A misery index...with a memory

If we're going to track vibes, we need better metrics.

It’s hard enough tracking the many variables that comprise an advanced economy. As Chair of President Biden’s Council of Economic Advisers, the staff and I spent a lot of time on that. But we also tried to track how people feel about the economy, which took us into the uncharted territory known as “vibes.” We even had a vibrarian.

A prominent past effort to understand economic vibes led to the creation of the “misery index” (MI). Invented by Authur Okun in the 1970s, the index is a simple, sensible construct: the sum of unemployment and inflation (see blue line in the figure below; I’ll explain the other line in a moment). The idea is that if both parts of the MI are high (aka, “stagflation”), people are legitimately miserable, facing rising prices with job (and therefore, income) losses, and vice versa. The MI was the vibes indicator in the 1970s-80s, peaking under President Carter above 20% in the latter half of 1980, at the same time he was unsuccessfully running against President Reagan.

Since we have measures of economic sentiment, we can correlate them with the MI. You’d expect a high MI to be associated with low sentiment. The University of Michigan’s sentiment measure is available on a monthly basis starting in 1978, and in its first decade, it was, at -0.87, tightly and negatively correlated with the MI.

But if you eyeball the end of the MI series, you’ll note that when Joe Biden and then Kamala Harris were running for reelection, the MI was historically quite low, yet the economy was a big negative for them. When voters went to the polls in November, the MI was 6.9% (2.7% inflation plus 4.2% unemployment), well below the MI’s long-term mean/median of 9.3%/8.4%.

And yet, sentiment measures and the general vibes reporting suggested a lot more misery than that. In fact, the correlation between consumer sentiment and the MI from 2020-24 was a measly -0.37.

There are at least two reasons for the breakdown, both of which relate to measurement problems. Neither the MI nor the sentiment indicator mean quite what they used to.

The MI’s problem is that, unlike people, it doesn’t have much of a memory. When we replace year-over-year inflation with four-year inflation we capture more of what was, and is, unsettling people about the economy. It’s not so much inflation, as in the rate at which prices are changing over a year. It’s the level of prices compared to what they used to be before they spiked up. What’s needed is a misery index with a memory, an MIM, which I construct here using the 4-year change in the price level.

As you see, that line slopes up sharply towards the end of the series, and was well above its mean on election day. It’s correlation with sentiment over the past 4-years is -0.54 (versus -0.37 using the MI).[1]

Meanwhile, the UMich sentiment index has its own set of measurement problems. It is increasingly telling us less about how people feel about the economy and more about their political preferences (see Cummings, Harris, Mahoney for a very compelling deep dive). The best evidence for this is the extent to which it flips in a matter of days when political control changes hands, with virtually no bearing on actual economic conditions.

For those of us trying to track both the economy and how people feel about the economy, these measurement issues beg many questions, some of which are downright Talmudic. What is relationship between hard data and soft vibes, and what roles are mainstream and social media playing in this (d)evolution? Social media is particularly problematic, as it spews not just incessant negativity about the economy, but viral falsehoods. Have economic sentiment or confidence measures really lost their explanatory power such that they’re now just telling us which side is winning?

The MIM is a simple example of tweaking an existing measure to better capture what folks are feeling. Policymakers need a much deeper dashboard of such measures. The vibrary is open, but there’s not much material in the stacks…yet.

---

[1] By construction, the MIM upweights inflation relative to unemployment. That’s useful in a period like the current one, but less so in periods of weak labor markets. What’s needed imho is a statistical method to develop time-varying weights for an MI, based perhaps on dynamic factor analysis using both hard data and sentiment indices. Dissertation chapter, anyone?…anyone?…Bueller??