Page 1
Case Western Reserve University
Weatherhead School of Management
FNCE 435 – Empirical Finance Fall 2021 Assignment 4


Part I – Concepts

  1. An earnings restatement is a revision of one or more of a company’s
    previous financial statements to correct an error. According to Investopedia, such
    restatements “can result from accounting mistakes, noncompliance, fraud,
    misrepresentation, or a simple clerical error.” From those, only clerical error is a benign
    reason for a restatement; all others suggest either poor managerial quality or wrongdoing.
    As such, your friend says that investors should react negatively to earnings restatements.
    You want to conduct an experiment to test your friend’s assertion—more specifically, the
    assertion that the return at time t on the stock of the company announcing the earnings
    restatement will be negative on average (across all such announcements). Please explain
    what you should do—in terms of setting up your hypothesis, which data to collect, and
    how you would test your hypothesis. In particular, be specific about the exact statistical
    test to be implemented.
    You can assume you have access to a large sample of earnings restatements, and to any
    data you deem relevant to your analysis.
    Figure 1: Targets’ abnormal returns around merger announcements
    FNCE 435 Fall 2021 Assignment 4
    Page 2
  2. An event study of merger announcements tracks what happens to the targets’ returns
    around announcement dates. Merger announcements occur sometimes after the close of the
    market, so reactions to such announcements may take place in the next trading day. The
    pattern is shown in the figure below. Take the first row of the table, for example: it states
    that average abnormal returns (where abnormal return is measured as the target’s stock
    return minus the market return) across targets for the 10th trading day preceding the merger
    announcement is –11 basis points, or –0.11%.
    Discuss whether merger conveys good news, bad news or no news for the target. Define
    the null as the merger does not carry any news, and the alternative hypothesis as the merger
    conveys good news. Please write down the hypothesis and test it.
    Then discuss whether markets are efficient in incorporating the information in the merger
    announcement.
    Part II – Empirical Examination
    This part will empirically explore the market reactions to sell-side analysts’
    recommendations. Sell-side research is big business. It employs many graduates from
    business schools, and a good amount is money is involved: Most top tier banks in the
    United States spend more than $100 million per year with equity research. Sell-side
    analysts do research on specific firms and produce reports with estimates about firms’
    future earnings, long-term growth, sales, etc.
    Analysts also produce recommendations on the firms they follow. Recommendations come
    with extensive research reports, but in the end a recommendation amounts to a statement
    about whether someone should buy, hold, or sell. For example, a buy issued by analyst i to
    firm j’s stock at time t means analyst i recommends that one should buy firm j’s stock at
    time t because that analyst believes the stock is undervalued and expected to appreciate in
    the next 12 months.
    Of course recommendations might be meaningless, just statements without any support
    from the firm value’s fundamentals. In fact, there is a long literature discussing how sellside
    analysts face enormous conflicts of interest. They might issue, say, a buy
    recommendation not because they truly believe the stock will appreciate but due to a
    different motivation. Here are two stories, for example, on why a buy recommendation
    might not be really predictive of higher returns for the recommended stock:
  3. Besides issuing recommendations and forecasts, analysts engage in other activities.
    Most notably, they can assist the investment banking activities of their employers
    in their advisory role in equity offerings. Specifically, they can help the investment
    bankers in their “pitch” to become underwriters in equity offerings. Having an
    analyst cover a firm can be beneficial to the firm since it raises awareness about the
    firm. It is even more beneficial if the coverage carries a positive tone—that is, if
    the projections issued by the analyst are optimistic. One big selling point for an
    investment banker is thus her ability—though the sell-side research arm of her
    employer—to provide sell-side coverage for the firm involved in the offering—and
    perhaps a coverage with a positive tone!
  4. Analyst i may also issue unwarranted optimistic recommendation to firm j’s stock
    in order to curry favor with firm j’s management personnel.
    FNCE 435 Fall 2021 Assignment 4
    Page 3
    We want to examine whether recommendations are valuable. One way to do that is to track
    the path of the recommended stock in the year following the recommendation’s
    announcement date. That has many challenges, though. Instead, let us take an easier path
    and examine how investors interpret a recommendation, and thus examine the market
    reaction to the recommendation. We examine such reaction through two different
    hypotheses.
