model
Using a different model to specify normal returns

can affect the point estimates and the variance.
The Market Model described here has lower variance than the Constant Mean Return Model, as it explains the part of the share price that is driven by movements in the market.
In principle, further reductions in variance could be generated by accounting for additional factors.
However, the MM and CMRM remain the workhorse models.

Слайд 20

Sensitivity: Clustering
We have assumed that events do not overlap in calendar

time.
This may be a strong assumption as company earnings announcements may coincide, implying positive covariance across different stocks.
There are two common ways to deal with this problem.
We may construct an equally (market capitalization) weighted portfolio of the two stocks and compute abnormal returns for this portfolio.
We may circumvent aggregation by analyzing the abnormal returns for the two stocks separately, rather than aggregating them up.

Слайд 21

Further issues
Sampling interval: If we are sure that the market internalizes

all new information quickly, a smaller event window can enhance our power to detect abnormal returns.
Event date uncertainty: Newspapers may only report earnings announcements the next day, but the market may have received this information the day before the report. So we should bring the event window a day forward form the report.

Слайд 22

Example:
The FASB and the SEC regulations necessitate that firms report earnings

announcements periodically.
Event studies can analyze the information content of these announcements.
CLM analyze the quarterly earnings announcements of 30 firms from the Dow Jones from 1989-1993.
Total sample of 600 announcements.

Слайд 23

Example (cont’d):
Use Institutional Brokers Estimate System to proxy for market expectations.
Classify

announcements into ‘good news’, ‘no news’ and ‘bad news’:
Good news (189 obs): Earnings exceed expectations by >2.5%.
No news (173): Earnings within 5% of expected value (-2.5% to + 2.5%).
Bad news (238): Earnings below expectations by >2.5%.
1 day sampling interval.
Event window: 41 day event window (20 before, 20 after).
Estimation window: 250 trading days prior to event window.

Слайд 24

Abnormal returns: MM and CMRM (Source: CLM)

Слайд 25

Discussion
The information content of earnings announcements do indeed appear to drive

abnormal returns in these stocks.
There is some evidence of both
under-reaction (markets do not react instantaneously) and
over-reaction (bad news stocks partially recover in value after initial losses)
There is some evidence of information leakage / insider trading.
Both the Market Model and the Constant Mean Return Model yield qualitatively similar results.

Lecture 8-2022

Содержание

Lecture 8: Event Study Analysis 821L1: Financial and Time Series EconometricsSlides created

Lecture outlineExamples of the relationship between breaking news and share prices.Overview

March 16: News breaks that Cambridge Analytica harvested facebook user data

March 15: Rihanna retorts at snapchat’s ‘who would you rather slap…’

Feb 21: Kylie Jenner tweets asks “does anyone else not open

MotivationUnanticipated events affect stock prices, and other economic time series.These graphs

Overview of an event studyEvent definitionWhat is the event? Examples: Earnings

Overview (Cont’d)

Overview (Cont’d)Testing procedureDetermine technique of aggregating abnormal returns across firms and

TimelineT0-T1: Pre-event, or estimation windowT1-T2: Event window. Contains the date of