The Phillips Curve: Timelessly Misleading


Tom Palley has written a blog post politely requesting that Paul Krugman might give a bit of recognition to non-mainstream contributors to economics. It would be nice to see this happen but I doubt that it will (although Palley is getting a bit of blog play out of it which is nice). Anyway, I note that in the post he links to a discussion him and Krugman had regarding the Phillips Curve. Before I get to the arguments put forward here let us examine the Phillips Curve in some detail.

The curve was an empirical relationship that the economist William Phillips found in 1958. Phillips noted that movements in wages and unemployment were negatively correlated in the UK between 1913 and 1948. The neo-Keynesians then picked up on this and argued that there would be a negative relationship between unemployment and inflation. They figured that wages were the key driver of inflation and that, since unemployment was the key driver of wages, then it must be the level of unemployment that must drive inflation.

It was quite a heroic leap by the profession to formalise a General Law from a total of 35 observations, but formalise they did. After all, it was a neat theory and economics at this time was trying to emulate the hard sciences. Well, the Phillips Curve didn’t work so well over the next half century. Here is a scatterplot diagram with a fitted curve showing the relationship between unemployment and wages in the USA from 1947-2014 (all data from FRED).


Now, if we saw a negative correlation — that is, if inflation rose as unemployment fell — we would expect to see a strong downward-sloping relationship. Actually the relationship is slightly upward-sloping meaning that inflation more so increased when unemployment increased. We also see a very low R-squared which can also be seen by how far off the line the various datapoints are. In English: there is no firm relationship here and the extremely weak relationship we do find runs in the opposite direction to what the Phillips Curve would predict.

So much for Phillips Curve theory. But what about the relationship between unemployment and wages? After all, this is what Phillips himself tried to show. It was the neo-Keynesians that came after him that tried to generalise based on his observations.


Well, here we at least see a somewhat negative relationship, albeit extremely weak. But we see an even lower R-squared. Basically it seems that the relationship between wages and unemployment in this period is pretty chaotic. In English: again, this is bunk; the sought after relationship does not exist.

Is this all hopeless then? Can we say nothing about inflation at all? No, all is not lost. There is one relationship that does hold firm: namely, that between wages and inflation.


Ah, there we go! There is a nice strong relationship between wages and inflation. When wages rise, inflation tends to rise. This, however, does raise the question of causality. After all, it could be that rising inflation leads to rising wages. Or it could mean the other way around. Frankly, we cannot say.

Indeed, I would be very hesitant to make any generalisations here. Sometimes inflations are wage-led. But sometimes wages rise in response to other sources of inflation. Every inflation must be studied in its particularity. Trying to generalise abstract laws that mimic laws in engineering or hard sciences is just stupid and will just lead up a blind alley.

This brings me back to Palley’s post. He notes that the mainstream have basically tried to solve this problem by incorporating expectations. If workers expect inflation they will bid up wages and so on. This is obvious nonsense and will not lead anywhere at all. If, for example, workers were heavily unionised and radicals got control over the unions we could easily imagine them just bidding up wages no matter what they thought inflation might be. They would just be doing so in order to get more of the economic pie. It is not like this has never happened before in history (*cough*, Allende in Chile, *cough*).

Palley is more realistic in this regard when he writes:

In my view, the real issue is the extent to which inflation expectations are incorporated into wage behavior. Workers may have absolutely correct expectations of inflation but not incorporate them into nominal wage demands because of job fears. (My Emphasis)

Here Palley hints at the fact that there is an aspect of social power to the bargaining process. Hence, if workers fear getting fired they might not try to bid up their wages. But once you open this box it is pretty hard to close. The power dynamics of worker bargaining are enormously complex and require historical and institutional nuance. Palley appears to instead fall back on the idea that bargaining power depends on unemployment — that is, that workers will bid up wages when there is low unemployment. But, as we have seen above, this position does not have empirical support.

In reality power dynamics are complicated and need to be studied in and of themselves. There are also many different types of inflation, as John Harvey has noted elsewhere in an excellent piece. In my forthcoming book I note four key types of inflation. These are as follows:

  1. Demand-pull inflation.
  2. Cost-push inflation.
  3. Speculative inflation.
  4. Exchange-rate inflation.

In laying out a Phillips Curve, or any other contraption, we are likely to only capture one type of inflation. Indeed, in the case of the Phillips Curve we only capture one sub-type of one type of inflation — namely, a sub-type of cost-push inflation (wage inflation). This does not lead to cogent thought. Given that Palley is in a dialogue with Krugman it might be worth pointing out that the latter has extremely crude views on real-world inflation which I have noted before. In subordinating one’s thoughts to a contraption one always runs the risk of blocking out the real-world.

