Paper: arXiv 2511.00665
Authors: José Ángel Islas Anguiano, Andrés García-Medina
Abstract
In this note, we compare Bitcoin trading performance using two machine learning models-Light Gradient Boosting Machine (LightGBM) and Long Short-Term Memory (LSTM)-and two technical analysis-based strategies: Exponential Moving Average (EMA) crossover and a combination of Moving Average Convergence/Divergence with the Average Directional Index (MACD+ADX). The objective is to evaluate how trading signals can be used to maximize profits in the Bitcoin market. This comparison was motivated by the U.S. Securities and Exchange Commission’s (SEC) approval of the first spot Bitcoin exchange-traded funds (ETFs) on 2024-01-10. Our results show that the LSTM model achieved a cumulative return of approximately 65.23% in under a year, significantly outperforming LightGBM, the EMA and MACD+ADX strategies, as well as the baseline buy-and-hold. This study highlights the potential for deeper integration of machine learning and technical analysis in the rapidly evolving cryptocurrency landscape.
Complexity vs Empirical Score
- Math Complexity: 5.0/10
- Empirical Rigor: 6.0/10 (capped from 7.0: transaction costs only partially modelled; see penalty flags)
- Quadrant: Holy Grail — high math complexity, high empirical rigor (effective empirical rigor capped from 7.0)
Why this score: The paper employs advanced machine learning models (LSTM, LightGBM) requiring solid technical understanding, but lacks detailed derivations or heavy formulaic proofs, placing it in the mid-range for math complexity. Empirical rigor is relatively high due to the use of real Bitcoin data, explicit backtest periods, and specific performance metrics, though the absence of full reproducibility code or extensive statistical robustness checks prevents a top score.
Research Flowchart
flowchart TD A["Research Goal: Compare ML vs TA for Bitcoin Trading"] --> B["Key Methodology<br>LightGBM vs LSTM vs EMA vs MACD+ADX"] B --> C["Data Source<br>Bitcoin Price Data (Pre/Post ETF)"] C --> D["Computational Process<br>Model Training & Backtesting"] D --> E["Key Finding: LSTM Achieved 65.23% Return<br>Outperformed all others & Buy-and-Hold"]
Evidence Card
What the authors claim, on what data, and who has checked it. Read from the paper's own text by a language model: every quote is verbatim, nothing was re-run. How to read this card
Proposes
The paper proposes to compare Bitcoin trading performance of two machine‑learning models (LightGBM and LSTM) with two technical‑analysis strategies (EMA crossover and MACD+ADX).
Source sentence
In this note, we compare Bitcoin trading performance using two machine learning models-Light Gradient Boosting Machine (LightGBM) and Long Short-Term Memory (LSTM)-and two technical analysis-based strategies: Exponential Moving Average (EMA) crossover and a combination of Moving Average Convergence/Divergence with the
Assumes 6
- A fair comparison is made under realistic market conditions. execution
Source sentence
To ensure a fair comparison under realistic market conditions, both approaches are integrated into a statistical learning framework in which model parameters are optimized in-sample and evaluated out-of-sample.
- Technical‑analysis strategies do not require a warm‑up window, while ML models need an initial 95‑day window to generate forecasts. model
Source sentence
the former does not require a window period to begin generating predictions, whereas the latter necessitates an initial 95-day window period to produce forecasts on the testing data.
- Trading strategies are evaluated without transaction costs or slippage in the headline results. execution
Source sentence
It is important to mention that these trading strategies do not include transaction costs (brokerage fees, slippage, etc.), which significantly impact real‑world performance.
- Price data from Binance are accurate and represent the market. data
Source sentence
Raw price data were downloaded from the free Binance source.
Show 2 more
- The training period is 2021‑01‑19 to 2024‑01‑09 and the test period is 2024‑01‑10 to 2024‑12‑31. data
Source sentence
The training dataset consists of price data from 2021-01-10, to 2024-01-09, while the testing dataset covers the period beginning with the approval of the first spot Bitcoin exchange‑traded funds (ETFs)
- No risk‑management features such as stop‑loss orders are used. execution
Source sentence
Neither uses risk management features like stop‑loss orders.
