Event Studies: Constant return, market-adjusted, and CAPM (Excel)

[@NEDLeducation]

You can find the spreadsheets for this video and some additional materials here: drive.google.com/drive/folders/1sP40IW0p0w5IETCgo464uhDFfdyR6rh7 Please consider supporting NEDL on Patreon: www.patreon.com/NEDLeducation

[@bambangsutrisno_umj]

I have downloaded all files from the link above. But I can't find the spreadsheet for this video (event study).

[@NEDLeducation]

Hi Bambang, the Google Drive has been updated. Please check it out :)

[@tantatum5676]

@@NEDLeducation Hi I also can not find the file of this video in the link. I can only find the advanced event study file.

[@paolofu3226]

This should have waaaaaaay more views. Thank you

[@NEDLeducation]

Hi Paolo, glad you liked the video :)

[@melikebetul536]

You have a great way of explaining event studies. Best!

[@MrBetter96]

Hi Savva, brilliant video. Can you explain why there would be statistically significance using the constant return model but not in the market-adjusted model or CAPM?

[@NEDLeducation]

Hi, and thanks for the excellent question! This can occur due to the fact that the event affected the market as a whole and not a specific company, therefore there is no significant effect when the market is adjusted for.

[@MrBetter96]

@@NEDLeducation brilliant, thanks Savva!

[@mattiabachini4007]

Video was really helpful!! I think it will give help me a lot since I am currently writing my bachelor's thesis!

[@NEDLeducation]

Hi Mattia, and glad the video helped! All the best with your thesis!

[@heimlechihoudini416]

This is the best youtube video regarding event studies out there! Thanks so much. One question though, in which papers did you find the three methods used here? Cheers

[@zomaareenkijker286]

Awesome video, helps out a lot. I do have a question however, why do you take the standard deviation of the estimation window and the event window and not just of the estimation window?

[@NEDLeducation]

Hi Zomaar, and glad the you liked the video! This is to make the results more comparable with the regression-driven estimates from the next video. You could calculate standard errors based on the estimation window alone as well. Hope it helps!

[@zomaareenkijker286]

@@NEDLeducation Thanks a lot!

[@GothrixZ]

Great video!!! One question. When calculating the CAPM-adjusted return, why did you include the company's alpha? In my study iliterature capm is only a systematic risk measure of an asset. Why is it important to include the alpha value of the company? Thanks in advance

[@NEDLeducation]

Hi, and glad you enjoyed the video. Excellent question, it is always handy to think about event study specifications in analogy to regression models (this is why I prefer conceptualising event studies as dummy variable regressions, like in this video: kzbin.info/www/bejne/hH3Uon6sh7uHbbc). Adjusting for CAPM without taking alpha into account (just subtracting beta times market return) would be akin to using the same intercept for all sample stocks, while subtracting alpha is analogous to employing cross-sectional fixed effects (allowing stocks to have differing risk-adjusted returns even when no event is observed). Fundamentally, this would be another case of bias-variance trade-off, and I would generally (for my own research) include alphas if my sample is large and they can be precisely estimated and not include them if the sample is small.

[@j2blockchain759]

Hi Savva, thanks for the amazing video! I was wondering how to do a wilcoxon signed rank test (for robustness)? Would it be possible on sheets/excel?

[@Benjamin-um7nq]

Hi NEDL! First off: Great video, it really helps a lot! I've got a question for you though regarding the calculation of the BHAR. You are calcualting the abnormal returns right away via CAPM/MM and then put those returns into the BHR formula, right? Is that a valid approach because I've never seen that anywhere, usually the BHAR formula is used and you put in the expected returns there. You probably do the right thing, I was just wondering... Do you maybe have a source like a book or a working paper or anything where this approach is used that you can refer me to? I know its a lot to ask, sorry about that, but it what really be much appreciated! Thank you!

[@Aida12764]

Thank you so much for the video, it is really helpful. In the video you calculate the first standard deviation for all the days, and in the excel file you calculate the standard deviation for only the estimation period. Which one should I consider? thank you

[@dylanbenjamineberdin1250]

Hi! Most of the material I've seen on CAPM (and used in class) uses tbill rates for to calculate market risk premium and expected return. May I ask why it's not used here? Just curious, and my knowledge on doing CAPM on excel is admittedly lacking. Thank you!

