Do candlestick charts suffice? The incremental economic value of technical indicators in stock return prediction

Although technical analysis remains widely used in practice, the need for manually constructed indicators is increasingly questioned as deep learning models can extract features directly from OHLC price data. This paper examines the standalone economic value of trading volume and manually constructed technical-indicator groups and whether adding these groups to OHLC candlestick charts provides incremental economic value. Using daily data for 1707 Chinese A-share stocks from 2013 to 2025, we construct approximately 3.36 million stock-window samples and encode them as multi-channel images containing candlestick charts, trading volume, and seven widely used manually constructed technical indicators. The primary analysis uses a 2D CNN on these images; complementary tests apply a logistic regression, a 1D CNN, and an LSTM to numerical representations of the same samples. Out-of-sample performance from 2021 to 2025 is evaluated using decile portfolios, Sharpe ratios, Fama–French alphas, and formal pairwise tests. Our empirical results yield two central findings. First, specifications based separately on OHLC, trading volume, and each technical-indicator group produce significantly positive high-minus-low returns and factor-adjusted alphas, together with economically meaningful Sharpe ratios; the OHLC-only specifications generally perform best under the nonlinear models. Second, adding these technical-indicator groups to OHLC does not systematically improve any of these performance measures in our setting. Together, these results provide the first large-scale evidence from the Chinese A-share market on the standalone and incremental economic value of these technical-indicator groups across multiple representations and model architectures.