PARSING THE RELATIONSHIP BETWEEN DEMOGRAPHICS AND INFLATION

The following information was released by the Federal Reserve Bank of Atlanta:

Nikolay Gospodinov Research Economist and Senior Adviser

Policy Hub: Macroblog

Publication

Macroeconomy

It is widely recognized that demographics can have broad and consequential implications for factors including the neutral rate of interest, potential output growth, and transmission of monetary policy (see, for example, here, here, and here) by affecting households' decisions about consumption and savings over their life cycle (see here and here). As the size and the composition of the age distribution change slowly, demographic projections are readily available (see here) and can be used to elicit valuable information about the low-frequency dynamics of key latent variables that guide monetary policy as well as asset valuations (see here and here) and inflation (see here).

This Macroblog post highlights some recent evidence on demographics as a low-frequency driver of trend inflation. This evidence suggests thatin addition to the overall aging of the populationthe demographic effect on inflation depends crucially on the composition of the age distributionthat is, whether the effect is disinflationary or inflationary depends on how saving behavior varies across age cohorts as the population ages. To quantify these effects, I revisit and extend the empirical framework of Juselius and Takats (2015, 2018) (and for more comprehensive analysis, see Goodhart and Pradhan, 2020). I consider annual data for 22 advanced economies (Australia, Austria, Belgium, Canada, Denmark, Finland, France, Germany, Greece, Ireland, Italy, Japan, Luxembourg, the Netherlands, New Zealand, Norway, Portugal, Spain, Sweden, Switzerland, the United Kingdom, and the United States) for the period 19602024. The data are obtained from the OECD database for demographic variables and CPI (year-over-year) inflation rates, with complementary data for real GDP per capita and missing observations for inflation from the World Bank's World Development Indicators.

I start by examining how a commonly used summary statistic of the age distributionthe dependency ratio, defined as (100 times) the number of people aged 19 and less and people aged 65 or older, relative to the number of people of working age (20 to 64 years old)co-moves with the inflation rate. Figure 1 plots these two variables for Swedenfor which this co-movement is particularly pronouncedas well as their cross-sectional averages across the 22 countries. Although the inflation rate is volatile, the dependency ratio is very smooth and persistent. Note, however, that the dependency ratio is essentially serving as a low-frequency filter for inflation and is tightly correlated with the corresponding low-frequency component of inflation. But the graph with averaged data (the right plot in figure 1) also suggests that for the majority of countries, including the United States, the dependency ratio tends to lead inflation. It is important to emphasize, at this point, that the potential effect of demographic structure on inflation reflects secular forces that may only amplify or dampen the prevalent sources of (dis)inflationary pressure such as monetary and fiscal policy, as well as commodity or geopolitical shocks, among others.

To assess the aggregate effect, I ran a panel regression (22 countries over the 19602024 period) of inflation on the dependency ratio, controlling for the output gap of real GDP per capita to account for business-cycle dynamics, and population in 1959 to account for demographic heterogeneity. The estimated coefficient on the dependency ratio is 0.185 with a two-way clustered standard error of 0.043. Alternative techniques, including new panel data methods that are robust to the integration or smoothness properties of the dependency ratio, deliver similar results. As the plots suggest, the effect of demographics on inflation is positive, and the sharp drop of the dependency ratio after 1980 might have contributed to the disinflationary dynamics during the 19902019 period.

But how are changes in the structure of age distribution contributing to the dynamics of the dependency ratio and in turnvia differential saving behavior and wealth accumulationon inflation? Figure 2 presents a breakdown of the contributions of the young (under 20) and old-age (65 and above) components to the total dependency ratio for the United States. From the mid-1960s to the early 2010s, both young and old-age dependency ratios have declined with the young ratio declining faster since 1980. Since the early 2010s, the old-age ratio started to rise while the share of young-age to working-age (20 to 64) population continued its downward trend, but the rise in the old-age ratio dominated, so the overall dependency ratio has been increasing over this period. The US Census provides accurate long-run population projections (under various scenarios) that might prove particularly valuable for the future evolution of the age distribution and its effects on trend inflation, provided that the low-frequency relationship remains stable. The projections until 2030 indicate that the recent dynamics are likely to persist with the total dependency ratio going up further due to the rising share of people aged 65 and above.

The evidence in figure 2 tentatively suggests that the composition of the age distribution could be as important as the shift in the distribution over time. The changes in the shape and structure of the age distribution have also been accompanied by changes in the distribution of financial assets across age cohorts that inform their spending behavior over the life cycle (see, for example, here and here). For these reasons, I disaggregate the dependency ratio into age categories (see also Juselius and Takats (2015, 2018)) that would better accommodate the heterogeneity across the age distribution and shed further light on how the evolution of the different age cohorts affect trend inflation over time.

I consider the following six age groups: under 15, 1529, 3044, 4559, 6074, and 75 and above. These age groups are sufficiently representative to characterize the age-specific effects on inflation across the age distribution. I then perform a panel regression analysis of inflation on these six age groups, output gap per capita, and initial (in year 1959) population for the same 22 advanced economies over the 19602024 period. Figure 3 presents the coefficient estimates on the age groups along with their 95 percent confidence intervals (in shaded area), based on heteroskedasticity and autocorrelation robust standard errors. For robustness, I explored other methods, including one that is estimating individual time-series regressions (for each country) and averaging these estimates. Inference was also robustified with respect to potential multicollinearity and persistence of the regressors. The results are largely in line with those shown in figure 3.

Figure 3 reveals a distinct "bat-shaped" pattern of the effect of demographics on inflation. This pattern reflects the underlying consumption/saving behavior over the life cycle. Young age groups, especially those aged 1529, tend to consume more and exert upward pressure on inflation. The young demographic composition in the advanced economies in the 1960s and 1970s have likely contributed to the inflation dynamics of the 1970s. The impact of aging on trend inflation necessitates a more nuanced view.

As the younger age cohort transitions to middle age, it enters a period in which saving rates are peaking before being drawn down in old age. The regression results confirm this pattern, with 3044 and 4559 age groups exhibiting a negative impact on inflation. The rising share of these groups in the total population between the early 1970s to 2000s (from an average of 35 percent to around 43 percent toward the end of this period) likely played an important role in the secular disinflation dynamics during the Great Moderation and postGlobal Financial Crisis periods. Furthermore, the middle-age to young-age ratio has been shown to possess strong predictive power for future valuation ratios and asset returns (see here and here) which, in turn, could affect inflation via saving and wealth effects.

The consumption/saving behavior shifts again for the 6074 age cohort as the households in this group start to deplete their accumulated savings and wealth and increase their consumption as they retire. This age group benefited the most from the sustained rise in asset prices (see here), which was accompanied by a strong wealth effect on consumption. The estimated coefficient for this age group is almost as large as the one for those aged 1529 (0.45 versus 0.59, respectively) and highly significant. The double-peak pattern of the estimated coefficients takes another leg down for the age cohort 75 years and older as discretionary spending slows down, with predominantly disinflationary consequences.

The empirical evidence I've discussed in this post highlights the role of demographics as a low-frequency driver of trend inflation. Importantly, although these low-frequency variations in the mean and shape of the age distribution might be slow moving and relatively small in magnitude, they can help to reinforce or mute the impact of cyclical variations in the inflation process. The inflationary or deflationary impulse of demographics appears to reflect the compositional dynamics of the age distribution that have direct effects on savings, wealth accumulation, and discretionary spending. The readily available long-run population projections are valuable for assessing underlying inflation pressures and helping to inform future policy.