    Take the event of issuance of a buy recommendation. For such event, we want to examine
    two hypotheses. First, we want to check whether the announcement of a buy carries any
    implication about the value of the firm. Our null hypothesis is that there is no market
    reaction to the announcement, so we write:
    H0: No price reaction to the announcement of a buy
    Ha: There is price reaction to the announcement
    Our second hypothesis is conditional on the rejection of H0. If we do find reactions to
    announcements of a buy we want to further examine the efficiency of the market in
    impounding new information in the firm’s stock. For that, we look at the overall price
    pattern around (and more important, after) the event date.
    I have collected for you 5 subsamples of recommendations. They are randomly selected
    recommendations issued between 1993 and 2018. The file, ‘a4_rec_data.sas7bdat’
    (available on Canvas), contains the following variables: PERMNO (permno of the
    company for which the recommendation was written), AMASKCD (a code identifying the
    analyst issuing the recommendation), ANNDATS (the recommendation announcement
    date), REC_TYPE (the type of the recommendation), and DESCRIPTION (a textual
    description of the type of recommendation).
    The following table shows the possible values of REC_TYPE and DESCRIPTION and
    what they mean exactly:
    Figure 1: Description of the types of recommendations in the recommendations dataset
    Take the first record of the dataset. It says that analyst 856 (AMASKCD=856) issued on
    July 18, 2002 (ANNDATS=”18JUL2002”d) a buy recommendation (REC_TYPE=1,
    DESCRIPTION=”Buy”) for the firm with PERMNO=10032.
    Figure 1 shows that dataset “a4_rec_data.sas7bdat” has 5 different types of
    recommendations. The dataset has 1,000 recommendations of each type. You will run five
    REC_TYPE DESCRIPTION Comment
    1 Buy
    New recommendation is a buy and the previous
    recommendation by the same analyst was either a hold or a
    2 Sell
    New recommendation is a sell and the previous
    recommendation by the same analyst was either a buy or a
    3 Up to hold
    New recommendation is a hold and the previous
    recommendation by the same analyst was a sell
    4 Down to hold
    New recommendation is a hold and the previous
    recommendation by the same analyst was a buy
    5 Buy Reit
    New recommendation is a buy and the previous
    recommendation by the same analyst was also a buy
    FNCE 435 Fall 2021 Assignment 4
    Page 4
    different event studies—one for each type of recommendation. In particular, notice that not
    all buys may be created equal. Figure 1 separates between a buy that implies an upgrade
    (that is the buy was preceded by a hold or sell, REC_TYPE=1) and a buy that simply
    reiterates a previous buy (REC_TYPE=5). Investors may interpret these two types
    differently.
  5. Download daily stock returns and market returns for the firms in the recommendations
    dataset. Use a period that encompasses all the dates in your input file, let’s say, between
    between January 1st, 1992 and Dec 31, 2019. Recall that daily stock returns appear in the
    DSF dataset (located at ‘/wrds/crsp/sasdata/a_stock’), and daily market returns appear in
    the dataset DSIX (at ‘/wrds/crsp/sasdata/a_indexes’).
    Of course, you can optimize your data collection and collect returns only around the
    important dates (let’s say 50 days around ANNDATS_SAS and 50 days around
    ANNDATS_SAS). It is up to you.
    (Hint: You will have to download stock returns for the whole sample period, but you do not
    need to download returns for stocks that are not part of the sample of recommendations.
    That is, since you need returns only for the included firms you will need to upload the
    inclusions file into the WRDS server. Use the PROC UPLOAD, which as the name implies
    does exactly the reverse of the PROC DOWNLOAD we already know. So you will need to
    run a code like this:
    proc upload data=d;
    run;
    Important: to avoid rerunning the downloading of the data, a task that can be time
    consuming since it depends on the performance of the Unix server, create a code to run
    step 1, and once you are done, start a new code for steps 2 and onwards.)
  6. Now you have your events clearly defined and the data on returns, it is time to run the
    event studies—five of them! The first event is the issuance of buys. Compute average
    abnormal returns for the firms receiving a buy recommendations in the 11-day period
    around these announcement dates. You should fill and show a table similar to the first panel
    (under “Buys”) in Figure 2. In Figure 2, AR stands for market adjusted returns, that is, the
    abnormal return when we subtract VWRETD from the firm’s actual return (RET variable).
    (This step is very similar to what we did in class for the event study on dividend
    announcements. That is, we need to match the datasets, then compute abnormal returns,
    then average the abnormal returns according to the nature of the subsample. Lavoisier once
    explained that “in Nature, nothing is lost, nothing is created, everything is transformed”.
