© 2025 Robert Costa. All rights reserved. This project is licensed under the GNU General Public License v3.0. For more details, see the LICENSE file. We offer an open-source public demo of our charting and trading system called CandleSage that is under development. In addition to the open-source version, we plan to provide a licensed version with finely tuned models for enhanced performance with a monthly subscription. See Rob's page with related learning resources.

Trading System Architectural Concepts

Our trading system is designed to support a broad range of strategies:

• Day Trading & Swing Trading:
  These strategies last from a few minutes up to a few days. They typically focus on futures or options to capture short-term market inefficiencies. The system generates quick, actionable signals and executes trades with pre-set stop losses.

• Long-Term Strategies:
  This component can include approaches like buying deep in-the-money options (or LEAPS) as a form of "equity replacement." While these strategies might be managed separately, ideally they form a component of overall portfolio optimization.

• Recurring Options Income Strategies:
  For trades that need an underlying stock position (e.g., the wheel strategy), our platform is designed to accommodate and integrate these strategies, effectively bridging short-term and long-term trading approaches.

Day and Swing Trading

Diagram 1: Overall Short-Term Trading System Architecture

This diagram shows the process flow from generating trading strategies based on technical signals, through executing trades with fixed stop losses in a producer consumer model with the queue of triggered signals flushed regularly, and lastly ongoing monitoring positions with dynamic exits. New strategies are placed on a rolling basis within the hardcoded daily limits and availability of capital. See Symbol Selection Strategies for symbol selection mechanisms. A set of profit allocation rules can be configured to determine how much accumulated capital should be re-used for new strategies, saved in a money market or similar risk-free account, or distributed externally as realized gain.

flowchart LR
    S[Symbol<br>Selection]
    A[Technical<br>Signals]
    B[Trading Strategy Generation]
    C[Trade Execution with<br>Fixed Stop Losses]
    D[Trade Monitoring<br>Technical Signals<br>with Dynamic Exits]

    S --> A
    A --> B
    B --> C
    C --> D
    D -- "New strategies restarted depending on daily limits" --> A

Loading

Diagram 2: Short-Term Trading Signal Generation Pipeline

This diagram breaks down the process of generating trading signals by combining a live data feed and historical data. Simple indicators (e.g., RSI, EMA crossover) produce basic calculated states, while more complex processing (via HMM and LSTM models) generates advanced states. Both feed into the overall trading signal monitoring system. See Symbol Selection Strategies for symbol choosing.

flowchart TD
    S[Symbol Selection]
    A[Live Data Feed]
    B[Historical Data]
    C[Calculated States<br>RSI, EMA Crossover]
    D[Complex States<br>HMM, LSTM]
    E[Trading Signal Monitoring]

    S --> A
    S --> B
    A --> C
    A --> D
    C --> E
    D --> E
    B -- "Training Models<br>& Backtesting" --> D

Loading

Long-Term Portfolio Management and Optimization

This system combines classical portfolio optimization with advanced AI-based. signal generation. The idea is to leverage historical data to calculate expected returns and risk (covariance) while simultaneously incorporating real-time signals to dynamically adjust the portfolio. This dual approach aims to improve long-term portfolio construction and management. See Symbol Selection Strategies for symbol choosing.

System Components

1. Historical Data Pipeline:

   • Historical Price Data: Ingest historical prices.
   • Data Preprocessing & Cleaning: Prepare the data for analysis.
   • Expected Returns & Covariance: Calculate statistical metrics for portfolio optimization.
   • Mean-Variance Optimization: Generate baseline portfolio weights.

2. Real-Time AI Signals:

   • Live Data Feed: Capture current market data.
   • Real-Time Processing: Clean and format live data.
   • HMM Signal Generation: Generate predictive signals using Hidden Markov Models.
   • LSTM Signal Generation: Generate predictive signals using LSTM networks.
   • Aggregation: Combine the AI signals into an overall adjustment factor.

3. Portfolio Integration & Execution:

   • Integration: Merge historical optimization outputs with AI signal adjustments.
   • Adjusted Portfolio Weights: Produce final portfolio allocations.
   • Trade Execution: Place trades based on the adjusted portfolio.
   • Risk Management & Monitoring: Apply stop losses, rebalance, and update the model based on performance feedback.

Diagram 3: Portfolio Optimization System Flow Diagram

flowchart TD
    %% Symbol Selection Branch
    S[Symbol Selection & Current Positions]

    %% Historical Data Branch
    A[Historical Price Data]
    B[Data Preprocessing & Cleaning]
    C[Calculate Expected Returns]
    D[Calculate Covariance Matrix]
    E[Mean-Variance Optimization]
    F[Baseline Portfolio Weights]

    %% Real-Time AI Signals Branch
    G[Live Data Feed]
    H[Model Maintenance]
    I[HMM Signal Generation]
    J[LSTM Signal Generation]
    K[Aggregate AI Signals]

    %% Integration
    L[Integrate Historical & AI Technical Signals]
    M[Adjusted Portfolio Weights]

    %% Trade Execution & Monitoring
    N[Trade Execution & Order Placement]
    O[Risk Management & Constraint Checks]
    P[Portfolio Monitoring & Rebalancing]
    Q[Performance Feedback & Model Update]

    %% Flow Connections
    S --> A
    A --> B
    B --> C
    B --> D
    C --> E
    D --> E
    E --> F

    S --> G
    A -- "Training Models<br>& Backtesting" --> H
    G --> H
    H --> I
    H --> J
    I --> K
    J --> K

    F --> L
    K --> L
    L --> M

    M --> N
    N --> O
    O --> P
    P --> Q

Loading

Symbol Selection Strategies

Determining which assets to trade is crucial for strategy execution. There are three main approaches we will support:

• Watchlist-Based Selection: Focuses on a predefined set of symbols, ensuring consistency and reducing noise. This approach works well for traders who prefer to specialize in familiar assets.

• Scanner-Based Selection: Dynamically identifies tradeable assets based on technical criteria such as volatility, volume, trend strength, or unusual options activity. This method helps capture emerging opportunities.

• LLM Approach: Scans email updates and news to discover new trade opportunities while maintaining a core watchlist. Assets can be added or removed based on scanner signals and performance over time.

Diagram 4: Symbol Selection Process

This diagram illustrates the different pathways for symbol selection, showing how market data feeds into both static watchlists and dynamic scanners.

flowchart TD
    A[Market Data Feed]
    B[Watchlist-Based Filtering]
    C[Scanner-Based Selection]
    D[Filtered Tradeable Symbols]
    E[Trading Signal Generation]

    A --> B
    A --> C
    B --> D
    C --> D
    D --> E

Loading