Presentation on theme: "The Supply and Demand Side Impacts of Credit Market Information Discussion Atif Mian, Chicago GSB."— Presentation transcript:
The Supply and Demand Side Impacts of Credit Market Information Discussion Atif Mian, Chicago GSB
Broad Question The paper poses a fascinating question: How does the availability of public credit registry information alter lending relationships? –We have now a large crowd emphasizing the importance of stuff like law, property rights, institutions etc. –We understand them at an abstract level, but what do they mean operationally? –One way to think about this question is the mechanism design view. For example, institutions are mechanisms that societies are able to design and implement in order to achieve some common good. –A very important example of such mechanism design is availability of credit history at individual and corporate levels. Many countries still do not have such effective systems, even though they are quite simple in terms of technology and investment. –So how important are credit histories? And how exactly do they work to improve financial market efficiency? This is my read on the contribution of this paper.
Quick Outline They have loan level data from a large microfinance bank in Guatemala. Staggered entry of credit registry across the 39 branches from March 2002 to January 2003 However, borrowers had no knowledge of this. They were then randomly given this knowledge between June and November 2004. (only done with groups, not individuals) Exploit staggered entry (I think) to show that there large screening in and out of borrowers, which improves loan outcomes. Informing borrowers, improves payment behavior.
Main Comment I already love the question, and it appears the authors have the basic ingredients to write a very good paper. –But there is a lot of confusion right now in the paper (hopefully no big worries hiding behind this confusion) There are a lot of tangential discussions, and chatting in the paper. And a theory which does not really seem essential. –For example, insurance discussion, claim that nothing is lost of sharing bad information on your clients, Hirshleifer effect etc. –I would suggest keep the initial question as sharp and simple as possible, then go directly into data and tests, carefully describing where identification of coefficients is coming from, and then finally some discussion of what the results mean, interpretation, welfare question etc.
On the other hand, the paper lacks some first- order information. For example, there is no data description section! Also a lack of careful description of identification assumptions. –This makes my job a bit difficult as a discussant. Some basic questions: –What is the frequency and time-frame of the full data that you have access to? –Do you also observe registry information? –What are typical loan contracts like in terms of interest rates, amount, maturity, roll-over propensity, renegotiation, etc.
Table 1: measure changes in the lending contracts observed on first loans which were issued before and after bureau. –Is this the identification strategy? I thought they were going to exploit the staggered nature of entry and essentially do a diff- in-diff? –I think table 1 is exploiting the staggered entry (because otherwise time dummy absorbs everything). However, none of this is explained in the paper. Why should we see an effect at the time of entry of bureau? Does entry mean all past information? Or does it mean they will start sharing from that point on? If it includes past information then how long is the history? What are the means of dependent variables? Try log size? What is ITE doing here? Why not difference-in-difference over time at the branch level? This will also address the weighting issue. This result should be given along side the loan level results. (and in later tables too)
Table 2: The result that loan volumes increase on existing borrowers is mechanical I think. We do not observe guys with 0 (-ve) loan demand, so conditioning on guys that continue to receive, it is not too surprising. In any event, the correct model should be a Tobit here. (again a careful discussion of where identification is coming from will help clarify the thoughts here) The authors try to rationalize the result of a negative impact of bureau on default by mean reversion. But if that were the case it is easily testable using placebo treatments. Why cant we observe bureau data and directly test the selection / de-selection process of the bank? Puzzle: Bureau can only help the MFI if the lender already had some other lending relationship. Given the magnitude of in and out screening, is it believable that all these guys were borrowing from multiple sources? Do you have that information? –When are all the other MFIs implementing this bureau and do they inform the borrowers? –Keeping track of above is critical since (a) bureau by definition only effect multiple guys initially, and (b) competition effects might even make you lose your best guy.
It is a bit surprising that borrowers do not know about credit registry. –Isnt that illegal? –Isnt it in the banks own interest to tell borrowers about it? Perhaps worth doing a welfare gain through moral hazard and adverse selection gains (will need a structural model …. Perhaps this is where theory can really help)