Financial Infrastructure and Quant Research

Project context

Smartoption strategy research focused on turning options market structure, premium collection, liquidity, DTE, and risk-control assumptions into disciplined selling strategies rather than treating trader alerts as the main product.

Research goals

Design and validate option-selling strategy rules covering entry selection, premium/risk constraints, DTE and liquidity filters, sizing, stop/close policy, take-profit behavior, and scenario-level backtest reporting.

Methodology

• Researched short-premium option-selling setups across index, ETF, and single-name contracts, translating DTE, ticker liquidity, premium level, bid-ask spread, time-of-day, and market-regime constraints into systematic entry rules.
• Designed auditable strategy specifications for candidate selection, position sizing, stop/close policy, take-profit behavior, contract-quality filters, duplicate handling, and risk-control boundaries.
• Reviewed replay and backtest evidence, separating scenario-level PnL from aggregate diagnostics and flagging quote gaps, stale or wide-spread contracts, cash/equity chronology gaps, lookahead risk, and overfit risk.
• Worked with research constraints around fill/no-fill evidence, bid/ask freshness, partial fills, broker rejects, candidate deduplication, and consistency between research assumptions and executable strategy logic.

Findings

• Produced a clearer quantitative research framing for option-selling strategy development: frozen candidate tables, entry and exit contracts, account-level sizing, financial-grade replay, and risk-aware validation before promotion.
• Contributed to research artifacts that make strategy assumptions inspectable rather than treating headline historical returns as production-ready.

Project context

Traditional long-volatility products can suffer from roll costs, structural decay, timing difficulty, and path dependence. This project reframes volatility as an activity generator: when institutions hedge, rebalance, clear trades, request data, or demand liquidity, financial infrastructure firms can monetize the resulting transaction flow.

Research goals

Build a systematic thematic-investing framework for a global financial-infrastructure universe across exchanges, broker platforms, fixed-income trading venues, data and ratings firms, settlement infrastructure, and market makers.

Methodology

• Defined a 24-stock global universe across NYSE, NASDAQ, XETRA, LSE, Euronext, HKEX, ASX, and TSX, selected by volatility-transmission mechanisms rather than generic financial-sector exposure.
• Built market-controlled volatility-channel regressions separating broad SPY beta from stress, activity, realized-volatility, and VVIX acceleration exposure.
• Used VVIX as a proxy for hedging urgency and distinguished volatility-of-volatility acceleration from the raw VIX fear level.
• Added a Qlib LGBModel overlay trained on Alpha158 features for cross-sectional ranking while keeping the economic mechanism and interpretable regressions as the core thesis.
• Combined business-model analysis, factor evidence, crisis-window validation, earnings resilience checks, and portfolio construction.

Findings

• Found positive VVIXGamma for 18 of 24 stocks after controlling for SPY, suggesting broad excess-return sensitivity to volatility-of-volatility acceleration.
• Constructed a diversified financial-infrastructure basket with exposure to derivatives clearing, options activity, broker flow, credit-market rebalancing, market making, financial data, and settlement volume.
• Validated that the strategy behaves as an equity-based volatility-infrastructure strategy rather than a direct long-VIX product.
• Identified clear regime boundaries: strongest in acute volatility and liquidity-shock environments, weaker in slow rate-driven or valuation-compression bear markets.

Project context

Long-term portfolio decisions can be distorted by inconsistent data sources, manual notes, and performance metrics that ignore recurring cash contributions. Total account value alone does not separate investment performance from external cash flow.

Research goals

Build a portfolio intelligence pipeline that refreshes market, fundamental, and news data, uses provider fallback, integrates Qlib outputs where available, generates reports and health artifacts, supports unattended scheduling, and evaluates performance through NAV-based risk metrics.

Methodology

• Built a canonical daily command path for data refresh, optional Qlib warmup, report generation, archive rollover, and pipeline health artifacts.
• Designed provider fallback across Alpha Vantage, Yahoo, and downloaded Qlib data, with persisted source coverage, missing-source diagnostics, and Alpha-Vantage-vs-Yahoo mismatch counts.
• Separated wealth from NAV so external cash flows could be unitized for time-weighted performance analysis.
• Implemented XIRR, TWR CAGR, annualized volatility, Sharpe ratio, Calmar ratio, max drawdown, and drawdown intervals.
• Built an Alpha Vantage universe-expansion workflow to compare the original sample universe with broader coverage before thesis review.

Findings

• Reproducible research pipeline with explicit methodology, source provenance, risk-aware reporting, scheduler contracts, and email/PDF delivery paths.
• Provider reliability design with source coverage and mismatch diagnostics.
• Universe-expansion validation for candidate-selection behavior under broader data coverage.
• Observed project test footprint outside virtual environments: 97 test files.

Project context

Crypto signals can be overfit when rules are tuned on short windows without transaction costs, slippage, funding context, or risk controls. A useful system needs clear separation between research, validation, risk filtering, and execution simulation.

Research goals

Build a Python-owned semi-automated research system for BTC/USDT and ETH/USDT using public Binance USD-M futures data, deterministic offline samples, persisted signals, risk filters, paper trading, and read-only monitoring APIs.

Methodology

• Built modules for public Binance market data, funding-rate ingestion, OHLCV caching, freshness diagnostics, indicators, strategy candidates, signal storage, risk limits, backtesting, paper-trading reports, and dry-run behavior.
• Kept the v1 boundary public-data only: no private exchange credentials and no real orders.
• Added validation modules for factor mining, candidate holdout, negative controls, cross-time checks, temporal validation, cluster quality, dynamic leverage, risk estimation, and transaction-cost-aware assumptions.
• Exposed FastAPI read-only endpoints for latest signals, risk status, news, and diagnostics.

Findings

• Public-only Binance ingestion for BTC/USDT and ETH/USDT 15m/1h OHLCV and funding data.
• Signal generation, risk filtering, diagnostics, persistence, read-only APIs, and deterministic sample-data mode.
• Paper-trading simulation over 50 recorded trades with fees and slippage: 52.0% win rate, 1.211 profit factor, positive expectancy of 0.0556 USDT/trade, and max drawdown of 4.3769 USDT on a 200 USDT paper account.
• Validation modules for holdout, negative controls, temporal checks, risk estimation, and factor quality; observed footprint: 42 test files.