[@NEDLeducation]

Hi Dylan, and thanks for the excellent question! Generally, you do indeed perform CAPM regressions on excess returns (with risk-free rate subtracted from both the stock and market return), however on daily data the risk free rate is virtually zero so in practice even academic research often skips this step when dealing with event studies on daily returns. On lower frequency data, however, the impact of the risk free rate becomes material and cannot be neglected. Hope this helps!

[@Nozzyb44]

Very helpful video thanks mate. I'm really struggling with how to apply this to a large sample of firms. Surely I don't have to carry out the above calculations for every company (around 400 companies)? Would take forever using closing price data to calculate returns etc. If i try to use a quasi portfolio, for example, how do I measure CAR and BHAR for the entire portfolio? Also not sure dummy variables would work, I am going to use them for esg scores within my regressions (i.e. high vs low rated). Your advice and recommended literature would be massively appreciated! I am trying (but struggling) to do an event study with regressions looking at the effect of esg ratings on stocks during covid. thanks!

[@jannik93]

Hey chris, I guess you have solved this problem now. Can you help me out ? :)

[@jonashansen6391]

Did my bachelors in something akin to event-studies. This sure would have helped haha. Great video.

[@panagiotadeli2811]

Thank you for the great video, i have a question as well. When they say that the estimation window and the event window needs to not overlap, did we just achieve this by not taking the same date as the date that the anticipation window starts? or we need to leave more days in bettween?

[@peterc.2301]

Amazing video as always Sava! Some ideas about furure content are, event studies on a spread basis. Also a fama and french 3 - 5 factor model could be nice. Many thanks for your amazing work!

[@NEDLeducation]

Hi Peter, and glad you are enjoying the channel! Thanks for the suggestions, these videos are indeed in my long-term plans.

[@dudiaharon9800]

Thanks for this great video!! How do we do the event study for a sample of firms to which we want to estimate the average abnormal return or the average cumulative abnormal returns and their significance? Can you upload a video on this? Thanks!!

[@NEDLeducation]

Hi, and thanks for the comment! I am planning to record a tutorial on the quasi-portfolio approach for event studies at some point in the future.

[@maxcompier5855]

great video! But if using the CAPM, dont you have to first calculate the excess returns by subtracting the risk free rate (10-year yield government bond) from the daily returns, and with that outcome calculate the alpha and beta?

[@NEDLeducation]

Hi, and glad you enjoyed the video! Excellent question as well. For estimations on daily frequency, the risk-free rate is negligibly low so in practice subtracting it is often omitted.

[@gozdee.5207]

I couldn't get the formula on the cell F2. Why you choose from F34 to F137?

[@sandranwaizugbo3485]

Thank you for this, it is really helpful. However, is it possible to use monthly data to conduct event studies? What are the statistical tests to be conducted befor and after conducting the event study analysis... that is robustness, and stationarity tests.

[@jas1000]

Could you expand on this by calculating CAR and plotting? Also add sig test for say CAR(0,T)? Thanks

[@NEDLeducation]

Hi, and thanks for the question! I might do a video on more advanced event studies techniques shortly.

[@yoyu1155]

@@NEDLeducation This would be really helpful. Please do !! Will be looking forward to this

[@sandrastojanovksa6199]

Hey there! awesome video, has helped me with my accounting assignment so much!! I was just wondering where do you find the closing index for a company?

[@NEDLeducation]

Hi Sandra, and glad the video was helpful! The closing share price can be quite easily found for free on YahooFinance for example, where you would have closing prices (close) and total return indices (adjusted closes for splits and dividends).

[@josuasinaga4749]

Hello, thank you so much for the video. I have one question, when estimating standard error, why did you times the std dev with the square root of the window length? Isn't it supposed to be divided by the square root of the window length? Look forward to your response. Thanks again!

[@NEDLeducation]

Hi Josua, and thanks for the question! Here, the standard error is multiplied to account for the volatility of multiple-day cumulative return (CAR), and not to account for the sample size (where you indeed would typically divide by the sqrt for the sample size). Hope it helps!

[@aisyahtulfaqihah2685]

Hi :) May I know how to calculate the AAR and t-stat? And what are the differences between AAR and CAR?

