Research skills summary
Slides lecture 1 Part 1
First thing you do while you have your data is to test it with some summary statistics and graphs to assess data
quality, and then compare summary stats to other studies using same data.
Check data for data issues. (missing data, basis points, or jumps in data).
Regression can still be estimated with unbalanced data
Do not interpolate data -> interpolation may bias your results
Do not replace missing values by zero (or any other number) sometimes data seems missing but its actual
value is zero.
Bottom line: in most cases, missing data should be left missing, but use own judgement and check database manual
Dealing with outliers.
1. Use data transformation to “pull” extreme values to mean: e,g, log transformation -> generate logy
2. Winsorizing: replace extreme values with upper/ lower cut-off value. (typically the 1 st and the 99th percentile
values)
3. Truncating (-trimming) : delete extreme observations
Prices & Return
,Slides Lecture 1 part 2
Motivation for event studies
Event study is widespread tool in financial research to assess impact of corporate events on firm value
Corporate events are situations when information about the firm is released to the market
Idea: if markets are efficient, change in stock price around announcement should reflect impact of event on
firm value.
What type of corporate events can you think of that may affect firm value?
o Do stock prices behave differently around stock splits than in normal periods?
o How do earnings announcements affect stock price
o How does an M&A announcement affect stock price
Under (semi-strong form) EMH, stock price should immediately
jump to new equilibrium level at announcement
Evidence of over/underreaction implies violation of
EMH but this hinges on correct specification of the
normal return
if you suspect market inefficiency, long-run event study
can be useful to detect long-term reversal to
equilibrium price
What is an Event Study?
Examine empirically the stock price response to
corporate events/decisions/announcements (identify the event of interest and its timing).
Collect the dates when a sample of firms made similar announcements (e.g. initiating new dividends)
Examine stock returns on the announcements dates and the days immediately before and after the event,
averaged across all firms in the sample (or across subsets)
Returns are compared to “normal expected returns” to obtain abnormal returns
Statistical test can be used to test whether the average abnormal return is different from zero, or whether
abnormal returns correlate with other characteristics (graphs, test, regression)
Analysis is typically done in “event time” t (where t = 0 denotes event day).
CAR = compounded average return
, The Event Date
The event date (τ = 0) is normally the announcement date. E.g. stock prices react when a merger is announced,
rather than when the merger actually takes place.
• Some uncertainty about the exact announcement date is sometimes unavoidable.
• Note that firms may announce their events at strategic moments, e.g. when the stock price is high or very low, or
when the “market sentiment“ is good. This may show up in abnormal returns before the announcement.
• In addition, stock prices may react before event (information leakage)
• Also note that firms may announce their events jointly with other announcements (confounding events), e.g. a
change of CEO and an acquisition.
Choice of Event Window
In practice, there can be some delay in stock price reaction to new information due to
1. Illiquid markets (low trading volume)
2. Time required to process information and update beliefs
3. Limits to arbitrage
4. Misclassified dates/times
Therefore, in practice multiple event window lengths are used (robustness check)
Typical window lengths used for short-run event studies are -/+ 1 day (3-day announcement return) and -/+ 5 days
(11-day announcement returns).
Computing expected return
Slides lecture 1 Part 1
First thing you do while you have your data is to test it with some summary statistics and graphs to assess data
quality, and then compare summary stats to other studies using same data.
Check data for data issues. (missing data, basis points, or jumps in data).
Regression can still be estimated with unbalanced data
Do not interpolate data -> interpolation may bias your results
Do not replace missing values by zero (or any other number) sometimes data seems missing but its actual
value is zero.
Bottom line: in most cases, missing data should be left missing, but use own judgement and check database manual
Dealing with outliers.
1. Use data transformation to “pull” extreme values to mean: e,g, log transformation -> generate logy
2. Winsorizing: replace extreme values with upper/ lower cut-off value. (typically the 1 st and the 99th percentile
values)
3. Truncating (-trimming) : delete extreme observations
Prices & Return
,Slides Lecture 1 part 2
Motivation for event studies
Event study is widespread tool in financial research to assess impact of corporate events on firm value
Corporate events are situations when information about the firm is released to the market
Idea: if markets are efficient, change in stock price around announcement should reflect impact of event on
firm value.
What type of corporate events can you think of that may affect firm value?
o Do stock prices behave differently around stock splits than in normal periods?
o How do earnings announcements affect stock price
o How does an M&A announcement affect stock price
Under (semi-strong form) EMH, stock price should immediately
jump to new equilibrium level at announcement
Evidence of over/underreaction implies violation of
EMH but this hinges on correct specification of the
normal return
if you suspect market inefficiency, long-run event study
can be useful to detect long-term reversal to
equilibrium price
What is an Event Study?
Examine empirically the stock price response to
corporate events/decisions/announcements (identify the event of interest and its timing).
Collect the dates when a sample of firms made similar announcements (e.g. initiating new dividends)
Examine stock returns on the announcements dates and the days immediately before and after the event,
averaged across all firms in the sample (or across subsets)
Returns are compared to “normal expected returns” to obtain abnormal returns
Statistical test can be used to test whether the average abnormal return is different from zero, or whether
abnormal returns correlate with other characteristics (graphs, test, regression)
Analysis is typically done in “event time” t (where t = 0 denotes event day).
CAR = compounded average return
, The Event Date
The event date (τ = 0) is normally the announcement date. E.g. stock prices react when a merger is announced,
rather than when the merger actually takes place.
• Some uncertainty about the exact announcement date is sometimes unavoidable.
• Note that firms may announce their events at strategic moments, e.g. when the stock price is high or very low, or
when the “market sentiment“ is good. This may show up in abnormal returns before the announcement.
• In addition, stock prices may react before event (information leakage)
• Also note that firms may announce their events jointly with other announcements (confounding events), e.g. a
change of CEO and an acquisition.
Choice of Event Window
In practice, there can be some delay in stock price reaction to new information due to
1. Illiquid markets (low trading volume)
2. Time required to process information and update beliefs
3. Limits to arbitrage
4. Misclassified dates/times
Therefore, in practice multiple event window lengths are used (robustness check)
Typical window lengths used for short-run event studies are -/+ 1 day (3-day announcement return) and -/+ 5 days
(11-day announcement returns).
Computing expected return
