We simulate the evolution of stylised loan portfolios to assess the impact of IFRS 9 and US-GAAP expected loss model (ECL) on the pro-cyclicality of realised losses and capital ratios of banks, relative to the incurred loss model of IAS 39. We focus on the interaction between the changes in loan loss provisions (LLPs) charges (flow channel) and stocks (stock channel) under ECL. Our results show that ECL model smooths the impact of credit losses on profits and capital resources, reducing the pro-cyclicality of capital and leverage ratios, especially under US GAAP. However, when GDP is highly volatile, the large differences in lifetime probabilities of defaults (PDs) between booms and bust cause sharp increases in LLPs in deep downturns, as seen for US banks during the COVID-19 crisis. Volatile GDP makes capital and leverage ratios more pro-cyclical and cause sharper falls in both ratios in deep downturns under US GAAP, compared to IAS 39. IFRS 9 ECL shows less sensitivity to lifetime PDs fluctuations due to the existence of loan stages, and hence reduces the pro-cyclicality of capital and leverage ratios even when GDP is highly volatile.