Research Skills Part 2

Question 1

Name 3 ways to deal with outliers

Answer 1

1. Data transformation (e.g. logs)
2. Winsorizing
3. Truncating = deleting extreme observations

Question 2

Important notes on Event Date

Answer 2

1. Firms may announce their events at strategic moments, e.g., when the stock price is high > shows up in abnormal returns before the announcement.
2. Stock prices may react before the event due to information leakage
3. Firms may announce their events jointly with other announcements > confounding events
4. But there can also be a delay in stock price reaction due to illiquid markets, more time needed to process information, limits to arbitrage, misclassified timezones.

Question 3

Name 3 ways to compute the expected returns

Answer 3

1. Constant-mean-market model = historical average return
2. Market-adjusted-return model = market return as proxy for normal return
3. Market model

The choice of estimation window is a tradeoff between precision and timeliness

Question 4

Name the assumptions when calculating the S.E. of the CAR

Answer 4

1. It assumes homoskedasticity per firm and no cross-correlations
2. It assumes that volatility is not affected by the event

Question 5

Give the formula for t-statistic

Answer 5

t-stat = (mean - X) / S.E.

Question 6

Give the t-stat for CAR

Answer 6

t-stat (CAR) = avgCAR / SQRT(var(avgCAR))

Question 7

Give the formula for S.E.

Answer 7

S.E. = SD / SQRT(N)

Question 8

Further economic insight can be gained by relating CARs to firm characteristics, even when mean stock price effect is zero. How?

Answer 8

Run cross-sectional regression of CARs on characteristics, such as firm size, industry, Tobin’s Q.

Note, characteristics must be known before the announcements.

Such a regression is important to:
- understand the sources of abnormal returns
- see how abnormal returns can be different across alternative types of firms
- see how abnormal returns can depend upon the characteristics of the event

Question 9

Name 5 limitations / drawbacks of event studies

Answer 9

1. We assume that the benchmark model is correct. Otherwise AR will be incorrect
2. We assume that event windows of firms do not overlap. Otherwise there’s EVENT CLUSTERING and the observations are not independent across securities. The var(avgCAR) is underestimated. Solution = form portfolios.
3. We assume that a firm’s beta remains constant after the announcement
4. Choice of event window and estimation window is a bit arbitrary
5. W assume abnormal returns to follow a normal distribution > use non-parametric tests

Question 10

Things to worry about in event studies…

Answer 10

1. Anticipated events > leakage or announcement timing
2. Confounding effects
3. Event day uncertainty
4. Thin or non-synchronous trading
5. Event-induced variance > vola of returns is assumed the same in event and non-event periods
6. Clustering / cross-sectional dependence > if event windows overlap, t-tests may reject too often, because ARs exhibit small correlations across securities.

Question 11

Why can’t we use CARs in long-horizon event studies? Propose an alternative.

Answer 11

The CAR is not the abnormal long-run return of buy-and-hold strategies, because it ignores the cumulative effect of returns. Instead, use buy-and-hold abnormal returns (BHARs).

BHAR is the buy-and-hold return of the event firm minus the buy-and-hold return on a benchmark portfolio.

Question 12

What’s best to use as a ‘benchmark’ in long-horizon event studies?

Answer 12

Instead of simple benchmark model, it is more common to use the return on characteristics-matched portfolio as benchmark. Typical characteristics are size, B/M ratio, leverage, industry…

Pro: no need to estimate factor loadings
Con: firms with similar characteristics can still be different

Question 13

Issues with long-horizon event studies

Answer 13

1. Results of long-horizon event studies critically depend on model for expected return, because errors are cumulated. This is not really the case for short-horizon event studies, since the discrepancy is negligible.
2. Finite-sample test-statistics have lots of problems:
   - Cross-correlation (many event windows overlap in time due to long event windows, causing cross-correlation)
   - Skewness (long run return of stock is positively skewed due to compounding, long run return of (market) portfolio is not)
   »> these issues can lead to large biases in t-statistics!!!

Question 14

What is the most interesting time window in an event study?

Answer 14

The event window

Question 15

Why does an event study focus on abnormal returns?

Answer 15

To control for overall market movements

Question 16

Why do we often need a cross-sectional analysis in an event study?

Answer 16

Because it helps to understand the sources of the abnormal returns

Because it helps to understand how abnormal returns vary across firms

Because it helps to understand how abnormal returns depend upon characteristics of the event

Because aggregate abnormal returns may be zero, despite clear announcement effects