Core Concepts¶

Understanding sigc's core concepts will help you build better trading strategies.

Overview¶

sigc is built around four fundamental concepts:

Signals - Numerical scores that rank assets
Portfolios - Weight allocations based on signals
Backtests - Historical simulations of portfolio performance
Type System - Compile-time checking of operations

graph LR
    A[Data] --> B[Signal]
    B --> C[Portfolio]
    C --> D[Backtest]
    D --> E[Metrics]

Quick Concept Map¶

Concept	What It Is	Example
Signal	Score per asset	Momentum, value, quality
Portfolio	Weight allocation	Long/short, long-only
Backtest	Historical simulation	2020-2024 test
Type System	Safety checking	Shape, dtype inference

The sigc Workflow¶

1. Load Data¶

Text Only

data:
  prices: load csv from "data/prices.csv"

Data flows into the system as time-series matrices (dates × assets).

2. Define Signals¶

Text Only

signal momentum:
  returns = ret(prices, 20)
  emit zscore(returns)

Signals transform data into scores that rank assets at each point in time.

3. Construct Portfolios¶

Text Only

portfolio main:
  weights = rank(momentum).long_short(top=0.2, bottom=0.2)

Portfolio construction converts scores into tradeable weights.

4. Run Backtests¶

Text Only

  backtest from 2020-01-01 to 2024-12-31

Backtesting simulates the strategy over historical data.

5. Analyze Results¶

Text Only

=== Backtest Results ===
Sharpe Ratio:    1.45
Max Drawdown:    8.12%

Metrics quantify strategy performance.

Key Principles¶

Reproducibility¶

sigc guarantees deterministic results:

Same inputs → same outputs, always
Content-addressed caching with blake3 hashes
No hidden randomness or state

Type Safety¶

The type system catches errors at compile time:

Text Only

signal bad:
  x = ret(prices, 20)      // Returns: time-series
  y = zscore(x)            // Error: zscore expects cross-sectional

Learn more in Type System.

Composability¶

Build complex strategies from simple components:

Text Only

fn vol_adj(x, lookback):
  x / rolling_std(ret(x, 1), lookback)

macro momentum_signal(px: expr, lookback: number = 20):
  let r = ret(px, lookback)
  emit zscore(vol_adj(r, 60))

signal combo:
  mom = momentum_signal(prices)
  rev = -zscore(ret(prices, 5))
  emit 0.7 * mom + 0.3 * rev

Section Overview¶

Page	Description
Signals	What signals are and how to build them
Portfolio Construction	Converting signals to weights
Backtesting	Simulation methodology
Type System	Shape-aware type checking
Architecture	System components and design

Next Steps¶

Start with What Are Signals? to understand the foundation of sigc strategies.