[@NEDLeducation]

Hi Aisyah, and thanks for the question! AAR and CAR are average and cumulative (simple sum) abnormal returns, respectively. Both can be tested for significance using a t-stat by dividing the returns onto respective standard errors. I have also got a video that naturally presents event studies as a dummy variable regression, check it out as well if you are interested: kzbin.info/www/bejne/hH3Uon6sh7uHbbc

[@Yearjohn]

Great work as always, Savva! but why was stdev used instead of steyx?

[@NEDLeducation]

Hi Yerzhan, and glad you liked the video! Calculating stdev on residuals (which abnormal returns in the CAPM model effectively are) is equivalent to calculating steyx. This was to maintain the analogy between the three models.

[@Yearjohn]

@@NEDLeducation thanks Savva, you rock! Indeed both approaches provide similar results, however steyx uses sqrt (n-2) as a denominator while stdev n-1....

[@aisyahtulfaqihah2685]

Hi :) What if I have 10 companies for this analysis ? May I know how to get the average of all companies for AAR, CAAR and t-test ? I already watched a lot of tutorial but most of them only showed with only one company. Let me know if there is references or tutorial on that.

[@NEDLeducation]

Hi Aisyah, and thanks for the excellent question! For 10 companies (or any number of companies), the same analysis can be performed using a quasi-portfolio of these stocks, calculating its return with respect to the dates of events corresponding to these stocks. Alternatively, it can be tested in a dummy variable regression framework with panel data (this generalisation is easier). I have got a video on event studies with dummy variables, to include more stocks there you can simply stack data under each other to form a panel: kzbin.info/www/bejne/hH3Uon6sh7uHbbc

[@iammallesh3043]

Did you get solution?? Actually I stuck with same problem in market model

[@aisyahtulfaqihah2685]

@@iammallesh3043 I get the AAR by using average of AR for all the companies by particular day. For t-test, I was using SSPS. More easy. I can send you my excel for your references. Let me know your email.

[@jannik93]

@@aisyahtulfaqihah2685 Hi, could you send this to me too?

[@aisyahtulfaqihah2685]

@@jannik93 let me know ur email

[@matthewsmith6697]

Is there anyway you could link a study that utilizes this buy and hold method for calculating abnormal returns? Every study I found is quite complicated to follow and utilizes many additional methods to reduce biases. It seems your method is much more simple and straight-forward

[@matthewsmith6697]

Hi. For my research I am conducting an event study for a sample of Pharmaceutical firms partaking in M&A in the U.S. Specifically, I want to include whether the event being a horzontal or vertical acquisition plays a part in the returns. However, I wanted to focus on a longer horizon, say, 1-year, 2-year and 3-years out from the event. Is this method still applicable? If not, how should I go about this? Thank you so much, love your channel!

[@NEDLeducation]

Hi Matthew, and thanks for the question! Sounds like a worthy research topic. Yes, this is possible to do event studies on a longer time horizon, studying monthly instead of daily returns, and the application is exactly the same as here but on a different frequency. Theoretically, it is possible to even calculate 3-year buy-and-hold abnormal returns on daily frequency, it would just look bulky in Excel :) Alternatively, other techniques, for example calendar time portfolios, can be applied.

[@MrTobisonson]

Can I use this to calculate the CAR or BHAR an then to analyze the over-or underperformance of IPOs in comparison to a market index? I am not quite sure since there is obviously no data available pre IPO.

[@kajalmittal5009]

Hi sir.. This video is very helpful.. I have one query.. How to calculate CAR for all 500 companies (as i have calculated abnormal return and CAR values for individual companies)

[@NEDLeducation]

Hi Kajal, glad the video helped and thanks for the question! For 500 companies, event studies can be performed at least in two different ways. You can either construct a quasi-portfolio of stocks that went through the event (a portfolio that invests in companies sometime before the event was realised for each of the portfolio companies) and calculate the CAR for this quasi-portfolio, or arrange your data into a panel and estimate CAR using dummy variable regressions (covered in the next video). Hope it helps! There are some papers on the subject at Google Drive, please check them out as well if you are interested!

[@TheTrialRun]

Wonder video. can you please make any video on creating Calendar time portfolios, I wants to learn it and found no video on whole youtube.

