Purpose – Aims to investigate the accuracy of parametric, nonparametric, and semiparametric methods in predicting the one-day-ahead value-at-risk (VaR) measure in three types of markets (stock exchanges, commodities, and exchange rates), both for long and short trading positions. Design/methodology/approach – The risk management techniques are designed to capture the main characteristics of asset returns, such as leptokurtosis and asymmetric distribution, volatility clustering, asymmetric relationship between stock returns and conditional variance, and power transformation of conditional variance. Findings – Based on back-testing measures and a loss function evaluation method, finds that the modeling of the main characteristics of asset returns produces the most accurate VaR forecasts. Especially for the high confidence levels, a risk manager must employ different volatility techniques in order to forecast accurately the VaR for the two trading positions. Practical implications – Different models achieve accurate VaR forecasts for long and short trading positions, indicating to portfolio managers the significance of modeling separately the left and the right side of the distribution of returns. Originality/value – The behavior of the risk management techniques is examined for both long and short VaR trading positions; to the best of one's knowledge, this is the first study that investigates the risk characteristics of three different financial markets simultaneously. Moreover, a two-stage model selection is implemented in contrast with the most commonly used back-testing procedures to identify a unique model. Finally, parametric, nonparametric, and semiparametric techniques are employed to investigate their performance in a unified environment.