| Summary: | 碩士 === 中華大學 === 資訊管理學系碩士在職專班 === 100 === Combining multiple factors may construct a more accurate stock return prediction model, but
a trial and error approach is clearly inefficient to discovery the best multi-factor model. In this
study, sort normalization was employed to normalize the independent variables and the
dependent variable, and regression analysis and regression trees were employed to establish
the rate of return prediction models to identify the most important variables affecting the rate
of return. The U.S. stock market was employed as the market to implement empirical analysis.
The results showed (1) the principal component analysis and variable clustering analysis
showed that stock selection factors can be divided into five ingredients: value (such as B/P),
growth (such as ROE), scale (such as total market capitalization), inertia (such as quarter
stock price rate of change), others (such as S/P). Univariate sorting method showed that the
value stocks earn high returns relying on high-risk, but the growth stocks, not. It also showed
that the value-oriented stocks intend to select small-cap stocks, but the growth-oriented stocks
have no such tendency. (2) When using regression analysis to establish the rate of return
prediction models, adopting rank values for independent and dependent variables is better
than adopting original values. When using backward elimation method, the first four most
important variables are the stock price, total market value, EPS, and debt-equity ratio. When
using forward selection method, the first four most important variables are the stock price,
total market value, GVI (0.06), and EPS. Using stepwise regression to construct the
regression model containing only a small number of independent variables can improve the
predictive ability. (3) Using regression tree to establish the rate of return prediction models,
the GVI (0.125) is the best factor in term of predictive ability except for stock price. The
RMSE and error rate in the test period of integrating multiple regression trees is lower than
the single regression trees. (4) The results of weighted scoring stock selection model showed
that the combination of multiple factors can increase the rate of return, and the ability that
ROE weight increases the rate of return is much lower than the B/P and S/P factors. The
combination of multiple factors can reduce the risk, and the ability that ROE weight reduces
the risk is much higher than the other two factors. The combination of multiple factors can
increase the Sharpe Ratio. Especially, the effects of interactions, B/P*ROE and ROE*S/P,
were very obvious.
|
|---|