Why is it so controversial to make this point? Why do we have to try to form a single theory of inflation? Why is it that a theory is only valid when it has been laid down in black and white form in a model? After WWII the profession tried to formulate a General Law of inflation. It was a disaster. Why on earth would we want to try to do this again? If economists — especially Keynesians — have not yet grasped that history will melt their attempts at timeless construction, they learned nothing from the lessons of the 1970s.

But that comes back again and again to that old timey question: why does every aspect of economic theory need to be cast in terms of a model? Sometimes modelling clarifies thought, but sometimes it obfuscates it. Inflation is such a complex, historical phenomenon it is definitely one that is obfuscated by modelling. So, why the need for models? Frankly, I think it is on those that build them to justify this, not on me to engage in conjecture in criticising it.


About pilkingtonphil

Philip Pilkington is a macroeconomist and investment professional. Writing about all things macro and investment. Views my own.You can follow him on Twitter at @philippilk.
This entry was posted in Economic History, Economic Theory. Bookmark the permalink.

28 Responses to The Phillips Curve: Timelessly Misleading

  1. Auburn Parks says:

    Great piece Phil.

    “Trying to generalise abstract laws that mimic laws in engineering or hard sciences is just stupid and will just lead up a blind alley.”

    you mean ridiculous notions like…. ‘Central Banks control inflation through minor interest rate adjustments’.

  2. JS says:

    I agree with most of your discussion, particularly regarding the futility of searching for timeless, precise quantitative relationships. Perhaps I am a bit more optimistic about finding qualitative regularities though. We do observe PC relationships in basically every sub period (there are a few graphs which break your 1st graph into different sub periods by color and you can see it), at least once controlling for expectations, but it shifts around due to the type of institutional and other exogenous changes you mention (worker bargaining power and exchange rates for example, which would be nice to control for).

    Curious to hear your thoughts about this work for example which got some attention not too long ago

    • Yes, you can break down sub-periods and generate Phillips Curves. But then I can go and pull data from, maybe, 20 countries around the world and generate no Phillips Curve. We can go round and round on this if we would like — ultimately we will get the result: Phillips Curve phenomena appear SOMETIMES. Or, put differently (as I did above): the Phillips Curve captures ONE type of inflation and misses all the others. Why on earth do we want to narrow our vision in this regard in search of Timeless Truths? Why do we need an Idol of Inflation? Why not accept inflation for what it is?

      • JS says:

        Well, I think the point of some of the empirical works with expectations augmented PCs is basically concluding that you do get an at least qualitatively consistent relationship, I’m not saying time invariant, once accurately controlling for inflation expectations (which can control for some other influences on inflation besides the labor market such as exchange rates, and cost push). If you could find 20 countries where the PC doesn’t hold after incorporating inflation expectations and testing for some parameter instability that would be quite interesting (and of course nailing done how expectations are formed is very difficult, if not even more subject to change over time). How do we know what inflation is though if we don’t investigate, for example to see how stable/unstable expectations augmented PCs are? I don’t think anyone is saying the unemployment and expectations are the only things that impact inflation, just that they are quite important, and that it is worthwhile to examine the historical connections (again without assuming they are not subject to some change) and if you can supplement this by controlling for more factors (unionization, exchange rates etc.) that is interesting too, in addition to more say historical analyses that you are promoting. This paper is a good one in my opinion related to the PC also.

      • Nah, not interested in the search for Timeless Truths. I’ve long grown weary of the game that economists play of “show me your model and I’ll show you mine”. This is not a pursuit of academic truth. Its a pissing contest, frankly, and its not a good way to train economic practitioners. Let the academics see who can reach the highest point on the wall. The rest of us have a real world to deal with.

  3. Marcelo. says:

    Phil, actually the original Phillips curve measured a long-term relationship: 1861-1957. Almost a hundred-year span.
    It probably has to do with the exchange rate regime as well.
    Best regards,


    • You’re quite right. It’s been a while since I read that paper. It’s actually very good. Phillips clearly breaks down the cost-push and demand-pull aspects of the inflation. Anyway, it doesn’t make that much of a difference to the thesis put forward above.

  4. circuit says:

    Nice post! Using a chart similar to your first one, Arthur Okun described the Phillips relationships as an “unidentified flying object”. In the 1970s, Okun and other Keynesians were able to refine the idea by taking into account the lag between unemployment rates and inflation, and by acknowledging shifts on the supply side, such as the food and energy shocks in the 1970s and productivity increases in the 1990s. Of course, the obvious problem is knowing how to exploit the PC under conditions of uncertainty, which is pretty much always, as Keynes taught us and Skidelsky and Davidson constantly remind us.

