Abstract:
This main purpose of this paper is to investigate the causal relationship
between knowledge (research output) and economic growth in US over 1981–2011. To
overcome the issues of ignoring possible instability and hence, falsely assuming a constant
relationship through the years, we use bootstrapped Granger non-causality tests with fixedsize
rolling-window to analyze time-varying causal links between two series. Instead of
just performing causality tests on the full sample which assumes a single causality relationship,
we also perform Granger causality tests on the rolling sub-samples with a fixedwindow
size. Unlike the full-sample Granger causality test, this method allows us to
capture any structural shifts in the model, as well as, the evolution of causal relationships
between sub-periods, with the bootstrapping approach controlling for small-sample bias.
Full-sample bootstrap causality tests reveal no causal relationship between research and
growth in the US. Further, parameter stability tests indicate that there were structural shifts
in the relationship, and hence, we cannot entirely rely on full-sample results. The bootstrap
rolling-window causality tests show that during the sub-periods of 2003–2005 and 2009,
GDP Granger caused research output; while in 2010, the causality ran in the opposite
direction. Using a two-state regime switching vector smooth autoregressive model, we find
unidirectional Granger causality from research output to GDP in the full sample.