Research

Our Methodology

Transparency is a core value. Here is exactly how we source data, run backtests, control for biases, and measure performance.

Data Sources

We use institutional-quality data feeds with point-in-time accuracy, covering 8 markets worldwide.

MarketData ProviderCoverage
US EquitiesNYSE, NASDAQ via premium data feedsFull S&P 500, NASDAQ 100, and 3,000+ additional tickers
Hong Kong EquitiesHKEX official dataHSI constituents, HSCEI, GEM board, all listed securities
China A-SharesSSE, SZSE licensed dataCSI 300, CSI 500, full A-share universe
Taiwan EquitiesTWSE, TPEx feedsTAIEX constituents and all listed securities
Korean EquitiesKRX market dataKOSPI 200, KOSDAQ 150, and broader universe
CryptocurrencyMultiple exchange APIsTop 100 by market cap, major trading pairs
CommoditiesCME, LME, ICE dataGold, Silver, Oil, Natural Gas, Agricultural
Global ETFsMulti-exchange ETF data500+ ETFs across all asset classes and geographies

Backtesting Methodology

Our backtesting engine follows a rigorous, multi-step process designed to produce realistic results.

Walk-Forward Analysis

Each strategy is validated through a walk-forward framework. The historical data is split into rolling in-sample (training) and out-of-sample (validation) windows. Parameters are optimized on in-sample data, then tested on unseen out-of-sample data. This prevents overfitting and provides a realistic estimate of future performance.

In-Sample (70%) → Optimize → Out-of-Sample (30%) → Validate → Roll Forward → Repeat

Transaction Cost Modeling

All backtests incorporate realistic transaction costs including brokerage commissions, exchange fees, stamp duty (for HK and UK markets), and estimated slippage. Market impact is modeled using historical volume data -- larger positions in less liquid securities incur higher estimated costs.

Cost = Commission + Exchange Fee + Stamp Duty + Slippage(Volume, Spread)

Position Sizing

Default backtests use equal-weight position sizing. Advanced reports include alternative sizing methods: volatility-targeting (risk parity), Kelly criterion, and maximum diversification. Each method's assumptions and limitations are clearly documented in the strategy report.

Benchmark Comparison

Every strategy is benchmarked against the appropriate market index and a risk-free rate. We report alpha, beta, tracking error, and information ratio to help you assess whether a strategy genuinely adds value or is merely taking on market risk.

Bias Controls

We take bias control seriously. Here are the specific biases we address and how we mitigate them.

Survivorship Bias

Critical

We use point-in-time databases that include delisted securities. Our universe at each backtest date includes only securities that were actually tradeable at that time, preventing the inflation of returns by excluding companies that failed or were acquired.

Look-Ahead Bias

Critical

All data used in signal generation is strictly timestamped. Financial statement data uses the filing date (not the period end date) to ensure strategies only act on information that was publicly available at the time of the trade decision.

Data-Snooping Bias

High

We employ walk-forward analysis with expanding or rolling windows. Strategy parameters are optimized on in-sample data and validated on out-of-sample data to prevent overfitting to historical patterns.

Transaction Cost Bias

High

All backtests include realistic transaction costs: brokerage commissions, exchange fees, stamp duty (where applicable), and estimated market impact based on historical volume and spread data.

Selection Bias

Medium

We report all strategies we test, not just the ones that perform well. Each strategy report includes a clear disclaimer about the number of strategies tested and the risk of multiple comparison bias.

Performance Metrics

Every strategy report includes 50+ metrics across five categories.

Return Metrics

  • Total Return
  • Annualized Return (CAGR)
  • Monthly Returns
  • Rolling Returns (1Y, 3Y, 5Y)
  • Risk-Adjusted Return
  • Excess Return vs Benchmark

Risk Metrics

  • Annualized Volatility
  • Maximum Drawdown
  • Drawdown Duration
  • Value at Risk (95%, 99%)
  • Conditional VaR (Expected Shortfall)
  • Downside Deviation
  • Ulcer Index

Risk-Adjusted Ratios

  • Sharpe Ratio
  • Sortino Ratio
  • Calmar Ratio
  • Information Ratio
  • Treynor Ratio
  • Omega Ratio

Trading Metrics

  • Win Rate
  • Profit Factor
  • Average Win / Average Loss
  • Maximum Consecutive Losses
  • Turnover Rate
  • Average Holding Period

Statistical Tests

  • T-Statistic of Returns
  • Skewness & Kurtosis
  • Jarque-Bera Test
  • Autocorrelation Analysis
  • Regime Analysis
  • Monte Carlo Simulation

Questions About Our Methodology?

We welcome scrutiny. Reach out to our research team for detailed documentation.