Getting Standard Error of Regression of previous observations

[DTHORNBLAD]

Hi all,

I am fairly new to STATA so I would appreciate help. To simplify, I have a dataset with two columns: year and money as shown below (fictitious data, for illustration):

2000	95256
2001	98531
2002	65465
2003	54612
2004	54624
2005	54652
2006	64671
2007	32132
2008	13211
2009	25896

What I would like to do is for the year 2009, generate a variable that displays the standard error if I were to regress money onto year for the five previous years. So, the new column in 2009 would be the standard error of regressing money onto year for 2008-2004. I can do this in excel fairly easily, but I cannot seem to do this is STATA. If it helps, I have been able to create five columns for each row that obtains the values for 'money' from the previous five years, but I do not think that will help.

Thanks everyone. Really appreciate the assistance.
-David

[Rich Goldstein]

I don't find the question very clear, but try the following code

gen se=.
forval i=2005/2009 {
regress money year if year>`i'-6 & year<`i'
replace se=_se[year] if year==`i'

}
rich

[Joe Canner]

David,

Here is a very simple-minded way to do what you are asking for:

regress money year if inrange(year,2004,2008)
gen se=_se[mpg] if year==2009

However, if you want to repeat this process for multiple years (e.g., for each year, generate the SE of the regression for the previous five years), that would require more work. If that is the case, let us know more about your data and we can suggest a more general solution.

Regards,
Joe

[DTHORNBLAD]

I would need it to do it for multiple years, but I can do that with a loop correct?

[DTHORNBLAD]

Hello all,

First I would like to thank everyone for the previous help, but I am still struggling. I have done this in the past in excel, but I am attempting to push my skills with Stata and to remove the chance of manual error. My goal is to find a way to record the standard error of a regression into a variable, while only regressing the previous five observations. Nick Fox suggested that I look at rolling commands, which I have been trying, but is not working. I would appreciate it if you could give me some advice.

Since I seem to have been unclear previously, my data looks like below. I have different airline names with different revenues for different years. For some airlines I have more years of data than others. My goal is to regress the yearly operating revenues (yr_op_revenues) onto the year variable, for the five previous observations (not including the current observation, which rolling seems to do). For example, for a 2006 observation I would like to regress the data from 2001-2005 and record the standard error estimate of the year (only independent variable) into 2006. If there is not five years of data, I want the standard error to be missing data.

I have tried by statements, but the last line of this code gives me a 'weights not allowed' error so I cannot put the se variable into the observation I desire.

gen se = .
by unique_carrier_name: regress yr_op_revenues year if inrange(year,year[_n],year[_n-6])
replace se[_n] = _se[year]

I have tried nested foreach commands to get within the airline, and then do the regression for each year, but the code below has not been working.

foreach airline of varlist unique_carrier_name {
foreach yearnum of varlist year {
regress yr_op_revenues year if `yearnum' == `yearnum'[_n-6]+6
gen se=_se[year]
}
}

I have also tried a foreach with a rolling command. I needed to do a foreach before the rolling or else there were non-unique years of data. However I keep getting invalid syntax.

foreach airline of varlist unique_carrier_name {
local egen min_year = min(year)
local egen max_year =max(year)

tsset year
time variable: year min_year to max_year
delta: 1 unit

rolling _b _se, window(5) saving(betas, replace): regress yr_op_revenues year
}

I know this is a lot to ask, but how in the world do I do this? I do not want to go back to making datasets in excel, but STATA is driving me a little nuts. Thanks so much everyone.

unique_carrier_name	year	yr_op_revenues	se
American Airlines	1996	23541	.
American Airlines	1997	26526	.
American Airlines	1998	15486	.
American Airlines	1999	23654	.
American Airlines	2000	58962	.
American Airlines	2001	25426	.
American Airlines	2002	35214	.
American Airlines	2003	25632	.
American Airlines	2004	25415	.
American Airlines	2005	26842	.
American Airlines	2006	26547	.
American Airlines	2007	21456	.
United Airlines	1993	36514	.
United Airlines	1994	25632	.
United Airlines	1995	21456	.
United Airlines	1996	23698	.
United Airlines	1997	57852	.
United Airlines	1998	21453	.
United Airlines	1999	26987	.
United Airlines	2000	24563	.
United Airlines	2001	25426	.
United Airlines	2002	26541	.
Delta	1985	26523	.
Delta	1986	25413	.
Delta	1987	26532	.
Delta	1988	24123	.
Delta	1989	25412	.
Delta	1990	26351	.
Delta	1991	23652	.
Delta	1992	23652	.
Delta	1993	25413	.
Delta	1994	25632	.
Delta	1995	25874	.
Delta	1996	26531	.
Delta	1997	25412	.
Delta	1998	36521	.
Delta	1999	35632	.
Delta	2000	96542	.
Delta	2001	25632	.

[Nick Cox]

Lots of confusion here! Frankly, the impression is that you are making lots of wild guesses about what the syntax might be and then are surprised and disappointed when you get lots of error messages in return. Still, you are trying something that isn't trivial at your stage of learning Stata.

On the rolling suggestion that I made (not Nick Fox, whoever he may be), it's an error to tsset your data just in terms of the time variable: it's panel data. You need to declare panel identifier too.

This code, for example,

local egen min_year = min(year)
local egen max_year = max(year)

is fairly bizarre. It is important that you learn not to write code randomly like this. (That may well sound patronising, but it's true.) It happens to be legal. But what does it do?

The first statement creates a local macro egen containing the text following. The second statement then replaces the macro contents with different contents. But you never use the macro, so nothing is achieved. Fortunately, if you just tell tsset what your set-up is you don't need anything like it.

You should be able to adapt this example to get (closer to) what you want.

webuse grunfeld
tsset company year
rolling  _b _se,  window(6) : regress mvalue year

[DTHORNBLAD]

Hi Nick,

Sorry about getting your name wrong previously. To answer your question, I wrote the min_year and max_year variables to calculate the starting and ending year of data for each airline. As I read rolling, it seemed that I needed to pre-define a time period, but since the years for airlines were different, I was trying to make the time period variable with min_year/max_year.

[Nick Cox]

OK, but I am not clear that you take my small point. Your local definitions just define the contents of a little macro, meaning here one piece of text. They would not themselves do anything to calculate minimum and maximum years. The egen statements, applied separately, would themselves fail for other reasons.

The positive advice remains. As I understand it, your central problem yields easily to rolling.