I want to calibrate different models by minimizing RMSE. When I use data from Schoutens (2003) everything is OK i.e I get reasonable parameters. The problem appears when I try to calibrate models to the data from yahoo finance. As the $S_0$ I take adjusted Close price and the option price is the average of the bid and ask prices. Out of over $1000$ options, I choose about $100$ options with different expiry times and strike prices. Ufortunately, when I calbrate different model I get very strange values. For example in BS model $\sigma=0.47$ (with $\sigma=0.18$ for Schoutens data) and in Kou model $\sigma$ is the same as in BS model and $\lambda=0$ so I get the same result as for BS model (and for Schoutens: $\sigma=0.136, \lambda=0.104, p=0.621, \eta_1=48.778, \eta_2=0.046$ and much better RMSE than for BS model). In Merton model I get very good fit but $\lambda$ is $20$ so very unrealistic. Where is the problem? How to choose data to get normal parameters?