IDX AI Trading System
Full-stack AI trading platform for the Indonesia Stock Exchange — FastAPI backend, 13-page Streamlit dashboard, IMSS multi-agent market simulator, ML prediction pipelines, and research workspaces with Playwright E2E coverage.
Evidence & Verification
Metrics and claims on this page are tied to the linked artifacts below (repository docs, experiment outputs, and deployment pages when available).
Tags
Technologies
Architecture
Financial ML Pipeline
Multi-model comparison for stock price prediction
LSTM
SVM
KNN
Ensemble
Model Performance (MAPE)
Overview
IDX AI Trading System consolidates prior finance work from financial_eng, quant-trading-id, and Predictive-Analytics-for-Indonesia-s-Economic-Forecasting into a unified full-stack trading platform. Unlike the earlier ML-only experiments, this is a production-oriented system with a FastAPI backend, Streamlit dashboard with 12 feature pages, multi-agent market simulation (IMSS), and research workspaces — designed for daily IDX equity operations.
Architecture
The system runs as a dual-service architecture: a FastAPI backend handling data, analysis, and ML inference, paired with a Streamlit multipage dashboard for interactive operations. Data persists in SQLite (trading.db, imss.db), with the IMSS simulator running as a self-contained package with background job lifecycle management.
Daily Jobs + SQLite (trading.db, imss.db)
| |
v v
FastAPI Route Layer IMSS Simulation Engine
| |
+----------> Streamlit Dashboard <----------+
(12 pages)
Dashboard (13 Pages)
The Streamlit dashboard provides a complete operator interface:
- Home — system health, data freshness, operator overview
- Screener — filter-driven stock screening and signal scan workflows
- Stock Detail — chart, technical, sentiment, flow, and risk views per symbol
- Sentiment — sentiment fetch, cleanup, and thematic analysis
- Virtual Trading — paper-trading simulation sessions with order/replay controls
- Settings — model operations, training readiness, artifact management
- Market Overview — market breadth, leaders, and flow screener surfaces
- ML Prediction — prediction, correlation, Monte Carlo, and technical overlays
- Backtesting — async backtest launch, polling, and results visualization
- Research Presets — promote accepted research candidates into durable presets
- Portfolio — positions, trade history filters, and reconciliation analytics
- Market Enrichment — market-to-sector-to-symbol enrichment workflow
- IMSS — background simulation run launch, monitoring, logs, and summary inspection
API Backend (15+ Route Groups)
The FastAPI backend exposes structured endpoints for:
/health— system health, freshness, manual update hooks/stocks— symbols, snapshots, charts, foreign flow, broker datasets/analysis,/fundamental— technical, signal, risk, and LLM-assisted analysis/signals— scan and active signal feeds/portfolio— summary, positions, and trade history/simulation— paper-trading session lifecycle and metrics/prediction— training, artifact management, inference, correlation, Monte Carlo/backtest— synchronous runs and background backtest jobs/market-enrichment— foreign/domestic flows, industries, brokers, symbol diagnostics/imss— background IMSS run lifecycle (create, list, status, logs, summary)
IMSS: Multi-Agent Market Simulator
The IDX Market Swarm Simulator (IMSS) is a self-contained multi-agent simulation package. It models market participant behavior through agent archetypes, runs background simulation jobs with lifecycle management APIs and WebSocket streams, and provides summary inspection and log analysis through the dashboard. This enables strategy stress-testing against simulated market dynamics beyond historical replay.
ML Pipeline
Signal generation uses a BiLSTM + SVR + GradientBoosting ensemble with multi-seed variance reduction for short-horizon IDX equity prediction. The ensemble combines sequential pattern detection (BiLSTM via TensorFlow/Keras), support vector regression, and gradient boosting for robustness across market regimes. Backtesting implements walk-forward validation with Monte Carlo uncertainty quantification using volatility-adjusted features. Sentiment analysis integrates news and social media signals as exogenous features alongside technical indicators.
Research Workspaces
Three automated research tools feed discoveries back into the main system:
- autoscreener — automated stock screening with configurable filters
- automontecarlo — Monte Carlo simulation batches for strategy validation
- autoresearch — systematic research pipeline with preset promotion flow
Testing
Python unit/integration tests plus Playwright E2E coverage across dashboard pages.
Historical Context
Earlier work included crypto forecasting (Bitcoin, XRP, ETH) and Indonesia macro-forecasting with 22+ economic indicators (R² = 0.933). Those experiments informed the consolidated architecture and evaluation strategy.