Due to the availability of high-frequency intraday data for the financial variables, most of the financial researchers are mainly concerned with modelling and forecasting volatility that is the key input to financial risk management. This paper, closer to Bandi and Russel (2008) and Engle et al (1990), aims to compare the performance of support vector machine, a new semi-parametric tool for regression estimation, heterogeneous autoregressive (SVM-HAR) models with the classical heterogeneous autoregressive (HAR) models in the context of profits from option pricing and trading. Using forecasts from these models for realized volatility obtained using 5-minute, 10-monute, 15-minute and optimally sampled intraday return, agents price short-term options on Nikkei 225 index before trading each other at average prices. It is observed that the support vector machine heterogeneous autoregressive models produced comparatively better performances on the basis of higher Sharpe ratios and average profits.