This thesis examines an artificial neural network option pricing model with an emphasis on its ability to capture the volatility surface and the effect of supply and demand of options on option prices. I compare the empirical pricing errors of the artificial neural network model to the Black-Scholes-Merton model and the Practitioner Black-Scholes model on S&P 500 and Dow Jones Industrial Average index options for the period 2010 to 2017. I find that the lowest pricing errors occur when the VIX (VXD) volatility index is used as volatility parameter in the artificial neural network model. Further, I find that the artificial neural model prevails both of the comparison models across option moneyness and time-to-maturity implying that the ANN model captures the volatility surface better than the comparison models. Finally, I find that the model’s pricing error is reduced further by including the proportional bid-ask spread as a proxy for the liquidity premium resulting from an imbalance between supply and demand, and the slope of the yield curve as a proxy the perceived recession risk by the market. I attribute the improvement of the artificial neural network model to its ability to learn the dynamics of option markets without imposing restrictive assumptions like those of the Black-Scholes-Merton model. However, I recognize that the artificial neural network model suffers from opaqueness which may render the model impractical to some users.
|Educations||MSc in Finance and Investments, (Graduate Programme) Final Thesis|
|Number of pages||100|