Long Memory and Economic Growth in the World Economy Since the 19th Century

G.P. Silverberg, B. Verspagen

Research output: Chapter in Book/Report/Conference proceedingChapterAcademic


We discuss the relevance of long memory for the investigation of long-term economic growth and then briefly review the state-of-the-art of statistical estimators of long memory on small samples and their application to economic datasets. We discuss theoretical mechanisms for long memory such as cross-sectional heterogeneity. We argue that this endogeneity should be explained endogenously and not simply assumed. Evolutionary models of growth appear to offer one natural explanation of such heterogeneity. Using the maddison (1995) [1] data on 16 countries starting in 1870, supplemented by more recent data down to the year 2001, we then apply different estimators to test the hypothesis of long memory on individual country gdp and gdp per capita. These estimators are beran’s fgn nonparametric test based on an approximate whittle ml estimator, robinson’s semiparametric log periodogram regressor, sowell’s parametric ml arfima estimator and the ml far estimator. The results are mixed and somewhat ambiguous between methods. Moving from the nonparametric to the parametric methods (i.e., controlling for short memory) we find less evidence of long memory. We find that robinson’s semiparametric method also suffers from severe sensitivity to the cuto. Parameters. We compare our results with those of michelacci and zaffaroni [2] and criticize their methodology. We conclude that the lack until now of a single test that deals successfully with all known problems (small sample bias, short memory contamination, specification error, parameter sensitivity) precludes the formulation of a definitive statement about long memory in economic growth.keywordsunit rootmaximum likelihood estimatorhurst parameterorder selectionshort memorythese keywords were added by machine and not by the authors. This process is experimental and the keywords may be updated as the learning algorithm improves.
Original languageEnglish
Title of host publicationProcesses with Long Range Correlations: Theory and Applications
EditorsG Rangarajan, M Ding
Place of PublicationBerlin
PublisherSpringer Verlag
Publication statusPublished - 1 Jan 2003

Publication series

SeriesLecture Notes in Physics

Cite this