Stock Price Prediction With Long Short-Term Memory Recurrent Neural Network

Abstract

In this paper, we investigate the prediction of daily stock prices of the top five companies in the Thai SET50 index. A Recurrent Neural Network (RNN) with Long Short-Term Memory (LSTM) is applied to forecast the next daily stock price (High, Low, Open, Close). Deep Belief Network (DBN) is applied to compare the result with LSTM. The test data are CPALL, SCB, SCC, KBANK, and PTT from the SET50 index. The purpose of selecting these five stocks is to compare how the model performs in different stocks with various volatility. There are two experiments of five stocks from the SET50 index. The first experiment compared the MAPE with different length of training data. The experiment is conducted by using training data for one, three, and five-year. PTT and SCC stock give the lowest median value of MAPE error for five-year training data. KBANK, SCB, and CPALL stock give the lowest median value of MAPE error for one-year training data. In the second experiment, the number of looks back and input are varied. The result with one look back and four inputs gives the best performance for stock price prediction. By comparing different technique, the result show that LSTM give the best performance with CPALL, SCB, and KTB with less than 2% error. DBN give the best performance with PTT and SCC with less than 2% error.

Cite this work

No public copy is stored for this record, so only the metadata is cited.

Stored only when both papers are in the corpus — 24 citations exist upstream.

Papers that cite the same references — the relational fallback while embeddings are still backfilling.