Do Technical Indicators Improve Deep Learning Forecasts? An Empirical Ablation Study Across Asset Classes

  1. Nabeel Saidd

Reviewed by

Read the full article See related articles

Discuss this preprint

Start a discussion What are Sciety discussions?

Listed in

Log in to save this article

Abstract

Technical indicators derived from historical price data have long been central to quantitative trading strategies, yet their actual contribution to modern deep learning forecasting models remains an open empirical question. This study presents a large-scale ablation analysis examining whether technical indicators improve next-day price prediction when used as inputs to recurrent neural networks. We conduct 500 controlled experiments across 10 assets spanning five asset classes—commodities (Crude Oil, Gold), cryptocurrencies (Bitcoin, Ethereum), equities (Apple, Microsoft), foreign exchange (EUR/USD, USD/JPY), and market indices (NASDAQ, S&P 500)—using daily OHLCV data from 2010 to 2025. Five feature configurations are evaluated: a raw OHLCV baseline and four indicator-augmented variants incorporating momentum (RSI, Stochastic Oscillator), trend (SMA, EMA, ADX, MACD), volatility (ATR, Bollinger Bands), and a combined all-indicator set. Each configuration is tested with both LSTM and GRU architectures across five random seeds to ensure statistical robustness. Our results show that technical indicators do not improve—and frequently degrade— forecasting performance relative to raw price data. The baseline OHLCV configuration achieves the lowest mean RMSE (0.166 ± 0.148) and highest mean directional accuracy (55.7% ± 5.5%). Every indicator-augmented configuration produces higher prediction error, with the comprehensive all-indicators variant exhibiting statistically significant degradation (34.6% RMSE increase, p < 0.001, Cohen’s d = −0.29). All four indicator categories show significant performance reduction at α = 0.05. GRU models achieve marginally higher directional accuracy than LSTM (55.3% vs. 51.0%), although RMSE differences between the two architectures are not statistically significant (p = 0.846). Foreign exchange stands out as the only asset class where volatility indicators improve performance (4.2% RMSE reduction), while high-volatility assets (cryptocurrencies, commodities) exhibit 83% higher mean prediction error than their low-volatility counterparts. These findings suggest that deep recurrent architectures implicitly learn the patterns captured by conventional technical indicators, making explicit indicator features redun- dant or even harmful. The results carry practical implications for feature engineering in neural network-based trading systems and highlight the importance of rigorous baseline comparisons in applied financial machine learning.

Article activity feed

  1. PREreview

    This Zenodo record is a permanently preserved version of a PREreview. You can view the complete PREreview at https://prereview.org/reviews/18778665.

    Summary of the Research

    This paper looks at how energy companies communicate their ESG information in two places, the European Union and China. The author studies how companies choose between global reporting rules like GRI and SASB, and local rules like the EU CSRD or China's CSDS. The study shows that using global standards does not help companies much anymore because almost everyone uses them. The paper also shows that following strict local rules can improve company performance because it builds trust with local regulators and communities. The author also finds that using too many reporting standards at the same time can confuse investors and hurt company value.

    Overall, the paper helps explain how companies can balance investor needs with local expectations. It also introduces the idea that companies should aim for simple and smart reporting instead of using as many standards as possible.

    Major Issues

    1. The paper uses very complex language and ideas The study is interesting, but the writing is very dense. Readers who are not familiar with ESG rules or economic theory may struggle to follow the argument.

    2. The method for coding disclosure choices needs clearer explanation The author says they used manual coding of company reports. It would help to explain more about how this was done and how accuracy was checked.

    3. The sample is small Only 25 companies were studied. This limits how much we can generalize the results. The author notes this, but it is still an important concern.

    4. The study focuses only on the energy sector The findings may not apply to other industries that face different rules or market pressures.

    5. The idea of a complexity penalty needs stronger evidence The inverted U shape is interesting, but more explanation is needed on why the turning point occurs where it does and whether it holds across different types of firms.

    Minor Issues

    1. Some graphs and tables are hard to read A simpler layout or clearer labeling would help readers understand the results more quickly.

    2. The introduction is long The first few pages include many references and details that could be shortened to improve flow.

    3. Some terms could be defined earlier For example, terms like strategic interoperability or double materiality may be new to some readers.

    4. The paper repeats similar points about signaling and legitimacy Combining these sections could reduce repetition.

    5. A few sentences are too long Shorter sentences would improve clarity and make the paper easier to read.

    Competing interests

    The author declares that they have no competing interests.

    Use of Artificial Intelligence (AI)

    The author declares that they did not use generative AI to come up with new ideas for their review.

  2. Version published to 10.20944/preprints202602.0692.v1