Prof. Stephen Figlewski, Finance Department, Stern School of Business, New York University 

Valuation models for options and other derivative securities require volatility forecasts for the underlying assets. So do quantitative methods for assessing market risk exposure, such as Value-at-Risk, that are being widely adopted in the securities industry. The tutorial will describe the different forecasting approaches and present a critical discussion of the tactics, strategy, and underlying philosophy behind them. Among the issues to be addressed are the following. 

• Modeling philosophy: What do we actually mean by volatility? How should we think about the general problem of forecasting it? 
• Forecasting volatility from historical data: If the real world consisted of lognormal diffusions with constant volatility parameters, or even if volatilities were time-varying but followed constant volatility processes, appropriate statistical estimation techniques would be unambiguous. But the real world is more random and less predictable than that. So what works best in practice: The classical unconditional volatility estimator? Generalized Autoregressive Conditional Heteroskedasticity (GARCH)? Or maybe a tweaked classical model based on "Optimized Unconditional Conditional Heteroskedasticity" (OUCH)?
• Implied volatility as a forecast: Both academics and practitioners believe strongly that implied volatilities derived from observed market option prices contain more information than do historical estimates. Yet they differ completely in what they think that information is and how they use it. Moreover, empirical studies find that implied volatilities do not pass standard unbiasedness tests. So what information is obtainable from implied volatilities and how do we get it?

Prof. Benjamin Van Roy, Engineering Economic Systems Department, Stanford University  

In principle, a wide variety of financial decision problems - ranging from dynamic asset allocation to derivatives pricing and hedging to transaction cost optimization - can be formulated in terms of stochastic control and solved by the algorithms of dynamic programming. Unfortunately, due to the curse of dimensionality, the associated computational requirements become intractable in most practical contexts. This tutorial will overview the main ideas and state-of-the-art of neuro-dynamic programming (a.k.a. reinforcement learning), a new methodology that offers a tractable approach to approximating dynamic programming solutions. Application areas in finance including derivatives pricing and asset allocation will be discussed, as will past experience with the use of such a methodology. 

Prof. Halbert White  Economics Department, University of California at San Diego 

This tutorial will explore issues of data snooping (or data mining in the negative sense) in evaluating the performance of asset trading/investment strategies. Particular attention will be devoted to new bootstrap-based methods for avoiding falling prey to data snooping/mining biases in selecting market strategies. The tutorial assumes a basic understanding of probability and statistics.