The shifting Scully curve: international evidence from 1871 to 2016

ABSTRACT

Scully curves estimating growth-maximizing public-sector size are constructed using panel data covering 17 industrialized nations from 1870–2016. Results suggest that government expenditure-to-GDP ratios between 24% and 32% were historically growth maximizing. Instrumental variable estimates support the quadratic Scully curve relationship as causal over this period. We allow for a shifting Scully curve and find that the economic growth-maximizing size changed throughout history, from 11% pre-WWI to 21% post-WWII and 41% following the OPEC crisis. The Scully curve disappears after the mid-1990s suggesting a structural change in the relationship between central government expenditure and economic growth.

KEYWORDS:

• Scully curve
• public-sector Size
• economic growth

JEL CLASSIFICATION:

• E62
• H10
• H50

Acknowledgments

The authors wish to thank the participants at the October 2017 Meetings of the Canadian Network for Economic History held at University of Toronto for their suggestions on an earlier draft of this article as well as comments provided by Alex Chernoff, Vincent Geloso, Byron Lew and Angela Redish.

Disclosure Statement

No potential conflict of interest was reported by the authors.

Notes

¹ State sector size was also shown to correlate negatively with output per-head given a fixed input ratio.

² Government’s role in the nineteenth century was laissez-faire, providing basic legal machinery. Governments subsequently grew following the influence of Marx and Keynes. After the Great Depression, governments focused on stabilization.

³ These periods represent the first great globalization (1871 to 1913), the war and depression (1914 to 1945), the postwar boom (1946 to 1973), the productivity slowdown era (1974 to 1995), and the modern post-Soviet era (1996 onward). Years used in estimation correspond imperfectly because of missing data during war years.

⁴ Accessed June 2018. Data and documentation available at: http://www.macrohistory.net/datahttp://www.macrohistory.net/data.

⁵ Appendix Table A1 summarizes missing values.

⁶ Source variable rgdpmad, real GDP per-capita in PPP terms.

⁷ Central government expenditure is scaled-up over three periods, 1870–1938, 1939–1945 and 1946–2013, using estimates of total public-sector spending shares at the State and Provincial level. Sources include historical GDP estimates by Mac Urquhart for Canada, Historical Statistics of the United States from 1890–1970 and OCED numbers post 1990. Unfortunately, suitable estimates are not available for other countries.

⁸ PP tests use four lags as selected by the Schwert criterion. IPS tests use bandwidths selected using the Bayesian Information Criterion.

⁹ The Scully relationship is also found using HP de-trended $G$.

¹⁰ Real interest rates are calculated from nominal rates by first calculating inflation from the available price index and subsequently employing the Fisher equation. Robustness checks using nominal interest rates are available upon request.

¹¹ Since Canada, Switzerland and the US are federations, a redistribution of the balance between central and regional governments might also affect economic growth in these countries.

¹² The war-time periods reflect heterogeneous economic shocks across countries, which could either be positive or negative. These may also be persistent. For example, Caplan (2002) finds positive output growth effects of wartime when fighting occurs on foreign soil but negative effects otherwise using data on 66 countries spanning 1881 to 1988. Vonyó (2008) argues that differential economic growth across western countries in the 1950s and 1960s can be explained by postwar reconstruction in some nations and structural modernization in others. We account for the unique effects of wars with war fixed effects and through period breaks in our estimates $-$ namely 1919 to 1938 and 1946 to 1973.

¹³ Military expenditure and urban population shares unavailable from 2013$-$2016 and for Australia prior to 1919, Finland prior to 1918 and several other European countries during the world wars.

¹⁴ Social expenditure data for the 17 countries are too sparse to create an annual series. Data at 10-year intervals were available covering 1870 to 1930 from Lindert (1994) and annual data post 1980 from the OECD (Ademai, Froni, and Ladaiquei 2011). Binary indicators for high and low social spenders across sub-periods were collinear with other covariates and thus coefficients could not be estimated. Social expenditure as a share of $G$ has grown from a below 1% average during the years 1870 to 1930 to 22% from 1980 to 2016.

¹⁵ Results available upon request. Endogenous growth models such as Barro (1990) also predict a Scully curve relationship in the long-run. Our data suggest that the Scully curve may be found in longer-term measures of GDP growth. Appendix Table A8 presents results showing a similar quadratic relationship in 5-year averaged data.

¹⁶ The predicted population growth rate is not affected by changes in the birth rate accompanying national wealth changes, for example.

¹⁷ In fact, these quadratic terms were significant in robustness exercises that excluded some of the insignificant covariates.

¹⁸ Including Australia delivers very similar results. Estimates for coefficients of $G$ and $G^2$ including Australia are $48.143$ and $-239.074$, respectively. The resulting Scully curve peaks at an economic growth rate of 2.4%, associated with a $G^{\ast }$ of 0.101.

¹⁹ Robustness analysis, available upon request, suggests a similar Scully curve excluding the 5 years post WW2. The same exercise for the interwar period is inconclusive because of small sample sizes.

²⁰ Acemoglu (1998, 2002) notes that skill-biased technology change was most rapid since the early 1980s.

²¹ This has been referred to as the “flat of the curve hypothesis”. See Schoder and Zweifel (2011) for discussions with respect to healthcare.