Using a Deep Learning Model to Simulate Human Stock Trader's Methods of Chart Analysis

TL;DR

The paper tackles stock-market predictability by simulating human chart analysis with a deep learning model. It introduces a $5\times600$ OHLCV input processed by a modified $1$D ResNet with skip connections to classify price movements into three labels defined by a $10\%$ or $20\%$ move within a horizon $D$ days, using thresholding on the top softmax logit to select trades. Training occurs on data from 2006–2015 with validation 2016–2019, and backtesting on 2020–2022 across Korea and US markets demonstrates notable outperformance against benchmarks, especially for Korea, while US results are mixed and sensitive to market conditions. The work claims novelty in leveraging softmax logits for trading decisions and provides a comprehensive backtest-based evaluation across market regimes, though it acknowledges limitations such as slippage, survivorship bias, and overfitting risk when extrapolating to real-world trading.

Abstract

Despite the efficient market hypothesis, many studies suggest the existence of inefficiencies in the stock market leading to the development of techniques to gain above-market returns. Systematic trading has undergone significant advances in recent decades with deep learning schemes emerging as a powerful tool for analyzing and predicting market behavior. In this paper, a method is proposed that is inspired by how professional technical analysts trade. This scheme looks at stock prices of the previous 600 days and predicts whether the stock price will rise or fall 10% or 20% within the next D days. The proposed method uses the Resnet's (a deep learning model) skip connections and logits to increase the probability of the prediction. The model was trained and tested using historical data from both the Korea and US stock markets. The backtest is done using the data from 2020 to 2022. Using the proposed method for the Korea market it gave return of 75.36% having Sharpe ratio of 1.57, which far exceeds the market return by 36% and 0.61, respectively. On the US market it gives total return of 27.17% with Sharpe ratio of 0.61, which outperforms other benchmarks such as NASDAQ, S&P500, DOW JONES index by 17.69% and 0.27, respectively.

Figures (14)

• Figure 1: Model architecture of wang2017time
• Figure 2: the modified model architecture used in this paper
• Figure 3: Box Plot of Sharpe Ratio with different thresholds during validation period using KR Best Model
• Figure 4: Box Plot of Sharpe Ratio with different thresholds during validation period using US Best Model
• Figure 5: Barplot of Quarterly return of KR market during the test period