Interest Rates and Privater, Non-Residential Fixed nvestment

In a very interesting post, Robert Waldman writes:

In a recent post, I noted that actual non-residential fixed capital investment doesn’t show the pattern one would expect based on optimizing models at all. Ugh that sentence was convoluted and so is the post it describes. In fact, the puzzling pattern is really very simple. Non residential fixed capital investment (nrfinv) is high when interest rates are high. 

 And

This correlation is strange for two reasons. First the sign is surprising. Other things equal, one would expect high interest rates to cause low investment. note the brick red curve in the graph is the Federal Funds rate — a policy instrument. Second the interest rates are nominal [but the investment data are real---inflation-adjusted: DC]. Sooner or later, I will try to understand what was going on.

This rather intrigued me, so using data from FRED,  I put together a data set including the following data:
Real Private Non-Residential Fixed Investment (PNRFI)
Real GDP (and the average annual rate of growth of RGDP over the subsequent 2 years) (RGDP and GrRGDP)
Real Moody's AAA corporate bond rate (RAAARate)
(Details of how I calculated all this if anyone is interested enough to ask)

This gave me 255 usable quarterly observations.  I regressed RPNRFI on GrRGDP and RAAARate.  The resulting regression is shown in the following table:

Variable	Coefficient	T-statistic
Constant GrRGDP RAAARate	+847.4 -69.6 +48.7	+10.88 -4.09 +5.41
R2 F D-W	0.422 23.35 0.018

Even adjusting nominal interest rates to real interest rates, I find, essentially the same thing Waldmann found.  (The D-W--Durbin-Watson--statistic indicate that there is serial correlation in RPNRFI.)  The simplest explanation for this is that the changes in the real interest rate are a result of shifts in the demand for funds to purchase new capital equipment.  The argument that high real interest rates will be associated with reduced investment assumes that the demand for funds curve is unchanged, and that changes in interest rates move the economy along that curve. 

This is, in a sense, a very Keynesian result--that changes in investment are a result of changes in expectations of future profits, much more than simply being a response to rising or falling interest rates.
*************************************************
Addendum.  I played with some alternative specifications.  In every case, the D-W statistic continued to show autocorrelation.  In most of the alternatives, the coefficient on the GDP variable was positive.  And the coefficient on the real interest rate was always positive.  Here's one example.  In this instance, I computed the percentage change in PRNRFI from 8 quarters prior, the percentage change in RGDP from 8 quarters prior, and the actual change in the RAAARate from 8 quarters prior.

Variable	Coefficient	T-statistic
Constant %ChRGDP ChRAAARate	-3.64 +1.84 +0.22	-4.03 +15.78 +1.63
R2 F DW	0.497 124.5 0.209

Note:  I was unable to figure out how to post this in the comments on Waldmann's blog.