Profit Optimization Strategies

This document outlines the three key strategies for maximizing profitability and efficiency in the AI Trading System.

Dynamic Budget & Tax Enhancements (Dec 2025)

Auto-Scaling Daily Input

  • scripts/autonomous_trader.py now inspects data/system_state.json (or SIMULATED_EQUITY) before the orchestrator spins up.
  • The new calc_daily_input() helper maps equity into a $10–$50/day deployment, adding extra basis points at the $2k / $5k / $10k gates while preserving the risk cap.
  • The resulting number is exported via DAILY_INVESTMENT, so every downstream component (tier budgets, risk manager, trade gateway) automatically receives the right daily budget without editing .env.

Tax Sweep Guidance

  • scripts/generate_telemetry_report.py now prints the latest equity/P&L plus a recommended tax sweep (default 28%) when cumulative profit is positive.
  • Use --tax-rate to align with your local marginal bracket; the recommendation is stored in the JSON output for dashboards/Slack pings.
  • This keeps the HYSA/tax bucket honest and enforces the “sweep 28% of realized gains quarterly” directive automatically.

1. Alpaca High-Yield Cash (3.56% APY)

Discovered: October 30, 2025 (Alpaca rate update email)

Key Features

  • APY: 3.56% (nearly 10x national average of 0.40%)
  • FDIC Insurance: Up to $1M
  • Liquidity: Full - can use as buying power anytime
  • Cost: $0 - automatically earns on idle cash

How This Boosts Fibonacci Strategy

  1. Profit Buffer Growth: When we make $1 profit in Phase 1, that $1 earns 3.56% APY while we wait to accumulate enough to scale to $2/day
  2. Passive Income Layer: Even uninvested cash generates returns
  3. Compounding Multiplier: Every dollar waiting to be deployed is earning interest

Example Math

Phase 1: Make $60 profit → Move to $2/day Phase
While waiting (30 days): $60 × 3.56% APY = ~$0.18 extra
Phase 2: Make $90 profit → Move to $3/day Phase
While waiting (30 days): $150 × 3.56% APY = ~$0.44 extra
...continues exponentially

When We Can Use This

  • ❌ NOT available during paper trading (current phase)
  • ✅ Only available with LIVE account + real money
  • ✅ Must meet balance or income requirements
  • ✅ Cannot be a Pattern Day Trader

Timeline

  • Now (Days 1-30): Paper trading - High-Yield Cash NOT available
  • Month 2+: Switch to live $1/day Fibonacci strategy → THEN enroll in High-Yield Cash
  • Benefit Starts: When we have real cash accumulating between Fibonacci phases

Action Items (Future - After Going Live)

  • Switch from paper to live account (Month 2)
  • Enroll in Alpaca High-Yield Cash program
  • Track High-Yield Cash earnings in daily reports
  • Include passive interest in profit calculations

Margin Rate Warning ⚠️

  • Alpaca margin costs 6.5% (base rate + 2.5%)
  • High-Yield Cash earns 3.56%
  • Negative spread = NEVER use margin
  • Strategy: Only trade with cash we have

2. OpenRouter Multi-LLM Strategy (Cost-Benefit Analysis)

Current Status: October 30, 2025 - Security incident resolved

Security Incident

  • ❌ Old API key exposed in git history (commit d3fa92d)
  • ✅ OpenRouter automatically disabled exposed key
  • ✅ New API key created and secured in .env
  • ✅ .env verified in .gitignore (will NOT be committed)
  • ⚠️ Old key still in git history but disabled by OpenRouter

Integration Status

  • ✅ MultiLLMAnalyzer fully built and integrated into CoreStrategy
  • ✅ Configured with 3 models: Claude 3.5 Sonnet, GPT-4o, Gemini 2 Flash
  • ✅ Code supports use_sentiment=True flag
  • ❌ Currently NOT calling AI during trades (making simple buy orders)

Cost Analysis

Model Cost per 1M tokens Input Output
Claude 3.5 Sonnet $3/$15 In/Out Deep analysis
GPT-4o $2.50/$10 In/Out Reasoning
Gemini 2 Flash $0.075/$0.30 In/Out Fast sentiment

