Monday, 27 June 2016

It's all in your head

The Reserve Bank came out with a new discussion paper last week, 'Inflation expectations and low inflation in New Zealand'. While Discussion Papers are "mainly for academic and professional economists" - and, I should add, represent the staffers' views and not the Reserve Bank's - this one is relatively easy going, and worth a look, because the authors (Özer Karagedikli and Dr John McDermott) have had a go at investigating one of the major puzzles in modern macroeconomics: why inflation has stayed unusually low.

There are other biggies - why has productivity growth (assuming we've measured it right) slowed down and what if anything can we do about it, and what can (or should) policymakers do if they run out of fiscal space and/or hit the zero lower bound for interest rates - but the unexpectedly low inflation puzzle is front and centre across the developed world.

I'd love to say they nailed it, but once I'd got my head around what they'd done, I wasn't totally convinced.

The heart of their argument runs like this. Inflation expectations affect actual inflation: if (for example) people expect low inflation, they'll settle for low wage rises, which will feed into low actual inflation. Fair enough. And then they say: what if inflation expectations don't just arrive out of the blue but are (partially or largely) influenced by actual inflation? Suppose actual inflation is, unexpectedly, only 1.5% instead of 2.0%. People revise their expectations down in line with the lower actual rate, and their lowered expectations (and the price and wage behaviours that follow) drive actual inflation lower again, to say 1.0%. Expectations are revised again ... you get the idea. Voilà - a self-fulfilling circle driving inflation down (or up, if the initial surprise had been higher than expected actual inflation) and one that has nothing to do with the strength of the economy or other things you might have expected to dominate what happened to inflation.

So they built a nice little model, which worked just as they argue. If you'd like to go through the details I've got a summary below, though you'll find the paper accessible enough in its own terms if you'd prefer to go to the source.

It is an interesting paper. But I was left with several questions at the end of the exercise.

The first is around how they model expectations. They say, people's expected rate of inflation will be some blend of (a) the latest actual outcome over some recent period, a backward looking measure, and (b) the expected inflation rate as measured in a survey, a forward looking measure. And they find that, modelled this way, not only do expectations have a strong influence on actual inflation but the weight that people apparently place on the latest actual inflation rate has increased markedly since 2008-09. So you have an explanation for the persistence of low inflation: a self-validating and strengthening feedback loop from low inflation to even lower expectations to even lower inflation again.

But this is all rather odd. The survey that asks about people's expected inflation rate is their inflation expectation, by definition. Subsequently saying that expectations are actually a mixture of those expectations and actual inflation is a bit of logical gymnastics I can't quite follow. But, as they say, "The empirical treatment of inflation expectations is crucial for the purpose of this paper", and if you're not convinced by their formula (and I'm not), some of the results fall over.

Even if you go along with their approach, though, you're still left with other questions.

One is: why? Why did people change in recent years from putting more weight on what they expect to happen, to putting more weight on what's actually happened? Have they suddenly stopped believing that they can get a handle on what lies around the corner - which wouldn't surprise me, in a post-GFC, post Brexit world? The researchers may well have unearthed an interesting mechanism or process, but we're still left with an unsolved, if different, problem.

Another question is: in the world they've modelled, what's happened to central bank credibility? If everybody believed their local central bank would indeed keep inflation around 2% (or wherever), then their expectations would stay around 2% irrespective of any wobbles in actual inflation along the way. Perhaps people in New Zealand (and the eurozone, and Japan) have indeed thrown in the towel and now prefer to believe the evidence of their own lying eyes rather than subscribe to what central bank governors say. If true, that's important, but again it raises a whole new research agenda to unpick the next layer of answers.

Bottom line? Because of the somewhat idiosyncratic modelling of expectations in this research, I wouldn't get too hung up on its exact outcomes. But I think it does make a good, wider point. Expectations have always mattered: that's seen most obviously in hyperinflations and deflations. But clearly they matter in more normal times, too, and they may not have got the policy attention from central banks that they should have.

That's changing. In the US, the Fed has been paying more attention to the financial markets' view on five year forward inflation, for example, and recent Monetary Policy Statements from the RBNZ have been zeroing in on expectations, too: they've included an 'inflation expectations curve'. So far so good: the big issue, though, is having realised that expectations matter, and possibly matter a lot, do central banks know how to manage expectations back towards levels more consistent with the banks' inflation targets?

The economics behind it

The authors start with the well-known Phillips Curve - an inverse relationship between inflation and some sort of measure of slack in the economy, often an unemployment rate - but not any old Phillips Curve. They've used a New Keynesian Phillips Curve, which adds in an extra factor, people's expectations of inflation. In this version, expectations have an independent life of their own in influencing inflation: if people, or firms, expect inflation to rise, they'll get their retaliation in first in their own wage and price setting, and inflation will rise even if the unemployment rate doesn't move. The authors also added in an extra term, to allow for the effect of import prices on overall inflation.

If you prefer symbols to words, here they are:

πt = βEtπt+1 + κyt + γΔpm,t + εt

where πis  inflation at time t; Etπt+1  is expected inflation in the following period, which has a weight of β; yis a measure of capacity utilisation (the coefficient κ will be negative if you use the traditional Phillips Curve unemployment rate and positive if you use an output gap); Δpm,t is the change in import prices, which has a weight of γ; and εis the usual stochastic error term.

The final wrinkle of importance is they look at that expected inflation term, and ask where does it come from? And they say that people will come to a view based partly on what they've recently experienced and partly on what they think will come next (as shown in consumer surveys, for example). So it will be something like this:

Etπt+1 = θπat + (1 - θ)πst

where the 'a' superscript means some measure of recent actual inflation and the 's' superscript means some survey measure of expected inflation. I haven't used their exact notation here, because if you can write π with a bar over it in Blogger, then you're a cleverer operator than I am.

And then they estimate the whole thing. It fits pretty well, with R2's around the 0.7 mark, and expectations do indeed play a big role. All good: then comes the interesting bit. They look at how the coefficients vary (or not) over time. And they find that θ, the weight on recent actual inflation when people come to take a view on what next, has risen substantially from 2008-09, as shown below.

And that's where I part company with the analysis. People form expectations with half a view on what's recently happened, and half a view on what might happen next? Sure. But once they've done that, then they've settled on a view. That's it. Expectations aren't a combination of that view and current inflation - that's already been factored in.


  1. It all sounds a little post hoc.

  2. "Inflation expectations affect actual inflation: if (for example) people expect low inflation, they'll settle for low wage rises, which will feed into low actual inflation."

    So cost push inflation? I thought we had gotten pass such odd notions.