Nick Rowe has raised an awkward point in the philosophy of economics. He writes:

If you want to increase the natural rate of interest, starting now, for the next 3 years [as fiscal expansionst want to], it does not matter what you do to the level of government spending right now. The only thing that matters is that you get people to believe that the growth rate of government spending will be lower than they would otherwise expect it would be.

This short paragraph appeals to not one but two unobservable variables: the natural rate of interest and expectations of government spending. Which poses the question: what should we make of a theory which rests so heavily upon unobservables*?

The issue here isn't, of course, confined to Nick. I'd argue that concepts such as the output gap or NRU, and therefore structural budget deficits, are also unobservable.

There's a long tradition, including positivists and operationalists, which says that an appeal to non-observables is just mumbo jumbo. I'm not sure I'd go so far. If we throw out all non-observables, then we must ditch talk of causality and mechanisms. And sometimes a reliance upon non-observables is justified - for example physicists' belief, until recently, in the Higgs particle.

Even so, I'm unhappy with such talk. First, insofar as expectations can be observed - as in the case of inflation expectations - empirical evidence suggests that governments cannot easily manipulate them because people form them on the basis of rules of thumb based upon current experience rather than in response to policy announcements. Partly for this reason, such expectations don't affect consumer spending much.

Secondly, the threshold for basing action upon unobservables should be high, because the costs of such policies can be great. For example, if you believe the official history of the 1979 Thatcher government, it hoped that announcing monetary targets would reduce inflation expecations and so bring down inflation easily. In reality, though, this didn't happen and we got a massive recession.

Osborne did not learn from that experience. Just as the Thatcher government destroyed jobs in pursuit of an unobservable - credibility - so he is wrecking lives in pursuit of another unobservable, the structural budget balance.

There is, though, an alternative here. We should use low-level empiricism rather than high theory. Let me take a case not from macroeconomics but from finance. Strictly speaking, the efficient market hypothesis is untestable because such tests rely upon unobservables such as investors' risk preferences, which leaves us with the Duhem-Quine problem. However, for practical purposes we don't need to bother with such high-falutin issues. We know - often from bitter experience - that most people who try to beat the market fail (pdf). This doesn't tell us that the EMH is true, but it does mean we should act as if it were. It's a useful guide to action.

I'd apply a similar principle to fiscal policy. Sure, high theory might give us reasons to doubt fiscalism. But low-level facts such as the UK's slowdown in 2011-12 and cross-country evidence suggest the fiscalists have a point. Even if they don't have a cast-iron theoretical case, their remedies do less human damage that a reliance upon unobservables.

Herein, though, lies a paradox. Back in the cold war era, mainstream western politicians and their hangers-on accused Marxists of sacrificing lives to an unempirical theory. Today, it is we Marxists who are making that accusation at the mainstream.

* Actually, there's another problem with Nick's paragraph;consumption Euler equations, on which Nick's claim depend, don't fit the facts.