Performance & Methodology
How to interpret NeuralTrend signals and results
NeuralTrend is designed as a market-research and paper-simulation platform. This page explains the information customers need to evaluate its signals, historical results, assumptions, and limitations—without disclosing the proprietary model recipe.
318
publicly listed assets
303
crypto assets
15
traditional assets
Daily
signal frequency
Historical simulations are not a live brokerage record
Results shown on NeuralTrend are historical model simulations. They are not brokerage statements, real-money results, or guarantees of future performance. A separately labeled live-since-launch record will be introduced as sufficient forward results accumulate.
Illustrative public assets
Recent one-year historical performance snapshot
A consistent one-year window is shown for the four public crypto assets rather than selecting the strongest horizon for each asset. Results include the fixed transaction-cost assumptions described below and exclude slippage.
| Asset | Coverage | AI strategy | Buy & Hold | Return spread | AI max drawdown | B&H max drawdown | Sharpe | Trades | Exposure | Data freshness | Latest signal |
|---|---|---|---|---|---|---|---|---|---|---|---|
| BTC-USD 366 observations | 2025-07-24 – 2026-07-24 | -14.4% | -45.3% | +30.9 pts | -28.7% | -53.1% | -0.55 | 3 | 49.2% | Current Through 2026-07-24 | HOLD |
| ETH-USD 366 observations | 2025-07-24 – 2026-07-24 | -10.8% | -49.7% | +38.9 pts | -35.4% | -67.6% | -0.24 | 1 | 36.6% | Current Through 2026-07-24 | BUY |
| SOL-USD 366 observations | 2025-07-24 – 2026-07-24 | +12.3% | -58.9% | +71.3 pts | -9.0% | -74.9% | 0.72 | 1 | 11.7% | Current Through 2026-07-24 | HOLD |
| XRP-USD 366 observations | 2025-07-24 – 2026-07-24 | -12.0% | -65.0% | +53.0 pts | -14.8% | -68.7% | -0.60 | 4 | 19.1% | Current Through 2026-07-24 | SELL |
Generated from the currently deployed data files. Coverage and data-through dates are shown per asset.
How signals work
Completed data becomes a BUY, HOLD, or SELL state
NeuralTrend intentionally avoids using incomplete same-day market information when producing the daily signal shown to users.
1
Completed market data
Daily price and volume data are collected after a market day is complete. Technical and calendar indicators are derived from that information.
2
Model evaluation
The asset’s model evaluates information available through the previous completed day. It does not use today’s high, low, close, or volume.
3
Published signal
The resulting BUY, HOLD, or SELL state is associated with the current signal date and is used by the site’s historical and paper simulations.
Data freshness
Each signal shows the completed data it is based on
NeuralTrend displays a data-through date, the time the deployed site data file was updated, and a simple freshness state. These indicators describe the website data available to the model; they are not real-time exchange timestamps or trade confirmations.
Within the expected cadence
Crypto is checked against calendar days. Stocks use a weekday-aware check so weekends do not automatically make data stale.
Later than usual
The latest completed row is one update beyond the normal tolerance. Users should verify the data-through date before acting.
Materially behind
The latest completed row is several expected updates behind. The signal remains visible for transparency but should not be treated as current.
Stock-market holidays are not embedded in the lightweight freshness calculation, so an exchange holiday can occasionally make a current stock file appear delayed.
Simulation assumptions
What is included in displayed results
Execution price
Historical and current paper simulations use the daily closing price stored for the relevant signal date. They are not real-time exchange executions.
Transaction costs
The current fixed assumptions are 1.00% per executed crypto side and 0.10% per executed stock side. Actual venue costs may differ.
Open positions
Positions still open at the end of a selected horizon are marked to market. No artificial final SELL or exit fee is added without a SELL signal.
Buy & Hold benchmark
The benchmark uses the same period and includes an entry cost. Residual cash is retained when a whole-share stock purchase cannot invest every dollar.
Not modeled separately
Slippage, bid–ask spread variation, DEX gas costs, liquidity-dependent price impact, market impact, taxes, borrow costs, and exchange-specific fee tiers are not modeled separately. Actual trading results can differ materially.
Evidence labels
Historical results and future live records serve different purposes
Historical strategy simulation
Shows how the signal series would have behaved under the website’s stated assumptions. The complete chart can include periods used during model development as well as later evaluation and production inference.
Live-since-launch performance
Will track signals and model versions forward from a defined publication date without later rewriting the record. It will be displayed separately once enough observations have accumulated to be meaningful.
Model maintenance
Monitoring, retraining, and historical revisions
Model quality is monitored regularly. Individual asset models are periodically retrained when performance and data-quality checks indicate that an update is appropriate. This process is currently operator-reviewed and is planned to become more automated over time.
Asset-specific updates
Each asset can be updated independently. A model may remain deployed for weeks, months, or longer depending on its observed behavior and available data.
Published history
Previously published signal rows are normally preserved when new daily signals are appended. Internal release checks flag unexpected historical changes.
Material revisions
If a correction or data cleanup changes earlier signals, NeuralTrend will identify the affected asset, period, reason, and scope in the public change log.
Technical methodology Optional details for users who want a deeper explanation
Inputs and model scope
The independent market inputs are daily open, high, low, close, and volume observations. NeuralTrend derives technical and calendar indicators from those inputs. Each currently supported asset uses its own model.
Chronological evaluation
During model development, observations are ordered by time and divided into 60% training, 20% validation, and 20% testing partitions. The current process uses a fixed chronological holdout rather than continuous walk-forward testing.
NeuralTrend does not publish the complete feature list, model architecture, optimization search space, label construction, retraining thresholds, or source code. Those elements form part of the proprietary product methodology.
Public disclosure
Model and signal-history change log
Material methodology changes, historical-signal revisions, and data corrections are recorded here. Routine appending of a new daily signal does not require an entry.
Public methodology and signal-history disclosure introduced
NeuralTrend added a public methodology page, clarified the difference between historical simulations and a future immutable live track record, and introduced a public change-log policy. This website release did not rewrite existing market CSV signal history.
- Historical signals changed
- No
- Affected period
- None
Limitations
Use the results as research evidence—not a promise
- Historical performance does not guarantee future performance.
- A fixed historical split is not the same as continuous walk-forward validation.
- The website does not yet provide real-time execution or an immutable live record.
- Upstream data providers may revise source data or discontinue assets.
- NeuralTrend does not remove assets solely because their model performed poorly.