Data
2024-01 – 2024-12 · Bitcoin · crypto · daily
- Bitcoin price data market-prices public
Source sentence
Raw price data were downloaded from the free Binance source4. As mentioned in the Introduction, the study period extends from January 19, 2021, to December 31, 2024.
Costs
Partial
A flat 0.1% fee per trade is applied in a robustness analysis (Table 3).
Source sentence
Table 3 summarizes the numbers of trades, the numbers of winning trades, the loss trades, and the profits using a 0.1% fee for each trade.
Finds 1
total return 65.2%
- After accounting for a 0.1% transaction fee per trade, only LSTM still outperformed buy‑and‑hold (53.23% vs 42.31%). performance
Source sentence
Interestingly, the only strategy that outperformed buy-and-hold after accounting for transaction fees was the LSTM.
Checked by others
No other paper in the archive has been found to support or contradict these findings (compared with 8 related papers). Coverage is partial, so this is neither replication nor refutation.
Code
No repository by the authors found in the paper.
Extracted 2026-11-14 · same card as JSON · A language model chose these items: check the source sentence before relying on one.
Cited References (33)
- Markowitz meets technical analysis: Building optimal portfolios by exploiting information in trend-following signals (2022)
- Technical trading rules in the cryptocurrency market (2020)
- LSTM–GARCH hybrid model for the prediction of volatility in cryptocurrency portfolios (2024)
- What drives bitcoin? An approach from continuous local transfer entropy and deep learning classification models (2021)
- On the convergence of adam and beyond (2019) — link
- Implementing deep neural networks for financial market prediction on the Intel Xeon Phi (2015)
- Deep learning with long short-term memory networks for financial market predictions (2018)
- Deep learning (2016)
and 25 more
- Quantitative Trading: How to Build Your Own Algorithmic Trading Business (2021)
- Long short-term memory (1997)
- Technical analysis strategy optimization using a machine learning approach in stock market indices (2021)
- A systematic review of fundamental and technical analysis of stock market predictions (2020)
- Stock market forecasting using deep learning and technical analysis: a systematic review (2020)
- Technical analysis and the London stock exchange: testing the MACD and RSI rules using the FT30 (2008)
- A comparative study of technical trading strategies using a genetic algorithm (2020)
- Revisiting the Performance of MACD and RSI Oscillators (2014)
- Application of neural network to technical analysis of stock market prediction (2022)
- Cryptocurrency trading: a comprehensive survey (2022)
- Deep learning and technical analysis in cryptocurrency market (2023)
- Optimization of Cryptocurrency Algorithmic Trading Strategies Using the Decomposition Approach (2023)
- Bitcoin: Medium of exchange or speculative assets? (2018)
- Is Bitcoin a Commodity? On price jumps, demand shocks, and certainty of supply (2019)
- Is Bitcoin a currency, a technology-based product, or something else? (2020)
- The bitcoin as a virtual commodity: Empirical evidence and implications (2020)
- Bitcoin as a digital commodity (2022)
- Forecasting the volatility of stock price index: A hybrid model integrating LSTM with multiple GARCH-type models (2018)
- The intelligent investor (2003)
- Lightgbm: A highly efficient gradient boosting decision tree (2017)
- Greedy function approximation: a gradient boosting machine (2001)
- An introduction to statistical learning: with applications in R (2013)
- The elements of statistical learning: data mining, inference, and prediction (2009)
- The M5 uncertainty competition: Results, findings and conclusions (2022)
- M5 accuracy competition: Results, findings, and conclusions (2022)
0 of 33 cited works are in the corpus; resolved via arXiv source bundles, Semantic Scholar and OpenAlex. Corpus-wide: 2596 linked papers.