Statistical Modeling
Development and validation of statistical models for financial data analysis. Spans regression frameworks, distributional modeling, and multi-factor decomposition with emphasis on out-of-sample robustness.
Systematic, data-driven research on quantitative methodology, market structure, and screening frameworks. All published research represents historical analysis — not investment advice.
Deep dives into the assumptions, biases, and validation techniques that determine whether systematic research holds up outside the backtest.
Risk models compress the structure of a portfolio into a single number -- expected volatility, value-at-risk, expected shortfall -- that the operator uses to size, hedge, and budget for adverse outcomes. The number is also wrong in characteristic ways, and the failure modes are systematic rather than random. This note examines where risk models break, why the breaks cluster around the moments the model matters most, and how to defend against them.
Correlation is the input to almost every portfolio-construction decision, but correlation is not stable. It compresses sharply during the regimes where diversification is most needed. This note examines how correlation instability undermines naive position-sizing, why average correlation is a misleading statistic, and how to model the regime dependence directly.
Position sizing is often treated as a downstream optimization, but it is more accurately a statement about how much the researcher trusts their own evidence. This note examines how confidence in a research result -- not its expected magnitude -- should govern the size at which it is deployed.
Most backtests draw conclusions from fewer independent observations than the researcher assumes. This note examines why sample size is routinely overstated in financial research, how statistical power governs the reliability of any conclusion, and practical methods for determining whether a dataset can support the claims being made.
A strategy that only works at one specific parameter setting is not a strategy -- it is a coincidence. This note examines how to distinguish genuine robustness from narrow optima, how to detect cliff effects, and how to build a parameter sensitivity workflow into the research process.
Backtested performance almost always exceeds live results. This note examines the structural reasons behind that gap -- slippage, market impact, fill assumptions, and timing delays -- and outlines practical approaches to modeling and measuring execution quality.
How repeated interaction with the same dataset contaminates research results, why hold-out samples don't fully solve it, and practical standards for maintaining research integrity.
A backtest that fits the data too well is not a good backtest — it is a good mirror. This note covers how to detect overfitting before it costs you, with practical diagnostics that separate genuine edge from memorized noise.
Markets shift between regimes — trending vs. mean-reverting, low-vol vs. high-vol, risk-on vs. risk-off. A strategy built for one regime will underperform or fail in another. This note examines how to detect shifts and what to do about them.
If you test enough strategies, some will look significant by accident. False discovery risk is the reason a disciplined researcher treats statistical significance as a starting point, not a conclusion.
Look-ahead bias is the use of information that was not yet available at the time a trading decision was supposedly made. It is the single most common reason a strategy that looks excellent in research fails in production.
A flat cost assumption is better than nothing — but not by as much as most researchers assume. This article examines how transaction cost modeling affects backtest reliability and outlines practical approaches for building more realistic estimates.
A strategy that looks strong in one backtest window can still fail in live trading. Walk-forward validation helps reduce false confidence by testing a system across multiple unseen periods under changing conditions.
Survivorship bias can make backtests look more stable, more profitable, and less fragile than they really are. This note outlines what the problem is, how it enters a research pipeline, and practical ways to contain it.
Weekly observations on volatility, breadth, rotation, and correlation. Scanner-informed, not prediction-based.
Week of May 4 market structure analytics: stable low-volatility regime confirmed across all input dimensions, breadth extending into majority participation, factor dispersion expanding broadly across momentum and size factors, and the post-disruption normalization cycle now fully transitioned into a baseline regime.
Week of April 27 market structure analytics: VIX settling deeper into the low-volatility band, breadth crossing the 50% threshold, sector rotation broadening to a fully distributed configuration, and correlation closing the remaining gap to the pre-disruption baseline.
Week of April 20 market structure analytics: VIX settling into the 17-19 low-volatility range, breadth crossing the 40% participation threshold, sector rotation broadening materially, and correlation approaching pre-disruption baseline levels.
Week of April 13 market structure analytics: VIX settling into the 18-20 transitional range, breadth showing incremental improvement, sector participation broadening modestly, and correlation regime continuing its gradual normalization toward baseline.
Week of April 6 market structure analytics: VIX partial mean-reversion from prior week spike, breadth stabilization, sector rotation broadening tentatively, and correlation regime cooling from elevated levels.
VIX surges to 26.78 with front-end backwardation. Breadth collapses to 19% above the 50-day moving average, reversing three weeks of recovery. The cyclical rotation narrows to Energy-only dominance. Sixteen of 25 industry groups are in correction territory.
Implied-realized spread compresses for the third consecutive week. Breadth edges higher to 52%. The cyclical rotation that began two weeks ago extends into a third week, led by Energy follow-through. Single-stock dispersion remains elevated.
Volatility term structure re-steepened after last week’s flattening. Breadth stabilized but did not recover. Cyclicals showed early relative strength while Technology remained mixed. Dispersion is the theme.
Inaugural issue. Volatility term structure flattened after two weeks of compression. Breadth participation narrowed to a 3-month low. Sector rotation showed defensive tilt. This is what we observed.
Core research domains spanning quantitative finance, statistical inference, and systematic methodology.
Development and validation of statistical models for financial data analysis. Spans regression frameworks, distributional modeling, and multi-factor decomposition with emphasis on out-of-sample robustness.
Research into temporal dependencies, regime detection, and forecasting methodologies applied to financial time series. Includes autoregressive models and non-stationary process identification.
Quantitative study of market mechanics, order flow dynamics, and liquidity measurement. Focuses on execution costs, price formation, and structural characteristics of electronic markets.
Design and evaluation of rules-based research frameworks. Includes backtesting methodology, walk-forward validation protocols, and statistical robustness assessment.
Upcoming membership tiers will include weekly research notes and extended scanner outputs.
Request Access