Up next in 10
Earning Management in Stata | Accrual management in Stata
2K views
May 16, 2024
In this video, I share an easy method to estimate discretionary and non-discretionary accruals in Stata. I have developed a command ear_mgt that can be used to estimate Jones (1991), Dechow (1995), Kasznik (1999), Kothari (2005) models. It can also be used to estimate the discretionary and non-discretionary accruals model for each year i.e. for each year or for any other variable, it can estimate crossectional regression and estimate the discretionary and non-discretionary accruals. Download the files used in this video: https://payhip.com/b/3lxym Website: thedatahall.com As an Amazon Associate, I earn from qualifying purchases.
View Video Transcript
0:00
Welcome to the Data Hall YouTube channel
0:02
In today's video, we are going to talk about earning management. So earning management is an area that is well researched on
0:12
And within earning management, we have a specific area which is called accrual earning management
0:20
And what accrual earning management is it contains two different types of accrual earning management
0:27
One is called discretionary accrual. And the second one is called non-discretioning accruals
0:32
So discretionary accruals are those accruals that are intentionally managed by the managers
0:39
And non-discretioning accruals are the one that results due to certain accounting procedures
0:47
Now, we need to divide or disentangle these two accruals, right? So we have total accruals and then we need to divide them into these two parts so that we can use them as a dependent or
1:00
independent variable in our research. And to divide accruals, total accruals into or to disentangle this total accrual into the
1:10
discretionary and non-discretionary part, we need to use certain models. So we have these four main types of models that are used in research
1:19
We have John's model, DeKau, 1995, Kassniak model, and then we have Kutari model
1:25
So what we are going to do is we are going to look into the, the formulas or equations of these models
1:32
We are not going to talk about the theoretical aspect of it. That is what you can cover from the research papers
1:39
And then we are going to look into how to estimate these models in Stata
1:45
So basically this is the applied side of this accrual management. How do we estimate these accrual management models
1:53
So these are the four models that have placed them side by side
1:57
So we can compare them and we can compare them. we can understand what are the differences in these models
2:02
Again, why are these differences there? These papers do present that through article justifications in the research papers, right
2:13
So first thing is to estimate these models, we first need TA
2:19
first thing, it is panel data, so we need information, we need cross-sectional and time series data
2:25
So for each firm, we need different years of data. First, we need to calculate TA, which stands for total accruals
2:33
And to calculate total accruals, it is a difference between net income minus OCF or cash flow from operations, right
2:44
Operating cash flows. Once we have total accruals, what we do is we divide that with the lag of assets
2:53
So if you can check it this over here, this A. IT minus 1 stands for the lag of total assets
3:03
So this becomes the dependent variable of our, let's just say, JOLS model
3:08
On the independent side, we have the constant term, which is one divided by the lack of total assets
3:20
Again, all these variables are scaled by the lack of total asset
3:24
to control for heterostasticity. Then the second variable is change in rapid
3:28
avenue that is again scaled by total assets, lack of total assets
3:33
And the third variable is property plant and equipment, right? So what we do is we take our dependent variable
3:40
regress them on the independent variable. And the residual part of this equation
3:47
would become the discretionary accruals, right? And the fitted value of this equation would become the
3:55
the non-discretionary accruals. and how do we do that in Stata? That is what we are also going to discuss in this video
4:03
Then Decar model is somewhat similar. We just do the change in account receivable, right
4:10
REC stands for accounts receivable. So change in account receivables. Those are deducted from revenues
4:18
And the rest of the equation is exactly the same. Then the third model is also somewhat same to decarve 1995
4:26
except for we have change in cash flow from operations. And in Kodahari model, if you can see in other models, they have the constant, they use
4:39
the no constant. So this is a regression through origin and this is their constant
4:48
So not exactly the regression through origin, but when you execute this, you would not use
4:56
the constant. So if you are using Stata, you would use the no constant option. But in Kotaari
5:02
we do have a constant. This is what they have explicitly mentioned in their paper
5:08
Okay. And the additional thing is as compared to KASNIC model, instead of CFO, we have R way, right
5:18
So what we do is these are the steps that we use. We select the relevant models. So first we
5:23
need to decide which models we would use either we use John 1991 or DECA which is also called
5:30
modified John model etc so once we have decided on the model then we would calculate the variable that are used in the model So if we decide to use John model then we need to calculate TA
5:45
We need to have change in revenues. We need to have PPE, etc
5:50
And then we need to estimate the regression. The residual from the regression in step three that we have estimated would become the measure of discretionary accrual
5:59
The fitted value would be a measure of non-discretionary accrual. do remember that there are two ways of performing this step three
6:07
Either you perform a pooled regression, pooled or less regression, or what you can do is you can regress
6:17
you can do a cross-section regression for each year. So for each year we execute a cross-section regression instead of applying a regression
