Can anyone suggest a paper that deals with pricing correlation swaps on an arithmetric average realized correlation of a basket of stocks? I was thinking along the lines of modeling each stock using the Heston model and a nice big correlation matrix to specify the correlation between each brownian motion (and price via MC)...but it would be nice to see a paper that deals with these issues...Also need to find some papers on dividend swaps (not seen much from google on this)Thanks

Can anyone confirm if the following approach is sound?- given n stocks in the basket, calibrate a seperate Heston model to each stock- assume the dw_t terms for each stock's volatility process are correlated (i.e. have a nxn correlation matrix for these)- assume all the dw_t terms for dS_t are independent- simulate each stock evoluation via MC and work out correlation swap payoff then discount back to time 0.(this is a generalisation of the approch for valuing correlation swaps given in "Modeling of Variance and Volatility Swaps for Financial Markets with Stochastic Volatilities" by Swishchuk, A.)

Hi,Not sure between points (2) and (3). Are you saying dw(t,i) and dS(t,i) has a correlation(possibly option implied) for ith stock, dS(t,i) and dS(t,j) has a correlation (maybe historical) and dw(t,i) and dw(t,j) are uncorrelated?.Either ways my naive guess would be that multidim heston would produce a lower expectation. I mean when correlation is being played, I would think you would like a dynamic/time varying correlation matrix. Because you have problems of positive definiteness and bounded values [-1,1] for corr matrix I think people try to take a dynamic covariance matrix.One such type of a covar matrix is one where the matrix is driven by a diffusion process with the probabilities prescribed by a wishart distribution. The dynamics of the corr matrix process are then got from dynamics of the covar matrix through an ito manipuation. This matrix is able to meet the above constraints (under some conditions).Thanks

I was intending to follow the approach in the paper I quoted below, but (ahem) I managed to misquote it! Their model is of the following form:dS_i(t) = S_t^i ( mu_i(t) dt + sigma_i(t) dW_i(t) )d(sigma_i)^2(t) = k_i(theta_i^2 - (sigma_i)^2(t) )dt + gamma_i sigma_i(t) dV_i(t)They assume the W_i are correlated, the V_i are uncorrelated and W_i, V_j are uncorrelated (they deal with a 2 stock case, I am generalising here).Do you have a reference for the stochastic correlation matrix approach you suggested? I am keen to keep the model fairly simple as calibration (to 20 stocks in my case) will be difficult enough with the above model.Any views on whether correlation should be implied or historical?

maybe It's a constructive idea.

Hi tibbar,Did you get any information, papers, *anything* on these instruments? I am in the same predicament as you are. Basket of stocks and pricing of correlation swaps using an arithmetic average realized correlation. I also need info on dividend swaps as well..Thanks,Ram