Estimated Costs If Enabled

  • Daily: 3 models × 2 calls × ~1000 tokens = $0.50-2/day
  • Monthly: $15-60 for AI analysis
  • Annually: $180-720

Decision: When to Enable OpenRouter

Phase 1 (Now - Days 1-30): ❌ DISABLED

  • Current: Paper trading with simple buy orders ($10/day)
  • Profit: $0.02/day (essentially break-even)
  • Analysis: NOT worth spending $15-60/month for $0.02/day profit
  • Purpose: Testing infrastructure, not making real money yet

Phase 2 (Month 2-3): ❌ STILL DISABLED

  • Live $1/day Fibonacci strategy with RL system
  • Projected: $1-3/day profit
  • Analysis: NOT worth spending $2/day AI cost when making $1-3/day
  • Purpose: Validate RL system profitability first

Phase 3 (Month 4+): ✅ ENABLE WHEN PROFITABLE

  • Scaled to $5-13/day Fibonacci phases
  • Projected: $10-50/day profit
  • Analysis: Worth enabling - $2/day AI cost becomes negligible
  • Purpose: Multi-LLM consensus improves market regime detection
  • ROI: If AI improves returns by 10-20%, pays for itself immediately

Enable OpenRouter When

# Conditions to enable:
if (
    daily_profit > 10  # Making $10+/day consistently
    and fibonacci_phase >= 5  # At $5/day investment phase or higher
    and rl_sharpe_ratio > 1.0  # RL system validated
):
    use_sentiment = True  # Enable multi-LLM analysis

What OpenRouter Will Do (When Enabled)

  • Market sentiment analysis (3-model consensus)
  • Risk-off detection (market crashes, volatility spikes)
  • Sector rotation signals (which sectors to weight)
  • Entry/exit timing optimization
  • News sentiment integration

Action Items

  • Keep OpenRouter integrated but disabled (current)
  • Monitor profit levels in Month 4+
  • Enable when making $10+/day consistently
  • Track ROI: Does AI improve returns enough to justify cost?

3. Claude Batching Strategy (Prevent Token Exhaustion)

Problem: Running out of Claude tokens daily, interrupting progress

Solution: Agent parallelization via Claude Agents SDK

Key Strategies

Based on Agent Swarm Best Practices:

  1. Batch Agent Swarms: Spawn multiple agents in parallel using Task tool
    • Example: “Spawn 5 agents in parallel to work through remaining tasks”
    • Reduces total time and maximizes output per token
  2. Work Incrementally: Break work into smaller focused batches
    • One batch for planning
    • One batch for implementation
    • One batch for testing
  3. Upgrade Plan If Needed:
    • Pro: $20/mo
    • Elite: $200/mo (20x more than Pro)
    • Monitor usage and adjust
  4. Pause When Credits Exhausted:
    • Review output
    • Document progress
    • Clean up code
    • Resume when credits refresh

Implementation Guidelines

  • ALWAYS use parallel Task tool calls when possible
  • Launch 3-5 agents simultaneously for research
  • Use specialized agent types (general-purpose, Explore, Plan)
  • Work in focused batches rather than trying to complete everything at once

Claude Batching Skill (Future)

  • Could create a skill to manage agent spawning
  • Track token usage across tasks
  • Optimize batch sizes for efficiency
  • Auto-pause before exhaustion

4. Auto-Input Scaling (Compounding Accelerator)

Added: December 2, 2025 (theta-scale branch)

Overview

Dynamic daily input scaling based on account equity to accelerate compounding:

def calc_daily_input(equity: float) -> float:
    base = 10.0  # Minimum daily input
    if equity >= 10000:
        base += 4.0 * ((equity - 10000) / 1000)  # +$4 per $1k above $10k
        base += 4.0 + 0.4  # Tier bonuses
    elif equity >= 5000:
        base += 0.3 * ((equity - 5000) / 1000) * 10
        base += 0.4
    elif equity >= 2000:
        base += 0.2 * ((equity - 2000) / 1000) * 10
    return min(base, 50.0)  # Cap at $50/day

Scaling Tiers

Equity Level Daily Input Monthly Total Time Saved
$0-$2k $10 $300 Baseline
$2k-$5k $12-$14 $360-$420 1 month
$5k-$10k $16-$20 $480-$600 2 months
$10k+ $24-$50 $720-$1500 3+ months

Enabling Auto-Scale

# Via environment variable
export ENABLE_AUTO_SCALE_INPUT=true

# Via CLI flag
python scripts/autonomous_trader.py --auto-scale

5. Theta Harvest Execution (Options Premium)

Added: December 2, 2025 (theta-scale branch)

Overview

Automatic theta (time decay) harvesting through options premium selling when equity gates are met.

