Karl Smith at the Modeled Behavior blog has responded to my recent post on the problems in using some kinds of economic theory to guide policy interventions. His reaction is critical, to put it mildly. It also seems to me to be an almost perfect illustration of the attitude that I was trying to describe.
Smith begins by saying that I have some “wrong ideas” about economics, and then quotes this portion of my post:
In practice, the problem of excessive abstraction by economic theory that Brooks identifies becomes increasingly severe as we try to evaluate the effects of proposed interventions and programs over years and decades, rather than months and quarters.
He then immediately says this about it:
First, this is backwards. With cyclical policy its generally speaking easier to access the effect of interventions over longer and longer horizons because the economy increasingly resembles a frictionless market as you extend time in to the future.
For example, if you asked what effect would properly done fiscal stimulus today have on the economy 20 years from now, the fairly easy and straight forward answer is, none.
Stimulus is not central planning or industrial policy. It should have no lasting effects. If it does, then you did it wrong. What’s more difficult is the short run.
Unless he is using “properly done” as tautologically implying his conclusion, then I’m not so sure he’s right about this.
A stimulus action is designed to change behavior in the real world – some investment decisions must change. Imagine, as an illustrative example, that a stimulus program in country X executed in 1820 resulted in the digging of a large number of canals that otherwise would not have been constructed. This makes incremental investment in shipping capacity and development of improved shipping technology more attractive over the next several decades than it would otherwise have been, as compared to the development of land-based transport systems. X then develops a greater reliance on ship than rail transport. This leads to some gains beyond what was unanticipated in 1820. Over the succeeding decades, X becomes a world leader in the shipping industry; by 1850, total output in country X is higher than it would have been had no canals been dug.
Countries that had not built early canal networks invest relatively more building out rail networks, just as X would have done had it not built all those canals in 1820. When, hypothetically, a set of technical innovations occur, it becomes suddenly apparent in the second half of the 19th century that rail is now a superior form of transport. Country X finds itself behind, and by 1880, total economic output is significantly below where it otherwise would have been. Further, increased competition with a neighboring country over shipping lanes leads to escalating tensions, finally resulting in a full-scale war in 1890 that would not have occurred had X’s shipping industry been smaller, which further reduces total output.
Of course, the exact opposite situation could have obtained, in which the decisions changed by the stimulus action turned out to position the society unexpectedly well for future developments through the end of the 19th century. It’s also possible that, as per Smith’s assertion, that building all those canals would have turned out to have had no material effect after decades.
But how could any of this be predictable in 1820 when making the stimulus decision? For that matter, how could we even know after the fact in 1890 what the effect of this decision was on total output, since this would require that we know what the counterfactual path of development would have been in the absence of the stimulus?
Changing, for example, the rate of economic activity today will lead to changes in actual investment decisions taken today in the context of the opportunities that are perceived to be available today. This will to some extent, and potentially to a material extent, change the array of choices available to us in the future. Professor Smith’s confident assertions notwithstanding, this change in the future option set may or may not change total future output materially. I believe the fancy name for all of this is path dependence.