[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

st: Econometric/estimation issues

From   kokootchke <>
To   statalist <>
Subject   st: Econometric/estimation issues
Date   Tue, 14 Jul 2009 22:28:07 -0400

Hello!! I would like to get some advice regarding my econometric model and how I can improve it. I have tried to think about all the major problems with my model and I've done my best to address these issues, but I'd like to get your advice, if possible and perhaps comments on how I can improve my model...

I have a big data set containing information on bonds issued by both sovereign and corporate entities in developing countries at different points in time. I am basically interested in investigating how bond spreads' dependence on both domestic (i.e., related to the issuing country) vs. global factors has changed over time.

Note: the spread of a bond is defined as the difference between the interest rate promised by
the bond issuer and the rate on a risk-free asset of similar maturity
and issued in the same currency.

I am running the following reduced-form model for two separate periods ("before" and "after"):

s = M'A + X'B + Z'C + e


* s is the (log) spread of a bond over a risk-free asset, 

* M is a set of issue- and issuer-specific characteristics (like the currency of the bond, the total dollar amount of the bond issuance, the term length or maturity of the bond, etc.), 

* X is a set of domestic factors (like GDP growth, total external debt/GDP, reserves/imports, etc.) and 

* Z is a set of global factors (variables that proxy for liquidity and credit conditions abroad, like US interest rate, volatility indices, TED spreads, etc.)

Now, the most severe estimation issues I think I may face are: 

* Endogeneity: spreads may affect domestic conditions; e.g., return on bond will depend on domestic interest rate which will affect GDP growth, debt ratios, etc. -- I address this problem via instrumental variables.

* Selection: we only observe the spread when firms and governments actually issue a bond, otherwise we don't observe it. Use Heckman model to address this problem.

* Autocorrelation and heteroskedasticity in errors: use robust variances.

* Sovereign and corporate issues may be different: Sovereign dummies? SUR estimation?

What other problems are evident from my specification? Would you recommend some particular model/technique that I haven't talked about?

For example, I thought I could use latent factor models to explore the effects of domestic vs. global factors on spreads -- do you think this would be relevant? I've never used these models and I don't know where to begin... so if you know of any references, I'd appreciate the pointers...

Finally, if I find significant differences in the coefficients in the "before" and "after" periods, is there a way to identify when the "structural break" takes place?

Thank you very much in advance!


Insert movie times and more without leaving Hotmail®.
*   For searches and help try:

© Copyright 1996–2017 StataCorp LLC   |   Terms of use   |   Privacy   |   Contact us   |   What's new   |   Site index