Quantitative Markets & Institutions Lab

Independent Research, Clearly Classified#

Kyle Wisniewski built and maintains the Quantitative Markets & Institutions Lab as an independent demonstration of developing quantitative-finance research capability. It is an educational research environment—not an employer, asset-management firm, investment track record, financial-institution affiliation, or source of investment advice.

The mission is direct: examine how financial models behave when exposed to real data, estimation error, changing market regimes, and institutional conditions. Each research path begins with a practical investor question, states the finding plainly, and then opens into evidence, history, mathematics, implementation, assumptions, and limitations.

The live research environment and open-source repository expose different layers of the same record.

Question-First Research Architecture#

The Lab separates an accessible finding from the technical material needed to inspect it. A reader can begin with a decision question, then move through progressively deeper evidence without being asked to accept a model on presentation alone.

The recurring evidence ladder is:

1. Question — the practical decision or misconception under examination.
2. Finding — the narrow result supported by the study.
3. Measured evidence — the observed data, sample, method, and output produced in the research.
4. Established context — documented historical, institutional, or theoretical material from external sources.
5. Interpretive implication — what the evidence may change about a decision, stated without exceeding the design.
6. Assumptions and limitations — where estimation, specification, data, or regime uncertainty constrains the conclusion.
7. Reproducible artifact — code, notebook, tests, data vintage, and research log sufficient for inspection.

Findings provide the plain-language entry. Method documents the validation protocol. The technical pages and repository carry the mathematics, implementation, and executed artifacts.

Current Evidence Base#

The frozen 15-ETF universe is SPY, QQQ, IWM, EFA, EEM, AGG, TLT, LQD, GLD, DBC, VNQ, USMV, MTUM, VLUE, and QUAL. This breadth supports cross-asset and factor questions while remaining a bounded sample rather than a claim about every investable market.

Six Executed Research Paths#

• Portfolio estimation error examines why an optimized portfolio can deteriorate outside its estimation window.
• Historical value-at-risk validation asks whether a risk number survives its own forecast backtest.
• Diversification regimes measures how stock–bond co-movement changed within the documented sample.
• Factor ETF attribution separates advertised exposures from claims of persistent alpha.
• Volatility-model limits compares constant-volatility assumptions with observed market behavior.
• Monte Carlo error distinguishes additional simulated paths from broader model truth.

Each path publishes a supported, rejected, or revised conclusion. A passing check establishes that a model or claim survived the declared validation procedure; it does not establish permanent truth or future market performance.

Stock–Bond Regime Dossier#

The stock–bond regime dossier asks why stocks and bonds can fall together and what that reveals about diversification.

Within the documented price sample from January 2, 2015 through July 6, 2026, the 126-trading-day rolling SPY–TLT correlation averaged −0.35 during 2015–2019, reached +0.37 during 2022–2023, and ended at +0.31. The observation supports one bounded conclusion: diversification protection was regime-dependent in this sample.

A separate counterfactual stress moved the full correlation matrix toward one while holding component volatilities fixed. Annualized volatility for the tested balanced multi-asset allocation rose from 11.6% to 15.8%. That allocation was 55% equities, 30% bonds, and 15% alternatives—not a literal 60/40 portfolio.

The historical observation and the static stress answer different questions. Neither identifies a universal cause, forecasts the next regime, proves persistence, assigns a probability to a future state, or supplies a timing signal.

Reproducibility and Verification#

The 204 quantitative tests are anchored to theory and numerical invariants across pricing, portfolios, empirical risk, stochastic simulation, and supporting mathematics. The separate 16 publishing checks protect the public research interface and its evidence paths. Together they form 220 total automated checks, but only the 204-model suite is described as theory-anchored quantitative verification.

The repository preserves the Python library, executed notebooks, test suite, and documented data provenance. The public Lab adds decision briefs, interactive tools, methods, limitations, and a research log. Code and prose remain connected so a public claim can be traced to the artifact that supports it.

Assumptions and Limitations#

The research uses historical samples, model specifications, estimation windows, and selected liquid instruments. Results can change with data vintage, portfolio definition, transaction assumptions, calibration, window length, market structure, and regime. Correlation establishes co-movement, not cause. Static stress tests simplify relationships in order to expose sensitivity; they are not forecasts.

The Lab therefore supports bounded confidence: a reader can inspect what was asked, what was measured, what validation was applied, and what the evidence does not establish. That boundary is part of the research result.