Title: Technical Note: Background, Data, and Parameters Behind BIST’s New Quantitative Trade Model (NQTM)

Introduction
This technical note presents a concise overview of the motivation, data landscape, and methodological choices that underpinned the development of BIST’s new quantitative trade model (NQTM). The aim is to provide practitioners, researchers, and stakeholders with a clear understanding of the model’s foundations, the data employed, and the critical parameters that shape its performance and interpretability.

1. Background and objectives
– Rationale: The burgeoning complexity of financial markets, coupled with the need for disciplined, data-driven trading strategies, motivated the creation of NQTM. The model is designed to synthesise diverse signals into actionable insights while emphasising robustness, transparency, and replicability.
– Scope: NQTM targets a broad set of tradable instruments within BIST’s ecosystem, incorporating equities, fixed income proxies, and liquid derivatives where appropriate. The focus is on generating quantitative signals that support systematic trade decision-making, risk-aware position sizing, and disciplined execution.
– Governance and ethics: Development followed rigorous governance standards, including reproducibility checks, version-controlled code, and documentation practices that enable auditability and external validation where feasible.

2. Data landscape and preprocessing
– Data sources: NQTM integrates multiple data streams, including:
– Price and volume data at high frequency for liquidity and microstructure insights.
– Fundamental and accounting data where timely and reliable.
– Macroeconomic indicators and granular market microstructure variables.
– Alternative data inputs, subject to quality and coverage constraints.
– Data quality and cleaning: Steps include:
– Alignment of timestamps across sources to a common clock.
– Handling of missing values via domain-appropriate imputation or exclusion rules.
– Outlier detection and capping procedures to reduce the impact of transient anomalies.
– Adjustment for corporate actions, dividends, and split events to maintain consistency.
– Feature engineering: Derived features capture pricing anomalies, momentum, mean-reversion tendencies, volatility regimes, liquidity proxies (e.g., bid-ask spreads, depth), and cross-asset interactions. Features are generated with attention to stationarity, normalisation, and interpretability.
– Dataset construction: A rolling, calendar-aware dataset is used to ensure temporal integrity. Train/validation/test splits are defined to mimic real-world deployment, with out-of-sample evaluation preserving temporal ordering.

3. Modelling approach and architecture
– Modelling philosophy: NQTM adopts a disciplined, multi-faceted modelling approach that blends traditional statistical methods with modern machine learning techniques, selecting methods that offer interpretability, stability, and performance in live trading conditions.
– Candidate methodologies:
– Statistical models: Linear and logistic regression, regularised (L1/L2) variants to promote sparsity and reduce overfitting.
– Time-series models: Autoregressive components, VAR/VARX structures where appropriate, and volatility modelling for risk-aware signals.
– Machine learning models: Tree-based methods (e.g., gradient boosting) and regularised neural components for nonlinear interactions, with attention to calibration and overfitting risk.
– Model selection criteria: Performance is evaluated using out-of-sample predictive accuracy, economic plausibility, robustness to regime shifts, and stability of signal timing. Interpretability and ease of deployment are weighed alongside raw metrics.
– Ensembling and decision logic: Signals from multiple models can be blended using rule-based weighting schemes or a meta-model to enhance robustness. Position sizing and execution logic are aligned with portfolio risk limits and liquidity considerations.

4. Parameterisation and hyperparameters
– Regularisation and sparsity: Where applicable, regularisation strengths (e.g., alpha in Lasso or Ridge) are chosen to balance bias-variance trade-offs and to facilitate interpretable feature selections.
– Windowing and look-back periods: Sliding windows are specified for retraining cadence, feature calculation horizons, and signal generation. These choices reflect data-generating processes and operational constraints.
– Thresholds and signal gates: Signal thresholds are set to control false positives/negatives, with consideration of transaction costs and slippage. Gating rules ensure signals are only acted upon when liquidity and market conditions meet minimum criteria.
– Calibration: Probabilistic outputs are calibrated to align with observed outcomes, using approaches such as isotonic regression or Platt scaling where appropriate. Calibration is validated in out-of-sample periods to ensure stability.
– Risk controls: Stop-loss, maximum daily drawdown, position limits by instrument, and overall risk budget constraints are embedded to prevent outsized losses and preserve capital.

5. Evaluation framework
– Backtesting methodology: Historical simulations respect live trading frictions, including transaction costs, slippage, and batch execution constraints. The evaluation uses walk-forward testing to mirror production dynamics.
– Metrics: A combination of statistical and economic metrics is used, such as:
– Predictive accuracy, Sharpe ratio, and information ratio.
– Sortino ratio, maximum drawdown, and turnover.
– Economic value added, including net profit after costs and risk-adjusted returns.
– Robustness checks: Sensitivity analyses assess the impact of feature perturbations, alternative data slices, and parameter changes. Stability over different market regimes is explicitly examined.
– Reproducibility: All experiments are tracked via version-controlled code and data snapshots, with clear documentation of random seeds, data versions, and model configurations to enable replication.

6. Deployment considerations
– Pipeline and operationalization: The modelling workflow is designed for repeatable, low-friction deployment, including automated data ingestion, feature computation, model retraining, signal generation, and risk-checks prior to execution.
– Monitoring and governance: Ongoing monitoring tracks model performance, data quality, and compliance with risk controls. An escalation process is defined for drift or material performance degradation.
– Transparency and explainability: The model’s signal generation and decision rules are documented to support auditability. Where feasible, interpretable components are highlighted to facilitate understanding by risk committees and stakeholders.

7. Limitations and caveats
– Data dependency: Model performance is contingent on the quality and continuity of input data. Gaps or changes in data coverage can affect signals and outcomes.
– Model risk: Like all quantitative models, NQTM assumes stationarity to an extent; regime shifts can alter signal efficacy. Regular review and retraining mitigate, but do not eliminate, risk.
– Execution risk: Real-world frictions such as latency, market impact, and operational bottlenecks must be managed through robust trading infrastructure and risk controls.

Conclusion
The development of BIST’s new quantitative trade model (NQTM) reflects a careful balance between statistical rigour, practical trading considerations, and transparent governance. By leveraging a rich data tapestry, thoughtful feature engineering, and a disciplined modelling framework, NQTM aims to deliver robust, interpretable signals that support disciplined trading decisions while maintaining stringent risk controls. Ongoing monitoring, validation, and iteration will continue to refine the model as market dynamics evolve.

Notes for practitioners
– When reviewing the model, pay close attention to data provenance, feature definitions, and the retraining cadence, as these are common sources of variability in production deployments.
– Engage with risk teams early to ensure alignment on thresholds, limits, and reporting requirements.
– Maintain clear versioning and documentation for any changes to data sources or modelling choices to uphold auditability and reproducibility.

August 7, 2026 at 12:00PM
研究：BIST 新定量贸易模型
https://www.gov.uk/government/publications/bist-new-quantitative-trade-model
技术说明，详细介绍用于创建 BIST 的新定量贸易模型（NQTM）的背景、数据和参数。

阅读更多中文内容： BIST 新量化交易模型（NQTM）的背景、数据与参数技术笔记