6:27
on whole data, and then we calculate the residual. And again, the residual becomes the
6:33
the discretionary crewers. Okay. So how do we do that in in status? There are multiple ways of doing that in
6:41
Stata. You can you can obviously write your own do file but what I have done is I have
6:49
made an easy to use command which is called EAR underscore MGT. This whole thing becomes the
6:57
command. This part is the command. Normally you would install the command using SSC install, but this command is not available using SSC install
7:08
What you can do is you can download it from the link given in the description of this video
7:14
Okay, so once you have written the command name, then you need to specify the variable in your
7:19
dataset that contains total accrual. Remember, you do not need to calculate this whole part
7:27
You just need total accruals and the program, the command would automatically need
7:33
divide it by the lag of total asset. So you just need to provide the total accruals
7:39
You can also use if and in parameters, conditional statements. Now we have these options
7:47
So first option is we need to specify a model. So within models we either write
7:52
John's 19901, DECO, etc. Whatever model we specify, we would have to specify the requirements for that model
8:03
example if we specify we want to use John's 1991 so what we do is we specify
8:08
John's 1991 over here and so this then we would have to so in John's
8:18
1999 we have total accruals which we have already specified in the equation in
8:23
the command we need revenues and we need PPE right so what we need to do is
8:28
there is this revenues option so we write revenue and and within parenthesis, we write the name of the variable in our data set that contains the revenues
8:41
And the second thing for John's 1991, we need property plan equipment
8:46
So we use the PPE option and we specify within parenthesis, we specify the name of the variable. Right
8:57
the name of the variable that contains the data related to property plant and equipment
9:06
Okay, so let's move to Stata and see how we are going to execute this in Stata
9:14
So first thing is that when you download that command from the link given in the description of this video
9:23
So what you would see is you would be able to download this ADU file
9:29
So this, what you need to do is you need to copy this ADU file and paste into the directory
9:35
which is your current working directory, right? And then you would be able to use this command
9:42
Okay, so my current working directory is the D earning management folder
9:49
So I'm going to set by current working directory. and then I'm going to import the data file
9:56
So within data, what we have, let me show you the data
10:03
So let me sort this data on symbol and ear. Okay, so we have different firms
10:14
And for each firm we have around four years of data. We have their property plant and equipment, revenue account to see will, total assets, return on assets, total accruals. Remember, this is not scaled. And then we have cash flow from operation, CFO, right? What we need to do is, first, we need to actually set this data. So we need to tell data that this is a panel data. And within our data, this variable identifies the firm information and this is the time variable
10:49
Normally, you would have yearly data to estimate earning management values, earning management models
10:58
So, X2Sat. Next, what we need to do is we need to let me remove all this and let me write it by myself
11:08
So we need to write the command name. You would write it exactly like this including the underscore without any spaces This is the exact command that you going to use Then specify the variable that contains the information regarding total accruals
11:25
Again, this is not scaled, divided by a lag of total assets because that is what the program would do by itself
11:34
Next, we need to specify the option, so which model we would use
11:40
So there are four options that we can use. So you can write either write John's 1991 or Decama 1995
11:49
Kazniak 1999 or Code Harder 2005. Whichever model you want to use, you would have to write it in exact same manner
11:58
all small caps and without any spaces within them. So if you do something like this, I mean you do not write the model
12:10
that I have specified over here or you misspell the string, then it is going to give you an error that the option model is
12:22
incorrectly specified, right? So you need to specify the correct model. So once you have, let's get back to that
12:30
once you have written the model, then what you need to do is you need to provide
12:36
the specific parameters that are required by the Johns And I have discussed that in my presentation file
12:45
So John's 1991 model requires that you specify the total accrual, the revenues and PPE
12:54
Remember, you don't need to calculate the change in revenue by yourself
12:58
That is what this program would do. You just need to specify. So what we do is within the syntax, we have this revenue option, right
13:08
So we would specify revenue and within this revenue we would write the name of the variable that in this case it happens to be exact same name, but that might not necessarily be the case in your data
13:25
The second thing that we need to specify in John's model is the PPE and for PPE we have this PPE option, right
13:35
So we specify a PPE option. Again, if we miss specify, or let's say if we forget to specify all the variables that are required
13:47
and we execute the model, what it is going to say is that for John's 1991 model, you need to specify
13:55
these three options, right? So we are going to specify PPE and our property plant and equipment variable is this one
14:10
And last thing it tells us to specify is total asset because we need to scale them using total assets
14:16
So this is T acid and this is our variable that contains the information regarding total assets
14:25
