06-04-2016 09:40 PM
I am running a time-series predictive regression, where left hand side is monthly returns and right hand side are some economic variables. According to Hodrick, R. (1992), I should not use the traditional standard error, but instead use the estimator of the standard errors that imposes the null hypothesis of no serial correlation in returns but does not impose an assumption of conditional homoskedasticity. i.e. Hodrick 1992 (IB)
How can I use SAS 9.4 to compute Hodrick 1992 (IB) standard error and two-sides p value?
Thanks very much.
Hodrick, R. (1992), "Dividend Yields and Expected Stock Returns: Alternative Procedures for Inference and Measurement", Review of Financial Studies, vol 5, 357-386.
06-07-2016 11:14 AM
reading the paper I believe we don't have a direct implementation of this method in SAS. You can, however, use the AUTOREG or REG procedures to run OLS, get the residuals with the OUTPUT statement and calculate standard errors (equation 8) in IML. Since the assymptotic distribution of the parameter estimates follows the normal distribution, large sample p-values for the two sided test are also easy to obtain in IML. The paper also suggests that the small sample empirical distribution is different from the assymptotic distribution and therefore if the sample is small the p-values might need to be obtain from a monte carlo simulation.
Thank you for posting,
06-16-2016 11:40 AM
There are many examples of using simulation in SAS/IML on my blog. If you are new to IML, you might want to look at Ten Tips for Learning the SAS/IML Language. If simulation is something important to you (dissertation or job requirement), you might want to invest in the book Simulating Data with SAS, which includes both DATA step and SAS/IML examples.