diff --git a/pfhedge/stochastic/merton_jump.py b/pfhedge/stochastic/merton_jump.py index e79943e5..40ed95a1 100644 --- a/pfhedge/stochastic/merton_jump.py +++ b/pfhedge/stochastic/merton_jump.py @@ -110,8 +110,8 @@ def generate_merton_jump( - (sigma ** 2) / 2 - jump_per_year * (math.exp(jump_mean + jump_std ** 2 / 2) - 1) ) - * dt - * torch.arange(n_steps).to(randn) - ) + * dt + * torch.arange(n_steps).to(randn) + ) brown = randn.cumsum(1) * math.sqrt(dt) return init_state[0] * (drift + sigma * brown + jump.cumsum(1)).exp()