The industry consensus on the implementation of the International Financial and Reporting Standard 9 - Financial Instruments (IFRS9) in the field of credit risk is that the estimation of credit risk parameters should be conditioned in the baseline, upside and downside macroeconomic scenarios presumed to be representative of the respective state of the economy. The existing approaches to scenario generation and probability weights assignment suffer from arbitrary inputs, e.g. expert judgment, quantiles selection, severity metric, the specification of a conditioned path. We present a pioneering forecasting approach using a Bayesian MS-VAR which is net of these arbitrary components. This method allows for the consistent contemporaneous formulation of the baseline and alternative scenarios and endogenously ties them to their respective probability weights. We propose to generate representative scenarios as unconditional regime-specific forecasts and to calculate the probability weights associated with representative scenarios as unconditional lifetime transition probabilities. We illustrate the method on artificial as well a real data and conduct an empirical backtest, in which generated scenarios are compared to the actual development during the financial crisis. The method is challenged with the DSGE model and conditional forecasting.