Research on value prediction and investment model based on machine learning
Download as PDF
DOI: 10.23977/FEIM2022.024
Author(s)
Yutong Dai, Hongyuan Zhang
Corresponding Author
Yutong Dai
ABSTRACT
In investment markets, high returns can be accompanied by high risks. Making more efficient use of limited information and making the best portfolio strategy is an eternal topic in the investment market. In this paper, by using the daily closing price information up to that day, a model is established to predict that day's price and make the best portfolio of gold and bitcoin according to the balance of returns and risks. First, we use the SVR prediction model to obtain information from price flow data to predict the closing price of the investment day. On this basis, we use the optimal portfolio decision model based on multi-objective programming to select the optimal portfolio by weighted distance under the careful consideration of investment return, potential maximum investment loss, and risk coefficient. According to our model, we made a total of 613 buy and sell adjustments between 2016 and 2021. Moreover, our investment strategies contribute to the appreciation of the initial principal's value from $1000 to $14,357.607 after five years. Second, from the aspects of the forecast model's accuracy and anti-interference ability, return-risk balanced viewpoint, we prove that our model provides the optimal strategy. Moreover, a disturbance verification test is made to test the strategy, and some detailed examples prove that our model of decision-making can still have a considerable income within 20% of the interference.
KEYWORDS
SVR prediction model, optimal portfolio decision model, multi-objective programming