The topic of this post might be “procrastination.” I have written a bit about aggregate consumption in the USA and I keep telling myself to write something similar about investment. I never do. I am using this post as a pre-commitment device (sorry to use the blog for my personal problems).

OK on investment there are two approaches. One is the old Keynesian ad hoc reduced form accidentally theoretical approach of looking for correlations in the data and guessing about causation. This lead to the flexible accelerator, which just says that the ratio of investment to GDP is high if GDP growth is high and if (nominal interest rates minus lagged inflation) are low.

Another approach starts with a well defined optimization problem which contains two elements. First investment is desirable because it produces capital which is combined with labor to produce some product which is sold for money. Second there are some adjustment costs which keep the variance of aggregate investment huge as it is and not megagigantic as it would be in the simplest model. This also means that investment depends on the future marginal product of capital as well as the current marginal product of capital.

I think the first approach is vastly superior to the second which has added nothing of value to our understanding. I conclude this based on the following claims (which I promise to document some day — see procrastination)

1) the Flexible accelerator fits data collected decades after it was abandoned by academic macroeconomists sometimes in the late 70s and early 80s.

Here the one issue is that the BEA (the national income and product accountants) has decided that the relative price of investment has decreased markedly since then. This means one has to decide whether to look at real investment divided by real GDP or nominal divided by nominal. The old accelerator fits nominal/nominal fine. To get to (real I)/(real Y)you have to uh multiply real investment by roughly the ratio of the price indices (that is get it back to nominal/nominal). This can be done various ways.

2) note in the second “micro founded” approach I snuck in the assumption that investment is non residential fixed capital investment. Houses provide services to their residents, they aren’t combined with labor to produce products to sell. The advance in macroeconomics included ignoring the housing sector completely. This might make it difficult to make sense of recent events. Second capital is assumed to be fixed capital not inventories.

This is very important, because interest rates are much more highly correlated with residential than with non residential investment.

The approach of someone who hasn’t been exposed to the macroeconomic theory of the past 3 or 4 decaddes would be to go to FRED and look at the series and check if they look about the same (they don’t) and have the same relationships with other series (not at all). Then the unschooled amateure macroeconomist would presumably not assume that they are all approximately the same.

Even someone expert in modern macro should test the assumption that all investment can be lumped together. I think the radical rejection of the assumption would leave only two choices. Either one might decide that the scientific method is based on ignoring the data, or one might decide that current DSGE macro models should be scrapped and one should attempt to find any pieces which might be useful in constructing a quite different model. I think academic macro is based on making the first choice.

In contrast the ad hoc accidentially theoretical flexible accelerator can perfectly well be a bit more flexible with three separate equations for inventory investment, non residential fixed capital investment, and residential fixed capital investment.

Aside from that, micro founded models of aggregate investment imply that future profitability is correlated with current investment. For the sake of argument, I have considered total investment and non residential fixed capital investment. I see no such pattern.

I think that, if I ever write up my investment regressions, I will conclude that the effort to micro found macro added nothing and subtracted key insights from old 1960s era paleo Keynesian macro.