Expert Dialog

Question #27: Difference in mean - pre/post financial regulation

If you are conducting a study where you test AAR/CAAR in association with corporate credit rating changes before and after a financial regulation (ie. test if the effect on the stock market from a credit rating change (Moody's) is more/less extreme after the regulation), how would you test for the difference in mean?

I have looked at multivariate regression models but i don't quite understand it. Could you be so kind to elaborate on how you would test for the difference in post/pre AAR/CAAR.

Answer by Dr. Simon Müller:
You may define two event windows (e.g. -10,..., 0 and 0,..., 10). If you have no significant effect on event window 1 and a significant effect on window 2, then you know that the financial regulation has an effect on your firms. This is option 1. Option 2: calculate abnormal returns and use a t-test for paired samples for examining two time points. How to calculate this test on CAAR? - I have to think about it.
Question #26: Possible Error in R Code

Just a short note: I guess there is a sort of small typo within the R code example provided within the 'local tools for event studies'-section. When calculating the simple returns, the variable filled should not be called firmSub$firmSub but rather firmSub$firmReturn.Otherwise it will not be found within the OLS estimation and further on.

Besides: I would appreciate more extensive event study R code examples a lot, but of course see your need to focus on other priorities :D

Answer by Dr. Simon Müller:
Dear Rabea,thank you for that tip. I fixed it.Best,Simon
Question #25: Adjusted significance tests

I'm having problems in calculating both the adjusted Patell test and the Adjusted BMP test. Could you explain me how to get the average of the sample cross-correlation of the estimation period abnormal returns? Is this the Pearson correlation coefficient or anything else? In the analysis report you provide it is reported the first-order autocorrelation, is this in any way utilised? I'm stuck on this step and your help would be really useful.

Answer by Dr. Simon Müller:
The reported autocorrelation is NOT the value used in the adjusted BMP test. The correlation in the adjusted BMP test is the correlation across firm time series and yes we use Pearson. Maybe you check the definiton of autocorrelation.
Question #24: Cross-Sectional Test

Ich wüsste gerne woher bei der Erklärung des Cross-Sectional-Tests die Wurzel N vor dem Bruch kommt. Die leitet sich für mich nirgends heraus ab und ist auch unterschiedlich zu allen gängigen Artikeln über ein und dieselbe Signifikanztest-Art.

Answer by Dr. Simon Müller:
Das hängt mit der Definition des AAR/CAAR-Terms zusammen. In unserem Fall wird hier der Mittelwert gebildet, in anderen Papern (z.B. in der Beschreibung von Eventus) wird im Zähler die Summe verwendet.
Question #23: bARC Error: Empty Estimation Window was Extracted

My data has missing values of stock price for some dates in estimation window of some events, therefore bARC result give an error for those events "Empty estimation window were extracted". As I don't have stock price data of firms for those missing dates, is there any way to perform analysis successfully for those events.

Answer by Anton Levchenko:
bARC perfrorms analysis even if estimation window is shorter, than expected, but calculations for events with empty estimation window are senseless.You should complete data to perform analysis for such events correctly. Also you may set other parameters for estimation window of the event.
Question #22: Validation Rules for Input Data

I am currently working on my master thesis in Finance, for which I need to conduct an event study, to see the effects M&As have on the stock performance of these companies. However when uploading the CSV files I get: ''Unrecoverable error during Event Data import''. Without any explanations of what is going on? I have soms cells for which there was no data (for example not public yet or non existent at that time), I named these ''NA''. Does it have something to do with this?

Answer by Dr. Markus Schimmer:
Your file structure must conform with the file structure examplified in the sample files the website provides. Also, each variable/column has a distinct data type... so when a number is expected for a variable (e.g., closing price) you must not put text, such as NA.
Question #21: Validation Rules for Input Data

I am currently working on my master thesis in Finance, for which I need to conduct an event study, to see the effects M&As have on the stock performance of these companies. However when uploading the CSV files I get: ''Unrecoverable error during Event Data import''. Without any explanations of what is going on? I have soms cells for which there was no data (for example not public yet or non existent at that time), I named these ''NA''. Does it have something to do with this?

Answer by Dr. Markus Schimmer:
Your file structure must conform with the file structure examplified in the sample files the website provides. Also, each variable/column has a distinct data type... so when a number is expected for a variable (e.g., closing price) you must not put text, such as NA.
Question #20: bARC Error : Empty Estimation Window were Extracted

Dear Prof,

I got around 5000 events for around 700 firms in my data. When calculated bARC, results show only 1600 events successfully and for the remaining events following error was found in Analysis report folder.

"Event (Id: 980) calculations stopped due to reason: Empty estimation window were extracted. Event (Id: 981) calculations stopped due to reason: Empty estimation window were extracted" similar error for all remaining events.

Can you please any advice, why this error occurred.

Thanks for your kind cooperation

Best regards,
Imran haider

Answer by Anton Levchenko:
Hello, Imran,This error means your firm, or market data have gaps, and the software were unable to extract estimation window. It is a critical error, so calculations for the event 981 are skipped.Please check your input for the time period, mentioned in the event 981, and complete data, if needed.Best regards,Dr. A. Levchenko
Question #19: Seemingly Unrelated regression in event studies

I would like to perform an event study analysis with multiple events per company. The event type that I am about to examine is enforcement actions, thus I am faced with a dilemma. Since my firms have received several enforcement actions during the a 10-year time span without seasonal or calendar effects, meaning that a particular firm can receive an enforcement action in i.e, January 2010 and receive another enforcement action in September 2010 or another firm could receive an action in March 2007 and the next action in April 2014. Given that, I have several firms that receive actions in a very short period of time, thus if I will employ the market model my estimation window will contain another enforcement action of the same firm. So, my main question here is should I treat every firm's event as separate in my event study (i.e., let us assume that a firm has 5 events, should I treat each event separately?) or as I have during the estimation window other events of the same type (and of the same firm) should I use another approach as the one described in Binder (1985), the seemingly unrelated regressions?

Answer by Dr. Markus Schimmer:
You should definitely continue treating each event individually. Also, you should investigate whether the confounding of your estimation period is material. If you have a long time series of returns making up your estimation window, the enforcement action included may have a minor/negligable effect on your alphas and beta - if so, you should go ahead with the market model. Yes, using the approach presented by Binder (1985) may be an alternative route... but my assumption is that you can go with the market model.
Question #18: Patell t-stat comparison

I hope all is well. I am currently undertaking an event study methodology on a merger between two firms. I have followed your brilliant steps by using the market model and I am testing the null hypothesis that AR=0. My patell t-stat has come to -0.628714625. I am slightly confused about what I compare this value to in order to reject or accept my null hypothesis.

Answer by Dr. Simon Müller:
Your value of -0.6271 is a t statistic. As a rule of thumb a test is significant to the 5% level if the absolute value of t test statistic is greater than 1.96. If you want to calculate the exact critical value or the p-value, you may do it in your favourite statistic package or in Excel. Degree of Freedom = sample size - 2 (with market model).

Pages