[@NEDLeducation]

Hi Parveen, and glad you liked the video! As for your suggestion, a calendar time portfolio would just be a type of event study when you investigate the impact of some common and repetitive event on corporate value (i.e., dividend announcements) and track the performance of a portfolio of stocks that experienced this event over the longer term. Hope it helps. Might do a video on that in distant future.

[@TheTrialRun]

@@NEDLeducation please do it , i am still waiting

[@NEDLeducation]

@@TheTrialRun Hi again Parveen, pleased to announce the video on calendar-time portfolios is in the pipeline and I will post it in the upcoming month. Thanks again for the suggestion!

[@TheTrialRun]

@@NEDLeducation Thank you so much

[@nourkarim2721]

Thank you so much for the explanation. I have a question regarding the St.dev. Why did you take the St.dev for the whole period including the days after the post-event window? is it ok if we take the St.dev only for the estimation period without including the pre-event and post-event period?

[@NEDLeducation]

Hi Nour, and glad the video helped! As for your question, yes, certainly, it is also a very common estimation technique! I did it the way I presented it here to make the results more comparable to the dummy variable estimation method from the next video. Hope it helps!

[@camillevanhuysse8838]

Thank you for the video. I have a question about the data selection. I see you use the data adjusted for splits but not adjusted for dividends and capital gains if I am correct. Could you tell me which is best to use because I am a bit confused? thanks in advance!

[@NEDLeducation]

Hi Camille, and thanks for the question! Adjusted closes in Yahoo Finance take into account both splits and dividends. For daily event studies, the choice is unlikely to affect the results materially, but it is preferable to use total return indices (or adjusted closes as they are called in Yahoo Finance).

[@musickingdom7402]

Amazing video! It really helps! If however, I'm doing an event study with 100 different events instead of 1, do I have to do this 100 times?

[@woutkoch7497]

Did you already find the answer to this? I am also having multiple events per year (24).

[@vikramgopinath4141]

Great analysis, could the same be performed for FX pairs ? Just thinking how I can use returns ?? Is it possible??

[@NEDLeducation]

Hi Vikram, yes of course! You can calculate daily returns as daily rates of change in respective FX rates.

[@CristianoDLX]

Hey, great video. I'm using the event study for my thesis at the university. But I'm struggling with citing the correct formula since there are many different ways of calculating the t-stat. What type of T-Testing is this ? I cant find any reference to certain literature. Is this the Crude Dependence Adjustment Test (?). Where can I find a formula that i can put into my thesis that proves your calculations? --> The abnormal return models are explained in certain books similar to your analysis. But for the t-stat everything is explained by using the AVERAGE abnormal return in a certain period. I hope you can help. Thanks in advance !

[@faizaahmed6488]

Hi Sava Its a great work. You are a very learned person. I have one question can I perform event study on regime switch if yes than how?

[@farhanshazia5610]

Very clear explanation. Thank you. Keep up the good work!

[@NEDLeducation]

Thanks! Glad the video helped :)

[@louisemagnusson5135]

Hi! I have a question about the calculation of the std over 10 and 21 days. Why do you take the std of 1 day*SQRT(10) and (21)? Why the square root and not just std 1 day*10 and *21?

[@NEDLeducation]

Hi Louise, and thanks for the excellent question! This due to variance scaling linearly with time (if we assume returns are independent). This would make standard deviation (square root of variance) scale as a square root of time period length.

[@TheOsayilir]

Excellent explanation, thank you.

[@ipintan3072]

Hello Sir, right now I am doing my thesis, and use this great video as reference. If you dont't mind, I've difficulties in analysed the result, Please help me. My thesis is about stock split and abnormal return.

[@NEDLeducation]

Hi Ipin, and glad you are using the video for your own research. All the best with your thesis! I might be able to advise in greater detail if you specify further what the issue is.

[@gokhanduzu]

Thank you for this great video! May i ask to be sure that your 10 day caculations referring anticpation and adjustment corresponding to estimation window of (10,10) .Is this right?

[@NEDLeducation]

Hi Gokhan, and glad you liked the video! Yes, the total return window is [-10; 10], while [-10; 1] is anticipation, [0;0] is the event itself, and [1;10] is adjustment.

[@gokhanduzu]

@@NEDLeducation thanks a lot again

[@filippodipalma8606]

Hi there I noticed that Stdev is calculated over the full sample the video (F34:F137), however, in the template that you have uploaded is only for the estimation period (F34:F90). Could you please advise on which one is correct?

