journal article
LitStream Collection
Home
Refenes, Apostolos Paul; White, Halbert
doi: 10.1002/(SICI)1099-131X(1998090)17:5/6<349::AID-FOR701>3.0.CO;2-Xpmid: N/A
In forecasting a financial time series, the mean prediction can be validated by direct comparison with the value of the series. However, the volatility or variance can only be validated by indirect means such as the likelihood function. Systematic errors in volatility prediction have an ‘economic value’ since volatility is a tradable quantity (e.g. in options and other derivatives) in addition to being a risk measure. We analyse the fidelity of the likelihood function as a means of training (in sample) and validating (out of sample) a volatility model. We report several cases where the likelihood function leads to an erroneous model. We correct for this error by scaling the volatility prediction using a predetermined factor that depends on the number of data points. Copyright © 1998 John Wiley & Sons, Ltd.
Refenes, Apostolos Paul; White, Halbert
doi: 10.1002/(SICI)1099-131X(1998090)17:5/6<369::AID-FOR702>3.0.CO;2-Spmid: N/A
In this paper we apply statistical inference techniques to build neural network models which are able to explain the prices of call options written on the German stock index DAX. By testing for the explanatory power of several variables serving as network inputs, some insight into the pricing process of the option market is obtained. The results indicate that statistical specification strategies lead to parsimonious networks which have a superior out‐of‐sample performance when compared to the Black/Scholes model. We further validate our results by providing plausible hedge parameters. Copyright © 1998 John Wiley & Sons, Ltd.
Refenes, Apostolos Paul; White, Halbert
doi: 10.1002/(SICI)1099-131X(1998090)17:5/6<389::AID-FOR703>3.0.CO;2-Npmid: N/A
In parameter estimation, we take advantage of the Cramer–Rao lower bound (CRLB) to evaluate the performance of estimation algorithms since the CRLB provides a theoretical upper bound on estimation accuracy. In pattern recognition, the same concept can be quite useful in terms of knowing the point of diminishing return. In this paper, we develop an innovative approach to quantifying the classification CRLB by combining the concepts of sufficient statistics and data compression with a metric that measures class separability. This approach allows us to assess the degree of performance optimality attained by each classifier. Instead of ranking performance of each classifier based on a confusion matrix, the proposed approach assigns a quantity called the optimality score that indicates the extent to which a classifier approximates the Bayes classifier. We illustrate the power of this approach with two interesting examples—two‐class prediction problems with known and unknown class‐conditional probability density functions. The latter case deals with prediction of S&P 500 price‐movement direction based on raw price data and technical indicators. Copyright © 1998 John Wiley & Sons, Ltd.
Refenes, Apostolos Paul; White, Halbert
doi: 10.1002/(SICI)1099-131X(1998090)17:5/6<401::AID-FOR704>3.0.CO;2-Cpmid: N/A
This paper uses the daily Dow Jones Industrial Average Index from 1963 to 1988 to examine the linear and non‐linear predictability of stock market returns with some simple technical trading rules. Some evidence of non‐linear predictability in stock market returns is found by using the past buy and sell signals of the moving average rules. In addition, past information on volume improves the forecast accuracy of current returns. The technical trading rules used in this paper are very popular and very simple. The results here suggest that it is worth while to investigate more elaborate rules and the profitability of these rules after accounting for transaction costs and brokerage fees. Copyright © 1998 John Wiley & Sons, Ltd.
Refenes, Apostolos Paul; White, Halbert
doi: 10.1002/(SICI)1099-131X(1998090)17:5/6<415::AID-FOR705>3.0.CO;2-Xpmid: N/A
This paper sets out to determine the strategic positioning of Spanish savings banks, using data drawn from published financial information. Its starting point is the idea of the strategic group, regularly employed in business management to explain the relationships between firms within the same sector, but with the characteristic that the strategic group is identified using financial information. In this way, groups of firms that follow a similar financial strategy—with similar cost structures, levels of profitability, borrowing, etc.—have been obtained. As the exploratory data analysis technique used to obtain these strategic groups, a combination of a non‐supervised neural network, the Self‐Organizing Feature Maps (SOFM) with Cluster Analysis (CA) is proposed. This methodology permits the visualization of similarities between firms in an intuitive manner. The application of the proposed methodology to the financial information published by the totality of Spanish savings banks allows for the identification of the existence of profound regional differences in this important sector of the Spanish financial system. Thereafter, a bivariate study of the financial ratios details the aspects that distinguish the savings banks that operate in the different Spanish regions. Copyright © 1998 John Wiley & Sons, Ltd.
