2
$\begingroup$

I calibrated (by minimizing RMSE) the Black Scholes, VG and CGMY models to data from 2005 (before the crisis) and to data from 2020. The results surprised me. I do not understand why for data from 2005 it gets a significant improvement in relation to the Black Scholes model, and in 2020 all model parameters are such that the RMSE value is the same for all models. What does it mean? What happened that there is no improvement when using the VG or CGMY models?

enter image description here

$\endgroup$
  • 4
    $\begingroup$ It's really surprising that the RMSE is exactly the same. Are you sure you're putting in the right number. $\endgroup$ – Bob Jansen Sep 22 at 14:41
  • $\begingroup$ Yes, I also ue data from Schotens(2003) and my result was very close so I think that my codse is OK. If someone have opportunity to calibrate these models, I can send my data (or someone can send me his data) and we can compare our result. At this point I'm stuck because I don't know what's going on :( $\endgroup$ – Mr.Price Sep 22 at 14:57
  • $\begingroup$ I only add that when calibrating to one maturity these models give better results than BS model, but the more maturity are added, the RMSE converges to the same as in BS model $\endgroup$ – Mr.Price Sep 22 at 15:06
  • $\begingroup$ Agree with Bob - your 2020 calibrations look off. The CGMY model is not defined by $Y = 2$ since the gamma function has a singularity at -2. The characteristic function blows up for $Y$ "too close" to 2 already. $C \approx 0$ means that the overall activity level of the process is close to zero. $\nu \approx 0$ for VG means that the variance of the gamma process is close to zero. $\endgroup$ – LocalVolatility Sep 22 at 20:21
  • $\begingroup$ I introduced constraints on the parameters C, G, M, Y and the algorithm came to the values ​​that give them the lowest RMSE. I should write $Y\approx 2$ because in reality $Y=1.9999...$. I will repeat the tests tomorrow to be 100% sure, and if someone was able to calibrate to my data or has the data himself and could send me, I would be very grateful because the easiest way to check where is the problem is comparing results. $\endgroup$ – Mr.Price Sep 22 at 21:12

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.