Browsing by Subject "Stock price forecasting"
Now showing 1 - 6 of 6
Results Per Page
Sort Options
Item Comparative approach of financial forecasting using neural networks(Texas Tech University, 1996-08) Saad, Emad W.Not availableItem Neural network model for stock forecasting(Texas Tech University, 1995-05) Tan, HongThe purpose of the thesis is to use the predictive abiUty of the ANN to analyze some financial time series. Different kinds of stock price data of more than 7 years are collected and used for prediction. Other kinds of time series also are used for the purpose of testing and making comparison. The probabihstic neural network, PNN, is used primarily because of its one-pass fast learning algorithm when dealing with large data sets. The short-term trend of the stock prices is predicted using the powerful classification ability of PNN. The modification of the PNN is made to forecast the real value of time series by first grouping real values into classes, and then converting the predicted class membership to some corresponding value afterward. The thesis will not pursue any trading strategy development, but it would rather attempt to provide necessary and useful information for making such kind of trading decision. Combining the trend and real change of stock prices along with some other knowledge will allow applicant to seek more benefits in the financial market.Item On the conditional forecast of the market risk premium and its economic significance from a long time series perspective(Texas Tech University, 2002-05) Peng, ZhuomingThe equity return data in Wilson and Jones (2002) used in this dissertation previously has not been available for research. This dissertation also is the first attempt in the literature to examine the forecastibility of the annual market risk premium with nonoverlapping observations. Forecasts of the market risk premium obtained with the regression models specified in Equations (3,6a) and (3,6b) and Equations (3,10a) and (3,10b) of this dissertation represent a new approach in the literature. Aggregate leverage variable and the levels of the previous market risk premiums are new variables employed in the regression models. By employing the longest equity return data for the last 130 years and the new regression models, the empirical evidence found in this dissertation generally indicates that the dividend yield series does posses the forecasting power towards the expected market risk premiums. The new testing models represented by Equations (3,6a) and (3,6b) appear to be good forecasting models. Especially, the conditional volatility estimated by EGARCH (1,1) specification can help to forecast the level of the expected market excess returns sampled with three different intervals, namely, annual, quarterly, and monthly. However, the relationship between the conditional mean, i,e,, the return, and the conditional volatility, i.e.,, the risk, of the expected return appears to be less correlated over time. The relationship between the monthly conditional mean and its conditional volatility from January 1914 to December 1956 remains negative, despite the fact that the short-term T-bill rates have been excluded from the set of explanatory variables. In addition, this relationship is not statistically significant in the last subperiod, January 1957 to December 1999, As such, it remains to be seen how financial economists may explain these puzzles in future research. Both quarterly and monthly ex ante market risk premiums appear to have mean-reverting tendencies. The monthly forecasting models do appear possessing economic significance. The evidence of the aggregate leverage {AL) having forecasting value is weak in the annual data, but there is some evidence that the AL variable may forecast the monthly and quarterly excess returns. Last but not least, the "manufactured" annual dividend yield data from 1802 to 1870 in Schwert (1990) does not appear having forecasting values.Item The detection and consequences of beta nonstationarity(Texas Tech University, 1986-12) Howe, Thomas StanleyThe development of return-generating models, some of which rely on beta, has provided a means of examining the abnormal performance of stock returns around the time of an event. One of the problems in using such models is that beta is apparently nonstationary. This study uses simulated daily stock returns to examine the ability of the cumulative sum of the squared recursive residuals (CSRR) and the Quandt log-likelihood ratio (QLLR) to identify a given level of beta change and the effect of a given level of beta change on the results of abnormal returns tests. This study also uses daily stock returns surrounding the listing and delisting of the firms' bonds on Standard and Poor's "CreditWatch" to examine the nature of capital asset pricing model (CAPM) parameter nonstationarity and the effect of allowing for CAPM parameter nonstationarity on abnormal returns test results for these firms. Analysis of the simulated security returns suggests that, given the range of error variances generally found in daily stock returns, the CSRR and QLLR show little ability to identify even a sudden 50 percent beta change and are highly sensitive to outliers. This low power appears not to present a problem. Even a 50 percent sudden beta change leaves the rejection frequencies and average p-values of the abnormal returns tests and the average abnormal return and mean square error of the CAPM regressions largely unchanged. In the CreditWatch samples, the CSRR indicates parameter nonstationarity for nearly every security over a 4-year period. Comparison of the CSRR results with the results of traditional parameter nonstationarity tests suggests that the significant CSRR findings are more often associated with heteroscedasticity or outliers than with a beta change. In the CreditWatch section, the cumulative average residuals appear sensitive to the periods used in estimating the CAPM parameters and the method used to allow for the apparent parameter nonstationarity. This sensitivity is apparently due primarily to instability in the alpha estimates.Item The detection and consequences of beta nonstationarity(Texas Tech University, 1986-12) Howe, Thomas StanleyNot availableItem Two essays on market behavior(2006) Glushkov, Denys Vitalievich; Titman, Sheridan; Wessels, RobertoMy dissertation consists of two essays which investigate how the reaction of market participants to aggregate and firm-specific information affects asset prices and firms’ corporate choices. The first essay studies the implications of investor sentiment for asset prices. It develops a novel stock-by-stock measure of investor sentiment which I call sentiment beta. Using this measure I test several hypotheses. First hypothesis postulates that sentiment affects stocks of some firms more than others due to differences in firm characteristics. Second hypothesis predicts that more sentiment sensitive stocks are more likely to be held by individual investors. Consistent with the first hypothesis, I find that more sentiment-sensitive stocks are smaller, younger, have greater short-sales constraints, idiosyncratic volatility and lower dividend yields. Given size and volatility, high sentiment beta stocks have greater analyst coverage and institutional ownership, higher likelihood of S&P500 membership, higher turnover and lower book-to-market ratios. Stocks that are more exposed to sentiment changes deliver lower future returns, which is inconsistent with the risk factor interpretation of investor sentiment. Institutional analysis reveals that institutions stayed away from sentiment-sensitive stocks in the 1980’s, but held more of these stocks since the early 1990’s. The second essay tests a catering hypothesis which predicts that firm managers concerned about the current stock price will deviate from the optimal policy in setting profitability and revenue growth targets due to the incentives to cater to the time-varying relative investor demand for firms with different composition between revenue growth and profit margins. I develop a measure which I call a revenue growth premium and document three results consistent with catering interpretation: 1) time periods when the premium is high tend to be followed by “higher-than-expected” sales and investment growth, advertising, acquisitions and R&D; 2) catering to the premium is more pronounced among firms where managers care more about the short-term stock price; 3) consistent with “bounded rationality” version of catering story, trading strategy based on longing stocks of firms with high margin surprises and shorting firms with low margin surprises when the premium is high yields 40/bp per month after adjusting for risk and post-earnings announcement drift.