Research reference
Equity research guide library
Follow a research question from its economic definition to an implementation that can be inspected.
Value signals start with the denominator
Define book-to-market and earnings yield before ranking stocks, and separate apparent cheapness from accounting comparability.
Profitability is not a single quality score
Distinguish gross profitability, operating profitability and leverage-sensitive quality measures before combining them.
Read asset growth before calling it investment
Calculate asset growth consistently and investigate acquisitions, disposals and denominator effects without rewriting the signal after the fact.
Momentum windows: what the skipped month changes
Specify a return window in calendar terms, compound it correctly, and keep a research convention separate from an executable trade.
Low volatility and minimum variance are different portfolios
Distinguish stock-level volatility ranking from portfolio-level risk optimization, with a two-stock covariance example.
Build the universe that existed, not the one that survived
Preserve historical eligibility, security identity and delistings so a stock-selection study does not select its sample using future outcomes.
Combine factor scores without hiding missing data
Make directions, scaling, missingness and tie-breaking explicit before a composite score becomes a portfolio.
Weighting rules change the question a portfolio answers
Compare capitalization, equal and constrained weights while distinguishing name count from economic concentration.
Sector neutrality is one constraint, not a risk certificate
Distinguish within-sector ranking from benchmark sector matching, then check the exposures that remain.
Point-in-time fundamentals require more than a reporting date
Separate period end, public release, vendor availability and research decision time, while preserving original and revised values.
An earnings surprise needs an unrevised expectation
Match actual results to the correct estimate basis, fiscal period and pre-announcement consensus snapshot.
Event returns are not automatically tradable returns
Separate the announcement reaction from the return available after information processing, order submission and an eligible fill.
Corporate actions belong in the return ledger
Reconcile shares, cash and economic value through splits, dividends and mergers without double-counting adjustments.
Rebalancing needs a clock and a boundary rule
Specify observation, decision and execution times, then use explicit buffers to distinguish intentional persistence from stale holdings.
Turnover needs a denominator before it becomes a cost
Reconcile one-way and two-way turnover, drifted weights and basis-point charges with a worked rebalance.
Build a cost ledger without charging slippage twice
Choose a price benchmark, separate execution components, and account for unfilled orders rather than assuming frictionless quotes.
Participation is a constraint, not a capacity certificate
Translate portfolio size into stock-level trading demand and distinguish a liquidity scenario from evidence of scalable performance.
A short return needs a borrowing and cash-flow ledger
Account for borrow availability, fees, distributions and recalls before treating a negative stock weight as an executable short position.
Validate in time and remove overlapping outcomes
Separate model fitting from later evaluation, including the less visible leakage created by forward-return labels and panel rows.
Count the experiments that did not make the chart
Treat windows, filters, weighting rules and implementation choices as research trials, not invisible adjustments around a selected winner.
A benchmark comparison is not the same as alpha
Separate active return, factor-explained return and residual performance, and make benchmark and funding conventions comparable.
What an inspectable equity research dossier should contain
Organize hypothesis, data provenance, decision rules, validation and unresolved limitations so a reviewer can trace a result back to its inputs.