---
tags:
- time-series
- forecasting
- temporal
- trading
- btc-usd
- ensemble
library_name: temporal-forecasting
---

# Temporal Trading Model: BTC-USD (Ensemble Member 3/5: balanced (lookback=90))

This is a pre-trained Temporal transformer model for time series forecasting of BTC-USD. This model is part of a consensus ensemble.

## Ensemble Context

This model is **member 3 of 5** in a consensus ensemble for BTC-USD (1d).

The ensemble uses 5 models with different lookback periods and focus strategies to generate diverse forecasts. These forecasts are then aggregated using multiple consensus strategies (gradient, confidence, timeframe, volatility, mean reversion, acceleration, swing, risk-adjusted) to produce robust trading signals.

### All Ensemble Members

1. [momentum (lookback=30)](../momentum_lookback30)
2. [balanced (lookback=60)](../balanced_lookback60)
3. **balanced** (lookback=90) - *This model*
4. [mean_reversion (lookback=60)](../mean_reversion_lookback60)
5. [momentum (lookback=45)](../momentum_lookback45)

### How the Ensemble Works

1. Each member model generates independent forecasts using its specific lookback window and focus
2. Multiple consensus strategies analyze the ensemble's forecasts from different perspectives
3. Each strategy produces action recommendations (BUY/SELL/HOLD) with confidence scores
4. Final consensus aggregates all strategy recommendations into a unified trading signal


## Model Details

- **Symbol**: BTC-USD
- **Interval**: 1d
- **Lookback Window**: 90 periods
- **Forecast Horizon**: 7 periods
- **Focus**: balanced
- **Training Date**: 2025-11-09 20:27 UTC
- **Training Epochs**: 21
- **Best Validation Loss**: 0.29258017241954803

## Training Dataset

- **Date Range**: 2020-12-30T00:00:00 to 2025-11-08
- **Training Samples**: 1146
- **Lookback Period**: 90 1d intervals

## Model Architecture

- **Model Type**: Temporal Transformer
- **d_model**: 512
- **Encoder Layers**: None
- **Decoder Layers**: None

## Features

Input features used: 17 features

## Usage

```python
from strategies.model_cache import get_model_cache

# Import from HuggingFace
cache = get_model_cache()
model_path, scaler_path, metadata = cache.import_from_huggingface(
    repo_id="YOUR_REPO_ID",
    symbol="BTC-USD",
    interval="1d",
    lookback=90,
    focus="balanced",
    forecast_horizon=7
)
```

## License

GPL-3.0-or-later

## Citation

```
@software{temporal_trading_model,
  title = {Temporal Trading Model: BTC-USD (balanced)},
  author = {Unidatum Integrated Products LLC},
  year = {2025},
  url = {https://github.com/OptimalMatch/temporal-trading-agents}
}
```