Testing for structural break for time series data

[Ashvinder SinghChaddha]

Hello everyone,

I have a time series data where I regress air transport traffic against GDP, oil price, and FX rates. I have observations for each variable from 2000 to 2019. I wanted to know if there was a structural break in the time series for GDP and undertook the following command

regress historicalpassengers gdp, vce(robust)

where I obtained the following output:

 
Linear regression

Number of obs =
20


F(1, 18) =
1742.79


Prob > F =
0.0000


R-squared =
0.9836


Root MSE =
3.4e+06





Robust

historical~s Coef.
Std. Err.
t
P>t [95% Conf.
Interval]




gdp 379.8068
9.097881
41.75
0.000 360.6928
398.9207

_cons -2.81e+07
2504966
-11.23
0.000 -3.34e+07
-2.29e+07

I then ran

estat sbsingle

and got the following:

. estat sbsingle
insufficient observations at the specified trim level
r(198);

Wondering if what I did was right? Thank you.

[Justin Niakamal]

Judging by the error, you don't have enough observations (N = 20). I'm not sure how appropriate it is to adjust the trim level, but you can try something like this to see what level allows the test to run:

forvalues x = 1/30 {
    cap estat sbsingle,  trim(`x')
    if _rc == 198 {
        continue 
    }
    di "`x'"
    continue, break
}

[Ashvinder SinghChaddha]

Judging by the error, you don't have enough observations (N = 20). I'm not sure how appropriate it is to adjust the trim level, but you can try something like this to see what level allows the test to run:

forvalues x = 1/30 {
cap estat sbsingle, trim(`x')
if _rc == 198 {
continue
}
di "`x'"
continue, break
}

Thanks Justin. For the syntax you provided above, is there a part I can change to adjust the trim level?

[Justin Niakamal]

That's what the code does essentially. It finds the trim level once the 'insufficient observations at the specified trim level' error is no longer encountered. Note that was for demonstration, I can't comment on whether or not that's an appropriate course of action. Here's a modified version

forvalues x = 1/30 {
    cap estat sbsingle,  trim(`x')
    if _rc == 198 {
        continue 
    }
    local tlevel = `x'
    continue, break
}

estat sbsingle,  trim(`tlevel')

[Ashvinder SinghChaddha]

I have a question on structural break based on this. After having tested my time series data for structural break, I found that in my n=20 observations, nearly every year has a structural break. This weakens my next step of using a dummy to overcome the structural break problem, to 'absorb’ the structural break. Example:

I am interested in b1 in regression y=a +b1X1 +b2X2

• I identify year=2 is a structural break. create dummy=1 if year=2
• New model: y=a+b1x1 + b2(x1*dummy) + b2x2
• Here, b1 is the estimate during ‘normal times’, and b2 is estimate during ‘break'

Given that I have a structural break in every year, the effect of including the dummy is weaker (I think). Wondering if there are any solutions to this?

[zakelo]

Hello everyone

I am trying to estimate a non-linear model with STATA 15, after the specification of the transition variable, when I launch the estimation of my model on the data which are spread over the period 1990-2020, I have the following error message :

"not enough observations at the specified trim level".

I have tried all the proposed solutions in this forum but I always get the same error message. Is there a solution to solve this problem?

Thank you in advance