Choosing Initial Parameters for Heston Model Calibration

Summary

The document raises a practical calibration problem for the Heston stochastic volatility model. The author fits the model in Matlab using Monte Carlo simulations, with 5,000 paths, two time steps per day, and a 360-day horizon. Calibration with the fminsearch optimizer is slow, and changing the initial parameter values appears to have little effect, prompting concern that the objective may contain many local minima.

The question asks how to find better starting values, but includes no answer, proposed initialization method, calibration target, or performance comparison. It therefore serves mainly as a statement of the optimization challenge rather than a guide to resolving it. Any useful prescription would need to account for the chosen objective function, parameter constraints, simulation noise, and whether calibration targets option prices or other market observations.

Key ideas

• Heston model calibration can be computationally slow when its objective is evaluated through Monte Carlo simulation.
• The author reports that changing starting values has little apparent effect with fminsearch.
• The document asks for a way to select initial parameters but provides no solution or evidence comparing methods.
• Calibration outcomes may depend on objective design, parameter constraints, and simulation noise.

Tags

# Initial values for Heston Model calibration


# Initial values for Heston Model calibration












I'm doing a Heston model in Matlab using simple Monte Carlo simulations (5.000 paths and 2 steps per day, simulating 360 days). When I try to calibrate the Heston parameters using fminsearch it takes a lot of time and initial values almost doesn't change, so I guess there are a lot local minimization points. So is there any algorithm to obtain good initial values?

Kind regards, Oliver.

Shown in full with attribution under the source's licence. Licence: CC BY-SA 4.0 (Stack Exchange)

This summary was written by Stratmill's research agent from the original; it is not a copy of the source.