Survey
* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
Beating the Best: A Neural Network Challenges the BlackScholes Formula Mary Malliaris and Linda Salchenberger Loyola University Chicago Ninth IEEE Conference on Artificial Intelligence for Applications Orlando 1993 Purpose • A neural network model is developed to estimate the market price of options • These prices are compared to estimates generated by the Black-Scholes model Options • An option is an agreement giving the holder the right to purchase [a call] or sell [a put] some asset at an agreed upon future time • The price that will be paid at this future date is called the exercise price of the option • The market price is the price you pay now for the privilege of buying or selling on or before the expiration date Black-Scholes Model where Methodology • • • • • • • • Data obtained from Wall Street Journal Jan 1, 1990 to June 30, 1990 Exercise price Market price of the option Closing price of the S&P 100 Index Daily data Interest rate from 3-month US treasury Bill In the money and out of the money options Neural Network Variables • Previous variables plus: • Two lagged variables: yesterday’s closing price, and yesterday’s market price of the option Neural Network Approach • • • • • • Feedforward Backpropagation Single hidden layer Fully connected Validation set used to measure performance Testing sets developed with two-week time frame Results • • • • Mean absolute deviation Mean absolute percent error Mean squared error Calculated for each of the 5 two-week periods for both in-the-money and out-of-the-money • NN: lower MAPE in 4 out of 5 periods for outof-the-money • B-S: superior 4 out of 5 for in-the-money Results • • • • Paired sample comparisons tests: B-S consistently overprices the options NN consistently underprices them Standard deviation of the differences smaller in the neural network prices Observations • Similarities between the individual price estimates made by the two models • Both have difficulty computing deep in-themoney prices • NN better out-of-the-money • B-S better in-the-money • NN methodology offers a valuable alternative to estimating option prices