In recent days, investors are facing higher market risk due to the pandemic situation, but this is not the only time, investors also experienced similar risk during the Global Financial Crisis in 2007. We are interested in the tool to accurately estimate the market risk and ways to keep the portfolio maintaining the good performance in the extreme situation. This paper particularly investigates the Value-at-risk (VaR), which is one of the simplest and helpful tools to estimate market risk. By using data of SET50 and Thai Baht Gold, this paper demonstrates the best way to compute VaR among the various models of joint distribution of SET50 return and Thai Baht Gold return. There are two important components of joint distribution. The first one is the marginal distributions of SET50 and Gold. The study compares between the normal distribution and the extreme value distribution. Another component is the dependence structure of SET50 and Gold which will is described as copula. The study compares the gaussian copula with the clayton copula, the dependence structure capable of capturing lower tail dependence during the extreme negative return. Nevertheless, the result shows that value-at-risk using extreme value distribution and gaussian copula is our best model. To maintain the portfolio performance. the study sets the optimization problem and find the optimal weight by maximization the risk-adjusted return and use value-at-risk each model represents the risk instead of the standard deviation. The study shows that the optimal weight improves the portfolio performance during crisis, but the portfolio performance is worse during the non-crisis period.