🧭 Market Study Hierarchy
Understanding, Prediction, Action.
This document formalizes the three primary layers of empirical market study within QLIR-style research pipelines.
Each layer has distinct epistemic goals, data requirements, and output forms.
Together, they define a reproducible progression from exploration → inference → simulation.
Level 1 — Non-Path-Dependent Characterization
“What does the landscape look like?”
Purpose
Quantify unconditional or quasi-conditional statistical features of the data — no sequencing, no temporal dependency, just structural description.
Examples
- Candle-size distributions by day-of-week or session.
- Bollinger-band width percentiles.
- Average wick-to-body ratio for volume-filtered bars.
- Volatility vs. time-of-day curves anchored on session open.
Properties
- Bar order is irrelevant (no temporal dependency).
- Output is distributional: histograms, percentiles, heatmaps.
- Used for calibration, feature scaling, or visualization (e.g., Tableau dashboards).
Typical Outputs
summary_table
distribution_plot
p(X | filter)
Level 2 — Path-Dependent Characterization
“Given that this sequence happened, what tends to happen next?”
Purpose
Describe conditional or sequential behavior — transition probabilities without trade logic.
Examples
- “70 % of the time when Bollinger width < threshold for ≥ 5 bars, expansion follows within 15 bars.”
- “After three consecutive up-closes with declining volume, median next-bar return = -0.12 %.”
- “Volatility spikes decay to baseline within N bars 80 % of the time.”
Properties
- Preserves chronological order.
- No exposure or capital logic; introduces temporal conditioning.
- Produces persistence curves, hit-rate tables, or event-anchored averages.
Typical Outputs
event_summary
transition_matrix
p(next_state | pattern)
Role in Pipeline
This is the bridge layer that converts descriptive structure (Level 1) into parameterized hypotheses for backtesting.
Level 3 — Backtests (Decision Simulations)
“If I acted on those patterns with real rules and capital, what happens?”
Purpose
Evaluate full trading logic under execution and capital constraints.
Adds New Dimensions
- Entry / exit definitions
- Position sizing and leverage
- Fees, slippage, funding
- Portfolio aggregation and risk metrics
Typical Outputs
equity_curve
PnL_distribution
drawdown_stats
Sharpe/Sortino
Conceptual Flow
raw_market_data
↓
Level 1: Descriptive statistics → what the terrain looks like
↓
Level 2: Conditional characterization → what tends to happen next
↓
Level 3: Simulation (backtest) → what happens if I act on it
Each layer feeds the next:
- L1 → L2: supply priors and filters
- L2 → L3: supply candidate conditions and parameters
- L3 → feedback: validate or falsify intuition, feeding back into new L1/L2 refinements
Why Maintain This Separation
| Dimension | Level 1 | Level 2 | Level 3 |
|---|---|---|---|
| Goal | Understanding | Prediction | Action |
| Data handling | Shufflable | Ordered | Causal simulation |
| Validation | Statistical significance | Conditional probability | Economic utility |
| Time budget | Interactive | Exploratory | Batch/offline |
| Output | Distributions | Conditional metrics | Performance curves |
Key Takeaway
A characterization study (Levels 1 & 2) transforms intuition into reproducible statistics. A backtest (Level 3) transforms those statistics into actionable performance under constraints.
Maintaining these layers separately keeps your research stack modular, falsifiable, and conceptually clean.