Sensitivity and Simulation Analysis of Granger Causality: An Empirical Investigation
Author: Zahid Asghar

Centrality of causality in Economics dates back at least to Hume and Adam Smith. Hume’s work on exploring relationship between money and prices, and Smith’s book “An Inquiry into the Nature and Causes of the Wealth of Nations” are the clear manifestations of the importance of causality in economics. Later on with the development of statistical tools like correlation and regression in the early 20th century, the issue of causal inference has become more important than before. In the second half of the 20th century, with the introduction of Granger causality, it was considered that the issue of causality could simply be resolved by using statistical tools. A lot of literature in economics in the last two three decades is based on this concept of causality despite some thoughtful criticism on this technique by economists. In this thesis, we have made an effort to analyse the mis(use) of Granger causality on empirical grounds so that one can understand what are the pros and cons of Granger causality .By using published papers in refereed journals, we analyse whether Granger causality helps in establishing any causal direction? Our analysis suggests that causal relationships established through Granger causality are not robust with respect to sample range, lag length selection, base year change etc which is against the basic Axiom C proposed by Granger(1980) that all causal relationships should maintain the same direction. Moreover, Monte Carlo and Bootstrap simulation evidence also indicate that Granger causality results are misleading, and the valid causal relationship can probably be determined only if the whole population, all the relevant variables and true model are known. We conclude that no short cut procedure, merely based upon statistical techniques can be considered as a final word on causality .Nevertheless, concept of structural causality, an idea borrowed from the idea of Simon, Hendry and Hoover, works well and it can be considered a preferred approach for testing causality because it is based on extra statistical information which comes through past historical events and helps in introducing asymmetry in the relationship. Supervisor: Dr.Asad Zaman Co-supervisor: Dr. Rehana Siddiqui

Meta Data

Keywords : Granger Causality, Sensitivity and Simulation, Simulation Analysis
Supervisor: Asad Zaman
Cosupervisor: Rehana Siddiqui

Related Thesis​