Market data contain far more movement than usable forecast signal. These guides follow the decisions required to define an observation, test it on unseen outcomes, determine whether it still matters after risk and implementation costs, and examine the open-source components behind the platform.
How a research idea becomes a reproducible signal, an out-of-sample result, and—only when the evidence warrants it—a position in a constrained portfolio.
Look up the definitions, formulas, units, and interpretation used throughout the research library, with links to the full technical guides.
Define the inputs, timing, universe, transformations, and controls before looking at performance.
Definition of an alpha factor, signal normalization, portfolio construction, and factor-library evaluation.
How dated observations become a security-level forecast, how signals differ from factors and trades, and how a crowding hypothesis reaches portfolio evaluation.
How cross-sectional ranks place securities on a comparable scale at one date, including ties, missing values, group ranks, and a worked factor example.
How z-scores and cross-sectional ranks preserve different information, with a numerical comparison, implementation guidance, and portfolio consequences.
Exclusion, neutral components, peer imputation, carrying a prior observation, and available-weight renormalization change factor coverage and meaning, with a worked three-input example.
How winsorization caps cross-sectional extremes, how it differs from deletion and ranking, and how the threshold changes factor scale and holdings.
How neutralization removes stated common exposures from a security score or portfolio, with sector residuals, equations, and a six-stock example.
How factor exposures are estimated, aggregated from securities to portfolios, neutralized during construction, and converted into risk contributions.
How an economic idea about crowded positioning becomes a dated security-level signal, a cross-sectional factor, and a risk-controlled portfolio.
A precise guide to combining registered signal definitions, recording timing and normalization, testing the composite, and keeping design metadata separate from executable output.
Test ranking quality, persistence, breadth, dependence, decay, and statistical reliability on later outcomes.
A practical definition of cross-sectional information coefficient, including rank IC, estimation noise, interpretation, and its relationship to portfolio construction.
A worked cross-sectional information coefficient calculation, with the rank formula, a Python example, timing rules, and implementation checks.
How rank and Pearson information coefficients differ, when each convention fits the research question, and what a disagreement between them reveals.
Why information coefficient magnitude has no universal cutoff, and how persistence, breadth, implementation, and costs determine whether an IC is useful.
How cross-sectional sample size, serial dependence, overlapping returns, confidence intervals, and repeated testing affect the reliability of an IC estimate.
How repeated specification search manufactures convincing backtests, why structural change weakens even genuine signals, and which research controls preserve honest evidence.
How rolling ICIR converts a time series of information coefficients into a stability-adjusted signal score, and how window choice affects its interpretation.
How to measure an IC horizon curve, distinguish cumulative from incremental forward returns, and use signal decay in research and portfolio timing.
Why mean information coefficient and IC hit rate can disagree, how each measure is calculated, and which definition of a hit belongs in factor evaluation.
How forecast-ranking IC differs from portfolio Information Ratio, where ICIR fits, and how breadth, constraints, and costs connect signal evidence to active returns.
When raw IC and sector-neutral IC answer different research questions, how to calculate each measure, and what a sign reversal reveals about a factor.
How signal scores and IC evidence become expected returns, how covariance and constraints change the allocation, and why IC alone cannot determine portfolio weights.
Why factor breadth counts independent investment decisions rather than securities, how correlation reduces effective breadth, and how breadth enters expected Information Ratio.
How the transfer coefficient measures the alignment between return forecasts and constrained active positions, with a sector-neutral example and the generalized Fundamental Law.
How forecast skill, independent breadth, and portfolio transfer combine into potential Information Ratio, with a worked numerical example and practical limitations.
Translate forecasts into weights while accounting for covariance, uncertainty, constraints, turnover, and costs.
Why a high factor score can still produce no trade, and how research evidence, risk, costs, and portfolio limits determine the final holding.
The mean-variance portfolio objective, its estimated return and covariance inputs, and the checks applied before using its weights.
How security-level factor scores become long and short portfolio weights, how common exposures are controlled, and how the resulting factor return is measured.
How an expected-return forecast is separated from common factor exposures, how residual alpha differs from a realized residual, and how it enters portfolio construction.
The definition and annualization of one-way turnover, with observed turnover and concentration for five allocation methods.