Equity Gates

Equity Strategy Target Premium Risk Level
$5k+ Poor Man’s Covered Calls $5-7/day Defined
$10k+ Iron Condors (calm regime) $10-15/day Defined
$25k+ Full Options Suite $20-30/day Mixed

IV Percentile Filter

Only sells premium when IV percentile > 50% (ensures we’re selling expensive options).

Integration

The ThetaHarvestExecutor in options_profit_planner.py now connects directly to Alpaca execution:

# Automatically executed in orchestrator's Gate 7
from src.analytics.options_profit_planner import ThetaHarvestExecutor

executor = ThetaHarvestExecutor(paper=True)
result = executor.evaluate_theta_opportunity(
    symbol='SPY',
    account_equity=5000,
    regime_label='calm',
)
if result.strategy != 'none':
    executor.execute_theta_order(result, alpaca_client)

6. VIX-Triggered Trade Auditor

Added: December 2, 2025 (theta-scale branch)

Overview

Adaptive audit frequency based on market volatility (VIX) for continuous system improvement.

Audit Frequency Rules

VIX Level Frequency Rationale
< 25 Weekly Normal conditions, standard review
25-35 Daily Elevated vol, catch issues faster
> 35 Twice Daily Crisis mode, maximize protection

Features

  • Analyzes closed trades for win rate, profit factor, patterns
  • Queries RAG for McMillan theta loss patterns
  • Generates actionable recommendations
  • Logs to telemetry for continuous improvement

Usage

from src.agent_framework.auditor import TradeAuditor

auditor = TradeAuditor()
result = auditor.run_audit(force=True)

print(f"Win Rate: {result.win_rate}%")
print(f"Recommendations: {result.recommendations}")

7. Quarterly Profit Sweep (Tax Reserve)

Added: December 2, 2025 (theta-scale branch)

Overview

Automatic quarterly profit sweep to reserve funds for estimated tax payments.

Configuration

export TAX_RESERVE_PCT=28.0  # Short-term capital gains rate
export QUARTERLY_SWEEP_ENABLED=true

Quarter-End Dates

  • March 31 (Q1)
  • June 30 (Q2)
  • September 30 (Q3)
  • December 31 (Q4)

Calculation

taxable_profit = end_equity - start_equity - deposits + withdrawals
tax_reserve = taxable_profit * 0.28  # 28% to HYSA

Integration

Runs automatically as Gate 9 in the orchestrator on quarter-end dates:

from src.orchestrator.telemetry import run_quarterly_sweep

result = run_quarterly_sweep(
    start_equity=100000,
    end_equity=105000,
    deposits=900,  # $10/day * 90 days
    force=False,
    dry_run=True,  # Set False for live execution
)
# Result: ~$1,148 to tax reserve (28% of $4,100 profit)

Summary

These seven strategies work together to maximize system profitability:

  1. Alpaca High-Yield Cash: Idle cash earns 3.56% APY passively
  2. OpenRouter Multi-LLM: Enable AI analysis when profit justifies cost (Month 4+)
  3. Claude Batching: Maximize development velocity through parallel agent execution
  4. Auto-Input Scaling: Accelerate compounding as equity grows (+2-3 months saved)
  5. Theta Harvest: Premium selling when equity gates met (+$5-30/day at scale)
  6. VIX-Triggered Audit: Adaptive critique frequency for continuous improvement
  7. Quarterly Tax Sweep: Automated 28% reserve for tax obligations

Key Principle: Every optimization should pay for itself. Don’t add costs until revenue justifies them.