• Add names of the tests in the CAR tests.
• Return a data.frame?
• Add an average return over the event window as an item of the list.

Hi Iegor,

I am using your package for my event sutdy and it is working great. However, I wanted to ask if there are possibilities to report also more statistical properties such as standard errors and such?

At least, I could not figure out how to get them through the package.

Many thanks for your great work.

Hi Iegor,

hope you are doing well.

I think I have spotted an error in the car_brown_warner_1985 calculation. In your function, you take the mean of daily brown warner:

function (list_of_returns, car_start, car_end, percentage = 90) 
{
    ...
    car_mean <- **mean**(daily_brown_warner_1985_statistics_tidy[, 
        4], na.rm = TRUE)
    statistic <- sum(daily_brown_warner_1985_statistics_tidy[, 
        5], na.rm = TRUE)/sqrt(nrow(daily_brown_warner_1985_statistics_tidy))
  ...
}

However, with this approach I think (not completely sure though) that you take the average across day t and the average across securities i, leading to the daily average abnormal return in the event window instead of the cumulative average abnormal return across the event window.

When calculating in excel, I get the right results editing your code the following way:

{
    ...
    car_mean <- **sum**(daily_brown_warner_1985_statistics_tidy[, 
        4], na.rm = TRUE)
    statistic <- sum(daily_brown_warner_1985_statistics_tidy[, 
        5], na.rm = TRUE)/sqrt(nrow(daily_brown_warner_1985_statistics_tidy))
  ...
}

Note that in the statistics you actually take the sum function.

Let me know what your thoughts are on this.

Kind regards

Hello,

I have cluster/categories of news that I wish to test the significance as a sample. Can you please let me know if it is possible to do such rank test through the package?

Thank you in advance.

Hi Iegor,

I've been playing with this package and it's really great - thank you for coming up with it. After some research, the only thing I was not able to understand how to do was adding multiple event dates and perform the analysis using the same number of days for the estimation period. Is there any way to use the same market index, the same estimation period, the same event window but use different stocks and different event dates for each stock? I would really appreciate your help.
Cheers!

• Fix brown_warner_1985() help page. The test takes into account a cross-sectional dependence.
• Elaborate on the meaning of weekday, percentage, and mean of returning data frames
• Fix car_brown_warner1985() help page. Looks like duplication of Lamb test.

Hi!,

Thanks for getting back to my email. As requested, I'm replying with my issue on here.

I have coded up a FamaFrench model on Python which is working, and I'd like to use the results in R and apply statistical tests to them from estudy2. I understand I need to create an object similar to that returned by the apply_market_model() function (ie, a list object with at least dates, normal, abnormal and predicted returns).

My question is this: How can I import the results into R (e.g. from a csv) and reformat them so that they are suitable to be passed to the parametric and non-parametric statistical tests?

I've attached some sample data (random) in the format I can export the CSVs in.

EDIT: I've got to the point where I can import all the relevant CSVs into R and reformat them as per the picture below, but the parametric/nonparametric test throws this error: Error in if (list_of_returns[[i]]$estimation_end >= event_start) { :
argument is of length zero --> I guess I need to reformat the data so the function knows how to deal with it. Any help greatly appreciated!

Thanks,
Will

ISIN3.csv
ISIN2.csv
ISIN1.csv

Screenshot of my object that is trying to mimic the S3 object returned by apply_market_model

Hello,

I wish to carry out non parametric tests but I am not able to generate class returns from my data. I wish to use autogression for abnormal return which will later be tested on the non parametric tests for significance. I tried the single_return code in the documentation but it is also not recognised as a class returns.

If all rates are NA, throw an error in returns().

• car_lamb()
• car_brown_warner_1985()

When calculating the car_mean in car_parametric_tests, only the mean of the ARs are calculated - there is no cumulative summation.

e.g., for the calculation of car_mean in the car_lamb process, in car_parametric.R :
car_mean <- mean(daily_lamb_statistics_tidy[, 4], na.rm = TRUE)

This takes the ARs from daily_lamb_statistics_tidy and calculates the mean.
There is no accumulation of the ARs prior to calculating the mean, and therefore it is not a cumulative average abnormal return.

Thus all that is provided is the longitudinal mean of ARs. These ARs are already the cross-sectional averages (which is correct - might be clearer to change the name to AARs as is typical in literature i.e. Average Abnormal Returns).
Thus it would be more useful to provide the CAAR (Cumulative Average Abnormal Return) by calculating it as:
caar <- sum(daily_lamb_statistics_tidy[, 4], na.rm = TRUE)

The test is described in Corrado in the extension by Cowan (1992, p. 6).

Hi Mr Rudnytskyi,
I hope you are well.

I am a master's student at UCT in South Africa, and am making use of your package estudy2 for a large part of my analysis for my dissertation.
Thank you for providing the package!

Please could you assist me with a problem I am finding with the car_lamb() testing function.

I have a fairly large dataset of equity prices.
I run your package process (as illustrated in the vignette), and record the results of apply_market_models(). I then take these returns objects and reorganise copies into the four different levels of ICB Industry classifications (in descending order; industry, supersector, sector, subsector).

Then I run the all tests on these new groupings, namely
parametric_tests(), nonparametric_tests(), car_parametric_tests(), car_nonparametric_tests().
This functions without error on the larger groupings (i.e. Industry & Supergroup).

However, when I attempt to apply car_parametric_tests() on the smaller groupings, the car_lamb() function runs into the following error:

Error in data.frame(..., check.names = FALSE) : arguments imply differing number of rows: 5, 4 In addition: Warning message: In '/.default' (event_means, sd) : longer object length is not a multiple of shorter object length

Despite my best efforts, I have not discovered why this does not work with the smaller groupings. This error is exactly the same for each subgrouping that throws it.

I need to demonstrate that this is something to do with the smaller groupings, and not the data since the same data and the same market models are used in each grouping.

P.S. I skip every group that is less than or equal to 2 in size.