[@NEDLeducation]

Hi Filippo, and thanks for the great question! Naturally, the classical approach is to calculate it over the estimation period only. However, in the regression approach for event studies the standard error is calculated for the whole sample period. Both are acceptable.

[@filippodipalma8606]

@@NEDLeducation Hi there, thank you for the prompt response much appreciated. I have a further question if I may, for instance, the event window needs to be adapted for each of the CAR or can be a standard one? i.e. from the one utilized in a (+20,-20) sample and (+5,-5) can we still utilize the one starting from -20 or shall we calculate it from -5? hope this makes sense

[@aliakitoluckman7111]

Hello Sir, thank you for your explanation video, I wanna ask again sir, Should the confounding event in the estimation period be excluded? or just exclude it from the event period only?

[@NEDLeducation]

Hi Aliakito, and thanks for the comment! If you have identified a confounding event that might affect the estimation window parameters, you can exclude it for additional robustness.

[@aliakitoluckman7111]

@@NEDLeducation Thank you so much for the answer sir

[@british021]

Brilliant video, really clear on the concepts of event studies and how to explain the significance, if any arises. I have a question regarding CAR and BHAR in particular. I am doing my Masters thesis and examining 12 nations stock indices (BRICS and G7 nations) on 7 financial crises, 7 terrorist attacks, and 6 natural disasters. the benchmark will use is the MSCI ACWI Index which captures all nations effectively. As such I will use both the mean adjusted model (or constant return as you name here), and the market model. I want to use both the CAR and BHAR as my estimation window is 90 days, and BHAR being useful for longer event periods. The anticipation and adjustment window will be (-10, +10) and (-5, +5), again as information efficiency might be drastically different from developed and emerging countries. Now to examine the CARs, I will also use the CAARs (Cumulative Average Abnormal Returns) given the amount of events and nations to determine the significance for each category of events (1) financial crisis, (2) terrorist attacks, and (3) Natural Disasters and (4) all events combined. I was wondering if it is possible to do the same with BHAR? If so, how would I go about this. Appreciate any insight or suggestions in regards to this. Cheers

[@NEDLeducation]

Hi Joseph, and really glad the video helped! All the best with your Masters thesis! Looks like you are on the way to a great dissertation :) As for your question, an average BHAR can be calculated using a geometric mean. Just as CAAR is CAR/number of days in a window, BHAAR can be calculated analogously as (1+BHAR)^(1/number of days in a window) - 1. This will account for continuous compound rate of return and work better in longer windows, as you mentioned. Hope it helps!

[@gillesdejonckheere]

@@NEDLeducation Hey NEDL, how about taking an average of multiple firms. Let's say you have calculated the BHAR of firm A of 100 days, how do you then calculate the average of all the firms in your sample? What about the t-stats and p-values? Greets Gilles

[@andreaocchipinti724]

Hi! If I am doing an event study on the price of Bitcoin after an event, and I have its returns from saturday and sunday, how can I solve the issue with the benchmark to calculate alpha and beta? Do i have to delete them and consider only benchmark trading days? Thank you so much 🙏🏼

[@lherath8658]

Who is the founder of the Market Adjusted Model? Can I please have the link to a study?

[@NEDLeducation]

Hi, and thanks for the question! I would point you towards Brown and Warner (1985) and MacKinlay (1997) which are classical event study sources to cite.

[@lherath8658]

@@NEDLeducation thank you so much!

[@aliakitoluckman7111]

Hello sir, thank you so much for your explanation video. I would ask you about paired sample t test to test the different before and after CAR. Is the CAAR Calculation for this test done by averaging the days in event window (average before and average after) or the company-wide average in each event window? And if I use the calculation CAAR for average the company wide it was hard to test the normality data for window -1,1 because the rows is too short. Hope you can explain about my question. Thank you very much 🙏🏻

[@NEDLeducation]

Hi Aliakito, and glad you liked the video! As for your question, using a paired sample t-test for an event study is quite exotic and unconventional, but it should work in principle. You can pair each of the observations from the period ahead of the event with the respective observation before the event and compare abnormal (or even raw) returns. I address the paired sample t-test further in this video: kzbin.info/www/bejne/epiTZHidZ9GhqKc

[@aliakitoluckman7111]

I wanna ask again sir, if I use SPSS to test the abnormal return significant should I test the normality data for the first?

