Hi,
I would like to estimate the impact of formal microcredit and informal microcredit on consumption as C = b1X1 + b2X2 + b3M + b4E + ... + e, where X1, X2 are exogenous variables; M is total amount microcredit borrowing (continuous variable), M includes formal and informal borrowing; E is education achievement (school gap).
I identify M and E are endogenous variables.
A household might borrow microcredit from formal or informal sources or both because they might borrow 3 times a year (Ex: one from informal, two from formal). Data is panel data (5 years)
As far as I know, IF M is a dummy variable for borrow =1 otherwise 0 (not borrow). we can use the bivariate discrete choice and endogenous treatment effects to estimate the impact of formal credit and informal credit separately. However, in my case, M is an amount of total microcredit borrowing, a household might borrow from formal, informal, or both, or not borrow.
One more problem is that E also an endogenous variable in the main equation. I do not know how to correct the endogeneity.
Sum up: My question is that
How can I estimate the impact of formal credit and informal credit separately and correct for the bias cause by M and E?
Thank you so much for your help!
Best Chung
Chung your analysis is clear but i suggest that IV GMM might not be good enough. Therefore, i propose you consider QVAR
Hi Chung,
I would like to try your question.
First, I suggest you take careful note of your topic and the data you have. Examining the impact of formal microcredit and informal microcredit on consumption implies that you need data on these two variables. Having data on M which is defined as total amount of microcredit borrowing indicate that you can only study "the impact of microcredit on consumption". Note that M is not discrete instead continuous. The only possible way to investigate your topic is to get disaggregated data on microcredit (formal and informal separately). If these two data sets are not available at the various data sites, then I do not think you can use the M data to study your current topic. Added to this, you must check the definition of M very well. If M is the sum of formal and informal (total) then you can study the proposed topic(see above). M can be total but it is the total formal microcredit say in a year or total informal microcredit. Either of these two, you can study the proposed topic. In the case of the above three data scenarios, you must clearly state how you measured the "microcredit" in the proposed topic.
The other possibility, which I recommend that you review empirical papers, is to get proxy variables for formal and informal microcredit. Then, you justify well how such proxies measure an individual's formal and informal microcredit.
Regarding endogeneity, I think your question lacks some information. You mentioned "panel data" without telling us the estimation technique you are using. The issue of endogeneity can be tested. And panel estimation techniques (e.g. GMM) can easily be used to solve if one finds such problem. Some of the techniques generate instruments which are considered exogenous under some assumptions and these instruments are used for the potential endogenous variables. I suggest you argue (from theoretical perspective) strongly how and why you suspect M and E endogenous in that model. If possible back with empirical papers that have done same. More importantly, test for endogeneity and if you find that, use the appropriate panel estimation technique that can solve it.
I hope this answers your question.
Hi John,
Thank you for your answer!
I would like to go point by point in your answer.
First, I have data on formal and informal amount of microcredit that household borrow.
Second, dis-aggregate total M into (Mf + Mi), where Mf, Mi is formal and informal credit respectively, is a good idea. However, I forget to mention that M in the equation is in form of log. the equation should be C = b1X1 + b2X2 + b3log(M) + b4E + ... + e
Therefore, I think it is not able to dis-aggregate logM into logMf + logMi. Also a household can borrow Mf or Mi or both, therefore, there should be a way to deal with the non-random selection of Mf and Mi. In case M is an dummy for borrow or not, the bivariate discrete choice and endogenous treatment effects approach can be used.
Third, the proxy for microcredit. So far, in literature, researcher use a dummy or amount of microcredit borrowing. I will keep searching for other proxies
Fourth, I am going to use IV GMM to estimate the equation. the suspect M and E are endogenous come from process to derive the estimable equation (Educational achievement appears not to be strictly exogenous. Other than the optimal level of consumption, a household also makes decision on its education. Hence, like the estimable equation, a household’s optimal education achievement is functions of the microcredit borrowing and other relevant factors. Therefore, it is likely that the education achievement (E) is correlated with the error term of the equation) .
So, the question is How to disagregate M with taking account for sample selection and the present of endogenous E?
Please let's me know if it is not clear
Best
Chung
- Badges
- Science topic
- Similar topics
- Philosophy
- Philosophy of Mathematics
- Informality