Paul Lajbcygier

Improving option pricing with the product constrained hybrid neural network

In the past decade, many studies across various financial markets have shown conventional option pricing models to be inaccurate. To improve their accuracy, various researchers have turned to artificial neural networks (ANNs). In this work a neural network is constrained in such a way that pricing must be rational at the option-pricing boundaries. The constraints serve to change the regression surface of the ANN so that option pricing accuracy is improved in the locale of the boundaries. These constraints lead to statistically and economically significant out-performance, relative to both the most accurate conventional and nonconventional option pricing models.

A hybrid neural network approach to the pricing of options

A new method of pricing options is introduced. The method is based on the augmentation of a conventional model with an artificial neural network (ANN) trained on the difference between the standard model and actual options data. The pricing accuracy has been demonstrated using the actual All Ordinaries Share Price Index (AO SPI) options on futures. This hybrid approach is shown to provide greater accuracy than either standard model or ANN used alone.

Optimizing Technical Trading Strategies with Split Search Genetic Algorithms

Genetic algorithms can be used to search a space and find a near optimal solution to a problem. Standard genetic algorithms are composed of three operators: reproduction, crossover and mutation. Mutation is the occasional random alteration of the value of a bit-string. The role of the mutation operator is to introduce some randomness into the search. Many researchers erroneously consider mutation unimportant when compared to reproduction and crossover. They argue that the mutation rate is so low that it may as well be non-existent. However, mutation can prevent the search from ending in a local optima. A variant of the standard genetic algorithm, that splits the population into two and applies a high mutation rate to one of the sub-populations, is proposed and tested in this study. The idea is that the high mutation rate will permit the sub-population to ‘jump’ out of a local minima. It is found that the approach optimizes well on a test-bed of functions. The same approach is then applied to a practical optimization problem in finance.

Estimating the number of mutual fund styles using the generalized style classification approach and the GAP statistic

A growing empirical literature has shown that funds managers can he classified by their investment style and that this classification is useful in predicting performance. Brown and Goetzmann (1997) created a generalized style classification (GSC) technology that has been used in various studies to decide what funds managers belong to what styles. The GSC technology relies on k-means clustering to group fund manager returns into styles. Since the GSC technology relies on k-means clustering, the number of styles must be known a priori. Conventional techniques can be used to estimate the number of groups/style but they are dependent on various unrealistic assumptions. A new method to estimate the number of styles, first proposed by Tibshirani, Walther, and Hastie, (2000), known as the Gap statistic is adapted to the GSC technology. The approach is used to verify the number of styles in Japanese mutual funds data.

Factors Affecting the Birth and Fund Flows of Commodity Trading Advisors

This paper investigates the birth of commodity trading advisers (CTAs) and their flow–performance relation. Specifically, we address three questions. First, we investigate the impact of existing CTAs’ performance on the number of new CTAs entering the market. Second, we investigate the importance of performance and other fund-specific factors to fund flows throughout the entire lives of CTAs. Third, we examine the smart money effect on CTAs. That is, we ask whether investors are successful in selecting subsequent well performing CTAs. Our results show that CTA managers tend to start up funds after a period of poor performance across the CTA industry. The flow-performance relation is strongly supported, with top-performing CTAs rewarded with high inflows. However, we find no evidence of ‘smart money’ effect, indicating that investors are generally unsuccessful in choosing subsequent well performing CTAs.

Option pricing with the product constrained hybrid neural network

It is well known that conventional option pricing models have systematic, statistically and economically significant errors or residuals. In this work an artificial neural network (ANN), which estimates the residuals from the most accurate conventional option pricing model, so as to improve option pricing accuracy, is constrained in such a way so that pricing must be rational at the option-pricing boundaries. These constraints lead to statistically and economically significant out-performance relative to both the most accurate conventional and non-constrained ANN option pricing models.

Factors affecting the birth and fund flows of CTAs

Our paper investigates the timing of the inception of commodity trading advisors and the relationship between their fund flows and performance. Our results show that commodity trading advisor industry performance has, over the long-run (short-run), a positive (negative) effect on new commodity trading advisors. The functional form of the flow-performance relation varies across commodity trading advisor subcategories. Also, we do not observe a ‘smart money’ effect, indicating that investors are generally unsuccessful in choosing subsequent high-performing commodity trading advisors.

Measuring Immediate Price Impact Using Nonparametric Models

As a consequence of recent technological advances and the proliferation of high-frequency trading and other forms of algorithmic trading, the cost of trading in financial markets has irrevocably changed. One important change relates to how trading affects prices; this is known as price impact. Understanding price impact is vital as it helps in evaluating different trading strategies and hence leads to optimal execution strategies that minimize trading costs. Besides evaluating established parametric price impact models in the literature, this paper proposes a novel nonparametric approach, known as Generalized Additive Models, to estimate price impact. This paper provides the first empirical analysis of the performance of different immediate price impact models for individual trades using out-of-sample predictions. The study finds that the nonparametric model outperforms all other models both in- and out-of-sample. The outperformance comes from (1) the appropriate price-impact normalization, (2) the greater data-fitting flexibility inherent in nonparametric frameworks, and (3) the incorporation of new explanatory variables which cannot easily be accommodated analytically.

Is Fundamental Indexation able to time the market? Evidence from the Dow Jones Industrial Average and the Russell 1000

Fundamental Indexation weights stock according to a firm’s economic size, not stock price or market capitalization. This means that, at least in theory, unlike traditional market capitalization weighted indexes, it prevents overinvestment in overpriced stock and vice-versa. It should effectively time the market by avoiding incorrect investment in cyclically mispriced stock. We ascertain if Fundamental Indexation outperforms traditional indexing and whether any outperformance can be attributed to market timing. Using almost fifty years of Dow Jones Industrial Average Index and Russell 1000 Index returns, we find some evidence of limited market timing but no evidence of overall positive abnormal performance.

Imputation Credits and Equity Returns

The provision of imputation tax credits can in principle lower the returns that investors require on equity. Whether in practice imputation credits lower the returns that investors require depends in large part on the impact of foreign investors on equity prices. This is because foreign investors in general cannot use the credits that domestic equities provide. We use a range of pricing models and monthly data from July 1987 to December 2009 to test whether, holding risk constant, equity returns are related to credit yields. We find no evidence that the provision of imputation tax credits lowers the returns investors require on equity.