Paper: arXiv 2410.21291
Authors: Angel Varela
Abstract
Predicting the stock market is a big challenge for the machine learning world. It is known how difficult it is to have accurate and consistent predictions with ML models. Some architectures are able to capture the movement of stocks but almost never are able to be launched to the production world. We present Achilles, with a classical architecture of LSTM(Long Short Term Memory) neural network this model is able to predict the Gold vs USD commodity. With the predictions minute-per-minute of this model we implemented a trading bot to run during 23 days of testing excluding weekends. At the end of the testing period we generated $1623.52 in profit with the methodology used. The results of our method demonstrate Machine Learning can successfully be implemented to predict the Gold vs USD commodity.
Complexity vs Empirical Score
- Math Complexity: 4.0/10
- Empirical Rigor: 6.0/10 (capped from 7.0: transaction costs not modelled; see penalty flags)
- Quadrant: Street Traders — empirical and applied, lower math complexity (effective empirical rigor capped from 7.0)
Why this score: The paper uses standard LSTM architecture and technical indicators (RSI/EMA) with minimal novel math, but includes a 23-day live trading bot test with specific profit figures, detailed data preprocessing, and operational implementation details.
Research Flowchart
flowchart TD
A["Research Goal<br>Can ML predict Gold vs USD<br>for profitable automated trading?"] --> B{"Data & Methodology"}
B --> C["Dataset: Gold vs USD<br>Minute-by-minute historical data"]
B --> D["Model Architecture<br>LSTM Neural Network"]
C --> E["Computational Process<br>Model Training & Minute-by-Minute Prediction"]
D --> E
E --> F["Trading Bot Implementation<br>23-day live test (weekdays only)"]
F --> G["Key Findings & Outcomes<br>Profit: $1623.52<br>ML successfully applied to commodity trading"]
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 Achilles, a simple LSTM‑based model and trading bot for minute‑by‑minute prediction of the Gold vs USD commodity.
Source sentence
We present Achilles, with a classical architecture of LSTM(Long Short Term Memory) neural network this model is able to predict the Gold vs USD commodity.
Assumes 4
- The trading experiments are run on a paper account without fees or commissions. execution
Source sentence
we do not have the fees and commissions real accounts could have.
- Predictions are generated minute‑by‑minute and the bot decides to invest based on each prediction. model
Source sentence
Our model will be predicting each minute and based on these predictions we’ll let our Trading Bot decide if to invest or not.
- A sliding‑window of the last 120 minutes is used to train the LSTM. data
Source sentence
We use N = 120 to use the last 120 minutes to train on the next minute.
- The inner loop fetches news data from specific websites and feeds it to FinBERT. data
Source sentence
The Inner loop is used to extract the data from different websites(Benzinga.com
Data
Gold vs USD · commodities · intraday
- ETH‑USD minute‑level price data market-prices unknown
Source sentence
The market we chose to train our model is ETH-USD (Ethereum), known by its volatility.
- Gold vs USD price data market-prices unknown
Source sentence
this model is able to predict the Gold vs USD commodity.
- News headlines from Benzinga, Investing.com, ft.com text-news unknown
Source sentence
The Inner loop is used to extract the data from different websites(Benzinga.com
Costs
Not modeled
Source sentence
we do not have the fees and commissions real accounts could have.
Finds 2
total return 162.3%
- The trading bot generated $1,623.52 profit over a 23‑day testing period starting from $1,000. performance
Source sentence
At the end of the testing period we generated $1623.52 in profit with the methodology used.
- The model has only 9.544 parameters, making it lightweight for training and inference. other
Source sentence
Note that our model only has 9.544 parameters, it is a small model
Checked by others
No other paper in the archive has been found to support or contradict these findings (compared with 15 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.