  5. circuit says:

    Phil, I just read your comment at 2:23. I totally agree with you that the search for timeless truth is futile and that the focus ought to be toward the practical use of ideas. That said, I’m curious to know more about how policymakers can achieve their goal without relying on a model. I only started your blog a few weeks ago but it seems to me that even you have a model, in that you have an idea of the way the economy functions and how policy affects it.

    I would say the problem is not the use of models, but rather the myopic avoidance to apply the model in an open, flexible way, sometimes recognizing that policy requires operatives to cautiously experiment and revisit their beliefs to ensure their model continues to correspond with reality and retains relevance for properly affecting change. Is that sort of what your are suggesting?

  6. Dantey says:

    Hey Philip!

    Dunno why but your ‘timeless truth’ comment reminds me of the many times when men have tried to figure out women.
    Actually, how do economists deal with women,especially their ‘non-linear,dynamical behaviour’? 😛

  7. hassran says:


    Sorry to appear nit-picky, but could you please provide the labels for all the axes in the graphs? For someone like me who is just an interested lay-person they are not as self-evident as they may be to a more proficient reader.

    Thank you and sorry to not have added anything worthwhile to this discussion.

  8. I have my doubts about the feasibility of building “the” inflation model. But when discussing inflation, we are discussing a numerical quantity, and it makes a big qualitative difference to behaviour whether inflation is 1% or 5% over a span of several years.

    Without some sort of quantities entering your thinking – which implicitly determines a mathematical model of some sort – how do you judge whether inflation will be 1% or 5%?

    And rightly or wrongly, central banks like the Bank of Canada have been given the task of keeping inflation near a specific target. They thus have a legal mandate to understand why inflation is where it is, and their needs are probably driving the academic research agenda.

    • First of all, we can still discuss a given inflation despite not having some sort of model. We just search out which variables are causing the inflation and why. See here, for example.

      Secondly, I don’t understand this:

      Without some sort of quantities entering your thinking – which implicitly determines a mathematical model of some sort – how do you judge whether inflation will be 1% or 5%?

      I can think about quantities without having a “model”. I can buy five apples and two bananas without having a “model” of my purchasing.

      • Why do you buy 5 bananas and not 500? Possible reasons I can think of:

        1) you find that you normally eat around 5 bananas each week, and you go shopping roughly once a week. This is some form of adaptive expectations based on your historical consumption. That can be specified mathematically.

        2) you would like to buy 500 bananas for some reason (you want to throw a big banana split party), but you can only afford 5. This then becomes some for of utility optimisation with a hard budget constraint.

        I will have to read your link later, but without having at least a order of magnitude approximation of the sensitivity, how can you discuss whether a variable influences inflation? It matters a lot whether a given increase of wages (for example) raises CPI inflation by 0.5%, or 15%. As soon as you commit yourself to those sensitivities, you end up with something that starts putting bounds on inflation, which is an inflation model.

      • I still do not require a model of banana-buying to figure out how many bananas I want. Sorry, I just don’t.

      • I now read your article about the 1970s inflation. The discussion implies a number of (partial) models. For example, your graph showing the relationship between unit labour costs and inflation could be interpreted as a regression relationship. I believe that the visual representation is better for interpretation, and so I would not bother determining regression coefficients.

        If you did not look at the data, and test whether it behaved the way simple models operated, there was no way for you to say that Monetarists or Austrians were wrong about the role of money in the determination of inflation.

      • I used no models in this analysis. I chose variables and studied their interaction within an intuitive framework. Just as I would when I bought bananas or whatever.

      • theDukeofURL says:

        Phil, don’t forget what Keynes said about those who claimed that their thinking was theory independent. In having said that, I would urge you to alter your use of “model” and replace it with “theory”. I realize that you are following the discipline standard, but this feature of the discipline standard, as for a number of others, can cause problems under certain circumstances. For a good discussion of this: Patrick Suppes, “A Comparison of the Meaning and Uses of Models in Mathematics and the Empirical Sciences”. In Freudenthal, ed., The Concept and the Role of the Model in Mathematics and Natural and Social Sciences (1961). In short, I would argue that it is not the best practice, while noting that it is the common practice.

      • I’m happy to let things lie in this regard and just draw a distinction between models and what I call ‘schemas’.

  9. -_-_-_-_-8 says:

    Are the axises labeled correctly?

    I think your graph should be labeled % change in CPI and % change in unit labor costs. Also, the axises should be each labeled telling which is which.

    I came here because I have been looking at the same relationships.

  10. -_-_-_-_-8 says:

    You gave this thing much more clarity.

Leave a Reply

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s