The Weakest Link: Wages and Inflation
A Far from Definitive Test in the Spirit of Chris Sims
RBA Governor Lowe’s speech to the AFR Business Summit this week was notable for the extent to which the RBA continues to hang its hat on the labour market and wages in the determination of the inflation outlook. This largely accounts for the slow-grind in returning inflation to the target range in the RBA’s forecasts.
The problem with this approach is that it situates inflation at the end of an implied causal chain running from monetary policy, to the unemployment rate, to wages and then inflation. This is fine as far as it goes, but it’s a problem if wages growth depends heavily on expected inflation outcomes and these in turn depend heavily on monetary policy. If wages growth is to be consistent with the inflation target, then so do inflation expectations. If inflation expectations are falling short, then monetary policy is not doing enough. This was part of the problem with pre-pandemic monetary policy. While the RBA has recalibrated its estimate of the NAIRU, raising the bar for monetary policy, it remains wedded to the same model of the inflation process.
It was not always so. Governors Macfarlane and Stevens were very dismissive of the NAIRU and maintained it played no role in monetary policy decision-making. The RBA’s inflation models have usually featured some relationship with wages. Historically, these were long-run equilibrium relationships of the type we should expect to see in the data, although these were sometimes only weakly accepted by the data. The NAIRU is more appropriately viewed as an equilibrium condition rather than a predictive model, but the way the RBA talks about these relationships can easily give the impression the latter is what they have in mind.
An alternative approach is the theoretically agnostic one due to Chris Sims (pictured above), which imposes minimal theory and instead allows the data to speak for itself. We can consider the relationship between inflation expectations, inflation, wages and the change in the unemployment rate in the context of a system in which all the variables depend on each other. We still need to impose some structure on these relationships in order to make them economically interpretable. In what follows, I simply order the data in terms of the assumed relative speed of adjustment, which is as I have listed the variables above. All this is saying is that inflation expectations adjust more quickly than inflation, which in turns adjusts more quickly than wages and then the unemployment rate. If you don’t like that ordering, substitute your own and see what you get.
We also need to specify the dynamics, for which I assume a two quarter lag, which is the most parsimonious specification that clears, if only weakly, tests for serial correlation while also being dynamically stable. Note that this only captures very short-run dynamics. We are not testing long-run relationships. We should also concede that different lag specifications can give very different results. I would not care to die in a ditch defending this lag order (or any other).
We can then test which variables predict which. Note that this is a very weak test of causality, in the sense that the same test would show that Christmas cards cause Christmas. But just as we have other information to tell us that Christmas cards do not cause Christmas, we have other information to suggest what relationships we should expect to see in these data.
The statistical tests are shown in the following table for those who know how to interpret them (the null being tested is Granger non-causality). Note that BEI is the bond market-implied break-even inflation rate, TM is the trimmed mean measure of inflation, which also determines the beginning of the sample period from Q1 2003, WCI is the wage cost index and UR is the unemployment rate. All variables except the BEI are first differenced.
The tests suggests that the breakeven inflation rate individually and jointly predicts the other variables in the system. The change in the unemployment rate predicts wages growth, as we would expect, which is the traditional Phillips curve relationship. However, we reject wages having predictive power for inflation.
To be clear, I do not consider these tests to be in any way definitive. I would argue these relationships from theory before leaning too heavily on what we can recover from the data in this framework, for which there may be problems of observational equivalence. We only narrowly reject a role for wages growth in predicting BEI rates, for example. And as already noted, we are not testing long-run relationships.
For me, what stands out is the role of inflation expectations as the most predictive variable in the system. Inflation expectations are the most forward-looking variable in an otherwise backward-looking system, so they are picking-up information not otherwise observable to the model, but that’s kind of the point. We get to choose some of the information we feed into the system via inflation expectations, not least about monetary policy.
This is why central bank credibility is so critical. It is worth recalling that the RBNZ once implemented an inflation targeting regime with no monetary policy operating instrument other than Don Brash’s mouth (albeit with the implicit backing of the quantity of settlement cash, but that was left mostly unchanged). While I’m not suggesting that as a model, if the RBA wants target-consistent rates of growth for wages and the price level, it needs to communicate in ways that condition expectations appropriately.
Governor Lowe also gave the business community a bit of a touch up on investment spending, suggesting that there were plenty of investment opportunities out there, implying that business are leaving some on the table. Hurdle rates of return got another outing. But we know that economic policy uncertainty raises the value of retaining rather than exercising the option to invest and there is a lot of policy uncertainty at the moment. That includes uncertainty about whether monetary policy will provide a predictable future flow of nominal income.
His speech somewhat strangely highlighted the role of small business in accounting for a much larger share of investment than its share of output. Of course, this only highlights the low capital productivity of small business. While there are good reasons for liking small business, its capital expenditure relative to output is not one of them. Everyone loves to hate on big business, but they are hard to beat for capital efficiency. So give big business some love too.
ICYMI
Reprising an earlier theme in this newsletter, Gauti Eggertsson and colleagues explain why the pandemic jump in household saving won’t be spent.
Think Dow 32,000 is impressive, try Dow one million. Kevin Hassett, they hardly knew ya!
Here are your bloody tweets:
Tiny point, but on your RBNZ observation probably fairer to say "with the explicit threat - rarely exercised - to adjust settlement cash. I used to use the analogy of a mugger with the baseball bat. The threat was quite explicit and real, but most people just gave up their wallets without being beaten.