matlab coding
Budget: $30 – $250 USD
Actually I am using a multimodal distribution based on extreme value theory and copula to make prediction of a portfolio of stock indices. The interior of the distribution is built using a non-parametric gaussian kernel, and the tails are based on a parametric Generalized Pareto distribution. Then a copula is calibrated assuming the Student-t distribution in order to induce a dependence structure between the generated residuals of each asset. Finally, the simulations estimate the Value-at-Risk (VaR) of the portfolio over a period of one month . All these steps are available in terms of matlab code. The code uses the Statistics and Machine Learning Toolbox for some commands.
In parallel, what i would like if possible is (and this is the required task), in a VaR backtesting framework to add the VaR or loss forecast of the model cited previously as part of a rolling window approach so as to calculate the loss forecast for the next day and compare it to the actual losses at the closing time in the end of the day. The aim being to compare the outcome with other model of VaR prediction (for which the code is already available);
To sum up, the objective is to write a code that would enable us to have VaR forecasts one day ahead assuming a rolling window (to be defined) to take into account progressively new observed data that would transform the shape of the distribution from which the simulation is drawn (the semi-parametric distribution , with Generalized Pareto distribution in the tails and a gaussian kernel in the interior part of the distribution), instead of the simulations.
PS: I can send you the graphs or web pages with code and illustrations that may make it quite straight forward for the viwer.
In parallel, what i would like if possible is (and this is the required task), in a VaR backtesting framework to add the VaR or loss forecast of the model cited previously as part of a rolling window approach so as to calculate the loss forecast for the next day and compare it to the actual losses at the closing time in the end of the day. The aim being to compare the outcome with other model of VaR prediction (for which the code is already available);
To sum up, the objective is to write a code that would enable us to have VaR forecasts one day ahead assuming a rolling window (to be defined) to take into account progressively new observed data that would transform the shape of the distribution from which the simulation is drawn (the semi-parametric distribution , with Generalized Pareto distribution in the tails and a gaussian kernel in the interior part of the distribution), instead of the simulations.
PS: I can send you the graphs or web pages with code and illustrations that may make it quite straight forward for the viwer.