[@NEDLeducation]

@@aliakitoluckman7111 Hi again, not necessarily, the results of event study tests are rather robust to non-normality.

[@aliakitoluckman7111]

@@NEDLeducation thank you so much sir

[@thomashirzinger]

So how is the risk-free rate incorporated in the capm here? And thank you for the video

[@NEDLeducation]

Hi Thomas, and thanks for the question! Generally, in daily event studies, one can omit the risk-free rate from the analysis as the value of daily risk-free rate is very small (for example, a 2% annual risk-free rate would be just 0.0079% daily, very negligible short-term). The inclusion of risk-free rate is more relevant for monthly or annual returns. Hope it helps!

[@Caiocaiocaio18]

@NEDL - this video is very useful. Are there any papers which reference the fact the risk free rate can be omitted for daily returns?

[@iammallesh3043]

Thank you sir I am stuck with ARR, CAAR and t value (standard deviation, standard error) I have 44 Abnormal Returns by using market model... What to do sir to get solution above things

[@NEDLeducation]

Hi, and thanks for the question! If you have got 44 individual abnormal returns, you can test for their significance separately, as I show in this video for example, or you can find whether they are significantly different from zero on average, using something like a t-test on the sample of 44 returns.

[@iammallesh3043]

@@NEDLeducation where I have to test This t value sir and how to test t value Could you tell me sir... Send me any video link

[@pashton]

Great work. thanks alot you just saved my day

[@DaljeetKaur-l2o]

Sir, I have a question that if we are taking different indices in event study methodology then the calculations for each indices will be on different sheets. If we have 15 indices and 5 different events under study, it is not possible to include huge tables or more than 100 diagrams in research paper then how can we summarize the results?

[@kergok5662]

great explanation. thanks so much.

[@NEDLeducation]

Hi Ker, and glad the video helped! Stay around for more content!

[@melisarosman]

Hello sir. Sorry for asking a lot. Actually, in this event study, the type of data is panel or cross-section? And if i research the impact on each 9 sector of the company, so i have to test in SPSS multiple times (9x) as the number of sectors right?

[@NEDLeducation]

Hi agan Melisa, and thanks for the comment! As for your question, this event studies example is a simple time series for one company. If you would like to perform event studies on multiple companies, you can either use pseudo-portfolios (a portfolio of all sample companies with returns recorded at a particular point in time with regards to the event of interest), or indeed perform multiple tests (in your case, nine). Hope it helps!

[@nayabarshad1524]

WELL-EXP-LAINED. this has been extremely helpful to me 👌

[@melisarosman]

Hello sir. my lecturer asked me, if the data aren't distributed normally then we used wilcoxon signed rank test method (median based), how can we make sure the conclusion is valid while data arent normal? Thankyou sir sir

[@NEDLeducation]

Hi Melisa, and thanks for the question! Overall, linear regression can still be used when the data is not normally distributed as normality is not required as per the Gauss-Markov assumptions. More technically, normality simply ensures that regression estimators are equivalent to maximum likelihood estimators. Non-parametric tests such as Wilcoxon signed rank test are used to provide additional robustness when non-normality is present and to ensure adequate statistical power. I might do a video on Wilcoxon test at some point in the future. Hope it helps!

[@HongNgocHo-if3zb]

How I calculate AR, CAR if I only have 1 firm but there are ten event days

[@Strongholdex]

22:20. Is it not necessary to divide with standart error instead of standart deviation?

[@melisarosman]

Sir, is it OK when we run OLS regression to get the alpha and beta in market model? but the data is not normally distributied, and doest meet the OLS assumption?

[@NEDLeducation]

Hi again Melisa, have responded to your comment earlier. Long story short, yes you can, as normality is not required as per the Gauss-Markov theorem.

[@parminderbajaj3]

Hello sir, Thanks for this video.I am performing event study analysis using CAPM model. I have calculated CAR for 21days event window and now I want to compute T statistics of CAR. Can you help me how to calculate it.

