• Login
  • Register
  • Search

Best Investment Strategy: Prediction Based on AR-LSTM & Decision Based on SVM

Changhong Zhao, Kang Li


In volatile asset market transactions, traders always want to maximize the total returns. This paper performs day-by-day forecasting of closing prices, quantitatively analyzing the market using economic indicators, and proposes the best daily investment strategy that maximizes total returns. Firstly, we use Python’s TensorFlow library to build an AR-LSTM model to make day-by-day predictions of the closing prices of gold and bitcoin. Secondly, we use the greedy algorithm, support vector machines(SVM) combined with three economic indicators, a total of two methods to analyze the market situation. On the one hand, we rely on the predicted results and use the greedy idea to get the maximum total return . On the other hand, we established the Support vector machine optimize specifications model. Firstly, 15 indicators that can express the characteristics of market oscillation, trend, and energy are selected for factor analysis. Secondly, three public factors were identified. The calculated factor scores were then fed into a SVM, which, combined with the linear programming results, can output one of 2 (trade) or 1 (continue to hold). Thirdly, the three technical indicators, EMA, RSI, and MACD, are calculated by applyin g the predicted closing price. Finally, the indicators are linearly weighted, and the weighted results are re-operated with the output of the support vector machine to obtain the trading intention for the day. Then, we use Monte Carlo and grid search methods to adjust the four parameters in the model, w1, w2, r, uplimit. We use grid search, fix w1, w2 and adjust r and uplimit. Subsequently, we fixed r, uplimit, and used the Monte Carlo method to adjust the size of w1, w2 and determined the optimal daily investment strategy.


AR-LSTM ; Investment Decision; SVM ; Factor Analysis; Parameter Optimization

Full Text:


Included Database


Liu L, Lou L, Liu XJ, and Shi SZ. Comparative study of comex gold futures price forecasting based on lstm. Journal of Changchun University of Science and Technology: Natural Science Edition, 44(2):7, 2021.

Fang PC. A machine learning based stock prediction and quantitative investment system. PhD thesis, Zhejiang University, 2018.

Zhou H. Quantitative investment model and practice based on dual technical indicators. PhD thesis, South China University, 2019.

DOI: http://dx.doi.org/10.18686/fm.v7i3.5354