How to measure slippage against a decision price, separate market impact from delay cost, and include it in backtests and turnover estimates.
How uncertainty in expected-return forecasts can be estimated, represented by an ellipsoid, and carried into robust portfolio construction.
How covariance eigenvectors define orthogonal portfolio directions, with a two-asset example, normalization choices, and uses in risk and statistical factor models.
A counterfactual test of whether explicit joint-loss penalties change the weights selected by a mean-variance portfolio optimizer.
How covariance blending, separate risk limits, and worst-model objectives combine competing estimates of portfolio risk.
A walk-forward test of a quadratic forecast-error penalty and ellipsoidal alpha uncertainty across eight registered factor portfolios.
A walk-forward comparison of a long-window covariance, an equal blend, and separate limits under long- and short-window risk models.
How a noisy covariance estimate can be combined with a stable target, where the combination belongs in a factor risk model, and how its weight should be tested.
A walk-forward test of fixed, learned, and changing covariance weights against separate short- and long-horizon risk limits.
Four hypothetical portfolio scenarios show how lot basis, tax timing, loss capacity, wash state, neutrality constraints, and tax-payment mode change an otherwise ordinary rebalance.
Follow market events and empirical studies from the original evidence through the realized portfolio result.
Factor scores and ETF returns around the April 2025 tariff announcement and February 2026 court ruling, followed by ex-ante allocations and realized paths at three probabilities for the tariff view.
A scenario study built on factor scores, the full energy complex, and Polymarket-implied probabilities through a live geopolitical shock, closing with a concrete test: what a risk-adjusted position on that view actually earned, at three levels of confidence, once the event played out.
How StrategyNet’s Apache-2.0 Hypercube project combines memory-mapped field slices with a deterministic dependency graph for low-latency analytics over changing data.
Volt compiles exact listed option contracts into canonical portfolios that can be evaluated, searched, and passed into optimization or machine-learning systems.
How Jörmungandr separates distributed experience, validation, checkpoints, and inference from replaceable reinforcement-learning algorithms and rollout selectors.
StrategyNet publications on portfolio construction, live analytical state, deterministic factor processing, reinforcement learning, options representation, portfolio control, and empirical event studies. Selected proprietary papers require reader details before download.
Entity-aligned Slice vectors and deterministic Hypercube graphs, with measured low-latency paths and a leakage-controlled robust, quantile, and triple-barrier stat-arb study.
Stateful post-generation callbacks for persistent thresholds, ordered bounded handoff, JSON Lines capture, deterministic replay, and effect-safe deployment.
A checkpointable actor–learner runtime with replaceable RL algorithms, quantile return models, auditable QUBO rollout selection, and controlled convergence studies.
The contract, payoff, lifecycle, compiler, and graph semantics that keep every representation of an option portfolio aligned.
A contract-level search method that evaluates feasible portfolios and trains order-invariant surrogates to rank the wider strategy space.
A sequential-control formulation for closes, hedges, rolls, and exercises over a versioned portfolio state and action graph.
Two empirical option studies combining event timelines, market state, strategy choice points, target-volatility scans, and full marked P&L rollouts.
Six annotated screen recordings follow live signal review, factor analysis, strategy construction, comparison, and paper execution. The side rail identifies the decision and evidence visible at each stage.
A shared cross-section ranks securities and isolates the long and short tails. Factor evidence identifies the market property carried by those names. The library retains the definition, history, release dates, and point-in-time cutoff.
Pulse ranks securities by current signal strength. The symbols view checks direction across horizons. The factor analyzer adds ICIR, spread Sharpe, and quantile separation for the selected signal.
Raw and risk-adjusted IC establish cross-sectional structure. A selected factor then moves into a time-series review of IC, ICIR, hit rate, spread behavior, and sample depth.
The analyst selects three high-ranked factors, checks their joint behavior, adds them to a strategy draft, runs candidate constructions, compares completed results, and stages one basket in the paper blotter.
Two completed constructions are selected from the run list. Daily returns, cumulative paths, and correlation show how their realized behavior differs.
A completed strategy is staged after its score, risk measures, and recent returns are reviewed. The paper blotter shows holdings, target deltas, readiness, simulated orders, and fills.
