MarketForge AI 0.5.0 — Research Workbench and Prospective Benchmark v3
Pre-release
MarketForge AI 0.5.0
MarketForge AI 0.5.0 expands the project from a forecasting and backtesting application into a broader local-first financial research workbench.
This release adds direct public market-data imports, additional transparent forecasting models, multi-asset portfolio analysis, market-regime classification, volatility forecasting, stress testing, saved research projects, experiment tracking, report generation, accessibility improvements and prospective frozen benchmark v3.
Research and educational software only.
MarketForge AI does not provide financial advice, guaranteed predictions, automated trading signals or promises of profit.
Major New Features
Public Market-Data Connectors
MarketForge can now import public historical candle data directly from:
- Binance Spot
- Coinbase Exchange
- Kraken Spot
These connectors:
- Use public market-data endpoints
- Do not require exchange API keys
- Do not access trading accounts
- Do not place orders
- Pass imported data through the MarketForge validation pipeline
Additional Forecasting Models
MarketForge now includes eight transparent forecasting choices:
- Naive persistence
- Robust drift
- Moving-block bootstrap
- Original ensemble
- Exponential smoothing
- Momentum
- Mean reversion
- Regime-aware ensemble
The new regime-aware ensemble changes its model mixture using recent market conditions such as trend, momentum, volatility and regime confidence.
Forecast Calibration
Forecast intervals can now use:
- No calibration
- Empirical calibration
- Conformal-style widening
Calibration information is recorded in forecast metadata so adjustments remain visible and auditable.
Multi-Asset Portfolio Analysis
The new Portfolio workspace can analyse several market datasets together.
It includes:
- Timestamp alignment
- Return correlations
- Covariance analysis
- Individual asset volatility
- Portfolio equity simulation
- Rebalancing
- Turnover
- Maximum drawdown
- Diversification measurements
Supported allocation methods include:
- Equal weight
- Inverse volatility
- Risk parity
- Minimum variance
Users can configure allocation limits, rebalancing frequency, transaction costs, initial capital and optional target volatility.
Market-Regime Classification
MarketForge can classify recent market conditions such as:
- Bullish trend
- Bearish trend
- Sideways market
- High-volatility market
- Low-volatility market
- Uncertain or transitional regime
The system returns the evidence behind each classification rather than displaying only a label.
Volatility Forecasting
The Research Lab now includes:
- Exponentially weighted volatility
- Parkinson high-low volatility
- Garman–Klass OHLC volatility
- Combined volatility estimates
- Horizon-adjusted volatility
- Volatility percentiles
- Risk-regime descriptions
Scenario Stress Testing
Users can test hypothetical market conditions using:
- Configurable price shocks
- Volatility multipliers
- Volume shocks
- Reproducible random seeds
- Adjustable confidence levels
- Multiple simulated scenarios
Stress-test results include:
- Median outcome
- Best and worst scenarios
- Loss probability
- Value at risk
- Expected shortfall
- Distribution quantiles
Saved Research Projects
MarketForge now includes a local SQLite research workspace.
Users can store:
- Project names and descriptions
- Dataset fingerprints
- Forecast settings
- Experiment settings
- Metrics
- Result data
- Tags
- Result hashes
- Creation dates
The database is stored locally at:
storage/marketforge.db