    Accordingly, use the code we discussed in class as a starting point for your analysis.)
    FNCE 435 Fall 2021 Assignment 4
    Page 5
    Figure 2: Summary statistics, abnormal returns
    Plot and show a graph of the cumulative abnormal return around buys’ announcement
    dates. Important: you are graphing cumulative abnormal returns, not market adjusted
    returns. A suggestion of the format of your graph (though not exactly its shape) is shown
    in Figure 3.
    Figure 3: Average cumulative abnormal returns around event dates
    How to generate the graph? We’ve used the PROC GPLOT in class. Suppose you have a
    dataset D with the following variables: REL_DAY (from t=-10 to t=+10), ACAR (average
    cumulative abnormal return for that specific day and subsample). You thus write:
    symbol1
    color=green interpol=spline value=square;
    proc gplot data=d;
    plot acar*rel_day;
    run;
    Finally, examine what happens with the cumulative abnormal return from days +2 thru +5.
    For the event study of buys, for example, you fill out and show the first line in the output
    Day # obs
    Average
    AR(%) t # obs
    Average
    AR(%) t # obs
    Average
    AR(%) t # obs
    Average
    AR(%) t # obs
    Average
    AR(%) t
    -5 000 0.0000 0.00 000 0.0000 0.00 000 0.0000 0.00 000 0.0000 0.00 000 0.0000 0.00
    -4 000 0.0000 0.00 000 0.0000 0.00 000 0.0000 0.00 000 0.0000 0.00 000 0.0000 0.00
    -3 000 0.0000 0.00 000 0.0000 0.00 000 0.0000 0.00 000 0.0000 0.00 000 0.0000 0.00
    -2 000 0.0000 0.00 000 0.0000 0.00 000 0.0000 0.00 000 0.0000 0.00 000 0.0000 0.00
    -1 000 0.0000 0.00 000 0.0000 0.00 000 0.0000 0.00 000 0.0000 0.00 000 0.0000 0.00
    0 000 0.0000 0.00 000 0.0000 0.00 000 0.0000 0.00 000 0.0000 0.00 000 0.0000 0.00
    1 000 0.0000 0.00 000 0.0000 0.00 000 0.0000 0.00 000 0.0000 0.00 000 0.0000 0.00
    2 000 0.0000 0.00 000 0.0000 0.00 000 0.0000 0.00 000 0.0000 0.00 000 0.0000 0.00
    3 000 0.0000 0.00 000 0.0000 0.00 000 0.0000 0.00 000 0.0000 0.00 000 0.0000 0.00
    4 000 0.0000 0.00 000 0.0000 0.00 000 0.0000 0.00 000 0.0000 0.00 000 0.0000 0.00
    5 000 0.0000 0.00 000 0.0000 0.00 000 0.0000 0.00 000 0.0000 0.00 000 0.0000 0.00
    Buys Sells Upgrades to hold Downgrades to hold Reiterations of buy
    FNCE 435 Fall 2021 Assignment 4
    Page 6
    in Figure 4. This analysis mimics the analysis in page 4-13 of the lecture notes on how to
    properly examine the efficiency with which the market reacts to announcements.
    Figure 4: Analyzing cumulative abnormal returns from days +2 thru +5
  7. Now, repeat step 2 for each of the other 4 events and fill in the rest of Figures 2 and 4
    accordingly (You may but do not need to create Figure 3 for the other events.)
    (Hint: You should create one single code to analyze and event and run the code separately,
    only changing one single line to define the event being analyzed. For example, if dataset
    D has the full dataset of recommendations for the five events, and you want to restrict the
    analysis to the event on buys, you run:
    data d;
    set d;
    where rec_type=1;
    run;
  8. Now, the important part: interpret your results!
    For each event, use the outputs from steps 2 and 3 to address the hypotheses in this study.
    Does the announcement of recommendations is interpreted as informative by investors?
    Does the information matches the recommendation’s advice? That is, do markets react
    positively to buys, negatively to sells, etc? Is the market efficient in reacting to the
    recommendations? Make sure you state your hypotheses clearly, and formally examine
    them from a statistical standpoint. The analysis should be carried out separately for each
    event.
    Event # obs
    Average
    CAR[+2,+5] t
    Buys 000 0.0000 0.00
    Sells 000 0.0000 0.00
    Upgrades to hold 000 0.0000 0.00
    Downgrades to hold 000 0.0000 0.00
    Reiterations of buy 000 0.0000 0.00