Chapter 7: Going to Production¶

Deploy strategies from research to live trading.

The Production Journey¶

Text Only

Research → Validation → Paper Trading → Live Trading → Monitoring

Each stage requires different considerations.

Pre-Production Checklist¶

Strategy Validation¶

Infrastructure¶

Risk Controls¶

Position limits set
Exposure limits configured
Circuit breakers enabled
Maximum loss defined

From Research to Production¶

Research Code¶

Text Only

// research/momentum.sig
data:
  source = "historical_prices.parquet"

params:
  lookback: range(40, 100, 20)  // Optimize

signal momentum:
  emit zscore(ret(prices, lookback))

portfolio research:
  weights = rank(momentum).long_short(top=0.2, bottom=0.2)
  backtest walk_forward(...) from 2010-01-01 to 2024-12-31

Production Code¶

Text Only

// production/momentum_live.sig
config: "config/production.yaml"

data:
  source = "s3://mybucket/live_prices/"
  format = parquet

// Fixed parameters (from optimization)
signal momentum:
  emit zscore(ret(prices, 60))

portfolio live:
  weights = rank(momentum).long_short(
    top = 0.2,
    bottom = 0.2,
    cap = 0.03
  )

  constraints:
    gross_exposure = 2.0
    net_exposure = 0.0
    max_sector = 0.20
    max_position = 0.03

  costs = tc.bps(10)

Key Differences¶

Aspect	Research	Production
Parameters	Optimized	Fixed
Data Source	Historical	Live feed
Execution	Simulated	Real orders
Monitoring	Manual	Automated
Error Handling	None	Comprehensive

Production Configuration¶

Main Config File¶

YAML

# config/production.yaml

# Environment
environment: production
log_level: info

# Data
data:
  source: "s3://mybucket/prices/"
  refresh_interval: 1h
  cache_dir: "/var/cache/sigc"

# Strategy parameters
params:
  lookback: 60
  top_pct: 0.20
  position_cap: 0.03

# Risk limits
risk:
  max_position: 0.03
  max_sector: 0.20
  max_daily_loss: 0.03
  max_drawdown: 0.15
  gross_exposure: 2.0
  net_exposure_range: [-0.1, 0.1]

# Transaction costs
costs:
  commission_bps: 5
  spread_bps: 5
  market_impact_bps: 2

# Execution
execution:
  broker: alpaca
  order_type: limit
  limit_offset_bps: 5
  max_order_size: 0.02

Daemon Mode¶

Setup¶

YAML

# config/daemon.yaml
daemon:
  enabled: true
  pid_file: "/var/run/sigc/strategy.pid"
  log_file: "/var/log/sigc/strategy.log"

  schedule:
    # Compute weights before market close
    - cron: "30 15 * * 1-5"
      action: compute_weights
      timezone: "America/New_York"

    # Execute trades
    - cron: "55 15 * * 1-5"
      action: execute_trades
      timezone: "America/New_York"

Start Daemon¶

Bash

# Start in paper mode first
sigc daemon start \
  --strategy production/momentum_live.sig \
  --config config/production.yaml \
  --paper

# Check status
sigc daemon status

Paper Trading¶

Why Paper Trade First?¶

Verify execution logic
Test broker integration
Validate data pipeline
Check alerting
Build confidence

Duration¶

Minimum 2-4 weeks of paper trading before going live.

What to Check¶

Going Live¶

Final Pre-Live Checklist¶

Bash

# Run pre-flight checks
sigc preflight --config config/production.yaml

Text Only

Pre-Flight Checks
=================
[✓] Strategy file valid
[✓] Data source accessible
[✓] Broker connection OK
[✓] Account has sufficient capital
[✓] Risk limits configured
[✓] Alerting configured
[✓] Logging enabled

Ready for live trading.

Start Live Trading¶

YAML

# Update config
execution:
  paper: false  # Switch to live

Bash

# Start live
sigc daemon start \
  --strategy production/momentum_live.sig \
  --config config/production.yaml \
  --live

Monitoring¶

Key Metrics¶

Metric	Monitor	Alert
Daily P&L	Continuous	>2% loss
Positions	Every trade	Divergence >10%
Exposure	Hourly	Outside limits
Data freshness	Hourly	>2h stale
System health	Continuous	Any failure

Dashboard Example¶

Text Only

MOMENTUM STRATEGY - LIVE
========================
Status: RUNNING

Positions:
  Long:  50 stocks, $500,000
  Short: 50 stocks, $500,000
  Net:   $0 (0.0%)
  Gross: $1,000,000 (200%)

Performance (Today):
  P&L: +$1,234 (+0.12%)
  vs SPY: +0.08%

Performance (MTD):
  P&L: +$8,456 (+0.85%)
  Sharpe: 1.2

Risk Metrics:
  Daily VaR: $15,000
  Current DD: -2.3%
  Beta: 0.05

Alerting¶

YAML

alerting:
  slack:
    webhook: ${SLACK_WEBHOOK}
    channel: "#trading"

  rules:
    - name: "Large Loss"
      condition: daily_loss > 0.02
      severity: warning

    - name: "Circuit Breaker"
      condition: daily_loss > 0.03
      severity: critical
      action: halt_trading

    - name: "Trade Failure"
      condition: trade_failed
      severity: critical

Daily Operations¶

Morning Routine¶

Check overnight status
Verify positions match targets
Review any alerts
Check data freshness
Review broker account

Bash

# Morning checks
sigc status
sigc positions --compare target
sigc alerts --last 24h

End of Day¶

Review today's trades
Check P&L
Verify reconciliation
Note any issues

Bash

# EOD checks
sigc trades --today
sigc performance --today
sigc reconcile

Weekly Review¶

Performance analysis
Compare to backtest
Check parameter drift
Review any incidents

Bash

# Weekly report
sigc report --period 1w --output reports/weekly.html

Error Handling¶

Common Issues¶

Data Stale:

Text Only

Alert: Data not updated in 2 hours
Action: Check data pipeline, S3 access

Trade Rejected:

Text Only

Alert: Order rejected - insufficient buying power
Action: Check account balance, reduce position sizes

Connection Lost:

Text Only

Alert: Broker connection lost
Action: Auto-retry, failover to backup

Recovery Procedures¶

YAML

recovery:
  auto_retry:
    max_attempts: 3
    delay_seconds: 60

  failover:
    enabled: true
    backup_broker: "backup_config.yaml"

  manual_intervention:
    - circuit_breaker_triggered
    - large_position_divergence

Best Practices¶

1. Start Small¶

YAML

# Week 1-2
max_position: 0.01
gross_exposure: 1.0

# Week 3-4
max_position: 0.02
gross_exposure: 1.5

# Month 2+
max_position: 0.03
gross_exposure: 2.0

2. Monitor Everything¶

Log all: - Trades executed - Signals computed - Data received - Errors encountered

3. Have Kill Switches¶

Bash

# Emergency stop
sigc daemon stop --immediate

# Reduce exposure
sigc reduce-exposure --target 0.5

4. Document Everything¶

Maintain: - Run books - Incident reports - Parameter change log - Performance records

5. Regular Reviews¶

Daily: Quick health check
Weekly: Performance review
Monthly: Deep analysis
Quarterly: Strategy review

Exercises¶

Create a production configuration file
Set up paper trading for your strategy
Configure alerting for key metrics
Create morning and EOD checklists

Next Chapter¶

Continue to Chapter 8: Advanced Analytics for factor models and attribution.