So once we have done that, if we execute this command, what it would do is it would generate these two variables
14:38
DAQ, which stands for a discretionary. So its name would be, would start with DAQ, either it would be DAQ or NDQ and then the name of the model that we have used
14:51
So in the description we have, in the label we have its description
14:55
This is the discretionary accruals using John's 1991 model and not discretionary
15:00
approval using John's 1991 model. So that was the first thing that we have done and let me sort that by year just so
15:09
that we can see what had happened. You would see that for the first year which is 2005 in our dataset
15:17
We do not have any values for discretionary and not discretionary equal
15:22
That is because we are going to take lag offroral assets. And that would cost us one year of values
15:29
From 2006 and onwards, we would have discretion and non-discretion accrual values. Right
15:36
Now remember, what it did is it executed, it regressed the total accruals scaled by obviously all the variable scaled by total assets
15:47
And then whatever the error, whatever the residual value was became our discretionary accrual
15:54
But most of the research, what most of the research would do is it would execute a cross-section regression
16:04
So for each year, right, so for example, for 2008, it would execute this regression, right
16:12
And then calculate the residual and so forth for each year. So you can also do like for each sector and year, but in this case I'm just going to do
16:24
for each year. So this command is viable, right? Byable means we can use
16:30
we can use by sort with this, by sort prefix with this command
16:36
So we use BYS, which stands for by sort, ear, and then colon
16:42
and then the same command that we already had. But one thing that you need to remember is that
16:47
because when this command would be executed let me show you the error when this command would be executed it is going to because we are still using John model so it is going to generate this variable so we need to first remove them
17:04
Otherwise, we would not be able to execute this command. So once we have executed this command
17:13
so the residual and the DAQ and NDQ values would be, be different but what it had done it has used cross-section regression for each year right
17:26
okay what if we specify the decar model right this model name but we didn't so in decar
17:38
models what we need is we we also need account receivable but we haven't specified the
17:44
account receivable option so what if we execute this it is going to give us an error
17:49
and the arrow says that for the COW model, what you need to do is you need to specify all these options
18:00
So we have already specified revenue, PP and total asset options, but we didn't specify the account receivable
18:06
So what we are going to do is we are going to write account receivable and specify the account receivable variable
18:16
right if we execute that again you can do that by year by sector year or simply remove the prefix
18:25
and that would generate again the the discretionary and not discretionary rule for the decomm
18:33
model we are going to do quickly the other two models as well we are going to use this casniac model
18:41
and within this casniac model you would see the everything is exactly
18:46
the same as DECO but we just have additional variable as change in CFO so what we
18:52
are going to do is we have this CFO option and we are going to specify our variable name that
18:59
contains the cash flow from operations information and also we need to change oh we have already
19:08
changed that the name of the model we execute that we have our discretion or
19:13
description of approval. And for the last thing, for the Koto Hary model, we need ROA instead of CFO, right
19:23
And we also, okay, I will come back to the constant thing in a minute. So we need ROA and as you can see that
19:35
although I am specifying the ROA option, but I am not removing the CFO, right? But that doesn't matter because
19:43
what this program would do is it would not take CFO into account, right
19:47
Even if you have specified that into the command. So let's change that to Cotahari
19:55
And if we execute that and let me just show you the..
20:05
Okay, let me just summarize this DAQ for a second. And you can see that these are the values and what if
20:17
Let me drop these two and let me re-execute this but now remove CFO and let's see if that changes or not
20:31
And then we do some and you can see that whether you write those additional variables or not
20:41
the model is correctly calculated. So in previous case, we had additionally written the CFO option
20:48
although CFO option is not there in Cotari model, but the command would work perfectly fine
20:55
Now, let's come to the last option that we have in our command that is called no constant
21:04
So remember when you do regress DV on different IVs, there is one option which is called no constant within this regression and what this means is that
21:16
you need to so the regression would be through origin there the constant term would not be included in the regression right
21:23
and why this is needed is because if you look at all these models except for the kotoari models a kotahari model has its constant but in all other three models we do not have consens so there is a debate whether there should be a constant or not
21:39
But if you want to use these models as they were designed with all these Johns Decau and Kassniak, you would have to use this option over here that is no constant
21:58
But for the Koto Hari model because there is a constant, you would not have to use this option
22:04
So to make it simple, use the no constant option with Jones, although I didn't used but you should use it except for the Koto Hari model
22:20
So I hope that was useful. Do subscribe to this channel and do hit the like button and you can download this command from the link given in the description
#Accounting & Auditing
#Business & Industrial