Transcript Do-Ahead Replaces Run-Time: A Neural Network Forecasts Options

Do-Ahead Replaces Run-Time:
A Neural Network Forecasts
Options Volatility
Mary Malliaris and Linda Salchenberger
10th IEEE Conference on Artificial
Intelligence for Applications
Overview
• We compare existing methods of estimating
the volatility of daily S&P 100 Index options
• Implied volatility (calculated using the BlackScholes model)
• Historical volatility
• A neural network is used to predict volatility
Volatility
• A measure of price movement
• Often used to ascertain risk
• Ability to forecast volatility gives a
trader a significant advantage in
determining options premiums
Calculating Volatility
• There are two main approaches to estimating
and predicting the non-constant volatility
• The historical approach
– However this assumes that future volatility will
not change and that history will repeat itself
• The implied volatility approach
– One solves the Black-Scholes model for the
volatility that yields the observed call price
Neural Networks
• Layers of interconnected nodes
• Constructed in three layers
• Sigmoid function applied to sum of weighted
inputs at each node
• Connection weights are learned by the
network through a training process by looking
at training set examples
Neural Network Architecture:
Nodes, Connections, & Weights
w1
w2
w3
F(sum inputs*weights)=node output
w19
F(sum inputs*weights)=output
w20
w21
w17
w16
w18
Each node in the hidden &
output layers applies a
function to the sum of the
weighted inputs.
Data
• S&P 100 (OEX)
• Daily closing call and put prices
• Associated exercise prices closest to at-themoney
• S&P 100 Index prices
• Call and put volume
• Call and put open interest
• 250 observations for six series of volatilities
Comparison of Historical and
Implied Volatility Estimates
Neural Network and Implied
Volatility Estimates
Results
• Historical volatility is only backward looking
• Implied volatility provides estimates which are
only valid at that current time
• Neural network volatility uses both short-term
historical knowledge and contemporaneous
variables in the estimate
• NN predictions can be made for a full trading
cycle and are more accurate