[@NEDLeducation]

Hi Parminder, and glad you liked the video! The t-statistic for the CAR would be the CAR divided by the standard error of your CAR estimation. As your CAR is over 21 days, you can take the standard deviation of daily abnormal returns in the estimation period and multiply it by the square root of 21. Hope it helps!

[@victorjansen783]

Thanks for the great video. I am looking however for a method to analyse multiple events trough time. I have about 330 events trough time, first i want to calculate for each event if there are abnormal returns, and after that i want to compute a test with take all the single events in consideration. Does anyone know how to do this in excel ?

[@NEDLeducation]

Hi Victor, and thanks for the great question! Yes, there is a technique for that: you can assemble a quasi-portfolio of assets based on the timing of the event (calculating average return across 330 assets, e.g., 10 days before the respective event, 9 days before the event, etc.) and then adjust the standard errors by the respective sample size. I might do a video on that at some point in the future.

[@CatNamedRio]

Hi, what will the formula be if my event window is -40 and +10? I see that the sd uses 15 days, which is the same for the anticipation and adjustment periods.

[@NEDLeducation]

Hi Asbi, and thanks for the question! In your case, the standard deviation would be multiplied by the square roots of 40, 10, and 51 for anticipation, adjustment, and full periods.

[@alexcod556]

Great video. I'm kind of new to event studies and working with a task analyzing public listed sports teams. Im analyzing how different match results affect the stock price, so there is a lot of events each year. Since I am working with a lot of events, lets say 1 event per week (1 match per week), how should I instead choose the anticipation, adjustment and event window? does it affect the results if an event I'm analyzing is withing another events anticipation, adjustment and event window? For example in your case, what if i did want to analyze the event of the Blizzard event 05/11-2018 but aswell as an event occurring 12/11-2018 and another one at 19/11-2018?. Can i do so by only changing the anticipation, adjustment and event window? Hope you understand what i mean.

[@NEDLeducation]

Hi Alex, and thanks for the excellent question! Happy to hear you are applying the insights from the tutorial to your own research. As you correctly stated, the major issue with event studies when events are frequent is confounding events (particularly, when event windows are too large they start overlapping). For this application, I would reduce the window lengths to the minimum, having one day anticipation and one day adjustment. Also, do make sure you correctly map matchdays to trading days - sometimes the game is on a weekend or occurs late in the evening when markets are closed already which means you need to effectively carry the event date forward to the next trading day. As a football fan myself, I did play around with this quite a bit and can wish you all the best with your research :) Here is a classical paper on the issue: www.tandfonline.com/doi/pdf/10.2753/REE1540-496X4705S404

[@rashedulhassanshimul1929]

Thank you

[@melisarosman]

Hello again sir. I wanna ask for my thesis, if the closing price firm are same from t-10 until t+10 and i get all abnormal return are zero (0), should i take that firm in my sample or just banish it??

[@NEDLeducation]

Hi again Melisa, it might be the case that your firm is quite illiquid or infrequently traded. If that is the case, excluding it can be a good idea. If it is reasonably frequently traded and the closing prices matched by chance, you can leave it in the sample. Hope it helps!

[@melisarosman]

@@NEDLeducation thank you sir

[@antoniomanuelcunha]

Hello. Did you find the t-statistic for the T-test by dividing abnormal return by standard deviation? Isn't it supposed to be diving by the standard error (not the same thing as standard deviation).

[@Dr_Robbi]

Hello did. You find the answer to this question yet? I am conducting an event study and stuck on that point

[@antoniomanuelcunha]

@@Dr_Robbi if you are using a sample, and, as such, you are doing inference (and not measuring the entire population), i have no doubt you should divide by the standard errors, not the standard deviations.

[@Ganieirfan]

What price or return should I take for event day if event day is a holiday for exchange?

[@NEDLeducation]

Hi Irfan, and thanks for the question! Generally, if the information is released onto the market on a non-trading day (e.g., a holiday), the next trading day after it is considered the event day.

[@Ganieirfan]

@@NEDLeducation Thank you so much

[@delaneyian4675]

so helpful

[@melisarosman]

Hello sir. Iam using market model to estimae expected return. I wanna ask, should we adjust the standard error of abnormal return, or we can deny it??

[@NEDLeducation]

Hi Melisa, and thanks for the question! If you are using the market-adjusted model, you can calculate the standard deviation of stock return minus market return in the estimation window to then derive the standard error for the event studies. Hope it helps!

