Skip to content

2025 - 12

Goal: Finish the year with profitable live trading and clear 2026 direction.

Plan 1: OKX Integration & Cross-Venue Arb

  • OKX Gateway: Implement exchange gateway with API integration
  • Account Setup: Configure API keys, risk limits, position tracking
  • Latency Comparison: Test Binance + OKX vs current Binance + Hyperliquid
  • Optimize v2 Strategy: Improve best performer (0.38 Sharpe) or deprecate cross-venue

Success Criteria: OKX operational with improved arb performance or clear decision to focus elsewhere

Progress 1

Status: ✅ Gateway Complete, ⏳ Awaiting Account Verification

Completed:

  • OKX API integration done (required new onboarding due to OKCoin→OKX rebranding)
  • Connected with OKX Institutional BD Director - direct line to leadership and dedicated account manager
  • Fee rate reduction to lowest tier + rebate negotiated
  • Binance.US institutional account ready, BD reconnected for fee reduction and rebate

Blocked:

  • OKX institutional account verification pending (external dependency)
  • Cross-venue arb launch waiting on account approval

Key Finding: Exchange BD relationships secured for both OKX and Binance.US with favorable fee structures. Ready to launch immediately once OKX verification completes.

Plan 2: Single-Venue Strategy

  • Design Strategy: Select approach (funding rate arb, basis trading, or mean reversion)
  • Implement: Build strategy logic for any single exchange
  • Paper Trading: Validate on simulation before live
  • Deploy Live: Launch with small capital on supported exchange

Strategy Options:

Type Description Priority
Mean Reversion Trade deviations from moving average High
Funding Rate Arb Long spot + short perp, earn funding Medium
Basis Trading Spot-futures spread on same venue Medium

Success Criteria: Profitable single-venue strategy running live

Progress 2

Status: ❌ Not Started

Deferred due to infrastructure disaster and data migration failure (see Learning).

Key Finding: Single-venue strategy blocked by infrastructure recovery. Remains high priority for Q1 2026.

Plan 3: ML Generalization

  • Reduce Overfitting: Stronger regularization, simpler model architecture
  • Achieve Positive R²: Beat baseline mean prediction on test set
  • Target >50% Accuracy: Realistic milestone toward 65% goal
  • Evaluate Sentiment: Test if LLM features improve regime detection

Success Criteria: Positive R² and >50% directional accuracy on test set

Progress 3

Status: 🔄 Progress on Understanding, Metrics Still Below Target

Metrics:

Metric Target Result Status
Accuracy >50% ~50% 🔄 Close
Up/Down Precision >50% ~50% 🔄 Close
>0 TBD ❌ Not yet

Completed:

  • Pivoted to classification model, deferred regression (need one working model first)
  • Developed empirical methods to evaluate SNR of dataset features and models
  • Tested simpler models - outperforming complex deep learning models
  • Achieved stable metrics across different dataset splits (consistency improved)

Key Finding: Problem now decomposed into two independent tracks: (1) dataset feature engineering, (2) model selection/tuning. Simpler models > complex deep learning for this task. Framework enables faster iteration for 2026.

Plan 4: Backtest System

  • Complete Framework: Enable rapid strategy iteration
  • Historical Data Pipeline: Clean data preparation for backtesting
  • Validation: Verify backtest results match paper trading

Success Criteria: Backtest system operational for 2026 strategy development

Progress 4

Status: ✅ Deployed with Limitations

Completed:

  • Framework development and deployment done
  • Historical data pipeline operational

Limitations:

  • Memory constraints limit supported time range
  • Testing speed slow, needs optimization

Key Finding: MVP backtest system ready for strategy iteration. Performance optimization needed for longer backtests.

Plan 5: Dashboard Fixes

  • Fix Summary Metrics: Correct misleading fill rate and P&L display
  • Verify Accuracy: Ensure summary matches detailed views
  • Improve multiple page UX: Better navigation and layout
  • [i] Latency Tracking: Better execution analytics and monitoring

Success Criteria: Accurate, trustworthy dashboard metrics

Progress 5

Status: ✅ Core Fixes Complete

Fixed misleading metrics, improved visual layout and usability. Latency tracking improvements in progress.

Key Finding: Dashboard now trustworthy for strategy monitoring. November's "0% fill rate" false alarm won't recur.

Year-End Success Criteria

Goal Target Result Status
Profitable Strategy At least one strategy with positive P&L Not achieved
Exchange Diversity OKX operational Gateway done, account pending 🔄
Single-Venue Strategy Deployed and validated Deferred
ML Progress Positive R², >50% accuracy ~50% accuracy, R² pending 🔄
Infrastructure Backtest ready, dashboard accurate ✅ Both operational

Learning

Infrastructure Disaster Cost 2 Weeks

Month started well with fast OKX integration. Then attempted to give backtest app access to paper trading's historical data in AWS. Due to Copilot limitations, tried various workarounds. Manual operation error corrupted the infrastructure.

Decision: Nuke and redeploy using IaC. Backed up database via RDS snapshot before nuke (good). Discovery: RDS snapshots can only restore to new database instances, not into existing Copilot-managed databases (bad). After trying multiple complex workarounds, accepted data loss and started fresh.

IaC Had Gaps

Even with IaC, redeployment required significant manual steps due to infra complexity grow. Fixed these gaps - can now redeploy entire infrastructure from scratch with one command. Data snapshot restore still has gaps with Copilot-managed infra.

Speed Under Pressure Causes Mistakes

When rushing to meet year-end goals, made avoidable manual operation error. Balance speed vs correctness. Lucky this happened with small capital, not client funds.