Macroeconomic Crises since 1870

We build on the Maddison GDP data to assemble international time series from before 1914 on real per capita personal consumer expenditure, C. We also improve the GDP data in many cases. The C variable comes closer than GDP to the consumption concept that enters into usual asset-pricing equations. (A separation of consumer expenditure into durables and non-durables is feasible for only a minority of cases.) We have essentially full annual data on C for 22 countries and GDP for 35 countries, and we plan to complete the long-term time series for a few more countries. For samples that start as early as 1870, we apply a peak-to-trough method for each country to isolate economic crises, defined as cumulative declines in C or GDP by at least 10%. The principal world economic crises ranked by importance are World War II, World War I and the Great Depression, the early 1920s (possibly reflecting the influenza epidemic of 1918-20), and post-World War II events such as the Latin American debt crisis and the Asian financial crisis. We find 87 crises for C and 148 for GDP, implying disaster probabilities around 3.6% per year. The disaster size has a mean of 21-22% and an average duration of 3.5 years. A comparison of C and GDP declines shows roughly coincident timing. The average fractional decline in C exceeds that in GDP during wartime crises but is similar for non-war crises. We simulate a Lucas-tree model with i.i.d. growth shocks and Epstein-Zin-Weil preferences. This simulation accords with the observed average equity premium of around 7% on levered equity, using a "reasonable" coefficient of relative risk aversion of 3.5. This result is robust to a number of perturbations, except for limiting the sample to non-war crises, a selection that eliminates most of the largest declines in C and GDP. We plan a statistical analysis that uses all the time-series data and includes estimation of long-run effects of crises on levels and growth rates of C and GDP. We will also study the bond-bill premium (empirically around 1%) and allow for time-varying disaster probabilities.