Files

zastian-dev 79816b9e2e Experiment with hourly timeframe-specific stops

- Added HOURLY_ATR_STOP_MULTIPLIER (1.8x) vs daily (3.5x)
- Added hourly-specific trail multipliers
- Strategy now uses timeframe field to select appropriate stops
- Tested multiple configurations on hourly:
  * 3.5x stops: -0.5% return, 45% max DD
  * 1.8x stops: -45% return, 53% max DD (worse)
  * Conservative regime (0.25x): -65% return, 67% max DD (terrible)
- Conclusion: Hourly doesn't work with this strategy
- Daily with relaxed regime remains best: +17.4% over 5yr, 24% max DD

Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>

2026-02-13 19:20:01 +00:00

3.0 KiB

Raw Blame History

CLAUDE.md

This file provides guidance to Claude Code (claude.ai/code) when working with code in this repository.

Project Overview

Vibe Invest is a Rust algorithmic trading bot using the Alpaca paper trading API. It executes a hybrid momentum + mean-reversion strategy across 50 tech/financial stocks. It has two execution modes: live paper trading and historical backtesting, plus a web dashboard.

Build & Run Commands

# Build
cargo build --release

# Run live paper trading (requires .env with ALPACA_API_KEY and ALPACA_SECRET_KEY)
cargo run --release
cargo run --release -- --timeframe hourly

# Run backtesting
cargo run --release -- --backtest --years 3
cargo run --release -- --backtest --years 5 --capital 50000
cargo run --release -- --backtest --years 1 --months 6 --timeframe hourly

# Run backtesting with custom date range
cargo run --release -- --backtest --start-date 2007-01-01 --end-date 2008-12-31
cargo run --release -- --backtest --start-date 2020-03-01 --end-date 2020-12-31 --timeframe hourly

# Lint and format (available via nix flake)
cargo clippy
cargo fmt

There are no tests currently in the project.

Architecture

main.rs          CLI parsing (clap), logging setup, mode routing
├── bot.rs       Live trading loop: market hours detection, per-symbol analysis,
│                order execution, position tracking (JSON persistence)
├── backtester.rs  Historical simulation: processes bars sequentially, tracks
│                  positions, calculates metrics (CAGR, Sharpe, Sortino, drawdown)
├── alpaca.rs    Alpaca REST API client with rate limiting (200 req/min via governor)
├── indicators.rs  Technical indicators: EMA, SMA, RSI, MACD, ADX, ATR,
│                  Bollinger Bands, ROC. Signal scoring algorithm in generate_signal()
├── dashboard.rs   Axum web server (default port 5000) with Chart.js frontend
├── config.rs    Strategy parameters, stock universe (50 symbols), risk limits
├── types.rs     Domain types: Signal, TradeSignal, Trade, BacktestResult, Bar, etc.
└── paths.rs     XDG-compliant file paths (~/.local/share/invest-bot/)

Key data flow: Both bot.rs and backtester.rs call indicators::generate_signal() which scores multiple indicators into a composite buy/sell signal. The bot executes via alpaca.rs; the backtester simulates internally.

Strategy Parameters (config.rs)

Hourly mode scales all indicator periods by 7x
Risk: max 22% position size, 2.5% stop-loss, 40% take-profit, trailing stop at 7% after 12% gain
Signal thresholds: StrongBuy ≥ 6.0, Buy ≥ 3.5, Sell ≤ -3.5, StrongSell ≤ -6.0
Backtester restricts to top 4 momentum stocks; live mode does not

Environment

Nix flake provides dev tools: rustc, cargo, clippy, rustfmt, rust-analyzer, cargo-watch
.env file required: ALPACA_API_KEY, ALPACA_SECRET_KEY, optional DASHBOARD_PORT
Persistent state stored in ~/.local/share/invest-bot/ (positions, equity history, logs)
Backtest outputs: backtest_equity_curve.csv, backtest_trades.csv in working directory

3.0 KiB Raw Blame History