Investment and Interest Rates II

In the post below, I discuss a very strong correlation in the data which surprised me — a high ratio non residential fixed capital investment to GDP is correlated with high nominal interest rates on corporate bonds.

I think the discussion in comments was very interesting and I have promised to pull back 2 comments (I didn’t promise to pull them back above the jump).

In a third post on a correlation coefficient, I will get to my current thoughts on what is going on. Here some general thoughts (after the jump)

First it is odd that so striking a correlation is not widely discussed. Partly, I think this is do to the leading role theory has in the academic macroeconomic literature — the correlation is neither implied by a prominent hypothesis nor does it falsify a prominent hypothesis. Partly, I think it is related to the prominent position hypothesis testing has in econometrics. “A atatistic” is read to mean “a test statistics” and statistics whose distribution is not implied by a null hypothesis aren’t considered. I’m about to go off on one of my usual tangents noting that, on the other hand, hypothesis testing is allowed no role, since models aren’t hypotheses and all models are false. So epirical work must be hypothesis testing and hypothesis testing is irrelelevant, so evidence and data are irrelevant. ùù

Whoah back on topic. the non polemical point is that in the post below, I show graphs and correlation coefficients, but no standard errors, t-statistics or confidence levels. The reason is that the correlation is what is called a spurious regression. It is a regression of one non stationary time series on another. The coefficient of the regression divided by the conventionally calcuated standard error does not have a t – distribution. The conventional test of the hypothesis that it is zero is completely invalid. I think the pulled back comments show that this isn’t a point about purely mathematical statistics — there are a large number of possible partial explanations of the pattern, many of which amount to saying it was a coincidence.

The point (if any) is that problems with my earlier super simple old fashioned approach to data help explain why macroeconomists don’t talk about such striking simple patterns in the data. Now I don’t think the conventional contemporary approach is OK. The problem with what I did is that there are too many candidate explanations. This doesn’t mean that a model which implies that such a pattern almost certainly won’t be observed are OK.

OK finally sort of keeping my promise

After a bit of back and forth (see the comment thread to the post below)

Marko
December 19, 2014 5:46 pm
Have you looked at the possibility of causation running from residential investment to gdp to nonresidential investment ?

It looks to me like housing starts generally leads gdp , then nonresidential investment follows with a considerable overall lag – a few years (relative to housing starts). Here’s one paper (2001) I found that looks at this :

ftp://ftp.aefweb.net/AefArticles/aef020208.pdf

“….What we have found instead is that capital formation in the residential sector (housing) causes GDP growth, which in turn causes capital formation in the business sector (plant and equipment). ”

The two plunges in real rates shown in DeLong’s graph around ’73 and ’79 correspond nicely to the rising housing starts.

,

December 19, 2014 9:06 pm
Yes , I can see how it may be that the Fed viewed vigorous nonresidential investment as a sign that it was time for “last call”. I also agree that era from Volcker to the GFC represents a significant break or regime change from the rest of the post-WWII period.

Here’s a couple of interesting reviews of Fed policy that describe the “credit rationing” that was in place through most of the earlier period , as well as the changing tolerance of inflation in the 60′s and 70′s. Uncle Milt put a stop to all that , helping to bring us to the happy space we occupy today :

http://www.hbs.edu/faculty/Publication%20Files/jep%202014%20-%20monetary%20policy%20at%20the%20fed_e5b57972-373c-41e6-a01d-cd1589a026ca.pdf

http://www.people.hbs.edu/jrotemberg/workcurr/fed4b.pdf

Jonny Bakho
December 20, 2014 10:21 am
Reverse causation? Maybe interest rates follow investment?

When greater investments are needed to meet demand across the economy, demand for investment capital increases and interest rates will rise. I would expect interest rates to follow investments and interest rates to be only a minor factor in the decision to invest.

At the micro-level, a business will invest in expansion when demand is expected to exceed capacity or when efficiency or regulatory considerations require investment.

In the 70s and early 80s large investments in fixed capital were needed to 1) meet the demands of the boomer market, 2) address the rapid rise of energy prices by investment in conservation, fuel switching and process changes 3) move production from the rust belt to the air conditioned, cheap-labor south and 4) meet stricter pollution control and environmental standards by changes in practices that required fixed investment. The changes in fixed investment between 1976 to 1985 are not easily explained by interest rates.

High interest rates are correlated with high inflation rates. This may make future investment more costly even if future interest rates are lower! If investment today is cheaper but at a higher interest rate, the project can be refinanced at a lower rate and repaid from inflated dollars. Note that recessions are correlated with drop in demand and are followed by drop in investment. For these reasons, interest rates may effect the financing of an investment, but have little influence on the decisions to invest and thus have only a minor effect on overall investment.

Investment follows demand. Interest rates are a problem to be solved.

If interest rates are an afterthought in the investment decisions, the interest rates should follow the investment, rather than investment following the interest rate? This would make sense in expanding micro considerations to the macro level and more easily explain the observed correlation.
-jonny bakho

In case anyone read this far, the reason the Bakho comment is relevant to this post is the list of explanations 1-5. They are all plausible. A statistic with so many plausible explanations isn’t very interesting (although I think worth more than the zero attention it has received in the literature).