Natural Language Trading Strategy Engine

This project converts natural language trading rules into a fully executable backtesting engine.
It follows a compiler-style pipeline:

Natural Language → JSON → DSL → AST → Python Code → Backtest Results

This project is built exactly according to the specifications of the assignment PDF.

---

Features

✔ Convert English trading rules into structured JSON

Input example:

“Buy when the close price is above the 20-day moving average and volume is above 1 million.
Exit when RSI(14) is below 30.”

Output JSON:

{
  "entry": [
    {"left": "close", "operator": ">", "right": "sma(close,20)"},
    {"left": "volume", "operator": ">", "right": 1000000}
  ],
  "exit": [
    {"left": "rsi(close,14)", "operator": "<", "right": 30}
  ]
}

Human-readable DSL representation

ENTRY: close > sma(close,20) AND volume > 1000000.

EXIT: rsi(close,14) < 30

DSL → AST (Abstract Syntax Tree)

The DSL is parsed using a Lark-based grammar, producing structured AST nodes:

• Series nodes

• Indicator nodes

• Comparison nodes

• Logical operators (AND/OR)

• Cross events

AST → Python code generation

Fully functional backtest engine

The system produces:

• Detailed list of trades

• Total return

• Max drawdown

• Win rate

• Average trade return

• Equity curve

Project Structure

DSL/
│
├── nlp/
│   └── nl_to_json.py           # Phase 1: Natural Language → JSON
│
├── dsl/
│   ├── grammar.lark            # Phase 2: DSL grammar
│   ├── parser.py               # Phase 3: DSL → AST
│   └── ast_nodes.py            # AST node classes
│
├── generator/
│   └── codegen.py              # Phase 4: AST → Python code
│
├── backtest/
│   └── simulator.py            # Phase 5: Backtest engine
│
└── demo/
    └── run_demo.py             # Phase 6: End-to-end pipeline demo

Installation

Install dependencies:

pip install pandas numpy lark-parser

Running the Demo

Execute the full pipeline:

python DSL/demo/run_demo.py

Example output includes:

• Input NL text

• JSON rules

• DSL

• AST

• Generated Python code

• Backtest results