The Reproducibility Problem

Algorithmic trading has a fundamental problem: when someone publishes a β€œprofitable strategy,” reproducing the results is virtually impossible. The reasons:

  • Unspecified parameters – the author forgot to mention key settings
  • Different data – data sources give slightly different prices
  • Hidden assumptions – commissions, slippage, execution time
  • Platform differences – the same algorithm yields different results on different backtesting engines

The PLUTUS framework was created to solve this problem.

What Is PLUTUS

PLUTUS is an open-source framework for the standardized description, testing, and publication of trading strategies.

Developed by an international research group and published on GitHub under the MIT license.

Architecture

PLUTUS defines four standardized components:

1. Strategy Specification

A formal description of the strategy in YAML/JSON format:

strategy:
  name: "Mean Reversion RSI"
  version: "1.0"
  author: "researcher@university.edu"

  signals:
    entry_long:
      condition: "RSI(14) < 30 AND SMA(50) > SMA(200)"
    exit_long:
      condition: "RSI(14) > 70 OR stop_loss(-2%)"

  parameters:
    rsi_period: 14
    sma_fast: 50
    sma_slow: 200
    stop_loss_pct: -2.0

  universe:
    type: "equity"
    market: "US"
    filter: "S&P 500 constituents"

  execution:
    order_type: "market"
    slippage_model: "fixed_bps(5)"
    commission_model: "per_share(0.005)"

2. Data Specification

Standardized data description:

  • Source (Yahoo Finance, Polygon, MOEX)
  • Period (start, end)
  • Frequency (1 minute, 1 hour, 1 day)
  • Processing (adjusted/unadjusted, fill method)
  • Data hash for verification

3. Backtest Engine

A standardized backtesting engine with:

  • Defined order processing logic
  • Fixed intra-bar calculation order
  • Transparent slippage model
  • Report with 50+ metrics

4. Report Format

A unified report format that includes:

  • Equity curve
  • All metrics (Sharpe, Sortino, Max DD, Calmar, etc.)
  • Trade distribution
  • Time-period analysis
  • Walk-forward results

Why This Matters

For Researchers

Publishing a strategy in the PLUTUS format allows other researchers to exactly reproduce the results. This is what the scientific world has long had for experiments, but what algorithmic trading has been missing.

For Practitioners

A standardized format simplifies:

  • Strategy comparison – all metrics are calculated the same way
  • Auditing – every parameter can be verified
  • Portability – transferring a strategy between platforms

For AI Agents

PLUTUS is especially useful for LLM agents that generate trading strategies. The standardized format enables:

  • Automatic validation of the specification
  • Running backtests without manual setup
  • Comparing results against a benchmark

Current Status

  • Version: 0.8 (beta)
  • Languages: Python (primary), adapters for C# and Java
  • Supported markets: US, EU, China, Crypto
  • Integrations: Backtrader, Zipline, VectorBT, QuantConnect

PLUTUS is a step toward making algo trading more transparent and scientific. If you are developing trading strategies, it is worth paying attention to.