This paper shows that the impact of changes in budgetary variables on major macroeconomic variables varies in sign and magnitude in times of crisis and non-crisis in France. We find that these nonlinearities are both frequent (as they exist on all behaviors analyzed: real GDP, private consumption, business investment and private employment) and significant. For this, we estimate time-varying probability Markov-switching models (TVPMS) in order to take into account two budgetary regimes, on the one side periods of severe recessions or depressions (crises), and, on the other side "normal" periods (expansions or moderate recessions). These two regimes are identified endogenously, so that we do not need to preliminary separate episodes of huge contractions and expansions of the business cycle. Further, we are able to identify the variables influencing the probability of a switch between regimes. Searching for nonlinear fiscal impacts in the form of regime-switching effects, we assume temporary variations in the budgetary variables, both on the revenue side (taxes on consumption, on firm's profit, lump sum transfers) and on the expenditure side (traditional public boosts of aggregate demand, transfers, and subsidies). Our results show that if one considers the aggregate GDP, public expenditure has a stronger impact during crisis and the expenditure multiplier is greater than the tax multiplier. Also, when households are sensitive to the unemployment situation, tax cuts do not increase consumption spending, while transfers are playing a significant role. On the firms side, our results show that direct taxes changes induce a (stimulus) effect in the investment rate only during non-crisis periods. A rise in subsidies has a negative influence during crises. Finally, the estimates suggest that employment policies should be asymmetric: fiscal measures aiming at reducing unit labor costs could be efficient in good times, while an increase in public employment is preferable during crisis.