[@melisarosman]

@@NEDLeducation sorry i mean, i am using Market Model sir. And i wanna ask again, can we use OLS to estimate the alpha and beta for market model? Because we know that one of the assumption in OLS method the data should be normally distributed. While the daily stock price data usually doesnt normally distributed. Iam a little bit confusing to examine the abnormal return using market model 🙏

[@NEDLeducation]

@@melisarosman Hi again Melisa, and thanks for the follow-up question! Yes, you can. The normality assumption is not required for the alpha and beta estimates to be reliable. As per the Gauss-Markov theorem, normality implies OLS estimates are equivalent to maximum likelihood estimates, but even if errors are non-normal, the estimators are still best linear unbiased. Hope it helps!

[@samin3742]

Legendary video

[@TheDharkos]

Awesome video thank you very much ! Does the formulas used for t-stat and p-value are also supposed to work if returns are computed using log ? Also why are you not using the log returns ?

[@NEDLeducation]

Hi and glad you liked the video! As for your question, yes, you can theoretically use log-returns as long as you do not apply pseudo-portfolios in your event studies (it is a technique that is used when you estimate an impact of similar events that occurred to different assets on different dates, and averaging a portfolio return over log-returns leads to biased results). I stick to simple returns as I use the product function to calculate BHARs and as it is generally a convention in the literature. Hope it helps!

[@HonISfirE]

how to find t significance of CAR of t days?

[@NEDLeducation]

Hi, and thanks for the question! To calculate the t-stat for the CAR over a particular window, you can estimate the standard error based on the daily standard deviation times the square root of the window length. Hope it helps!

[@shivamazad2712]

Product function is not giving any value Please help

[@kumariversha6985]

if instead of a Activision Blizzard a country (India) is used than do we still have SP500. If yes then why?

[@NEDLeducation]

Hi Kumari, and thanks for the question! For country-wide event studies, you can use a broad global stock market index (e.g., S&P Global 1200, MSCI World). For India specifically you can also use a global emerging market index (e.g., MSCI Emerging Markets). Using S&P 500 is not wrong, however, as the US market does proxy the global stock market reasonably well (although there are better alternatives available as mentioned above).

[@kumariversha6985]

@@NEDLeducation is your estimation window 56 days or 57 days. i got 57 after including 31st july

[@paolofu3226]

What if a stock is not traded on a particular date but the market index is? Should I use vlookup using as the search_value the days of the stock?

[@NEDLeducation]

Hi Paolo, and thanks for the question. Ideally, the trading days for the stock and for the relevant market index should coincide. If, however, it is the case they do not, you can either simply treat the next trading day as the relevant event day for your stock (technically, you can use vlookup to do that) or look at the opening price jump. Hope it helps!

[@paolofu9881]

@@NEDLeducation thank you 😁 Im doing an event study for 115 companies. After following your procedure, how can i combine the p-values of all the events to obtain a single p-value, in order to assess the statistical significance of the entire study

[@NEDLeducation]

@@paolofu9881 Excellent question! You can approach it from a number of perspectives. First, you can form a quasi-portfolio of these stocks (calculating the simple average or market cap-weighted average of returns for all stocks on the event day, and before and after the event days, respectively), applying the event study technique to the quasi-portfolio. Alternatively, you can use some multiple testing adjustment procedure to figure out if any of your 115 p-values are genuinely significant or it is random chance due to a large number of hypotheses being tested simultaneously. Among the most popular and simplest such methods would be the Bonferroni adjustment (multiplying all p-values by the number of tested hypotheses, so here you would need to multiply all p-values by 115 and see whether any remain below 5% or 10% threshold), or Wilson harmonic mean p-value (calculating the harmonic mean of p-values and comparing it to 5% or 10%). Hope it helps!

[@furqaanahmed8964]

Is there an easy way to apply this to a sample of 100 firms?

[@NEDLeducation]

Hi Furqaan, and thanks for the question! Yes, you can apply the same tests to a quasi-portfolio of 100 firms with returns recorded with respect to firm-specific event days, or, for a regression framework, in a panel estimation with dummy variables. Might do a video in the future to clarify it further!

[@furqaanahmed8964]

@@NEDLeducation please could you provide a quick run down of how to achieve this