# How to interpret coefficients in a dynamic OLS model?

I am trying to understand how to interpret the dynamic and static effect from coefficients in regression models.

$$GDP\_growth\_rate_{t,i} = \beta_1GCF_{t,i} +\beta_2GCF_{t-1,i}+\beta_3GCF_{t-2,i} +\beta X_{t,i} +u_{t,i}$$

where GCF is Gross Capital Formation and the model is estimated using OLS.

My question is am I correct in interpreting $$\beta_1$$ as the impact multiplier /immediate effect of GCF on GDP and $$\beta_1+\beta_2+\beta_3$$ as the long-run multiplier/effect?

yes the way how your model is set up $$\beta_1$$ would be immediate effect/multiplier and $$\beta_1+\beta_2+\beta_3$$ the long-run one.

However, an important caveat is that this is due to the way how you set up your model and not a general result. For example, in an ARDL model with stationary variables of the following form:

$$y_t = \alpha + \beta_1 y_{t-1} + \gamma_1 x_t + \gamma_2 x_{t-1}+ e_t$$

the long run multiplier would actually become: $$\frac{\gamma_1 + \gamma_2}{1 - \beta_1}$$

or in more general case

$$y_t = \alpha + \sum_{p=1} \beta_p y_{t-p} + \sum_{q=1} \gamma_q x_{t-q+1} +e_t$$

the long run multiplier would be given by: $$\frac{\gamma_1+\gamma_2+...+ \gamma_q}{1-\beta_1-\beta_2-...-\beta_p}$$.

In your case you do not include any lags of dependent variable so you have a special case where the denominator is 1 and hence it is enough to add in the coefficients but I thought it might be good to mention that as long as you include lagged dependent variable the calculation of the long run multiplier changes (see Verbeek (2008) guide to modern econometrics for more details).

• Thank you so much! your explanation cleared up my confusion. I want to confirm one more thing if that's okay. if I now include another lagged independent variable for example debt, then would that change the long-run effect of FDI? Thanks again, your answer was very helpful Aug 15 '20 at 22:15
• @hardy i am glad that I could be of help. Regarding your question it would not change the formula but of course all coefficients in regression are calculated conditional on each other so the point estimates would change.
– 1muflon1
Aug 15 '20 at 22:19