Update: Based on some reader feedback, I realise that I need to fill out parts of my logic in more detail. To repeat what I wrote below, I am unsure how my thinking fits in with the existing academic literature in this area. I am looking at this from the point of view of building a model that fits observed data, and there's some purist theoretical issues about price indices that I am not concerned about. Merijn Knibbe left this reference, which overlaps some of my thinking. The more detailed version will probably have to follow this weekend, since Professor Wren-Lewis caused yet another internet discussion that I want to weigh in on...
As a disclaimer, I have no idea how my comments fit in with the existing academic literature. I had a very brief discussion with Professor Randall Wray at the Modern Monetary Theory Conference on this topic, and it was an issue that he was well aware of. I would guess that it would cause greater anguish among mainstream economists, as it suggests that their preferred policy of inflation targeting is theoretically incoherent.
The Aggregation ProblemI originally started pondering this topic when I ran across the writings of a non-economist who simultaneously argued that we ought to look at the empirical data and not start with theory, and meanwhile did time series analysis of aggregate time series such as the CPI. (Since these origins are not critically important to my argument, I will not try to track down references.) My view was that this was incoherent: the empirical data was the millions of price measurements that get boiled down to the CPI time series, and we need to use to theory to justify how the aggregation is done. If one wants to build a theory of macroeconomics in the absence of theory, you need to start with disaggregated data, which most researchers do not have access to.
For example, I have recently purchased some craft beer in Kansas City, and bulk kitty litter in the Greater Montréal area. Since I am a consumer, why not directly aggregate time series of the prices of these items to create a consumer price index? We have good theoretical reasons to argue that these items should not be directly aggregated (left as an exercise to the reader), and so we instinctively would not try to create such a price index. However, we are relying on theory to determine which items we can aggregate.
Not Completely AustrianI believe that this sounds somewhat similar to the Austrian economist critique of aggregation. (I have only seen the internet Austrian version of the argument, so I am unsure exactly how it is expressed.) The argument as I have seen it presented was that any aggregation in economics was not justified; we are adding unlike quantities together (such as apples and oranges).
I do not think that we can can take this argument too far; it is reasonable to aggregate similar items together. Defining similar is tricky; take the previously mentioned apples and oranges (which may be aggregated into "fresh fruits"). For a Canadian, oranges are imported, mainly from Florida, where the prices are susceptible to extreme weather. Meanwhile, apples are available domestically. We have reasons to suspect that apple and orange prices have different drivers. However, we have less reason to be concerned by processed cereal prices.
The belief that we cannot aggregate items at all is a curious argument, since we do it all the time. This is why we have a unit of account (that is, money) in the first place. Firms need to plan, and they are highly reliant on monetary aggregation in their financial plans. The issue is that we cannot lose sight of the underlying real constraints on activity. For example, if you are an owner of department stores in Montréal, having $1 million of clothes on display is not helpful if those clothes are parkas and it is July.
In other word, my complaint is not that the sub-indices of the CPI are theoretically meaningless, rather the act of aggregating them into a single number. We obviously can do the final aggregation calculation, but we should not expect that single aggregate to correspond to any useful theoretical concept.
Why Aggregation Breaks DownThe 1970s experience is suggestive of the problems. Why would we expect the price of crude oil to move in a coordinated fashion with the prices of domestically produced services? Yes, energy prices are embedded into the cost structure of almost everything, but at the same time, doubling the price of oil would only add a few percent to the cost of most items, so there is no reason to believe that they will move proportionally (which a unitary price level would suggest).
For example, Warren Mosler argues that the deregulation of natural gas prices created a massive demand shock for oil (as U.S. utilities switched from oil to natural gas), which destroyed OPEC's pricing power. He argues that this deregulation accounted for the breaking of the back of the 1970s inflation, and that Volcker's rate hikes were actually counter-productive. (Please note that I have not studied his arguments, and so I cannot say that I agree with him, but I find no reason to reject them on a theoretical basis. However, his argument would likely cause most mainstream economists to become apoplectic.)
More Than One Hidden VariableThe entire premise of macroeconomics is that we can relate low-level observations to a few aggregate variables. We can make definite statements about some aggregates on the basis of accounting logic, but in order to make useful forecasts and statements about policy, we need to make assumptions about the behavioural relationships between aggregates.
