Moving average rules, volume and the predictability of security returns with feedforward networks
Article Abstract:
The linear prediction of current returns in linear conditional mean specifications with past buy-sell signals does not bring any forecast improvements in comparison with linear models which use past returns as regressors. This was gleaned from an examination of the linear and nonlinear predictability of stock market returns with some simple technical trading rules, using the daily Dow Jones Industrial Average Index from 1963 to 1988. Results showed that models with past returns offer an average of 2.5% forecast improvement in non-linear conditional mean specifications over the benchmark linear model with past returns.
Publication Name: Journal of Forecasting
Subject: Mathematics
ISSN: 0277-6693
Year: 1998
User Contributions:
Comment about this article or add new information about this topic:
Weight space analysis and forecast uncertainty
Article Abstract:
A study of the symmetries of feedforward networks according to their corresponding groups has found that these groups naturally act on and partition weight space into disjunct domains. An algorithm to generate representative weight vectors in a fundamental domain was obtained. When the metric structure of the fundamental domain was analyzed, a clustering method that uses the natural metric of the fundamental domain was derived, which obtains several clusters of weight vectors, even for large networks and large datasets.
Publication Name: Journal of Forecasting
Subject: Mathematics
ISSN: 0277-6693
Year: 1998
User Contributions:
Comment about this article or add new information about this topic:
Non-linear prediction of security returns with moving average rules
Article Abstract:
The use of buy-sell signals elicited from moving average rules with a band between the short and long moving averages as inputs in models for forecasting linear and nonlinear stock market returns was investigated. Specifically, the prediction performance of autoregressive, generalized autoregressive conditionally heteroscedastic-in-mean and feedforward network models were evaluated. Daily returns between Jan. 2, 1963 and Jun. 30, 1988 from the Dow Jones Industrial Average index were used as the data set.
Publication Name: Journal of Forecasting
Subject: Mathematics
ISSN: 0277-6693
Year: 1996
User Contributions:
Comment about this article or add new information about this topic:
- Abstracts: Evaluating the rationality of fixed-event forecasts. Cross-correlations and predictability of stock returns. Evaluating the predictive accuracy of volatility models
- Abstracts: The structure of the pseudo-equilibrium manifold in economies with incomplete markets. Hamilton approach to multi-dimensional screening
- Abstracts: A stopping rule for the computation of generalized method of moments estimator. The limiting distribution of the maximum rank correlation estimator
- Abstracts: Estimating deterministic trends in the presence of serially correlated errors. Money, prices, interest rates and the business cycle
- Abstracts: Bootstrapping multivariate spectra. Modeling nonlinearity of business cycles: choosing between the CDR and STAR models