i'm having problem in deciding how to run the regression.
i want to see whether before and after covid, banks look at different characteristics when pricing loans. If for example in the period before covid loan-specific characteristicss were the ones driving the interest rate charged, I would like to understand wheter the same relationship still holds after covid, or whether other variables become more relevant, in my case whether firm-specific characteristics weight more in the period after covid
i created a dummy variable for covid period (2020-2021) and my dataset goes from 2011 to 2021.
To sum up, I would like to understand whether and how the indipendent variables of my model changed their relevance after the event covid-19
how can i investigate this hypothesis?
@Maria Rabbi are your data in monthly or yearly. If yearly, you have only 11 observations which is not good enough for running a linear regression. If you have monthly data, run a linear regression and see association between loan characteristics (independent variables) and prices of loan (dependent var) before and after covid-19 (year dummy). Also intersect year dummy and loan characteristics and see the coefficients.
Proloy Barua thank you for your reply.
this is the result that i obtain but how can i interpret them? they look a bit confusing to me
Maria Rabbi ,
I would say:
- Include time-fixed effects (month or year dummies).
- Include the covariates of interest.
- Include the covariates of interest interacted with a dummy variable capturing the period after COVID. You do no have to include the dummy variable alone, as you already include month fixed effects.
Regarding the interpretation. Let's assume that b0 is the coefficient of "Years to maturity" and b1 is the coefficient of "Years to maturity x COVID".
b0 is the effect of years to maturity in the pre-covid period.
b0 + b1 is the effect of years to maturity in the post-covid period.
b1 captures the differential effect of years to maturity in the second period compared to the first period.
I hope it helps,
José-Ignacio
Your variables remain the same after and before covid. Regression follow the same pattern. Your topic must be researsherable.
The result of the regression Will not change once your variables are correct.
you can use the Kintos (1995) approach , divide your sample into two sub-samples after identifying the eve@nt as a stuctural beak.
Maria, obviously you wanted to confront us with a riddle. You gave no information about your variables (definition, dimension). Therefore, it is almost impossible to give advice. It is clear, that you have much more data than just 11, because you seem to have data on individual credit cases (how many?) with information about the enterprise (company) which gets the credit. Your dependent variable seems to be the interest rate on new credits measured in basic points (%/100). You should have a more thourough look on the definitions and dimensions of your date. Obviously, you have chosen your Covid-dummy with 1 before 2020 and -1 for 2020 and 2020, and I think that you have quarterly or even monthly data. In principle, that would mean enough data for estimation, but one must assume a very strong (auto)correlation between them. Therefore, their (statistical) value is questionable. I think, it is not feasible, to introduce a dummy and to add all independent variables multiplied by that dummy in one equation. This must lead to multicollinearity.
The main problem is that one cannot consider the fall of the interest rates as effects of a change of bank's behaviour as a consequence of Covid. It is, of course, mainly the interest policy of central banks together with low inflation, which is the main cause of the fall in (not only) credit interest rates. It is clear, that for low interest rates most supplements are lower. Another problem is that Covid is not over and, therefore, we do not have valid comparable after-Covid-data. Maybe that in 4 or 5 years, you can try two separate regressions for the before and after Covid period, but even then I recommend a more thorough specification.
If you are interested in a 'change in regime', then you will want to run a Chow Test. As others have mentioned, you never mentioned how many observations you have. You didn't define the unit of measure. And you didn't mention how many independent variables you have. There are serious issues to regression analysis if you have insufficient degrees of freedom. All of these issues are far more important in determining if your model will be sound, than the appropriate regime change test. A Chow Test will require that you have 30+ observations on each side of the 'expected' regime change. All of this assumes that you are using a time series dataset and not a cross-sectional dataset.
- Badges
- Science topic
- Similar topics
- Mathematics
- Statistics
- Statistical Software
- R