The usual assumption is that the calculated (observed) CPI is a proxy for a theoretical price level, and we can can then relate this singular theoretical price level to other hidden variables, such as the output gap. (Note that post-Keynesians generally object to the particular definitions of the output gap used by the mainstream, but there generally is the belief in a similar concept, which can be also viewed as a hidden variable that is to be estimated. I discuss this in Interest Rate Cycles: An Introduction.)
If we argue that there are at least two "price levels" in a theoretical model, we cannot cover both with a single time series proxy. Furthermore, the idea of a single "generalised output gap" that drives all of these price levels appears to break down.
The various outlandish assumptions that mainstream economists make in their "microfounded" models are there for one reason: to make it look respectable to pretend that there is only one good in the economy, with a single price level, which can then be controlled by manipulating a single price expectations series and a single output gap series. (Wages effectively disappear as a concept. The wage level is determined entirely by the price level and the marginal productivity of labour, and hence is a redundant variable.) Needless to say, this entire edifice collapses if we believe there is more than one price level in the economy.
(I am not arguing that I have a theory that can magically take the mainstream's place, and offer an easy way to forecast economic outcomes. Instead, my argument is that we should not expect economic forecasting (as it is currently understood) to be feasible, and we can demonstrate this by looking at the properties of theoretical models. All we can hope for is to extract general principles, and use them to guide policy.)
The rest of this article looks at various implications of these arguments.
WagesWages and salaries are a significant portion of the cost structure of domestically-produced goods and services. To what extent labour costs are correlated, we would expect them to act as a single unified variable that looks like a singular price level. As a result, it is not a mistake that the Fed is singularly interested in a tightening job market, even though their (loose) mandate is presumably to look at consumer price inflation.
However, the market for labour is regional and fragmented; there is little relationship between the bonuses paid to traders on Wall Street and the people who toil at minimum wage jobs in retail. Any realistic depiction of the labour market needs to take this into account. Even if we ignore the extreme top end of the income distribution, there is a big gap between the pay trends for workers in information technology and the minimum wage.
Nevertheless, if a tight labour market manages to buoy the wages across the employment spectrum, we would expect inflationary pressures. This was certainly the case in the 1970s, and during the mild cyclical upswing in the late 1990s.
IndexationThe widespread practice of indexing prices to some variable would tend to eliminate the divergences between multiple theoretical price levels. Such indexation could be tied to a domestic price index, or it could be through tying prices to those in a foreign (hard) currency (in which case the value of the domestic currency acts as the indexation variable).
In such an environment, we would expect that it would be much easier to move the entire price structure in one direction or another. Hence, much more susceptible to widespread inflation (as traditionally defined). Correspondingly, the decline of indexation (via union bargaining) may be the main explaining factor for the stability of observed CPI inflation, and not the switch to inflation-targeting.
ForecastingInflation-linked bond market participants have an extremely large vested interest in getting inflation forecasts right. Based on my observations, the usual practice is to forecast each component of the CPI separately, and then aggregate those forecasts to get the CPI forecast. Mainstream academic economists would presumably sneer at such a methodology, since they do not glorify the role of expectations in the steering of "the" price level. However, it is exactly the right procedure to follow if there is no unified price level as a theoretical concept.
PolicyInflation control is the key goal of mainstream economics, a common view regardless of their political leanings. Even Modern Monetary Theory argues that one of its advantages is that it allows for control of inflation.
However, if there is no single theoretical price level, there is no single inflation series. What exactly is the policy objective, and why?
I am certainly not a spokesman for Modern Monetary Theory. However, I would phrase the objective as follows (which I think is compatible with existing formulations). The role of money in the economy is to allow the government to provision itself. (The alternative is to seize what it needs, such as conscripting soldiers or seizing goods and services.) The government should aim to keep the prices it pays for wages, goods, and services stable, and have those prices act as a pricing anchor for other domestic prices. As a result, the hope is that citizens would face stable prices during their day-to-day lives, although this cannot be guaranteed. (In this case, stable prices could either be stable levels, or a stable inflation rate.)
The Job Guarantee wage would be a critical price anchor. It would act as the effective minimum wage in the economy, and one would normally expect that at least 2% of the labour force would be employed at that wage. This would keep private sector wages for unskilled labour near that wage, although it may require the automatic stabilisers to induce a recession if private sector wages were outstripping that level.
Concluding RemarksWe need to be very careful when attempting to infer what is happening in the economy by just looking at the aggregate CPI.
(c) Brian Romanchuk 2017