Refenes, Apostolos Paul; White, Halbert
doi: 10.1002/(SICI)1099-131X(1998090)17:5/6<429::AID-FOR706>3.0.CO;2-Hpmid: N/A
This paper addresses an extensively studied problem, which is a particular case of long‐term forecasting. In many practical situations, one has to predict a complete curve, i.e. the set of the 24 hourly values for the next day, of all the daily values for the next month or for the next year. For example, it is the case if the matter is to forecast the daily half‐hour electricity consumption. Many methods have been developed, standard linear methods (e.g. ARIMA) as well as neural ones. In this paper we present a very simple method that we call the K‐method. We assume the forecasting problem can be split up into three subproblems: the forecast of the mean (level of the values), of the standard deviation (scattering) and of the normalized profile (which essentially represents the shape). The profiles are classified using a Kohonen map with the neighbourhood preservation property and the mean and variance are fitted using any convenient short‐term forecasting method. Then, for some future curve, a strategy is defined in order to compute its expected normalized profile, the mean and the variance are forecast and the expected curve is computed. This method is low computation time consuming and is easy to develop. Two applications are presented: an example using artificial data and the prediction of the daily half‐hour electrical power curves in France. Copyright © 1998 John Wiley & Sons, Ltd.
Refenes, Apostolos Paul; White, Halbert
doi: 10.1002/(SICI)1099-131X(1998090)17:5/6<441::AID-FOR707>3.0.CO;2-#pmid: N/A
We propose to train trading systems and portfolios by optimizing objective functions that directly measure trading and investment performance. Rather than basing a trading system on forecasts or training via a supervised learning algorithm using labelled trading data, we train our systems using recurrent reinforcement learning (RRL) algorithms. The performance functions that we consider for reinforcement learning are profit or wealth, economic utility, the Sharpe ratio and our proposed differential Sharpe ratio. The trading and portfolio management systems require prior decisions as input in order to properly take into account the effects of transactions costs, market impact, and taxes. This temporal dependence on system state requires the use of reinforcement versions of standard recurrent learning algorithms. We present empirical results in controlled experiments that demonstrate the efficacy of some of our methods for optimizing trading systems and portfolios. For a long/short trader, we find that maximizing the differential Sharpe ratio yields more consistent results than maximizing profits, and that both methods outperform a trading system based on forecasts that minimize MSE. We find that portfolio traders trained to maximize the differential Sharpe ratio achieve better risk‐adjusted returns than those trained to maximize profit. Finally, we provide simulation results for an S&P 500/TBill asset allocation system that demonstrate the presence of out‐of‐sample predictability in the monthly S&P 500 stock index for the 25 year period 1970 through 1994. Copyright © 1998 John Wiley & Sons, Ltd.
Refenes, Apostolos Paul; White, Halbert
doi: 10.1002/(SICI)1099-131X(1998090)17:5/6<471::AID-FOR708>3.0.CO;2-Upmid: N/A
The usage of location information of weight vectors can help to overcome deficiencies of gradient‐based learning for neural networks. We study the non‐trivial structure of weight space, i.e. symmetries of feedforward networks in terms of their corresponding groups. We find that these groups naturally act on and partition weight space into disjunct domains. We derive an algorithm to generate representative weight vectors in a fundamental domain. The analysis of the metric structure of the fundamental domain leads to a clustering method that exploits the natural metric of the fundamental domain. It can be implemented efficiently even for large networks. We used it to improve the assessment of forecast uncertainty for an already successful application of neural networks in the area of financial time series. Copyright © 1998 John Wiley & Sons, Ltd.
Refenes, Apostolos Paul; White, Halbert
doi: 10.1002/(SICI)1099-131X(1998090)17:5/6<481::AID-FOR709>3.0.CO;2-Qpmid: N/A
Despite increasing applications of artificial neural networks (NNs) to forecasting over the past decade, opinions regarding their contribution are mixed. Evaluating research in this area has been difficult, due to lack of clear criteria. We identified eleven guidelines that could be used in evaluating this literature. Using these, we examined applications of NNs to business forecasting and prediction. We located 48 studies done between 1988 and 1994. For each, we evaluated how effectively the proposed technique was compared with alternatives (effectiveness of validation) and how well the technique was implemented (effectiveness of implementation). We found that eleven of the studies were both effectively validated and implemented. Another eleven studies were effectively validated and produced positive results, even though there were some problems with respect to the quality of their NN implementations. Of these 22 studies, 18 supported the potential of NNs for forecasting and prediction. Copyright © 1998 John Wiley & Sons, Ltd.
Showing 1 to 10 of 10 Articles