over Bai and Kao (2006), since it does not make the restrictive assumption of factor
stationarity in equation (13), which has substantial implications for the modelling
approach pursued in this paper. Bai et al. (2007) construct two estimators that jointly
estimate parameter coefficients and stochastic trends: Continuously-updated and Bias
Corrected (CupBC) and Continuously-updated and Fully Modified (CupFM). Monte
Carlo results from Bai, Kao and Ng (2007) suggest that CupBC and CupFM have
good finite sample properties and are distinctly superior in terms of mean bias in all
cases considered compared to Fixed Effects. As T increases we see bias reduction in
CupBC and CupFM, but no bias reduction as we increase N.10
4. Results
4.1 Data
Using annual data, we analyze debt sustainability for fifteen industrial
countries during the period of 1978 and 2005 and for twenty-seven emerging markets
during 1990 and 2005. The data for industrial countries comes from the OECD
Economic Outlook and Statistical Compendium data set and consists of primary
government balances as percent of GDP and gross government debt as a percent of
GDP. We have data between 1978 and 2005 for 15 countries: Austria, Belgium,
Canada, France, Germany, Greece, Ireland, Italy, Netherlands, Norway, Portugal,
Spain, Sweden, UK and US.
The data on emerging markets are from Budina and Fiess (2004), extended
and updated with the help of official statistics and IMF and World Bank country desk
officers. The 27 countries include: Argentina, Brazil, Bulgaria, Chile, China,
10 We consider whether we have a cointegrating relationship between yit and xit in equation (11) by
applying Bai and Ng’s (2004) PANIC to the regression. If both the factor and residual uit are stationary
in equation (13) then we can utilise Bai and Kao (2006). However, if the factor Ft is nonstationary and
the residual ut is stationary then Bai, Kao and Ng (2007) is the appropriate approach to estimating
equation (11).
14