Author(s)

Ziqi Wang

Corresponding Author

Ziqi Wang

ABSTRACT

Quantitative investment is an investment strategy based on machine learning, which uses a specific mathematical model to find and obtain excess returns from historical data. From September 11, 2017, to September 10, 2021, the price of gold and bitcoin fluctuated, containing many attractive value investment opportunities. This paper considers the value increment, commission cost, opportunity cost, and transaction risk and discusses maximizing the value increment. To begin with, according to the historical price data of the last 50 days so far, we predict the price rise and fall of the next day based on the idea of Time Series Analysis. For gold, the price trend is found to be non-stationary time series, and the difference method is used to eliminate the trend so that it meets the condition of stability, and then ARIMA Model is used for price prediction. A three-level neural network structure, including input, hidden, and output, is designed for bitcoin. The Back Propagation Neural Network method is used to predict the price of the next day, and then the price of each day is predicted by cyclic statements. After testing, the price prediction accuracy of gold and bitcoin has reached about 99%. Next, aiming at maximizing the total return of assets, an investment model is established from the perspective of Dynamic Programming according to the predicted price of gold and bitcoin the next day, considering transaction commission, opportunity cost, and potential risk. Finally, according to the daily trading decision model, we calculated that the quantitative investment of $1000 on September 16, 2016, would be worth $305,579.152 on September 10, 2021, with an annual return of 314.69%.

KEYWORDS

ARIMA Model, Back Propagation Neural Network Dynamic Programming