During the global financial turmoil of 2007 and 2008, no major derivative clearing house in the world encountered distress while many banks were pushed to the brink and beyond. An important reason for this is that derivative exchanges have avoided using value at risk, normal distributions and linear correlations. This is an important lesson. The global financial crisis has also taught us that in risk management, robustness is more important than sophistication and that it is dangerous to use models that are over calibrated to short time series of market prices. The paper applies these lessons to the important exchange traded derivatives in India and recommends major changes to the current margining systems to improve their robustness. It also discusses directions in which global best practices in exchange risk management could be improved to take advantage of recent advances in computing power and finance theory. The paper argues that risk management should evolve towards explicit models based on coherent risk measures (like expected shortfall), fat tailed distributions and non linear dependence structures (copulas).