Instrumental variables with S&P 500 Health Care Index

#1
Hello everyone,
I am new to this forum and I would need some help on a time series regression model with financial variables.

I am regressing Y=S&P500 Health Care index c X=bookmaker odds for the victory of Hillary Clinton x=S&P500 x=USD/EUR

everything is in dlog, and I use the lag of the odds as instrumental variable for the first X.

I was wondering about an instrument for the S&P500.. the lag does not work, so that I thought that MSCI world would be a nice idea, being the S&P500 Health Care components just 4.25% of the MSCI world index; but I fear reverse causality in this case too..could you please help me? is the MSCI world a good instrument?
 
#2
I work in econometrics so could give you a hand but I need a little more information.

You are using quite a bit of jargon that makes your post a little difficult to read. What is MSCI world?

And do you have a statistical question? Because this almost sounds like something that is suited to an expert in your field since it seems to require a bit of knowledge of the literature rather than a statistical expert.
 
#3
Hello,
thanks for the answer and sorry for my short explanation.

MSCI World is an equity index created by Morgan Stanley which comprehends the most important listed companies at a global level. The S&P 500 Health Care index accounts for 4.25% of the MSCI World and 13% of the S&P 500 general Index.

You are right, my question is probably more regarding the literature.. I am not an expert though on financial series and on internet I did not find anything similar to the problem I had.

But I have another question, probably more technical:
Yesterday night I probably found a solution.

my regression model is DLOG(S&P500 HEALTH CARE) c DLOG(HILLARY CLINTON ODDS) DLOG(USD/EUR) DLOG(S&P 500)

I am using as instrumental variable for DLOG(HILLARY CLINTON ODDS) its first lag, and for the S&P 500 I am using its lag (-1) as well... do you think that exogeneity conditions hold using the own lags of the variables?
 
#4
yes, the lag variable will most certainly fulfill the criteria. That's the foundation of dynamic causal models.

That is, if I am understanding your question correctly.