Borrowing constraints and international risk sharing: evidence from asymmetric error-correction

Abstract

We analyse the adjustment process of consumption growth after disequilibrating output shocks in a sample of Organization for Economic Cooperation and Development (OECD) countries. In particular, we test the hypothesis that consumption is smoothed to a lesser degree after negative shocks, whereas the impact of a positive shock is delayed for a longer period of time. Our analysis is based on an error-correction framework that allows for asymmetric adjustment. We find that the mean adjustment lag after a negative shock is significantly shorter than after a positive shock, especially since the beginning of the 1980s. This result is consistent with the interpretation that borrowing constraints limit the degree to which the impact of negative shocks on consumption can be smoothed.

Notes

¹ A related branch of the literature studies risk sharing within countries (e.g. Asdrubali et al., 1996; Scorcu, 1998; Buettner, 2002; Borge and Matsen, 2004).

² Several explanations have been proposed in the literature. See, e.g. Backus et al. (1992), Obstfeld (1994), Baxter and Crucini (1995), Stockman and Tesar (1995), Lewis (1996), Imbs (2006) and Hoffmann (2008).

³ In principle, we can also identify periods characterized by more extreme deviations by using, e.g. two SDs. However, since such periods rarely occur, we would not have enough observations for statistically identifying the effect.

⁴ Our approach requires the series to be cointegrated. Since evidence in favour of cointegration has already been reported in Leibrecht and Scharler (2008) for our data set, we start directly with the estimation of the ECM.

⁵ The mean adjustment lags are calculated as 1/|| and 1/||, respectively.

⁶ Note that we only report the estimation results for the case where the speed-of-adjustment coefficient depends on the sign of the deviation of idiosyncratic output growth from its average, but not on the size of the deviation. Taking the size of the deviation into account leaves our conclusions qualitatively unaffected. Also note that the system is stable in both time periods. However, to preserve space we do not report the results for Equation 3 estimated over the subsamples. Detailed estimation results are available upon request.